Account for ipv6 in kernel_get{sock,peer}name compat.

These 2 kernel wrapper functions need to be able to properly handle
the ipv6 addrlen, instead of returning -EAFNOSUPPORT(-97).

Signed-off-by: Auke Kok <auke.kok@versity.com>

ReleaseNotes.md

@@ -1,6 +1,133 @@
 Versity ScoutFS Release Notes
 =============================
 
+---
+v1.27
+\
+*Jan 15, 2026*
+
+Switch away from using the general VM cache reclaim machinery to reduce
+idle cluster locks in the client.  The VM treated locks like a cache and
+let many accumulate, presuming that it would be efficient to free them
+in batches.  Lock freeing requires network communication so this could
+result in enormous backlogs in network messages (on the order of
+hundreds of thousands) and could result in signifcant delays of other
+network messaging.
+
+Fix inefficient network receive processing while many messages are in
+the send queue.  This consumed sufficient CPU to cause significant
+stalls, perhaps resulting in hung task warning messages due to delayed
+lock message delivery.
+
+Fix a server livelock case that could happen while committing client
+transactions that contain a large amount of freed file data extents.
+This would present as client tasks hanging and a server task spinning
+consuming cpu.
+
+Fix a rare server request processing failure that doesn't deal with
+retransmission of a request that a previous server partially processed.
+This would present as hung client tasks and repeated "error -2
+committing log merge: getting merge status item" kernel messages.
+
+Fix an unneccessary server shutdown during specific circumstances in
+client lock recovery.  The shutdown was due to server state and was
+ultimately harmless.  The next server that started up would proceed
+accordingly.
+
+---
+v1.26
+\
+*Nov 17, 2025*
+
+Add the ino\_alloc\_per\_lock mount option.  This changes the number of
+inode numbers allocated under each cluster lock and can alleviate lock
+contention for some patterns of larger file creation.
+
+Add the tcp\_keepalive\_timeout\_ms mount option.  This can enable the
+system to survive longer periods of networking outages.
+
+Fix a rare double free of internal btree metadata blocks when merging
+log trees.  The duplicated freed metadata block numbers would cause
+persistent errors in the server, preventing the server from starting and
+hanging the system.
+
+Fix the data\_wait interface to not require the correct data\_version of
+the inode when raising an error.  This lets callers raise errors when
+they're unable to recall the details of the inode to discover its
+data\_version.
+
+Change scoutfs to more aggressively reclaim cached memory when under
+memory pressure.  This makes scoutfs behave more like other kernel
+components and it integrates better with the reclaim policy heuristics
+in the VM core of the kernel.
+
+Change scoutfs to more efficiently transmit and receive socket messages.
+Under heavy load this can process messages sufficiently more quickly to
+avoid hung task messages for tasks that were waiting for cluster lock
+messages to be processed.
+
+Fix faulty server block commit budget calculations that were generating
+spurious "holders exceeded alloc budget" console messages.
+
+---
+v1.25
+\
+*Jun 3, 2025*
+
+Fix a bug that could cause indefinite retries of failed client commits.
+Under specific error conditions the client and server's understanding of
+the current client commit could get out of sync.  The client would retry
+commits indefinitely that could never succeed.  This manifested as
+infinite "critical transaction commit failure" messages in the kernel
+log on the client and matching "error <nr> committing client logs" on
+the server.
+
+Fix a bug in a specific case of server error handling that could result
+in sending references to unwritten blocks to the client.  The client
+would try to read blocks that hadn't been written and return spurious
+errors.  This was seen under low free space conditions on the server and
+resulted in error messages with error code 116 (The errno enum for
+ESTALE, the client's indication that it couldn't read the blocks that it
+expected.)
+
+---
+v1.24
+\
+*Mar 14, 2025*
+
+Add support for coherent read and write mmap() mappings of regular file
+data between mounts.
+
+Fix a bug that was causing scoutfs utilities to parse and change some
+file names before passing them on to the kernel for processing.  This
+fixes spurious scoutfs command errors for files with the offending
+patterns in their names.
+
+Fix a bug where rename wasn't updating the ctime of the inode at the
+destination name if it existed.
+
+---
+v1.23
+\
+*Dec 11, 2024*
+
+Add support for kernels in the RHEL 9.5 minor release.
+
+---
+v1.22
+\
+*Nov 1, 2024*
+
+Add support for building against the RHEL9 family of kernels.
+
+Fix failure of the setattr\_more ioctl() to set the attributes of a
+zero-length file when restoring.
+
+Fix support for POSIX ACLs in the RHEL8 and later family of kernels.
+
+Fix a race condition in the lock server that could drop lock requests
+under heavy load and cause cluster lock attempts to hang.
+
 ---
 v1.21
 \

kmod/Makefile

@@ -5,13 +5,6 @@ ifeq ($(SK_KSRC),)
 SK_KSRC := $(shell echo /lib/modules/`uname -r`/build)
 endif
 
-# fail if sparse fails if we find it
-ifeq ($(shell sparse && echo found),found)
-SP =
-else
-SP = @:
-endif
-
 SCOUTFS_GIT_DESCRIBE ?= \
 	$(shell git describe --all --abbrev=6 --long 2>/dev/null || \
 		echo no-git)
@@ -36,9 +29,7 @@ TARFILE = scoutfs-kmod-$(RPM_VERSION).tar
 all: module
 
 module:
-	$(MAKE) $(SCOUTFS_ARGS)
-	$(SP) $(MAKE) C=2 CF="-D__CHECK_ENDIAN__" $(SCOUTFS_ARGS)
-
+	$(MAKE) CHECK=$(CURDIR)/src/sparse-filtered.sh C=1 CF="-D__CHECK_ENDIAN__" $(SCOUTFS_ARGS)
 
 modules_install:
 	$(MAKE) $(SCOUTFS_ARGS) modules_install

kmod/scoutfs-kmod.spec.in

@@ -4,9 +4,6 @@
 %define kmod_git_describe @@GITDESCRIBE@@
 %define pkg_date %(date +%%Y%%m%%d)
 
-# Disable the building of the debug package(s).
-%define debug_package %{nil}
-
 # take kernel version or default to uname -r
 %{!?kversion: %global kversion %(uname -r)}
 %global kernel_version %{kversion}
@@ -56,6 +53,18 @@ Source:		%{kmod_name}-kmod-%{kmod_version}.tar
 %global flavors_to_build x86_64
 %endif
 
+# el9 sanity: make sure we lock to the minor release we built for and block upgrades
+%{lua:
+  if string.match(rpm.expand("%{dist}"), "%.el9") then
+    rpm.define("el9 1")
+  end
+}
+
+%if 0%{?el9}
+%define release_major_minor 9.%{lua: print(rpm.expand("%{dist}"):match("%.el9_(%d)"))}
+Requires: system-release = %{release_major_minor}
+%endif
+
 %description
 %{kmod_name} - kernel module
 

kmod/src/Makefile.kernelcompat

@@ -6,26 +6,6 @@
 
 ccflags-y += -include $(src)/kernelcompat.h
 
-#
-# v3.10-rc6-21-gbb6f619b3a49
-#
-# _readdir changes from fop->readdir() to fop->iterate() and from
-# filldir(dirent) to dir_emit(ctx).
-#
-ifneq (,$(shell grep 'iterate.*dir_context' include/linux/fs.h))
-ccflags-y += -DKC_ITERATE_DIR_CONTEXT
-endif
-
-#
-# v3.10-rc6-23-g5f99f4e79abc
-#
-# Helpers including dir_emit_dots() are added in the process of
-# switching dcache_readdir() from fop->readdir() to fop->iterate()
-#
-ifneq (,$(shell grep 'dir_emit_dots' include/linux/fs.h))
-ccflags-y += -DKC_DIR_EMIT_DOTS
-endif
-
 #
 # v3.18-rc2-19-gb5ae6b15bd73
 # 
@@ -178,21 +158,12 @@ ifneq (,$(shell grep 'sock_create_kern.*struct net' include/linux/net.h))
 ccflags-y += -DKC_SOCK_CREATE_KERN_NET=1
 endif
 
-#
-# v3.18-rc6-1619-gc0371da6047a
-#
-# iov_iter is now part of struct msghdr
-#
-ifneq (,$(shell grep 'struct iov_iter.*msg_iter' include/linux/socket.h))
-ccflags-y += -DKC_MSGHDR_STRUCT_IOV_ITER=1
-endif
-
 #
 # v4.17-rc6-7-g95582b008388
 #
 # Kernel has current_time(inode) to uniformly retreive timespec in the right unit
 #
-ifneq (,$(shell grep 'extern struct timespec64 current_time' include/linux/fs.h))
+ifneq (,$(shell grep 'struct timespec64 current_time' include/linux/fs.h))
 ccflags-y += -DKC_CURRENT_TIME_INODE=1
 endif
 
@@ -307,6 +278,14 @@ ifneq (,$(shell grep 'int ..mknod. .struct user_namespace' include/linux/fs.h))
 ccflags-y += -DKC_VFS_METHOD_USER_NAMESPACE_ARG
 endif
 
+#
+# v6.2-rc1-2-gabf08576afe3
+#
+# fs: vfs methods use struct mnt_idmap instead of struct user_namespace
+ifneq (,$(shell grep 'int vfs_mknod.struct mnt_idmap' include/linux/fs.h))
+ccflags-y += -DKC_VFS_METHOD_MNT_IDMAP_ARG
+endif
+
 #
 # v5.17-rc2-21-g07888c665b40
 #
@@ -413,3 +392,107 @@ endif
 ifneq (,$(shell grep 'blkdev_put.struct block_device .bdev, void .holder' include/linux/blkdev.h))
 ccflags-y += -DKC_BLKDEV_PUT_HOLDER_ARG
 endif
+
+#
+# v6.4-rc4-163-g0d625446d0a4
+#
+# Entirely removes current->backing_dev_info to ultimately remove buffer_head
+# completely at some point.
+ifneq (,$(shell grep 'struct backing_dev_info.*backing_dev_info;' include/linux/sched.h))
+ccflags-y += -DKC_CURRENT_BACKING_DEV_INFO
+endif
+
+#
+# v6.8-rc1-4-gf3a608827d1f
+#
+# adds bdev_file_open_by_path() and later in v6.8-rc1-30-ge97d06a46526 removes bdev_open_by_path()
+# which requires us to use the file method from now on.
+ifneq (,$(shell grep 'struct file.*bdev_file_open_by_path.const char.*path' include/linux/blkdev.h))
+ccflags-y += -DKC_BDEV_FILE_OPEN_BY_PATH
+endif
+
+# v4.0-rc7-1796-gfe0f07d08ee3
+#
+# direct-io changes modify inode_dio_done to now be called inode_dio_end
+ifneq (,$(shell grep 'void inode_dio_end.struct inode' include/linux/fs.h))
+ccflags-y += -DKC_INODE_DIO_END
+endif
+
+#
+# v5.0-6476-g3d3539018d2c
+#
+# page fault handlers return a bitmask vm_fault_t instead
+# Note: el8's header has a slightly modified prefix here
+ifneq (,$(shell grep 'typedef.*__bitwise unsigned.*int vm_fault_t' include/linux/mm_types.h))
+ccflags-y += -DKC_MM_VM_FAULT_T
+endif
+
+# v3.19-499-gd83a08db5ba6
+#
+# .remap pages becomes obsolete
+ifneq (,$(shell grep 'int ..remap_pages..struct vm_area_struct' include/linux/mm.h))
+ccflags-y += -DKC_MM_REMAP_PAGES
+endif
+
+#
+# v3.19-4742-g503c358cf192
+#
+# list_lru_shrink_count() and list_lru_shrink_walk() introduced
+#
+ifneq (,$(shell grep 'list_lru_shrink_count.*struct list_lru' include/linux/list_lru.h))
+ccflags-y += -DKC_LIST_LRU_SHRINK_COUNT_WALK
+endif
+
+#
+# v3.19-4757-g3f97b163207c
+#
+# lru_list_walk_cb lru arg added
+#
+ifneq (,$(shell grep 'struct list_head \*item, spinlock_t \*lock, void \*cb_arg' include/linux/list_lru.h))
+ccflags-y += -DKC_LIST_LRU_WALK_CB_ITEM_LOCK
+endif
+
+#
+# v6.7-rc4-153-g0a97c01cd20b
+#
+# list_lru_{add,del} -> list_lru_{add,del}_obj
+#
+ifneq (,$(shell grep '^bool list_lru_add_obj' include/linux/list_lru.h))
+ccflags-y += -DKC_LIST_LRU_ADD_OBJ
+endif
+
+#
+# v6.12-rc6-227-gda0c02516c50
+#
+# lru_list_walk_cb lock arg removed
+#
+ifneq (,$(shell grep 'struct list_lru_one \*list, spinlock_t \*lock, void \*cb_arg' include/linux/list_lru.h))
+ccflags-y += -DKC_LIST_LRU_WALK_CB_LIST_LOCK
+endif
+
+#
+# v5.1-rc4-273-ge9b98e162aa5
+#
+# introduce stack trace helpers
+#
+ifneq (,$(shell grep '^unsigned int stack_trace_save' include/linux/stacktrace.h))
+ccflags-y += -DKC_STACK_TRACE_SAVE
+endif
+
+#
+# v6.1-rc1-2-g138060ba92b3
+#
+# set_acl now passed a struct dentry instead of inode.
+#
+ifneq (,$(shell grep 'int ..set_acl.*struct dentry' include/linux/fs.h))
+ccflags-y += -DKC_SET_ACL_DENTRY
+endif
+
+#
+# v6.1-rc1-3-gcac2f8b8d8b5
+#
+# get_acl renamed to get_inode_acl.
+#
+ifneq (,$(shell grep 'struct posix_acl.*get_inode_acl' include/linux/fs.h))
+ccflags-y += -DKC_GET_INODE_ACL
+endif

kmod/src/acl.c

@@ -107,13 +107,22 @@ struct posix_acl *scoutfs_get_acl_locked(struct inode *inode, int type, struct s
 	return acl;
 }
 
+#ifdef KC_GET_INODE_ACL
+struct posix_acl *scoutfs_get_acl(struct inode *inode, int type, bool rcu)
+#else
 struct posix_acl *scoutfs_get_acl(struct inode *inode, int type)
+#endif
 {
 	struct super_block *sb = inode->i_sb;
 	struct scoutfs_lock *lock = NULL;
 	struct posix_acl *acl;
 	int ret;
 
+#ifdef KC_GET_INODE_ACL
+	if (rcu)
+		return ERR_PTR(-ECHILD);
+#endif
+
 #ifndef KC___POSIX_ACL_CREATE
 	if (!IS_POSIXACL(inode))
 		return NULL;
@@ -201,8 +210,15 @@ out:
 	return ret;
 }
 
+#ifdef KC_SET_ACL_DENTRY
+int scoutfs_set_acl(KC_VFS_NS_DEF
+		    struct dentry *dentry, struct posix_acl *acl, int type)
+{
+	struct inode *inode = dentry->d_inode;
+#else
 int scoutfs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 {
+#endif
 	struct super_block *sb = inode->i_sb;
 	struct scoutfs_lock *lock = NULL;
 	LIST_HEAD(ind_locks);
@@ -240,7 +256,11 @@ int scoutfs_acl_get_xattr(struct dentry *dentry, const char *name, void *value,
 	if (!IS_POSIXACL(dentry->d_inode))
 		return -EOPNOTSUPP;
 
+#ifdef KC_GET_INODE_ACL
+	acl = scoutfs_get_acl(dentry->d_inode, type, false);
+#else
 	acl = scoutfs_get_acl(dentry->d_inode, type);
+#endif
 	if (IS_ERR(acl))
 		return PTR_ERR(acl);
 	if (acl == NULL)
@@ -286,7 +306,11 @@ int scoutfs_acl_set_xattr(struct dentry *dentry, const char *name, const void *v
 		}
 	}
 
+#ifdef KC_SET_ACL_DENTRY
+	ret = scoutfs_set_acl(KC_VFS_INIT_NS dentry, acl, type);
+#else
 	ret = scoutfs_set_acl(dentry->d_inode, acl, type);
+#endif
 out:
 	posix_acl_release(acl);
 

kmod/src/acl.h

@@ -1,9 +1,18 @@
 #ifndef _SCOUTFS_ACL_H_
 #define _SCOUTFS_ACL_H_
 
-struct posix_acl *scoutfs_get_acl(struct inode *inode, int type);
-struct posix_acl *scoutfs_get_acl_locked(struct inode *inode, int type, struct scoutfs_lock *lock);
+#ifdef KC_SET_ACL_DENTRY
+int scoutfs_set_acl(KC_VFS_NS_DEF
+		    struct dentry *dentry, struct posix_acl *acl, int type);
+#else
 int scoutfs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
+#endif
+#ifdef KC_GET_INODE_ACL
+struct posix_acl *scoutfs_get_acl(struct inode *inode, int type, bool rcu);
+#else
+struct posix_acl *scoutfs_get_acl(struct inode *inode, int type);
+#endif
+struct posix_acl *scoutfs_get_acl_locked(struct inode *inode, int type, struct scoutfs_lock *lock);
 int scoutfs_set_acl_locked(struct inode *inode, struct posix_acl *acl, int type,
 			   struct scoutfs_lock *lock, struct list_head *ind_locks);
 #ifdef KC_XATTR_STRUCT_XATTR_HANDLER

kmod/src/alloc.c

@@ -86,18 +86,47 @@ static u64 smallest_order_length(u64 len)
 }
 
 /*
- * An extent modification dirties three distinct leaves of an allocator
- * btree as it adds and removes the blkno and size sorted items for the
- * old and new lengths of the extent.  Dirtying the paths to these
- * leaves can grow the tree and grow/shrink neighbours at each level.
- * We over-estimate the number of blocks allocated and freed (the paths
- * share a root, growth doesn't free) to err on the simpler and safer
- * side.  The overhead is minimal given the relatively large list blocks
- * and relatively short allocator trees.
+ * Moving an extent between trees can dirty blocks in several ways. This
+ * function calculates worst case number of blocks across these scenarions.
+ * We treat the alloc and free counts independently, so the values below are
+ * max(allocated, freed), not the sum.
+ *
+ * We track extents with two separate btree items: by block number and by size.
+ *
+ * If we're removing an extent from the btree (allocating), we can dirty
+ * two blocks if the keys are in different leaves. If we wind up merging
+ * leaves because we fall below the low water mark, we can wind up freeing
+ * three leaves.
+ *
+ * That sequence is as follows, assuming the original keys are removed from
+ * blocks A and B:
+ *
+ * Allocate new dirty A' and B'
+ * Free old stable A and B
+ * B' has fallen below the low water mark, so copy B' into A'
+ * Free B'
+ *
+ * An extent insertion (freeing an extent) can dirty up to five distinct items
+ * in the btree as it adds and removes the blkno and size sorted items for the
+ * old and new lengths of the extent:
+ *
+ * In the by-blkno portion of the btree, we can dirty (allocate for COW) up
+ * to two blocks- either by merging adjacent extents, which can cause us to
+ * join leaf blocks; or by an insertion that causes a split.
+ *
+ * In the by-size portion, we never merge extents, so normally we just dirty
+ * a single item with a size insertion. But if we merged adjacent extents in
+ * the by-blkno portion of the tree, we might be working with three by-sizex
+ * items: removing the two old ones that were combined in the merge; and
+ * adding the new one for the larger, merged size.
+ *
+ * Finally, dirtying the paths to these leaves can grow the tree and grow/shrink
+ * neighbours at each level, so we multiply by the height of the tree after
+ * accounting for a possible new level.
  */
 static u32 extent_mod_blocks(u32 height)
 {
-	return ((1 + height) * 2) * 3;
+	return ((1 + height) * 3) * 5;
 }
 
 /*
@@ -828,7 +857,7 @@ static int find_zone_extent(struct super_block *sb, struct scoutfs_alloc_root *r
 		.zone = SCOUTFS_FREE_EXTENT_ORDER_ZONE,
 	};
 	struct scoutfs_extent found;
-	struct scoutfs_extent ext;
+	struct scoutfs_extent ext = {0,};
 	u64 start;
 	u64 len;
 	int nr;

kmod/src/block.c

@@ -22,6 +22,8 @@
 #include <linux/rhashtable.h>
 #include <linux/random.h>
 #include <linux/sched/mm.h>
+#include <linux/list_lru.h>
+#include <linux/stacktrace.h>
 
 #include "format.h"
 #include "super.h"
@@ -38,26 +40,12 @@
  * than the page size.  Callers can have their own contexts for tracking
  * dirty blocks that are written together.  We pin dirty blocks in
  * memory and only checksum them all as they're all written.
- *
- * Memory reclaim is driven by maintaining two very coarse groups of
- * blocks.  As we access blocks we mark them with an increasing counter
- * to discourage them from being reclaimed.  We then define a threshold
- * at the current counter minus half the population.  Recent blocks have
- * a counter greater than the threshold, and all other blocks with
- * counters less than it are considered older and are candidates for
- * reclaim.  This results in access updates rarely modifying an atomic
- * counter as blocks need to be moved into the recent group, and shrink
- * can randomly scan blocks looking for the half of the population that
- * will be in the old group.  It's reasonably effective, but is
- * particularly efficient and avoids contention between concurrent
- * accesses and shrinking.
  */
 
 struct block_info {
 	struct super_block *sb;
-	atomic_t total_inserted;
-	atomic64_t access_counter;
 	struct rhashtable ht;
+	struct list_lru lru;
 	wait_queue_head_t waitq;
 	KC_DEFINE_SHRINKER(shrinker);
 	struct work_struct free_work;
@@ -76,28 +64,15 @@ enum block_status_bits {
 	BLOCK_BIT_PAGE_ALLOC,	/* page (possibly high order) allocation */
 	BLOCK_BIT_VIRT,		/* mapped virt allocation */
 	BLOCK_BIT_CRC_VALID,	/* crc has been verified */
+	BLOCK_BIT_ACCESSED,	/* seen by lookup since last lru add/walk */
 };
 
-/*
- * We want to tie atomic changes in refcounts to whether or not the
- * block is still visible in the hash table, so we store the hash
- * table's reference up at a known high bit.  We could naturally set the
- * inserted bit through excessive refcount increments.  We don't do
- * anything about that but at least warn if we get close.
- *
- * We're avoiding the high byte for no real good reason, just out of a
- * historical fear of implementations that don't provide the full
- * precision.
- */
-#define BLOCK_REF_INSERTED	(1U << 23)
-#define BLOCK_REF_FULL		(BLOCK_REF_INSERTED >> 1)
-
 struct block_private {
 	struct scoutfs_block bl;
 	struct super_block *sb;
 	atomic_t refcount;
-	u64 accessed;
 	struct rhash_head ht_head;
+	struct list_head lru_head;
 	struct list_head dirty_entry;
 	struct llist_node free_node;
 	unsigned long bits;
@@ -106,13 +81,15 @@ struct block_private {
 		struct page *page;
 		void *virt;
 	};
+	unsigned int stack_len;
+	unsigned long stack[10];
 };
 
 #define TRACE_BLOCK(which, bp)									\
 do {												\
 	__typeof__(bp) _bp = (bp);								\
 	trace_scoutfs_block_##which(_bp->sb, _bp, _bp->bl.blkno, atomic_read(&_bp->refcount),	\
-				    atomic_read(&_bp->io_count), _bp->bits, _bp->accessed);	\
+				    atomic_read(&_bp->io_count), _bp->bits);	\
 } while (0)
 
 #define BLOCK_PRIVATE(_bl) \
@@ -126,7 +103,17 @@ static __le32 block_calc_crc(struct scoutfs_block_header *hdr, u32 size)
 	return cpu_to_le32(calc);
 }
 
-static struct block_private *block_alloc(struct super_block *sb, u64 blkno)
+static noinline void save_block_stack(struct block_private *bp)
+{
+	bp->stack_len = stack_trace_save(bp->stack, ARRAY_SIZE(bp->stack), 2);
+}
+
+static void print_block_stack(struct block_private *bp)
+{
+	stack_trace_print(bp->stack, bp->stack_len, 1);
+}
+
+static noinline struct block_private *block_alloc(struct super_block *sb, u64 blkno)
 {
 	struct block_private *bp;
 	unsigned int nofs_flags;
@@ -176,11 +163,13 @@ static struct block_private *block_alloc(struct super_block *sb, u64 blkno)
 	bp->bl.blkno = blkno;
 	bp->sb = sb;
 	atomic_set(&bp->refcount, 1);
+	INIT_LIST_HEAD(&bp->lru_head);
 	INIT_LIST_HEAD(&bp->dirty_entry);
 	set_bit(BLOCK_BIT_NEW, &bp->bits);
 	atomic_set(&bp->io_count, 0);
 
 	TRACE_BLOCK(allocate, bp);
+	save_block_stack(bp);
 
 out:
 	if (!bp)
@@ -233,32 +222,85 @@ static void block_free_work(struct work_struct *work)
 }
 
 /*
- * Get a reference to a block while holding an existing reference.
+ * Users of blocks hold a refcount.  If putting a refcount drops to zero
+ * then the block is freed.
+ *
+ * Acquiring new references and claiming the exclusive right to tear
+ * down a block is built around this LIVE_REFCOUNT_BASE refcount value.
+ * As blocks are initially cached they have the live base added to their
+ * refcount.  Lookups will only increment the refcount and return blocks
+ * for reference holders while the refcount is >= than the base.
+ *
+ * To remove a block from the cache and eventually free it, either by
+ * the lru walk in the shrinker, or by reference holders, the live base
+ * is removed and turned into a normal refcount increment that will be
+ * put by the caller.  This can only be done once for a block, and once
+ * its done lookup will not return any more references.
+ */
+#define LIVE_REFCOUNT_BASE (INT_MAX ^ (INT_MAX >> 1))
+
+/*
+ * Inc the refcount while holding an incremented refcount.  We can't
+ * have so many individual reference holders that they pass the live
+ * base.
  */
 static void block_get(struct block_private *bp)
 {
-	WARN_ON_ONCE((atomic_read(&bp->refcount) & ~BLOCK_REF_INSERTED) <= 0);
+	int now = atomic_inc_return(&bp->refcount);
 
-	atomic_inc(&bp->refcount);
+	BUG_ON(now <= 1);
+	BUG_ON(now == LIVE_REFCOUNT_BASE);
 }
 
 /*
- * Get a reference to a block as long as it's been inserted in the hash
- * table and hasn't been removed.
- */ 
-static struct block_private *block_get_if_inserted(struct block_private *bp)
+ * if (*v >= u) {
+ * 	*v += a;
+ * 	return true;
+ * }
+ */
+static bool atomic_add_unless_less(atomic_t *v, int a, int u)
 {
-	int cnt;
+	int c;
 
 	do {
-		cnt = atomic_read(&bp->refcount);
-		WARN_ON_ONCE(cnt & BLOCK_REF_FULL);
-		if (!(cnt & BLOCK_REF_INSERTED))
-			return NULL;
+		c = atomic_read(v);
+		if (c < u)
+			return false;
+	} while (atomic_cmpxchg(v, c, c + a) != c);
 
-	} while (atomic_cmpxchg(&bp->refcount, cnt, cnt + 1) != cnt);
+	return true;
+}
 
-	return bp;
+static bool block_get_if_live(struct block_private *bp)
+{
+	return atomic_add_unless_less(&bp->refcount, 1, LIVE_REFCOUNT_BASE);
+}
+
+/*
+ * If the refcount still has the live base, subtract it and increment
+ * the callers refcount that they'll put.
+ */
+static bool block_get_remove_live(struct block_private *bp)
+{
+	return atomic_add_unless_less(&bp->refcount, (1 - LIVE_REFCOUNT_BASE), LIVE_REFCOUNT_BASE);
+}
+
+/*
+ * Only get the live base refcount if it is the only refcount remaining.
+ * This means that there are no active refcount holders and the block
+ * can't be dirty or under IO, which both hold references.
+ */
+static bool block_get_remove_live_only(struct block_private *bp)
+{
+	int c;
+
+	do {
+		c = atomic_read(&bp->refcount);
+		if (c != LIVE_REFCOUNT_BASE)
+			return false;
+	} while (atomic_cmpxchg(&bp->refcount, c, c - LIVE_REFCOUNT_BASE + 1) != c);
+
+	return true;
 }
 
 /*
@@ -290,104 +332,73 @@ static const struct rhashtable_params block_ht_params = {
 };
 
 /*
- * Insert a new block into the hash table.  Once it is inserted in the
- * hash table readers can start getting references.  The caller may have
- * multiple refs but the block can't already be inserted.
+ * Insert the block into the cache so that it's visible for lookups.
+ * The caller can hold references (including for a dirty block).
+ *
+ * We make sure the base is added and the block is in the lru once it's
+ * in the hash.  If hash table insertion fails it'll be briefly visible
+ * in the lru, but won't be isolated/evicted because we hold an
+ * incremented refcount in addition to the live base.
  */
 static int block_insert(struct super_block *sb, struct block_private *bp)
 {
 	DECLARE_BLOCK_INFO(sb, binf);
 	int ret;
 
-	WARN_ON_ONCE(atomic_read(&bp->refcount) & BLOCK_REF_INSERTED);
-
+	BUG_ON(atomic_read(&bp->refcount) >= LIVE_REFCOUNT_BASE);
+	atomic_add(LIVE_REFCOUNT_BASE, &bp->refcount);
+	smp_mb__after_atomic(); /* make sure live base is visible to list_lru walk */
+	list_lru_add_obj(&binf->lru, &bp->lru_head);
 retry:
-	atomic_add(BLOCK_REF_INSERTED, &bp->refcount);
 	ret = rhashtable_lookup_insert_fast(&binf->ht, &bp->ht_head, block_ht_params);
 	if (ret < 0) {
-		atomic_sub(BLOCK_REF_INSERTED, &bp->refcount);
 		if (ret == -EBUSY) {
 			/* wait for pending rebalance to finish */
 			synchronize_rcu();
 			goto retry;
+		} else {
+			atomic_sub(LIVE_REFCOUNT_BASE, &bp->refcount);
+			BUG_ON(atomic_read(&bp->refcount) >= LIVE_REFCOUNT_BASE);
+			list_lru_del_obj(&binf->lru, &bp->lru_head);
 		}
 	} else {
-		atomic_inc(&binf->total_inserted);
 		TRACE_BLOCK(insert, bp);
 	}
 
 	return ret;
 }
 
-static u64 accessed_recently(struct block_info *binf)
-{
-	return atomic64_read(&binf->access_counter) - (atomic_read(&binf->total_inserted) >> 1);
-}
-
 /*
- * Make sure that a block that is being accessed is less likely to be
- * reclaimed if it is seen by the shrinker.   If the block hasn't been
- * accessed recently we update its accessed value.
+ * Indicate to the lru walker that this block has been accessed since it
+ * was added or last walked.
  */
 static void block_accessed(struct super_block *sb, struct block_private *bp)
 {
-	DECLARE_BLOCK_INFO(sb, binf);
-
-	if (bp->accessed == 0 || bp->accessed < accessed_recently(binf)) {
+	if (!test_and_set_bit(BLOCK_BIT_ACCESSED, &bp->bits))
 		scoutfs_inc_counter(sb, block_cache_access_update);
-		bp->accessed = atomic64_inc_return(&binf->access_counter);
-	}
 }
 
 /*
- * The caller wants to remove the block from the hash table and has an
- * idea what the refcount should be.  If the refcount does still
- * indicate that the block is hashed, and we're able to clear that bit,
- * then we can remove it from the hash table.
+ * Remove the block from the cache.  When this returns the block won't
+ * be visible for additional references from lookup.
  *
- * The caller makes sure that it's safe to be referencing this block,
- * either with their own held reference (most everything) or by being in
- * an rcu grace period (shrink).
- */
-static bool block_remove_cnt(struct super_block *sb, struct block_private *bp, int cnt)
-{
-	DECLARE_BLOCK_INFO(sb, binf);
-	int ret;
-
-	if ((cnt & BLOCK_REF_INSERTED) &&
-	    (atomic_cmpxchg(&bp->refcount, cnt, cnt & ~BLOCK_REF_INSERTED) == cnt)) {
-
-		TRACE_BLOCK(remove, bp);
-		ret = rhashtable_remove_fast(&binf->ht, &bp->ht_head, block_ht_params);
-		WARN_ON_ONCE(ret); /* must have been inserted */
-		atomic_dec(&binf->total_inserted);
-		return true;
-	}
-
-	return false;
-}
-
-/*
- * Try to remove the block from the hash table as long as the refcount
- * indicates that it is still in the hash table.  This can be racing
- * with normal refcount changes so it might have to retry.
+ * We always try and remove from the hash table.  It's safe to remove a
+ * block that isn't hashed, it just returns -ENOENT.
+ *
+ * This is racing with the lru walk in the shrinker also trying to
+ * remove idle blocks from the cache.  They both try to remove the live
+ * refcount base and perform their removal and put if they get it.
  */
 static void block_remove(struct super_block *sb, struct block_private *bp)
 {
-	int cnt;
+	DECLARE_BLOCK_INFO(sb, binf);
 
-	do {
-		cnt = atomic_read(&bp->refcount);
-	} while ((cnt & BLOCK_REF_INSERTED) && !block_remove_cnt(sb, bp, cnt));
-}
+	rhashtable_remove_fast(&binf->ht, &bp->ht_head, block_ht_params);
 
-/*
- * Take one shot at removing the block from the hash table if it's still
- * in the hash table and the caller has the only other reference.
- */
-static bool block_remove_solo(struct super_block *sb, struct block_private *bp)
-{
-	return block_remove_cnt(sb, bp, BLOCK_REF_INSERTED | 1);
+	if (block_get_remove_live(bp)) {
+		list_lru_del_obj(&binf->lru, &bp->lru_head);
+		block_put(sb, bp);
+	}
 }
 
 static bool io_busy(struct block_private *bp)
@@ -396,37 +407,6 @@ static bool io_busy(struct block_private *bp)
 	return test_bit(BLOCK_BIT_IO_BUSY, &bp->bits);
 }
 
-/*
- * Called during shutdown with no other users.
- */
-static void block_remove_all(struct super_block *sb)
-{
-	DECLARE_BLOCK_INFO(sb, binf);
-	struct rhashtable_iter iter;
-	struct block_private *bp;
-
-	rhashtable_walk_enter(&binf->ht, &iter);
-	rhashtable_walk_start(&iter);
-
-	for (;;) {
-		bp = rhashtable_walk_next(&iter);
-		if (bp == NULL)
-			break;
-		if (bp == ERR_PTR(-EAGAIN))
-			continue;
-
-		if (block_get_if_inserted(bp)) {
-			block_remove(sb, bp);
-			WARN_ON_ONCE(atomic_read(&bp->refcount) != 1);
-			block_put(sb, bp);
-		}
-	}
-
-	rhashtable_walk_stop(&iter);
-	rhashtable_walk_exit(&iter);
-
-	WARN_ON_ONCE(atomic_read(&binf->total_inserted) != 0);
-}
 
 /*
  * XXX The io_count and sb fields in the block_private are only used
@@ -488,7 +468,7 @@ static int block_submit_bio(struct super_block *sb, struct block_private *bp,
 	int ret = 0;
 
 	if (scoutfs_forcing_unmount(sb))
-		return -EIO;
+		return -ENOLINK;
 
 	sector = bp->bl.blkno << (SCOUTFS_BLOCK_LG_SHIFT - 9);
 
@@ -543,6 +523,10 @@ static int block_submit_bio(struct super_block *sb, struct block_private *bp,
 	return ret;
 }
 
+/*
+ * Return a block with an elevated refcount if it was present in the
+ * hash table and its refcount didn't indicate that it was being freed.
+ */
 static struct block_private *block_lookup(struct super_block *sb, u64 blkno)
 {
 	DECLARE_BLOCK_INFO(sb, binf);
@@ -550,8 +534,8 @@ static struct block_private *block_lookup(struct super_block *sb, u64 blkno)
 
 	rcu_read_lock();
 	bp = rhashtable_lookup(&binf->ht, &blkno, block_ht_params);
-	if (bp)
-		bp = block_get_if_inserted(bp);
+	if (bp && !block_get_if_live(bp))
+		bp = NULL;
 	rcu_read_unlock();
 
 	return bp;
@@ -712,8 +696,8 @@ retry:
 
 	ret = 0;
 out:
-	if ((ret == -ESTALE || scoutfs_trigger(sb, BLOCK_REMOVE_STALE)) &&
-	    !retried && !block_is_dirty(bp)) {
+	if (!retried && !IS_ERR_OR_NULL(bp) && !block_is_dirty(bp) &&
+	    (ret == -ESTALE || scoutfs_trigger(sb, BLOCK_REMOVE_STALE))) {
 		retried = true;
 		scoutfs_inc_counter(sb, block_cache_remove_stale);
 		block_remove(sb, bp);
@@ -1078,100 +1062,106 @@ static unsigned long block_count_objects(struct shrinker *shrink, struct shrink_
 	struct super_block *sb = binf->sb;
 
 	scoutfs_inc_counter(sb, block_cache_count_objects);
-
-	return shrinker_min_long(atomic_read(&binf->total_inserted));
+	return list_lru_shrink_count(&binf->lru, sc);
+}
+
+struct isolate_args {
+	struct super_block *sb;
+	struct list_head dispose;
+};
+
+#define DECLARE_ISOLATE_ARGS(sb_, name_) \
+	struct isolate_args name_ = { \
+		.sb = sb_, \
+		.dispose = LIST_HEAD_INIT(name_.dispose), \
+	}
+
+static enum lru_status isolate_lru_block(struct list_head *item, struct list_lru_one *list,
+					 void *cb_arg)
+{
+	struct block_private *bp = container_of(item, struct block_private, lru_head);
+	struct isolate_args *ia = cb_arg;
+
+	TRACE_BLOCK(isolate, bp);
+
+	/* rotate accessed blocks to the tail of the list (lazy promotion) */
+	if (test_and_clear_bit(BLOCK_BIT_ACCESSED, &bp->bits)) {
+		scoutfs_inc_counter(ia->sb, block_cache_isolate_rotate);
+		return LRU_ROTATE;
+	}
+
+	/* any refs, including dirty/io, stop us from acquiring lru refcount */
+	if (!block_get_remove_live_only(bp)) {
+		scoutfs_inc_counter(ia->sb, block_cache_isolate_skip);
+		return LRU_SKIP;
+	}
+
+	scoutfs_inc_counter(ia->sb, block_cache_isolate_removed);
+	list_lru_isolate_move(list, &bp->lru_head, &ia->dispose);
+	return LRU_REMOVED;
+}
+
+static void shrink_dispose_blocks(struct super_block *sb, struct list_head *dispose)
+{
+	struct block_private *bp;
+	struct block_private *bp__;
+
+	list_for_each_entry_safe(bp, bp__, dispose, lru_head) {
+		list_del_init(&bp->lru_head);
+		block_remove(sb, bp);
+		block_put(sb, bp);
+	}
 }
 
-/*
- * Remove a number of cached blocks that haven't been used recently.
- *
- * We don't maintain a strictly ordered LRU to avoid the contention of
- * accesses always moving blocks around in some precise global
- * structure.
- *
- * Instead we use counters to divide the blocks into two roughly equal
- * groups by how recently they were accessed.  We randomly walk all
- * inserted blocks looking for any blocks in the older half to remove
- * and free.  The random walk and there being two groups means that we
- * typically only walk a small multiple of the number we're looking for
- * before we find them all.
- *
- * Our rcu walk of blocks can see blocks in all stages of their life
- * cycle, from dirty blocks to those with 0 references that are queued
- * for freeing.  We only want to free idle inserted blocks so we
- * atomically remove blocks when the only references are ours and the
- * hash table.
- */
 static unsigned long block_scan_objects(struct shrinker *shrink, struct shrink_control *sc)
 {
 	struct block_info *binf = KC_SHRINKER_CONTAINER_OF(shrink, struct block_info);
 	struct super_block *sb = binf->sb;
-	struct rhashtable_iter iter;
-	struct block_private *bp;
-	bool stop = false;
-	unsigned long freed = 0;
-	unsigned long nr = sc->nr_to_scan;
-	u64 recently;
+	DECLARE_ISOLATE_ARGS(sb, ia);
+	unsigned long freed;
 
 	scoutfs_inc_counter(sb, block_cache_scan_objects);
 
-	recently = accessed_recently(binf);
-	rhashtable_walk_enter(&binf->ht, &iter);
-	rhashtable_walk_start(&iter);
+	freed = kc_list_lru_shrink_walk(&binf->lru, sc, isolate_lru_block, &ia);
+	shrink_dispose_blocks(sb, &ia.dispose);
+	return freed;
+}
 
-	/*
-	 * This isn't great but I don't see a better way.  We want to
-	 * walk the hash from a random point so that we're not
-	 * constantly walking over the same region that we've already
-	 * freed old blocks within.  The interface doesn't let us do
-	 * this explicitly, but this seems to work?  The difference this
-	 * makes is enormous, around a few orders of magnitude fewer
-	 * _nexts per shrink.
-	 */
-	if (iter.walker.tbl)
-		iter.slot = prandom_u32_max(iter.walker.tbl->size);
+static enum lru_status dump_lru_block(struct list_head *item, struct list_lru_one *list,
+					 void *cb_arg)
+{
+	struct block_private *bp = container_of(item, struct block_private, lru_head);
 
-	while (nr > 0) {
-		bp = rhashtable_walk_next(&iter);
-		if (bp == NULL)
-			break;
-		if (bp == ERR_PTR(-EAGAIN)) {
-			/*
-			 * We can be called from reclaim in the allocation
-			 * to resize the hash table itself.  We have to
-			 * return so that the caller can proceed and
-			 * enable hash table iteration again.
-			 */
-			scoutfs_inc_counter(sb, block_cache_shrink_stop);
-			stop = true;
-			break;
-		}
+	printk("blkno %llu refcount 0x%x io_count %d bits 0x%lx\n",
+		bp->bl.blkno, atomic_read(&bp->refcount), atomic_read(&bp->io_count),
+		bp->bits);
+	print_block_stack(bp);
 
-		scoutfs_inc_counter(sb, block_cache_shrink_next);
+	return LRU_SKIP;
+}
 
-		if (bp->accessed >= recently) {
-			scoutfs_inc_counter(sb, block_cache_shrink_recent);
-			continue;
-		}
+/*
+ * Called during shutdown with no other users.  The isolating walk must
+ * find blocks on the lru that only have references for presence on the
+ * lru and in the hash table.
+ */
+static void block_shrink_all(struct super_block *sb)
+{
+	DECLARE_BLOCK_INFO(sb, binf);
+	DECLARE_ISOLATE_ARGS(sb, ia);
+	long count;
 
-		if (block_get_if_inserted(bp)) {
-			if (block_remove_solo(sb, bp)) {
-				scoutfs_inc_counter(sb, block_cache_shrink_remove);
-				TRACE_BLOCK(shrink, bp);
-				freed++;
-				nr--;
-			}
-			block_put(sb, bp);
-		}
+	count = DIV_ROUND_UP(list_lru_count(&binf->lru), 128) * 2;
+	do {
+		kc_list_lru_walk(&binf->lru, isolate_lru_block, &ia, 128);
+		shrink_dispose_blocks(sb, &ia.dispose);
+	} while (list_lru_count(&binf->lru) > 0 && --count > 0);
+
+	count = list_lru_count(&binf->lru);
+	if (count > 0) {
+		scoutfs_err(sb, "failed to isolate/dispose %ld blocks", count);
+		kc_list_lru_walk(&binf->lru, dump_lru_block, sb, count);
 	}
-
-	rhashtable_walk_stop(&iter);
-	rhashtable_walk_exit(&iter);
-
-	if (stop)
-		return SHRINK_STOP;
-	else
-		return freed;
 }
 
 struct sm_block_completion {
@@ -1210,7 +1200,7 @@ static int sm_block_io(struct super_block *sb, struct block_device *bdev, blk_op
 	BUILD_BUG_ON(PAGE_SIZE < SCOUTFS_BLOCK_SM_SIZE);
 
 	if (scoutfs_forcing_unmount(sb))
-		return -EIO;
+		return -ENOLINK;
 
 	if (WARN_ON_ONCE(len > SCOUTFS_BLOCK_SM_SIZE) ||
 	    WARN_ON_ONCE(!op_is_write(opf) && !blk_crc))
@@ -1276,7 +1266,7 @@ int scoutfs_block_write_sm(struct super_block *sb,
 int scoutfs_block_setup(struct super_block *sb)
 {
 	struct scoutfs_sb_info *sbi = SCOUTFS_SB(sb);
-	struct block_info *binf;
+	struct block_info *binf = NULL;
 	int ret;
 
 	binf = kzalloc(sizeof(struct block_info), GFP_KERNEL);
@@ -1285,15 +1275,15 @@ int scoutfs_block_setup(struct super_block *sb)
 		goto out;
 	}
 
-	ret = rhashtable_init(&binf->ht, &block_ht_params);
-	if (ret < 0) {
-		kfree(binf);
+	ret = list_lru_init(&binf->lru);
+	if (ret < 0)
+		goto out;
+
+	ret = rhashtable_init(&binf->ht, &block_ht_params);
+	if (ret < 0)
 		goto out;
-	}
 
 	binf->sb = sb;
-	atomic_set(&binf->total_inserted, 0);
-	atomic64_set(&binf->access_counter, 0);
 	init_waitqueue_head(&binf->waitq);
 	KC_INIT_SHRINKER_FUNCS(&binf->shrinker, block_count_objects,
 			       block_scan_objects);
@@ -1305,8 +1295,10 @@ int scoutfs_block_setup(struct super_block *sb)
 
 	ret = 0;
 out:
-	if (ret)
-		scoutfs_block_destroy(sb);
+	if (ret < 0 && binf) {
+		list_lru_destroy(&binf->lru);
+		kfree(binf);
+	}
 
 	return ret;
 }
@@ -1318,9 +1310,10 @@ void scoutfs_block_destroy(struct super_block *sb)
 
 	if (binf) {
 		KC_UNREGISTER_SHRINKER(&binf->shrinker);
-		block_remove_all(sb);
+		block_shrink_all(sb);
 		flush_work(&binf->free_work);
 		rhashtable_destroy(&binf->ht);
+		list_lru_destroy(&binf->lru);
 
 		kfree(binf);
 		sbi->block_info = NULL;

kmod/src/client.c

@@ -435,8 +435,8 @@ static int lookup_mounted_client_item(struct super_block *sb, u64 rid)
 	if (ret == -ENOENT)
 		ret = 0;
 
-	kfree(super);
 out:
+	kfree(super);
 	return ret;
 }
 
@@ -479,7 +479,7 @@ static void scoutfs_client_connect_worker(struct work_struct *work)
 	struct scoutfs_sb_info *sbi = SCOUTFS_SB(sb);
 	struct scoutfs_mount_options opts;
 	struct scoutfs_net_greeting greet;
-	struct sockaddr_in sin;
+	struct sockaddr_storage sin;
 	bool am_quorum;
 	int ret;
 

kmod/src/counters.h

@@ -26,17 +26,15 @@
 	EXPAND_COUNTER(block_cache_alloc_page_order)		\
 	EXPAND_COUNTER(block_cache_alloc_virt)			\
 	EXPAND_COUNTER(block_cache_end_io_error)		\
+	EXPAND_COUNTER(block_cache_isolate_removed)		\
+	EXPAND_COUNTER(block_cache_isolate_rotate)		\
+	EXPAND_COUNTER(block_cache_isolate_skip)		\
 	EXPAND_COUNTER(block_cache_forget)			\
 	EXPAND_COUNTER(block_cache_free)			\
 	EXPAND_COUNTER(block_cache_free_work)			\
 	EXPAND_COUNTER(block_cache_remove_stale)		\
 	EXPAND_COUNTER(block_cache_count_objects)		\
 	EXPAND_COUNTER(block_cache_scan_objects)		\
-	EXPAND_COUNTER(block_cache_shrink)			\
-	EXPAND_COUNTER(block_cache_shrink_next)			\
-	EXPAND_COUNTER(block_cache_shrink_recent)		\
-	EXPAND_COUNTER(block_cache_shrink_remove)		\
-	EXPAND_COUNTER(block_cache_shrink_stop)			\
 	EXPAND_COUNTER(btree_compact_values)			\
 	EXPAND_COUNTER(btree_compact_values_enomem)		\
 	EXPAND_COUNTER(btree_delete)				\
@@ -90,6 +88,7 @@
 	EXPAND_COUNTER(forest_read_items)			\
 	EXPAND_COUNTER(forest_roots_next_hint)			\
 	EXPAND_COUNTER(forest_set_bloom_bits)			\
+	EXPAND_COUNTER(inode_deleted)				\
 	EXPAND_COUNTER(item_cache_count_objects)		\
 	EXPAND_COUNTER(item_cache_scan_objects)			\
 	EXPAND_COUNTER(item_clear_dirty)			\
@@ -117,15 +116,15 @@
 	EXPAND_COUNTER(item_pcpu_page_hit)			\
 	EXPAND_COUNTER(item_pcpu_page_miss)			\
 	EXPAND_COUNTER(item_pcpu_page_miss_keys)		\
+	EXPAND_COUNTER(item_read_pages_barrier)			\
+	EXPAND_COUNTER(item_read_pages_retry)			\
 	EXPAND_COUNTER(item_read_pages_split)			\
 	EXPAND_COUNTER(item_shrink_page)			\
 	EXPAND_COUNTER(item_shrink_page_dirty)			\
-	EXPAND_COUNTER(item_shrink_page_reader)			\
 	EXPAND_COUNTER(item_shrink_page_trylock)		\
 	EXPAND_COUNTER(item_update)				\
 	EXPAND_COUNTER(item_write_dirty)			\
 	EXPAND_COUNTER(lock_alloc)				\
-	EXPAND_COUNTER(lock_count_objects)			\
 	EXPAND_COUNTER(lock_free)				\
 	EXPAND_COUNTER(lock_grant_request)			\
 	EXPAND_COUNTER(lock_grant_response)			\
@@ -139,12 +138,13 @@
 	EXPAND_COUNTER(lock_lock_error)				\
 	EXPAND_COUNTER(lock_nonblock_eagain)			\
 	EXPAND_COUNTER(lock_recover_request)			\
-	EXPAND_COUNTER(lock_scan_objects)			\
 	EXPAND_COUNTER(lock_shrink_attempted)			\
-	EXPAND_COUNTER(lock_shrink_aborted)			\
-	EXPAND_COUNTER(lock_shrink_work)			\
+	EXPAND_COUNTER(lock_shrink_request_failed)		\
 	EXPAND_COUNTER(lock_unlock)				\
 	EXPAND_COUNTER(lock_wait)				\
+	EXPAND_COUNTER(log_merge_complete)			\
+	EXPAND_COUNTER(log_merge_no_finalized)			\
+	EXPAND_COUNTER(log_merge_start)				\
 	EXPAND_COUNTER(log_merge_wait_timeout)			\
 	EXPAND_COUNTER(net_dropped_response)			\
 	EXPAND_COUNTER(net_send_bytes)				\
@@ -159,6 +159,7 @@
 	EXPAND_COUNTER(orphan_scan)				\
 	EXPAND_COUNTER(orphan_scan_attempts)			\
 	EXPAND_COUNTER(orphan_scan_cached)			\
+	EXPAND_COUNTER(orphan_scan_empty)			\
 	EXPAND_COUNTER(orphan_scan_error)			\
 	EXPAND_COUNTER(orphan_scan_item)			\
 	EXPAND_COUNTER(orphan_scan_omap_set)			\
@@ -181,6 +182,7 @@
 	EXPAND_COUNTER(quorum_send_vote)			\
 	EXPAND_COUNTER(quorum_server_shutdown)			\
 	EXPAND_COUNTER(quorum_term_follower)			\
+	EXPAND_COUNTER(reclaimed_open_logs)			\
 	EXPAND_COUNTER(server_commit_hold)			\
 	EXPAND_COUNTER(server_commit_queue)			\
 	EXPAND_COUNTER(server_commit_worker)			\

kmod/src/data.c

@@ -560,7 +560,7 @@ static int scoutfs_get_block(struct inode *inode, sector_t iblock,
 	u64 offset;
 	int ret;
 
-	WARN_ON_ONCE(create && !inode_is_locked(inode));
+	WARN_ON_ONCE(create && !rwsem_is_locked(&si->extent_sem));
 
 	/* make sure caller holds a cluster lock */
 	lock = scoutfs_per_task_get(&si->pt_data_lock);
@@ -1551,13 +1551,17 @@ int scoutfs_data_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 	struct super_block *sb = inode->i_sb;
 	const u64 ino = scoutfs_ino(inode);
 	struct scoutfs_lock *lock = NULL;
+	struct scoutfs_extent *info = NULL;
+	struct page *page = NULL;
 	struct scoutfs_extent ext;
 	struct scoutfs_extent cur;
 	struct data_ext_args args;
 	u32 last_flags;
 	u64 iblock;
 	u64 last;
+	int entries = 0;
 	int ret;
+	int complete = 0;
 
 	if (len == 0) {
 		ret = 0;
@@ -1568,16 +1572,11 @@ int scoutfs_data_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 	if (ret)
 		goto out;
 
-	inode_lock(inode);
-	down_read(&si->extent_sem);
-
-	ret = scoutfs_lock_inode(sb, SCOUTFS_LOCK_READ, 0, inode, &lock);
-	if (ret)
-		goto unlock;
-
-	args.ino = ino;
-	args.inode = inode;
-	args.lock = lock;
+	page = alloc_page(GFP_KERNEL);
+	if (!page) {
+		ret = -ENOMEM;
+		goto out;
+	}
 
 	/* use a dummy extent to track */
 	memset(&cur, 0, sizeof(cur));
@@ -1586,48 +1585,93 @@ int scoutfs_data_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 	iblock = start >> SCOUTFS_BLOCK_SM_SHIFT;
 	last = (start + len - 1) >> SCOUTFS_BLOCK_SM_SHIFT;
 
+	args.ino = ino;
+	args.inode = inode;
+
+	/* outer loop */
 	while (iblock <= last) {
-		ret = scoutfs_ext_next(sb, &data_ext_ops, &args,
-				       iblock, 1, &ext);
-		if (ret < 0) {
-			if (ret == -ENOENT)
+		/* lock */
+		inode_lock(inode);
+		down_read(&si->extent_sem);
+
+		ret = scoutfs_lock_inode(sb, SCOUTFS_LOCK_READ, 0, inode, &lock);
+		if (ret) {
+			up_read(&si->extent_sem);
+			inode_unlock(inode);
+			break;
+		}
+
+		args.lock = lock;
+
+		/* collect entries */
+		info = page_address(page);
+		memset(info, 0, PAGE_SIZE);
+		while (entries < (PAGE_SIZE / sizeof(struct fiemap_extent)) - 1) {
+			ret = scoutfs_ext_next(sb, &data_ext_ops, &args,
+					       iblock, 1, &ext);
+			if (ret < 0) {
+				if (ret == -ENOENT)
+					ret = 0;
+				complete = 1;
+				last_flags = FIEMAP_EXTENT_LAST;
+				break;
+			}
+
+			trace_scoutfs_data_fiemap_extent(sb, ino, &ext);
+
+			if (ext.start > last) {
+				/* not setting _LAST, it's for end of file */
 				ret = 0;
-			last_flags = FIEMAP_EXTENT_LAST;
-			break;
+				complete = 1;
+				break;
+			}
+
+			if (scoutfs_ext_can_merge(&cur, &ext)) {
+				/* merged extents could be greater than input len */
+				cur.len += ext.len;
+			} else {
+				/* fill it */
+				memcpy(info, &cur, sizeof(cur));
+
+				entries++;
+				info++;
+
+				cur = ext;
+			}
+
+			iblock = ext.start + ext.len;
 		}
 
-		trace_scoutfs_data_fiemap_extent(sb, ino, &ext);
+		/* unlock */
+		scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
+		up_read(&si->extent_sem);
+		inode_unlock(inode);
 
-		if (ext.start > last) {
-			/* not setting _LAST, it's for end of file */
-			ret = 0;
+		if (ret)
 			break;
-		}
 
-		if (scoutfs_ext_can_merge(&cur, &ext)) {
-			/* merged extents could be greater than input len */
-			cur.len += ext.len;
-		} else {
-			ret = fill_extent(fieinfo, &cur, 0);
+		/* emit entries */
+		info = page_address(page);
+		for (; entries > 0; entries--) {
+			ret = fill_extent(fieinfo, info, 0);
 			if (ret != 0)
-				goto unlock;
-			cur = ext;
+				goto out;
+			info++;
 		}
 
-		iblock = ext.start + ext.len;
+		if (complete)
+			break;
 	}
 
+	/* still one left, it's in cur */
 	if (cur.len)
 		ret = fill_extent(fieinfo, &cur, last_flags);
-unlock:
-	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
-	up_read(&si->extent_sem);
-	inode_unlock(inode);
 
 out:
 	if (ret == 1)
 		ret = 0;
-
+	if (page)
+		__free_page(page);
 	trace_scoutfs_data_fiemap(sb, start, len, ret);
 
 	return ret;
@@ -1914,6 +1958,236 @@ int scoutfs_data_waiting(struct super_block *sb, u64 ino, u64 iblock,
 	return ret;
 }
 
+#ifdef KC_MM_VM_FAULT_T
+static vm_fault_t scoutfs_data_page_mkwrite(struct vm_fault *vmf)
+{
+	struct vm_area_struct *vma = vmf->vma;
+#else
+static int scoutfs_data_page_mkwrite(struct vm_area_struct *vma,
+				     struct vm_fault *vmf)
+{
+#endif
+	struct page *page = vmf->page;
+	struct file *file = vma->vm_file;
+	struct inode *inode = file_inode(file);
+	struct scoutfs_inode_info *si = SCOUTFS_I(inode);
+	struct super_block *sb = inode->i_sb;
+	struct scoutfs_lock *lock = NULL;
+	SCOUTFS_DECLARE_PER_TASK_ENTRY(pt_ent);
+	DECLARE_DATA_WAIT(dw);
+	struct write_begin_data wbd;
+	u64 ind_seq;
+	loff_t pos;
+	loff_t size;
+	unsigned int len = PAGE_SIZE;
+	vm_fault_t ret = VM_FAULT_SIGBUS;
+	int err;
+
+	pos = vmf->pgoff << PAGE_SHIFT;
+
+	sb_start_pagefault(sb);
+
+	err = scoutfs_lock_inode(sb, SCOUTFS_LOCK_WRITE,
+				 SCOUTFS_LKF_REFRESH_INODE, inode, &lock);
+	if (err) {
+		ret = vmf_error(err);
+		goto out;
+	}
+
+	size = i_size_read(inode);
+
+	if (scoutfs_per_task_add_excl(&si->pt_data_lock, &pt_ent, lock)) {
+		/* data_version is per inode, whole file must be online */
+		err = scoutfs_data_wait_check(inode, 0, size,
+					      SEF_OFFLINE,
+					      SCOUTFS_IOC_DWO_WRITE,
+					      &dw, lock);
+		if (err != 0) {
+			if (err < 0)
+				ret = vmf_error(err);
+			goto out_unlock;
+		}
+	}
+
+
+	/* scoutfs_write_begin */
+	memset(&wbd, 0, sizeof(wbd));
+	INIT_LIST_HEAD(&wbd.ind_locks);
+	wbd.lock = lock;
+
+	/*
+	 * Start transaction before taking page locks - we want to make sure we're
+	 * not locking a page, then waiting for trans, because writeback might race
+	 * against it and cause a lock inversion hang - as demonstrated by both
+	 * holetest and fsstress tests in xfstests.
+	 */
+	do {
+		err = scoutfs_inode_index_start(sb, &ind_seq) ?:
+			scoutfs_inode_index_prepare(sb, &wbd.ind_locks, inode,
+						    true) ?:
+			scoutfs_inode_index_try_lock_hold(sb, &wbd.ind_locks,
+							  ind_seq, false);
+	} while (err > 0);
+	if (err < 0) {
+		ret = vmf_error(err);
+		goto out_trans;
+	}
+
+	down_write(&si->extent_sem);
+
+	if (!trylock_page(page)) {
+		ret = VM_FAULT_NOPAGE;
+		goto out_sem;
+	}
+	ret = VM_FAULT_LOCKED;
+
+	if ((page->mapping != inode->i_mapping) ||
+	    (!PageUptodate(page)) ||
+	    (page_offset(page) > size))	 {
+		unlock_page(page);
+		ret = VM_FAULT_NOPAGE;
+		goto out_sem;
+	}
+
+	if (page->index == (size - 1) >> PAGE_SHIFT)
+		len = ((size - 1) & ~PAGE_MASK) + 1;
+
+	err = __block_write_begin(page, pos, PAGE_SIZE, scoutfs_get_block);
+	if (err) {
+		ret = vmf_error(err);
+		unlock_page(page);
+		goto out_sem;
+	}
+	/* end scoutfs_write_begin */
+
+	/*
+	 * We mark the page dirty already here so that when freeze is in
+	 * progress, we are guaranteed that writeback during freezing will
+	 * see the dirty page and writeprotect it again.
+	 */
+	set_page_dirty(page);
+	wait_for_stable_page(page);
+
+	/* scoutfs_write_end */
+	scoutfs_inode_set_data_seq(inode);
+	scoutfs_inode_inc_data_version(inode);
+
+	file_update_time(vma->vm_file);
+
+	scoutfs_update_inode_item(inode, wbd.lock, &wbd.ind_locks);
+	scoutfs_inode_queue_writeback(inode);
+
+out_sem:
+	up_write(&si->extent_sem);
+out_trans:
+	scoutfs_release_trans(sb);
+	scoutfs_inode_index_unlock(sb, &wbd.ind_locks);
+	/* end scoutfs_write_end */
+
+out_unlock:
+	scoutfs_per_task_del(&si->pt_data_lock, &pt_ent);
+	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_WRITE);
+
+out:
+	sb_end_pagefault(sb);
+
+	if (scoutfs_data_wait_found(&dw)) {
+		/*
+		 * It'd be really nice to not hold the mmap_sem lock here
+		 * before waiting for data, and then return VM_FAULT_RETRY
+		 */
+		err = scoutfs_data_wait(inode, &dw);
+		if (err == 0)
+			ret = VM_FAULT_NOPAGE;
+		else
+			ret = vmf_error(err);
+	}
+
+	trace_scoutfs_data_page_mkwrite(sb, scoutfs_ino(inode), pos, (__force u32)ret);
+
+	return ret;
+}
+
+#ifdef KC_MM_VM_FAULT_T
+static vm_fault_t scoutfs_data_filemap_fault(struct vm_fault *vmf)
+{
+	struct vm_area_struct *vma = vmf->vma;
+#else
+static int scoutfs_data_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+#endif
+	struct file *file = vma->vm_file;
+	struct inode *inode = file_inode(file);
+	struct scoutfs_inode_info *si = SCOUTFS_I(inode);
+	struct super_block *sb = inode->i_sb;
+	struct scoutfs_lock *inode_lock = NULL;
+	SCOUTFS_DECLARE_PER_TASK_ENTRY(pt_ent);
+	DECLARE_DATA_WAIT(dw);
+	loff_t pos;
+	int err;
+	vm_fault_t ret = VM_FAULT_SIGBUS;
+
+	pos = vmf->pgoff;
+	pos <<= PAGE_SHIFT;
+
+retry:
+	err = scoutfs_lock_inode(sb, SCOUTFS_LOCK_READ,
+				 SCOUTFS_LKF_REFRESH_INODE, inode, &inode_lock);
+	if (err < 0)
+		return vmf_error(err);
+
+	if (scoutfs_per_task_add_excl(&si->pt_data_lock, &pt_ent, inode_lock)) {
+		/* protect checked extents from stage/release */
+		atomic_inc(&inode->i_dio_count);
+
+		err = scoutfs_data_wait_check(inode, pos, PAGE_SIZE,
+					      SEF_OFFLINE, SCOUTFS_IOC_DWO_READ,
+					      &dw, inode_lock);
+		if (err != 0) {
+			if (err < 0)
+				ret = vmf_error(err);
+			goto out;
+		}
+	}
+
+#ifdef KC_MM_VM_FAULT_T
+	ret = filemap_fault(vmf);
+#else
+	ret = filemap_fault(vma, vmf);
+#endif
+
+out:
+	if (scoutfs_per_task_del(&si->pt_data_lock, &pt_ent))
+		kc_inode_dio_end(inode);
+	scoutfs_unlock(sb, inode_lock, SCOUTFS_LOCK_READ);
+	if (scoutfs_data_wait_found(&dw)) {
+		err = scoutfs_data_wait(inode, &dw);
+		if (err == 0)
+			goto retry;
+
+		ret = VM_FAULT_RETRY;
+	}
+
+	trace_scoutfs_data_filemap_fault(sb, scoutfs_ino(inode), pos, (__force u32)ret);
+
+	return ret;
+}
+
+static const struct vm_operations_struct scoutfs_data_file_vm_ops = {
+	.fault		= scoutfs_data_filemap_fault,
+	.page_mkwrite	= scoutfs_data_page_mkwrite,
+#ifdef KC_MM_REMAP_PAGES
+	.remap_pages	= generic_file_remap_pages,
+#endif
+};
+
+static int scoutfs_file_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	file_accessed(file);
+	vma->vm_ops = &scoutfs_data_file_vm_ops;
+	return 0;
+}
+
 const struct address_space_operations scoutfs_file_aops = {
 #ifdef KC_MPAGE_READ_FOLIO
 	.dirty_folio		= block_dirty_folio,
@@ -1945,6 +2219,7 @@ const struct file_operations scoutfs_file_fops = {
 	.splice_read	= generic_file_splice_read,
 	.splice_write	= iter_file_splice_write,
 #endif
+	.mmap		= scoutfs_file_mmap,
 	.unlocked_ioctl	= scoutfs_ioctl,
 	.fsync		= scoutfs_file_fsync,
 	.llseek		= scoutfs_file_llseek,

kmod/src/dir.c

@@ -11,11 +11,13 @@
  * General Public License for more details.
  */
 #include <linux/kernel.h>
+#include <linux/stddef.h>
 #include <linux/fs.h>
 #include <linux/slab.h>
 #include <linux/uio.h>
 #include <linux/xattr.h>
 #include <linux/namei.h>
+#include <linux/mm.h>
 
 #include "format.h"
 #include "file.h"
@@ -434,6 +436,15 @@ out:
 		return d_splice_alias(inode, dentry);
 }
 
+/*
+ * Helper to make iterating through dirent ptrs aligned
+ */
+static inline struct scoutfs_dirent *next_aligned_dirent(struct scoutfs_dirent *dent, u8 len)
+{
+	return (void *)dent +
+		ALIGN(offsetof(struct scoutfs_dirent, name[len]), __alignof__(struct scoutfs_dirent));
+}
+
 /*
  * readdir simply iterates over the dirent items for the dir inode and
  * uses their offset as the readdir position.
@@ -441,76 +452,112 @@ out:
  * It will need to be careful not to read past the region of the dirent
  * hash offset keys that it has access to.
  */
-static int KC_DECLARE_READDIR(scoutfs_readdir, struct file *file,
-			      void *dirent, kc_readdir_ctx_t ctx)
+static int scoutfs_readdir(struct file *file, struct dir_context *ctx)
 {
 	struct inode *inode = file_inode(file);
 	struct super_block *sb = inode->i_sb;
 	struct scoutfs_lock *dir_lock = NULL;
 	struct scoutfs_dirent *dent = NULL;
+/* we'll store name_len in dent->__pad[0] */
+#define hacky_name_len __pad[0]
 	struct scoutfs_key last_key;
 	struct scoutfs_key key;
+	struct page *page = NULL;
 	int name_len;
 	u64 pos;
+	int entries = 0;
 	int ret;
+	int complete = 0;
+	struct scoutfs_dirent *end;
 
-	if (!kc_dir_emit_dots(file, dirent, ctx))
+	if (!dir_emit_dots(file, ctx))
 		return 0;
 
-	dent = alloc_dirent(SCOUTFS_NAME_LEN);
-	if (!dent) {
+	page = alloc_page(GFP_KERNEL);
+	if (!page)
 		return -ENOMEM;
-	}
+
+	end = page_address(page) + PAGE_SIZE;
 
 	init_dirent_key(&last_key, SCOUTFS_READDIR_TYPE, scoutfs_ino(inode),
 			SCOUTFS_DIRENT_LAST_POS, 0);
 
-	ret = scoutfs_lock_inode(sb, SCOUTFS_LOCK_READ, 0, inode, &dir_lock);
-	if (ret)
-		goto out;
-
+	/*
+	 * lock and fetch dirent items, until the page no longer fits
+	 * a max size dirent (288b). Then unlock and dir_emit the ones
+	 * we stored in the page.
+	 */
 	for (;;) {
-		init_dirent_key(&key, SCOUTFS_READDIR_TYPE, scoutfs_ino(inode),
-				kc_readdir_pos(file, ctx), 0);
+		/* lock */
+		ret = scoutfs_lock_inode(sb, SCOUTFS_LOCK_READ, 0, inode, &dir_lock);
+		if (ret)
+			break;
 
-		ret = scoutfs_item_next(sb, &key, &last_key, dent,
-					dirent_bytes(SCOUTFS_NAME_LEN),
-					dir_lock);
-		if (ret < 0) {
-			if (ret == -ENOENT)
+		dent = page_address(page);
+		pos = ctx->pos;
+		while (next_aligned_dirent(dent, SCOUTFS_NAME_LEN) < end) {
+			init_dirent_key(&key, SCOUTFS_READDIR_TYPE, scoutfs_ino(inode),
+					pos, 0);
+
+			ret = scoutfs_item_next(sb, &key, &last_key, dent,
+						dirent_bytes(SCOUTFS_NAME_LEN),
+						dir_lock);
+			if (ret < 0) {
+				if (ret == -ENOENT) {
+					ret = 0;
+					complete = 1;
+				}
+				break;
+			}
+
+			name_len = ret - sizeof(struct scoutfs_dirent);
+			dent->hacky_name_len = name_len;
+			if (name_len < 1 || name_len > SCOUTFS_NAME_LEN) {
+				scoutfs_corruption(sb, SC_DIRENT_READDIR_NAME_LEN,
+						   corrupt_dirent_readdir_name_len,
+						   "dir_ino %llu pos %llu key "SK_FMT" len %d",
+						   scoutfs_ino(inode),
+						   pos,
+						   SK_ARG(&key), name_len);
+				ret = -EIO;
+				break;
+			}
+
+			pos = le64_to_cpu(dent->pos) + 1;
+
+			dent = next_aligned_dirent(dent, name_len);
+			entries++;
+		}
+
+		/* unlock */
+		scoutfs_unlock(sb, dir_lock, SCOUTFS_LOCK_READ);
+
+		if (ret < 0)
+			break;
+
+		dent = page_address(page);
+		for (; entries > 0; entries--) {
+			ctx->pos = le64_to_cpu(dent->pos);
+			if (!dir_emit(ctx, dent->name, dent->hacky_name_len,
+					le64_to_cpu(dent->ino),
+					dentry_type(dent->type))) {
 				ret = 0;
+				goto out;
+			}
+
+			dent = next_aligned_dirent(dent, dent->hacky_name_len);
+
+			/* always advance ctx->pos past */
+			ctx->pos++;
+		}
+
+		if (complete)
 			break;
-		}
-
-		name_len = ret - sizeof(struct scoutfs_dirent);
-		if (name_len < 1 || name_len > SCOUTFS_NAME_LEN) {
-			scoutfs_corruption(sb, SC_DIRENT_READDIR_NAME_LEN,
-					   corrupt_dirent_readdir_name_len,
-					   "dir_ino %llu pos %llu key "SK_FMT" len %d",
-					   scoutfs_ino(inode),
-					   kc_readdir_pos(file, ctx),
-					   SK_ARG(&key), name_len);
-			ret = -EIO;
-			goto out;
-		}
-
-		pos = le64_to_cpu(key.skd_major);
-		kc_readdir_pos(file, ctx) = pos;
-
-		if (!kc_dir_emit(ctx, dirent, dent->name, name_len, pos,
-				le64_to_cpu(dent->ino),
-				dentry_type(dent->type))) {
-			ret = 0;
-			break;
-		}
-
-		kc_readdir_pos(file, ctx) = pos + 1;
 	}
 
 out:
-	scoutfs_unlock(sb, dir_lock, SCOUTFS_LOCK_READ);
-
-	kfree(dent);
+	if (page)
+		__free_page(page);
 	return ret;
 }
 
@@ -1765,7 +1812,7 @@ retry:
 	}
 	old_inode->i_ctime = now;
 	if (new_inode)
-		old_inode->i_ctime = now;
+		new_inode->i_ctime = now;
 
 	inode_inc_iversion(old_dir);
 	inode_inc_iversion(old_inode);
@@ -1959,7 +2006,11 @@ const struct inode_operations scoutfs_symlink_iops = {
 #ifdef KC_LINUX_HAVE_RHEL_IOPS_WRAPPER
 	.removexattr	= generic_removexattr,
 #endif
+#ifdef KC_GET_INODE_ACL
+	.get_inode_acl	= scoutfs_get_acl,
+#else
 	.get_acl	= scoutfs_get_acl,
+#endif
 #ifndef KC_LINUX_HAVE_RHEL_IOPS_WRAPPER
 	.tmpfile	= scoutfs_tmpfile,
 	.rename		= scoutfs_rename_common,
@@ -1973,7 +2024,7 @@ const struct inode_operations scoutfs_symlink_iops = {
 };
 
 const struct file_operations scoutfs_dir_fops = {
-	.KC_FOP_READDIR	= scoutfs_readdir,
+	.iterate	= scoutfs_readdir,
 #ifdef KC_FMODE_KABI_ITERATE
 	.open		= scoutfs_dir_open,
 #endif
@@ -2005,7 +2056,14 @@ const struct inode_operations scoutfs_dir_iops = {
 	.removexattr	= generic_removexattr,
 #endif
 	.listxattr	= scoutfs_listxattr,
+#ifdef KC_GET_INODE_ACL
+	.get_inode_acl	= scoutfs_get_acl,
+#else
 	.get_acl	= scoutfs_get_acl,
+#endif
+#ifdef KC_SET_ACL_DENTRY
+	.set_acl	= scoutfs_set_acl,
+#endif
 	.symlink	= scoutfs_symlink,
 	.permission	= scoutfs_permission,
 #ifdef KC_LINUX_HAVE_RHEL_IOPS_WRAPPER

kmod/src/fence.c

@@ -25,6 +25,7 @@
 #include "sysfs.h"
 #include "server.h"
 #include "fence.h"
+#include "net.h"
 
 /*
  * Fencing ensures that a given mount can no longer write to the
@@ -79,7 +80,7 @@ struct pending_fence {
 	struct timer_list timer;
 
 	ktime_t start_kt;
-	__be32 ipv4_addr;
+	union scoutfs_inet_addr addr;
 	bool fenced;
 	bool error;
 	int reason;
@@ -171,14 +172,19 @@ static ssize_t error_store(struct kobject *kobj, struct kobj_attribute *attr, co
 }
 SCOUTFS_ATTR_RW(error);
 
-static ssize_t ipv4_addr_show(struct kobject *kobj,
+static ssize_t inet_addr_show(struct kobject *kobj,
 			      struct kobj_attribute *attr, char *buf)
 {
 	DECLARE_FENCE_FROM_KOBJ(fence, kobj);
+	struct sockaddr_storage sin;
 
-	return snprintf(buf, PAGE_SIZE, "%pI4", &fence->ipv4_addr);
+	memset(&sin, 0, sizeof(struct sockaddr_storage));
+
+	scoutfs_addr_to_sin(&sin, &fence->addr);
+
+	return snprintf(buf, PAGE_SIZE, "%pISc", SIN_ARG(&sin));
 }
-SCOUTFS_ATTR_RO(ipv4_addr);
+SCOUTFS_ATTR_RO(inet_addr);
 
 static ssize_t reason_show(struct kobject *kobj, struct kobj_attribute *attr,
 			   char *buf)
@@ -212,7 +218,7 @@ static struct attribute *fence_attrs[] = {
 	SCOUTFS_ATTR_PTR(elapsed_secs),
 	SCOUTFS_ATTR_PTR(fenced),
 	SCOUTFS_ATTR_PTR(error),
-	SCOUTFS_ATTR_PTR(ipv4_addr),
+	SCOUTFS_ATTR_PTR(inet_addr),
 	SCOUTFS_ATTR_PTR(reason),
 	SCOUTFS_ATTR_PTR(rid),
 	NULL,
@@ -232,7 +238,7 @@ static void fence_timeout(struct timer_list *timer)
 	wake_up(&fi->waitq);
 }
 
-int scoutfs_fence_start(struct super_block *sb, u64 rid, __be32 ipv4_addr, int reason)
+int scoutfs_fence_start(struct super_block *sb, u64 rid, union scoutfs_inet_addr *addr, int reason)
 {
 	DECLARE_FENCE_INFO(sb, fi);
 	struct pending_fence *fence;
@@ -248,7 +254,7 @@ int scoutfs_fence_start(struct super_block *sb, u64 rid, __be32 ipv4_addr, int r
 	scoutfs_sysfs_init_attrs(sb, &fence->ssa);
 
 	fence->start_kt = ktime_get();
-	fence->ipv4_addr = ipv4_addr;
+	memcpy(&fence->addr, addr, sizeof(union scoutfs_inet_addr));
 	fence->fenced = false;
 	fence->error = false;
 	fence->reason = reason;

kmod/src/fence.h

@@ -7,7 +7,7 @@ enum {
 	SCOUTFS_FENCE_QUORUM_BLOCK_LEADER,
 };
 
-int scoutfs_fence_start(struct super_block *sb, u64 rid, __be32 ipv4_addr, int reason);
+int scoutfs_fence_start(struct super_block *sb, u64 rid, union scoutfs_inet_addr *addr, int reason);
 int scoutfs_fence_next(struct super_block *sb, u64 *rid, int *reason, bool *error);
 int scoutfs_fence_reason_pending(struct super_block *sb, int reason);
 int scoutfs_fence_free(struct super_block *sb, u64 rid);

kmod/src/format.h

@@ -470,7 +470,7 @@ struct scoutfs_srch_compact {
  * @get_trans_seq, @commit_trans_seq: These pair of sequence numbers
  * determine if a transaction is currently open for the mount that owns
  * the log_trees struct.  get_trans_seq is advanced by the server as the
- * transaction is opened.   The server sets comimt_trans_seq equal to
+ * transaction is opened.   The server sets commit_trans_seq equal to
  * get_ as the transaction is committed.
  */
 struct scoutfs_log_trees {
@@ -1091,7 +1091,8 @@ enum scoutfs_net_cmd {
 	EXPAND_NET_ERRNO(ENOMEM)	\
 	EXPAND_NET_ERRNO(EIO)		\
 	EXPAND_NET_ERRNO(ENOSPC)	\
-	EXPAND_NET_ERRNO(EINVAL)
+	EXPAND_NET_ERRNO(EINVAL)	\
+	EXPAND_NET_ERRNO(ENOLINK)
 
 #undef EXPAND_NET_ERRNO
 #define EXPAND_NET_ERRNO(which) SCOUTFS_NET_ERR_##which,

kmod/src/inode.c

@@ -149,7 +149,14 @@ static const struct inode_operations scoutfs_file_iops = {
 	.removexattr	= generic_removexattr,
 #endif
 	.listxattr	= scoutfs_listxattr,
+#ifdef KC_GET_INODE_ACL
+	.get_inode_acl	= scoutfs_get_acl,
+#else
 	.get_acl	= scoutfs_get_acl,
+#endif
+#ifdef KC_SET_ACL_DENTRY
+	.set_acl	= scoutfs_set_acl,
+#endif
 	.fiemap		= scoutfs_data_fiemap,
 };
 
@@ -162,7 +169,14 @@ static const struct inode_operations scoutfs_special_iops = {
 	.removexattr	= generic_removexattr,
 #endif
 	.listxattr	= scoutfs_listxattr,
+#ifdef KC_GET_INODE_ACL
+	.get_inode_acl	= scoutfs_get_acl,
+#else
 	.get_acl	= scoutfs_get_acl,
+#endif
+#ifdef KC_SET_ACL_DENTRY
+	.set_acl	= scoutfs_set_acl,
+#endif
 };
 
 /*
@@ -1476,12 +1490,6 @@ static int remove_index_items(struct super_block *sb, u64 ino,
  * Return an allocated and unused inode number.  Returns -ENOSPC if
  * we're out of inode.
  *
- * Each parent directory has its own pool of free inode numbers.  Items
- * are sorted by their inode numbers as they're stored in segments.
- * This will tend to group together files that are created in a
- * directory at the same time in segments.  Concurrent creation across
- * different directories will be stored in their own regions.
- *
  * Inode numbers are never reclaimed.  If the inode is evicted or we're
  * unmounted the pending inode numbers will be lost.  Asking for a
  * relatively small number from the server each time will tend to
@@ -1491,12 +1499,18 @@ static int remove_index_items(struct super_block *sb, u64 ino,
 int scoutfs_alloc_ino(struct super_block *sb, bool is_dir, u64 *ino_ret)
 {
 	DECLARE_INODE_SB_INFO(sb, inf);
+	struct scoutfs_mount_options opts;
 	struct inode_allocator *ia;
 	u64 ino;
 	u64 nr;
 	int ret;
 
-	ia = is_dir ? &inf->dir_ino_alloc : &inf->ino_alloc;
+	scoutfs_options_read(sb, &opts);
+
+	if (is_dir && opts.ino_alloc_per_lock == SCOUTFS_LOCK_INODE_GROUP_NR)
+		ia = &inf->dir_ino_alloc;
+	else
+		ia = &inf->ino_alloc;
 
 	spin_lock(&ia->lock);
 
@@ -1517,6 +1531,17 @@ int scoutfs_alloc_ino(struct super_block *sb, bool is_dir, u64 *ino_ret)
 	*ino_ret = ia->ino++;
 	ia->nr--;
 
+	if (opts.ino_alloc_per_lock != SCOUTFS_LOCK_INODE_GROUP_NR) {
+		nr = ia->ino & SCOUTFS_LOCK_INODE_GROUP_MASK;
+		if (nr >= opts.ino_alloc_per_lock) {
+			nr = SCOUTFS_LOCK_INODE_GROUP_NR - nr;
+			if (nr > ia->nr)
+				nr = ia->nr;
+			ia->ino += nr;
+			ia->nr -= nr;
+		}
+	}
+
 	spin_unlock(&ia->lock);
 	ret = 0;
 out:
@@ -1620,10 +1645,14 @@ int scoutfs_inode_orphan_delete(struct super_block *sb, u64 ino, struct scoutfs_
 				struct scoutfs_lock *primary)
 {
 	struct scoutfs_key key;
+	int ret;
 
 	init_orphan_key(&key, ino);
 
-	return scoutfs_item_delete_force(sb, &key, lock, primary);
+	ret = scoutfs_item_delete_force(sb, &key, lock, primary);
+	trace_scoutfs_inode_orphan_delete(sb, ino, ret);
+
+	return ret;
 }
 
 /*
@@ -1705,6 +1734,8 @@ out:
 		scoutfs_release_trans(sb);
 	scoutfs_inode_index_unlock(sb, &ind_locks);
 
+	trace_scoutfs_delete_inode_end(sb, ino, mode, size, ret);
+
 	return ret;
 }
 
@@ -1800,6 +1831,9 @@ out:
  * they've checked that the inode could really be deleted.  We serialize
  * on a bit in the lock data so that we only have one deletion attempt
  * per inode under this mount's cluster lock.
+ *
+ * Returns -EAGAIN if we either did some cleanup work or are unable to finish
+ * cleaning up this inode right now.
  */
 static int try_delete_inode_items(struct super_block *sb, u64 ino)
 {
@@ -1813,6 +1847,8 @@ static int try_delete_inode_items(struct super_block *sb, u64 ino)
 	int bit_nr;
 	int ret;
 
+	trace_scoutfs_try_delete(sb, ino);
+
 	ret = scoutfs_lock_ino(sb, SCOUTFS_LOCK_WRITE, 0, ino, &lock);
 	if (ret < 0)
 		goto out;
@@ -1825,27 +1861,32 @@ static int try_delete_inode_items(struct super_block *sb, u64 ino)
 
 	/* only one local attempt per inode at a time */
 	if (test_and_set_bit(bit_nr, ldata->trying)) {
-		ret = 0;
+		trace_scoutfs_try_delete_local_busy(sb, ino);
+		ret = -EAGAIN;
 		goto out;
 	}
 	clear_trying = true;
 
 	/* can't delete if it's cached in local or remote mounts */
 	if (scoutfs_omap_test(sb, ino) || test_bit_le(bit_nr, ldata->map.bits)) {
-		ret = 0;
+		trace_scoutfs_try_delete_cached(sb, ino);
+		ret = -EAGAIN;
 		goto out;
 	}
 
 	scoutfs_inode_init_key(&key, ino);
 	ret = lookup_inode_item(sb, &key, &sinode, lock);
 	if (ret < 0) {
-		if (ret == -ENOENT)
+		if (ret == -ENOENT) {
+			trace_scoutfs_try_delete_no_item(sb, ino);
 			ret = 0;
+		}
 		goto out;
 	}
 
 	if (le32_to_cpu(sinode.nlink) > 0) {
-		ret = 0;
+		trace_scoutfs_try_delete_has_links(sb, ino, le32_to_cpu(sinode.nlink));
+		ret = -EAGAIN;
 		goto out;
 	}
 
@@ -1854,6 +1895,11 @@ static int try_delete_inode_items(struct super_block *sb, u64 ino)
 		goto out;
 
 	ret = delete_inode_items(sb, ino, &sinode, lock, orph_lock);
+	if (ret == 0) {
+		ret = -EAGAIN;
+		scoutfs_inc_counter(sb, inode_deleted);
+	}
+
 out:
 	if (clear_trying)
 		clear_bit(bit_nr, ldata->trying);
@@ -1962,6 +2008,8 @@ static void iput_worker(struct work_struct *work)
 		while (count-- > 0)
 			iput(inode);
 
+		cond_resched();
+
 		/* can't touch inode after final iput */
 
 		spin_lock(&inf->iput_lock);
@@ -2052,6 +2100,10 @@ void scoutfs_inode_schedule_orphan_dwork(struct super_block *sb)
  * a locally cached inode.  Then we ask the server for the open map
  * containing the inode.  Only if we don't see any cached users do we do
  * the expensive work of acquiring locks to try and delete the items.
+ *
+ * We need to track whether there is any orphan cleanup work remaining so
+ * that tests such as inode-deletion can watch the orphan_scan_empty counter
+ * to determine when inode cleanup from open-unlink scenarios is complete.
  */
 static void inode_orphan_scan_worker(struct work_struct *work)
 {
@@ -2063,11 +2115,14 @@ static void inode_orphan_scan_worker(struct work_struct *work)
 	SCOUTFS_BTREE_ITEM_REF(iref);
 	struct scoutfs_key last;
 	struct scoutfs_key key;
+	bool work_todo = false;
 	u64 group_nr;
 	int bit_nr;
 	u64 ino;
 	int ret;
 
+	trace_scoutfs_orphan_scan_start(sb);
+
 	scoutfs_inc_counter(sb, orphan_scan);
 
 	init_orphan_key(&last, U64_MAX);
@@ -2087,8 +2142,10 @@ static void inode_orphan_scan_worker(struct work_struct *work)
 		init_orphan_key(&key, ino);
 		ret = scoutfs_btree_next(sb, &roots.fs_root, &key, &iref);
 		if (ret < 0) {
-			if (ret == -ENOENT)
+			if (ret == -ENOENT) {
+				trace_scoutfs_orphan_scan_work(sb, 0);
 				break;
+			}
 			goto out;
 		}
 
@@ -2103,6 +2160,7 @@ static void inode_orphan_scan_worker(struct work_struct *work)
 
 		/* locally cached inodes will try to delete as they evict */
 		if (scoutfs_omap_test(sb, ino)) {
+			work_todo = true;
 			scoutfs_inc_counter(sb, orphan_scan_cached);
 			continue;
 		}
@@ -2118,13 +2176,22 @@ static void inode_orphan_scan_worker(struct work_struct *work)
 
 		/* remote cached inodes will also try to delete */
 		if (test_bit_le(bit_nr, omap.bits)) {
+			work_todo = true;
 			scoutfs_inc_counter(sb, orphan_scan_omap_set);
 			continue;
 		}
 
 		/* seemingly orphaned and unused, get locks and check for sure */
 		scoutfs_inc_counter(sb, orphan_scan_attempts);
+		trace_scoutfs_orphan_scan_work(sb, ino);
+
 		ret = try_delete_inode_items(sb, ino);
+		if (ret == -EAGAIN) {
+			work_todo = true;
+			ret = 0;
+		}
+
+		trace_scoutfs_orphan_scan_end(sb, ino, ret);
 	}
 
 	ret = 0;
@@ -2133,6 +2200,11 @@ out:
 	if (ret < 0)
 		scoutfs_inc_counter(sb, orphan_scan_error);
 
+	if (!work_todo)
+		scoutfs_inc_counter(sb, orphan_scan_empty);
+
+	trace_scoutfs_orphan_scan_stop(sb, work_todo);
+
 	scoutfs_inode_schedule_orphan_dwork(sb);
 }
 
@@ -2183,7 +2255,7 @@ int scoutfs_inode_walk_writeback(struct super_block *sb, bool write)
 	struct scoutfs_inode_info *si;
 	struct scoutfs_inode_info *tmp;
 	struct inode *inode;
-	int ret;
+	int ret = 0;
 
 	spin_lock(&inf->writeback_lock);
 

kmod/src/ioctl.c

@@ -58,25 +58,23 @@
  * key space after we find no items in a given lock region.  This is
  * relatively cheap because reading is going to check the segments
  * anyway.
- *
- * This is copying to userspace while holding a read lock.  This is safe
- * because faulting can send a request for a write lock while the read
- * lock is being used.  The cluster locks don't block tasks in a node,
- * they match and the tasks fall back to local locking.  In this case
- * the spin locks around the item cache.
  */
 static long scoutfs_ioc_walk_inodes(struct file *file, unsigned long arg)
 {
 	struct super_block *sb = file_inode(file)->i_sb;
 	struct scoutfs_ioctl_walk_inodes __user *uwalk = (void __user *)arg;
 	struct scoutfs_ioctl_walk_inodes walk;
-	struct scoutfs_ioctl_walk_inodes_entry ent;
+	struct scoutfs_ioctl_walk_inodes_entry *ent = NULL;
+	struct scoutfs_ioctl_walk_inodes_entry *end;
 	struct scoutfs_key next_key;
 	struct scoutfs_key last_key;
 	struct scoutfs_key key;
 	struct scoutfs_lock *lock;
+	struct page *page = NULL;
 	u64 last_seq;
+	u64 entries = 0;
 	int ret = 0;
+	int complete = 0;
 	u32 nr = 0;
 	u8 type;
 
@@ -107,6 +105,10 @@ static long scoutfs_ioc_walk_inodes(struct file *file, unsigned long arg)
 		}
 	}
 
+	page = alloc_page(GFP_KERNEL);
+	if (!page)
+		return -ENOMEM;
+
 	scoutfs_inode_init_index_key(&key, type, walk.first.major,
 				     walk.first.minor, walk.first.ino);
 	scoutfs_inode_init_index_key(&last_key, type, walk.last.major,
@@ -115,77 +117,107 @@ static long scoutfs_ioc_walk_inodes(struct file *file, unsigned long arg)
 	/* cap nr to the max the ioctl can return to a compat task */
 	walk.nr_entries = min_t(u64, walk.nr_entries, INT_MAX);
 
-	ret = scoutfs_lock_inode_index(sb, SCOUTFS_LOCK_READ, type,
-				       walk.first.major, walk.first.ino,
-				       &lock);
-	if (ret < 0)
-		goto out;
+	end = page_address(page) + PAGE_SIZE;
 
-	for (nr = 0; nr < walk.nr_entries; ) {
+	/* outer loop */
+	for (nr = 0;;) {
+		ent = page_address(page);
+		/* make sure _pad and minor are zeroed */
+		memset(ent, 0, PAGE_SIZE);
 
-		ret = scoutfs_item_next(sb, &key, &last_key, NULL, 0, lock);
-		if (ret < 0 && ret != -ENOENT)
+		ret = scoutfs_lock_inode_index(sb, SCOUTFS_LOCK_READ, type,
+					       le64_to_cpu(key.skii_major),
+					       le64_to_cpu(key.skii_ino),
+					       &lock);
+		if (ret)
 			break;
 
-		if (ret == -ENOENT) {
-
-			/* done if lock covers last iteration key */
-			if (scoutfs_key_compare(&last_key, &lock->end) <= 0) {
-				ret = 0;
+		/* inner loop 1 */
+		while (ent + 1 < end) {
+			ret = scoutfs_item_next(sb, &key, &last_key, NULL, 0, lock);
+			if (ret < 0 && ret != -ENOENT)
 				break;
+
+			if (ret == -ENOENT) {
+				/* done if lock covers last iteration key */
+				if (scoutfs_key_compare(&last_key, &lock->end) <= 0) {
+					ret = 0;
+					complete = 1;
+					break;
+				}
+
+				/* continue iterating after locked empty region */
+				key = lock->end;
+				scoutfs_key_inc(&key);
+
+				scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
+				/* avoid double-unlocking here after break */
+				lock = NULL;
+
+				ret = scoutfs_forest_next_hint(sb, &key, &next_key);
+				if (ret < 0 && ret != -ENOENT)
+					break;
+
+				if (ret == -ENOENT ||
+				    scoutfs_key_compare(&next_key, &last_key) > 0) {
+					ret = 0;
+					complete = 1;
+					break;
+				}
+
+				key = next_key;
+
+				ret = scoutfs_lock_inode_index(sb, SCOUTFS_LOCK_READ,
+							type,
+							le64_to_cpu(key.skii_major),
+							le64_to_cpu(key.skii_ino),
+							&lock);
+				if (ret)
+					break;
+
+				continue;
 			}
 
-			/* continue iterating after locked empty region */
-			key = lock->end;
+			ent->major = le64_to_cpu(key.skii_major);
+			ent->ino = le64_to_cpu(key.skii_ino);
+
 			scoutfs_key_inc(&key);
 
-			scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
+			ent++;
+			entries++;
 
-			ret = scoutfs_forest_next_hint(sb, &key, &next_key);
-			if (ret < 0 && ret != -ENOENT)
-				goto out;
+			if (nr + entries >= walk.nr_entries) {
+				complete = 1;
+				break;
+			}
+		}
 
-			if (ret == -ENOENT ||
-			    scoutfs_key_compare(&next_key, &last_key) > 0) {
-				ret = 0;
+		scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
+		if (ret < 0)
+			break;
+
+		/* inner loop 2 */
+		ent = page_address(page);
+		for (; entries > 0; entries--) {
+			if (copy_to_user((void __user *)walk.entries_ptr, ent,
+					 sizeof(struct scoutfs_ioctl_walk_inodes_entry))) {
+				ret = -EFAULT;
 				goto out;
 			}
-
-			key = next_key;
-
-			ret = scoutfs_lock_inode_index(sb, SCOUTFS_LOCK_READ,
-						key.sk_type,
-						le64_to_cpu(key.skii_major),
-						le64_to_cpu(key.skii_ino),
-						&lock);
-			if (ret < 0)
-				goto out;
-
-			continue;
+			walk.entries_ptr += sizeof(struct scoutfs_ioctl_walk_inodes_entry);
+			ent++;
+			nr++;
 		}
 
-		ent.major = le64_to_cpu(key.skii_major);
-		ent.minor = 0;
-		ent.ino = le64_to_cpu(key.skii_ino);
-
-		if (copy_to_user((void __user *)walk.entries_ptr, &ent,
-				 sizeof(ent))) {
-			ret = -EFAULT;
+		if (complete)
 			break;
-		}
-
-		nr++;
-		walk.entries_ptr += sizeof(ent);
-
-		scoutfs_key_inc(&key);
 	}
 
-	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
-
 out:
+	if (page)
+		__free_page(page);
 	if (nr > 0)
 		ret = nr;
-
 	return ret;
 }
 
@@ -409,8 +441,6 @@ static long scoutfs_ioc_data_wait_err(struct file *file, unsigned long arg)
 
 	if (!S_ISREG(inode->i_mode)) {
 		ret = -EINVAL;
-	} else if (scoutfs_inode_data_version(inode) != args.data_version) {
-		ret = -ESTALE;
 	} else {
 		ret = scoutfs_data_wait_err(inode, sblock, eblock, args.op,
 					    args.err);
@@ -524,7 +554,9 @@ static long scoutfs_ioc_stage(struct file *file, unsigned long arg)
 	}
 
 	si->staging = true;
+#ifdef KC_CURRENT_BACKING_DEV_INFO
 	current->backing_dev_info = inode_to_bdi(inode);
+#endif
 
 	pos = args.offset;
 	written = 0;
@@ -537,7 +569,9 @@ static long scoutfs_ioc_stage(struct file *file, unsigned long arg)
 	} while (ret > 0 && written < args.length);
 
 	si->staging = false;
+#ifdef KC_CURRENT_BACKING_DEV_INFO
 	current->backing_dev_info = NULL;
+#endif
 out:
 	scoutfs_per_task_del(&si->pt_data_lock, &pt_ent);
 	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_WRITE);
@@ -918,6 +952,9 @@ static int copy_alloc_detail_to_user(struct super_block *sb, void *arg,
 	if (args->copied == args->nr)
 		return -EOVERFLOW;
 
+	/* .type and .pad need clearing */
+	memset(&ade, 0, sizeof(struct scoutfs_ioctl_alloc_detail_entry));
+
 	ade.blocks = blocks;
 	ade.id = id;
 	ade.meta = !!meta;
@@ -1159,11 +1196,15 @@ static long scoutfs_ioc_get_allocated_inos(struct file *file, unsigned long arg)
 	struct scoutfs_lock *lock = NULL;
 	struct scoutfs_key key;
 	struct scoutfs_key end;
+	struct page *page = NULL;
 	u64 __user *uinos;
 	u64 bytes;
-	u64 ino;
+	u64 *ino;
+	u64 *ino_end;
+	int entries = 0;
 	int nr;
 	int ret;
+	int complete = 0;
 
 	if (!(file->f_mode & FMODE_READ)) {
 		ret = -EBADF;
@@ -1185,47 +1226,83 @@ static long scoutfs_ioc_get_allocated_inos(struct file *file, unsigned long arg)
 		goto out;
 	}
 
+	page = alloc_page(GFP_KERNEL);
+	if (!page) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	ino_end = page_address(page) + PAGE_SIZE;
+
 	scoutfs_inode_init_key(&key, gai.start_ino);
 	scoutfs_inode_init_key(&end, gai.start_ino | SCOUTFS_LOCK_INODE_GROUP_MASK);
 	uinos = (void __user *)gai.inos_ptr;
 	bytes = gai.inos_bytes;
 	nr = 0;
 
-	ret = scoutfs_lock_ino(sb, SCOUTFS_LOCK_READ, 0, gai.start_ino, &lock);
-	if (ret < 0)
-		goto out;
+	for (;;) {
 
-	while (bytes >= sizeof(*uinos)) {
+		ret = scoutfs_lock_ino(sb, SCOUTFS_LOCK_READ, 0, gai.start_ino, &lock);
+		if (ret < 0)
+			goto out;
 
-		ret = scoutfs_item_next(sb, &key, &end, NULL, 0, lock);
-		if (ret < 0) {
-			if (ret == -ENOENT)
+		ino = page_address(page);
+		while (ino < ino_end) {
+
+			ret = scoutfs_item_next(sb, &key, &end, NULL, 0, lock);
+			if (ret < 0) {
+				if (ret == -ENOENT) {
+					ret = 0;
+					complete = 1;
+				}
+				break;
+			}
+
+			if (key.sk_zone != SCOUTFS_FS_ZONE) {
 				ret = 0;
-			break;
+				complete = 1;
+				break;
+			}
+
+			/* all fs items are owned by allocated inodes, and _first is always ino */
+			*ino = le64_to_cpu(key._sk_first);
+			scoutfs_inode_init_key(&key, *ino + 1);
+
+			ino++;
+			entries++;
+			nr++;
+
+			bytes -= sizeof(*uinos);
+			if (bytes < sizeof(*uinos)) {
+				complete = 1;
+				break;
+			}
+
+			if (nr == INT_MAX) {
+				complete = 1;
+				break;
+			}
 		}
 
-		if (key.sk_zone != SCOUTFS_FS_ZONE) {
-			ret = 0;
-			break;
-		}
+		scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
 
-		/* all fs items are owned by allocated inodes, and _first is always ino */
-		ino = le64_to_cpu(key._sk_first);
-		if (put_user(ino, uinos)) {
+		if (ret < 0)
+			break;
+
+		ino = page_address(page);
+		if (copy_to_user(uinos, ino, entries * sizeof(*uinos))) {
 			ret = -EFAULT;
-			break;
+			goto out;
 		}
 
-		uinos++;
-		bytes -= sizeof(*uinos);
-		if (++nr == INT_MAX)
+		uinos += entries;
+		entries = 0;
+
+		if (complete)
 			break;
-
-		scoutfs_inode_init_key(&key, ino + 1);
 	}
-
-	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_READ);
 out:
+	if (page)
+		__free_page(page);
 	return ret ?: nr;
 }
 
@@ -1293,7 +1370,7 @@ static long scoutfs_ioc_get_referring_entries(struct file *file, unsigned long a
 			ent.d_type = bref->d_type;
 			ent.name_len = name_len;
 
-			if (copy_to_user(uent, &ent, sizeof(struct scoutfs_ioctl_dirent)) ||
+			if (copy_to_user(uent, &ent, offsetof(struct scoutfs_ioctl_dirent, name[0])) ||
 			    copy_to_user(&uent->name[0], bref->dent.name, name_len) ||
 			    put_user('\0', &uent->name[name_len])) {
 				ret = -EFAULT;

kmod/src/ioctl.h

@@ -366,10 +366,15 @@ struct scoutfs_ioctl_statfs_more {
  *
  * Find current waiters that match the inode, op, and block range to wake
  * up and return an error.
+ *
+ * (*) ca. v1.25 and earlier required that the data_version passed match
+ * that of the waiter, but this check is removed. It was never needed
+ * because no data is modified during this ioctl. Any data_version value
+ * here is thus since then ignored.
  */
 struct scoutfs_ioctl_data_wait_err {
 	__u64 ino;
-	__u64 data_version;
+	__u64 data_version; /* Ignored, see above (*) */
 	__u64 offset;
 	__u64 count;
 	__u64 op;

kmod/src/item.c

@@ -86,6 +86,8 @@ struct item_cache_info {
 	/* often walked, but per-cpu refs are fast path */
 	rwlock_t rwlock;
 	struct rb_root pg_root;
+	/* stop readers from caching stale items behind reclaimed cleaned written items */
+	u64 read_dirty_barrier;
 
 	/* page-granular modification by writers, then exclusive to commit */
 	spinlock_t dirty_lock;
@@ -96,10 +98,6 @@ struct item_cache_info {
 	spinlock_t lru_lock;
 	struct list_head lru_list;
 	unsigned long lru_pages;
-
-	/* written by page readers, read by shrink */
-	spinlock_t active_lock;
-	struct list_head active_list;
 };
 
 #define DECLARE_ITEM_CACHE_INFO(sb, name) \
@@ -1285,78 +1283,6 @@ static int cache_empty_page(struct super_block *sb,
 	return 0;
 }
 
-/*
- * Readers operate independently from dirty items and transactions.
- * They read a set of persistent items and insert them into the cache
- * when there aren't already pages whose key range contains the items.
- * This naturally prefers cached dirty items over stale read items.
- *
- * We have to deal with the case where dirty items are written and
- * invalidated while a read is in flight.   The reader won't have seen
- * the items that were dirty in their persistent roots as they started
- * reading.  By the time they insert their read pages the previously
- * dirty items have been reclaimed and are not in the cache.  The old
- * stale items will be inserted in their place, effectively corrupting
- * by having the dirty items disappear.
- *
- * We fix this by tracking the max seq of items in pages.  As readers
- * start they record the current transaction seq.  Invalidation skips
- * pages with a max seq greater than the first reader seq because the
- * items in the page have to stick around to prevent the readers stale
- * items from being inserted.
- *
- * This naturally only affects a small set of pages with items that were
- * written relatively recently.  If we're in memory pressure then we
- * probably have a lot of pages and they'll naturally have items that
- * were visible to any raders.  We don't bother with the complicated and
- * expensive further refinement of tracking the ranges that are being
- * read and comparing those with pages to invalidate.
- */
-struct active_reader {
-	struct list_head head;
-	u64 seq;
-};
-
-#define INIT_ACTIVE_READER(rdr) \
-	struct active_reader rdr = { .head = LIST_HEAD_INIT(rdr.head) }
-
-static void add_active_reader(struct super_block *sb, struct active_reader *active)
-{
-	DECLARE_ITEM_CACHE_INFO(sb, cinf);
-
-	BUG_ON(!list_empty(&active->head));
-
-	active->seq = scoutfs_trans_sample_seq(sb);
-
-	spin_lock(&cinf->active_lock);
-	list_add_tail(&active->head, &cinf->active_list);
-	spin_unlock(&cinf->active_lock);
-}
-
-static u64 first_active_reader_seq(struct item_cache_info *cinf)
-{
-	struct active_reader *active;
-	u64 first;
-
-	/* only the calling task adds or deletes this active */
-	spin_lock(&cinf->active_lock);
-	active = list_first_entry_or_null(&cinf->active_list, struct active_reader, head);
-	first = active ? active->seq : U64_MAX;
-	spin_unlock(&cinf->active_lock);
-
-	return first;
-}
-
-static void del_active_reader(struct item_cache_info *cinf, struct active_reader *active)
-{
-	/* only the calling task adds or deletes this active */
-	if (!list_empty(&active->head)) {
-		spin_lock(&cinf->active_lock);
-		list_del_init(&active->head);
-		spin_unlock(&cinf->active_lock);
-	}
-}
-
 /*
  * Add a newly read item to the pages that we're assembling for
  * insertion into the cache.   These pages are private, they only exist
@@ -1450,24 +1376,34 @@ static int read_page_item(struct super_block *sb, struct scoutfs_key *key, u64 s
  * and duplicates, we insert any resulting pages which don't overlap
  * with existing cached pages.
  *
- * We only insert uncached regions because this is called with cluster
- * locks held, but without locking the cache.  The regions we read can
- * be stale with respect to the current cache, which can be read and
- * dirtied by other cluster lock holders on our node, but the cluster
- * locks protect the stable items we read.  Invalidation is careful not
- * to drop pages that have items that we couldn't see because they were
- * dirty when we started reading.
- *
  * The forest item reader is reading stable trees that could be
  * overwritten.  It can return -ESTALE which we return to the caller who
  * will retry the operation and work with a new set of more recent
  * btrees.
+ *
+ * We only insert uncached regions because this is called with cluster
+ * locks held, but without locking the cache.  The regions we read can
+ * be stale with respect to the current cache, which can be read and
+ * dirtied by other cluster lock holders on our node, but the cluster
+ * locks protect the stable items we read.
+ *
+ * Using the presence of locally written dirty pages to override stale
+ * read pages only works if, well, the more recent locally written pages
+ * are still present.  Readers are totally decoupled from writers and
+ * can have a set of items that is very old indeed.  In the mean time
+ * more recent items would have been dirtied locally, committed,
+ * cleaned, and reclaimed.  We have a coarse barrier which ensures that
+ * readers can't insert items read from old roots from before local data
+ * was written.  If a write completes while a read is in progress the
+ * read will have to retry.  The retried read can use cached blocks so
+ * we're relying on reads being much faster than writes to reduce the
+ * overhead to mostly cpu work of recollecting the items from cached
+ * blocks via a more recent root from the server.
  */
 static int read_pages(struct super_block *sb, struct item_cache_info *cinf,
 		      struct scoutfs_key *key, struct scoutfs_lock *lock)
 {
 	struct rb_root root = RB_ROOT;
-	INIT_ACTIVE_READER(active);
 	struct cached_page *right = NULL;
 	struct cached_page *pg;
 	struct cached_page *rd;
@@ -1480,6 +1416,7 @@ static int read_pages(struct super_block *sb, struct item_cache_info *cinf,
 	struct rb_node *par;
 	struct rb_node *pg_tmp;
 	struct rb_node *item_tmp;
+	u64 rdbar;
 	int pgi;
 	int ret;
 
@@ -1493,8 +1430,9 @@ static int read_pages(struct super_block *sb, struct item_cache_info *cinf,
 	pg->end = lock->end;
 	rbtree_insert(&pg->node, NULL, &root.rb_node, &root);
 
-	/* set active reader seq before reading persistent roots */
-	add_active_reader(sb, &active);
+	read_lock(&cinf->rwlock);
+	rdbar = cinf->read_dirty_barrier;
+	read_unlock(&cinf->rwlock);
 
 	start = lock->start;
 	end = lock->end;
@@ -1533,6 +1471,13 @@ static int read_pages(struct super_block *sb, struct item_cache_info *cinf,
 retry:
 	write_lock(&cinf->rwlock);
 
+	/* can't insert if write has cleaned since we read */
+	if (cinf->read_dirty_barrier != rdbar) {
+		scoutfs_inc_counter(sb, item_read_pages_barrier);
+		ret = -ESTALE;
+		goto unlock;
+	}
+
 	while ((rd = first_page(&root))) {
 
 		pg = page_rbtree_walk(sb, &cinf->pg_root, &rd->start, &rd->end,
@@ -1570,12 +1515,12 @@ retry:
 		}
 	}
 
+	ret = 0;
+
+unlock:
 	write_unlock(&cinf->rwlock);
 
-	ret = 0;
 out:
-	del_active_reader(cinf, &active);
-
 	/* free any pages we left dangling on error */
 	for_each_page_safe(&root, rd, pg_tmp) {
 		rbtree_erase(&rd->node, &root);
@@ -1635,6 +1580,7 @@ retry:
 			ret = read_pages(sb, cinf, key, lock);
 		if (ret < 0 && ret != -ESTALE)
 			goto out;
+		scoutfs_inc_counter(sb, item_read_pages_retry);
 		goto retry;
 	}
 
@@ -2401,6 +2347,12 @@ out:
  * The caller has successfully committed all the dirty btree blocks that
  * contained the currently dirty items.  Clear all the dirty items and
  * pages.
+ *
+ * This strange lock/trylock loop comes from sparse issuing spurious
+ * mismatched context warnings if we do anything (like unlock and relax)
+ * in the else branch of the failed trylock.  We're jumping through
+ * hoops to not use the else but still drop and reacquire the dirty_lock
+ * if the trylock fails.
  */
 int scoutfs_item_write_done(struct super_block *sb)
 {
@@ -2409,40 +2361,35 @@ int scoutfs_item_write_done(struct super_block *sb)
 	struct cached_item *tmp;
 	struct cached_page *pg;
 
-retry:
+	/* don't let read_pages miss written+cleaned items */
+	write_lock(&cinf->rwlock);
+	cinf->read_dirty_barrier++;
+	write_unlock(&cinf->rwlock);
+
 	spin_lock(&cinf->dirty_lock);
-
-	while ((pg = list_first_entry_or_null(&cinf->dirty_list,
-					      struct cached_page,
-					      dirty_head))) {
-
-		if (!write_trylock(&pg->rwlock)) {
+	while ((pg = list_first_entry_or_null(&cinf->dirty_list, struct cached_page, dirty_head))) {
+		if (write_trylock(&pg->rwlock)) {
 			spin_unlock(&cinf->dirty_lock);
-			cpu_relax();
-			goto retry;
-		}
+			list_for_each_entry_safe(item, tmp, &pg->dirty_list,
+						 dirty_head) {
+				clear_item_dirty(sb, cinf, pg, item);
 
+				if (item->delta)
+					scoutfs_inc_counter(sb, item_delta_written);
+
+				/* free deletion items */
+				if (item->deletion || item->delta)
+					erase_item(pg, item);
+				else
+					item->persistent = 1;
+			}
+
+			write_unlock(&pg->rwlock);
+			spin_lock(&cinf->dirty_lock);
+		}
 		spin_unlock(&cinf->dirty_lock);
-
-		list_for_each_entry_safe(item, tmp, &pg->dirty_list,
-					 dirty_head) {
-			clear_item_dirty(sb, cinf, pg, item);
-
-			if (item->delta)
-				scoutfs_inc_counter(sb, item_delta_written);
-
-			/* free deletion items */
-			if (item->deletion || item->delta)
-				erase_item(pg, item);
-			else
-				item->persistent = 1;
-		}
-
-		write_unlock(&pg->rwlock);
-
 		spin_lock(&cinf->dirty_lock);
-	}
-
+	} while (pg);
 	spin_unlock(&cinf->dirty_lock);
 
 	return 0;
@@ -2597,24 +2544,15 @@ static unsigned long item_cache_scan_objects(struct shrinker *shrink,
 	struct cached_page *tmp;
 	struct cached_page *pg;
 	unsigned long freed = 0;
-	u64 first_reader_seq;
 	int nr = sc->nr_to_scan;
 
 	scoutfs_inc_counter(sb, item_cache_scan_objects);
 
-	/* can't invalidate pages with items that weren't visible to first reader */
-	first_reader_seq = first_active_reader_seq(cinf);
-
 	write_lock(&cinf->rwlock);
 	spin_lock(&cinf->lru_lock);
 
 	list_for_each_entry_safe(pg, tmp, &cinf->lru_list, lru_head) {
 
-		if (first_reader_seq <= pg->max_seq) {
-			scoutfs_inc_counter(sb, item_shrink_page_reader);
-			continue;
-		}
-
 		if (!write_trylock(&pg->rwlock)) {
 			scoutfs_inc_counter(sb, item_shrink_page_trylock);
 			continue;
@@ -2681,8 +2619,6 @@ int scoutfs_item_setup(struct super_block *sb)
 	atomic_set(&cinf->dirty_pages, 0);
 	spin_lock_init(&cinf->lru_lock);
 	INIT_LIST_HEAD(&cinf->lru_list);
-	spin_lock_init(&cinf->active_lock);
-	INIT_LIST_HEAD(&cinf->active_list);
 
 	cinf->pcpu_pages = alloc_percpu(struct item_percpu_pages);
 	if (!cinf->pcpu_pages)
@@ -2715,8 +2651,6 @@ void scoutfs_item_destroy(struct super_block *sb)
 	int cpu;
 
 	if (cinf) {
-		BUG_ON(!list_empty(&cinf->active_list));
-
 #ifdef KC_CPU_NOTIFIER
 		unregister_hotcpu_notifier(&cinf->notifier);
 #endif

kmod/src/kernelcompat.c

@@ -81,3 +81,69 @@ kc_generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
 	return written ? written : status;
 }
 #endif
+
+#include <linux/list_lru.h>
+
+#ifdef KC_LIST_LRU_WALK_CB_ITEM_LOCK
+static enum lru_status kc_isolate(struct list_head *item, spinlock_t *lock, void *cb_arg)
+{
+	struct kc_isolate_args *args = cb_arg;
+
+	/* isolate doesn't use list, nr_items updated in caller */
+	return args->isolate(item, NULL, args->cb_arg);
+}
+
+unsigned long kc_list_lru_walk(struct list_lru *lru, kc_list_lru_walk_cb_t isolate, void *cb_arg,
+				      unsigned long nr_to_walk)
+{
+	struct kc_isolate_args args = {
+		.isolate = isolate,
+		.cb_arg = cb_arg,
+	};
+
+	return list_lru_walk(lru, kc_isolate, &args, nr_to_walk);
+}
+
+unsigned long kc_list_lru_shrink_walk(struct list_lru *lru, struct shrink_control *sc,
+				      kc_list_lru_walk_cb_t isolate, void *cb_arg)
+{
+	struct kc_isolate_args args = {
+		.isolate = isolate,
+		.cb_arg = cb_arg,
+	};
+
+	return list_lru_shrink_walk(lru, sc, kc_isolate, &args);
+}
+#endif
+
+#ifdef KC_LIST_LRU_WALK_CB_LIST_LOCK
+static enum lru_status kc_isolate(struct list_head *item, struct list_lru_one *list,
+				  spinlock_t *lock, void *cb_arg)
+{
+	struct kc_isolate_args *args = cb_arg;
+
+	return args->isolate(item, list, args->cb_arg);
+}
+
+unsigned long kc_list_lru_walk(struct list_lru *lru, kc_list_lru_walk_cb_t isolate, void *cb_arg,
+				      unsigned long nr_to_walk)
+{
+	struct kc_isolate_args args = {
+		.isolate = isolate,
+		.cb_arg = cb_arg,
+	};
+
+	return list_lru_walk(lru, kc_isolate, &args, nr_to_walk);
+}
+unsigned long kc_list_lru_shrink_walk(struct list_lru *lru, struct shrink_control *sc,
+				      kc_list_lru_walk_cb_t isolate, void *cb_arg)
+{
+	struct kc_isolate_args args = {
+		.isolate = isolate,
+		.cb_arg = cb_arg,
+	};
+
+	return list_lru_shrink_walk(lru, sc, kc_isolate, &args);
+}
+
+#endif

kmod/src/kernelcompat.h

@@ -29,50 +29,6 @@ do {						\
 })
 #endif
 
-#ifndef KC_ITERATE_DIR_CONTEXT
-typedef filldir_t kc_readdir_ctx_t;
-#define KC_DECLARE_READDIR(name, file, dirent, ctx) name(file, dirent, ctx)
-#define KC_FOP_READDIR readdir
-#define kc_readdir_pos(filp, ctx) (filp)->f_pos
-#define kc_dir_emit_dots(file, dirent, ctx) dir_emit_dots(file, dirent, ctx)
-#define kc_dir_emit(ctx, dirent, name, name_len, pos, ino, dt) \
-	(ctx(dirent, name, name_len, pos, ino, dt) == 0)
-#else
-typedef struct dir_context * kc_readdir_ctx_t;
-#define KC_DECLARE_READDIR(name, file, dirent, ctx) name(file, ctx)
-#define KC_FOP_READDIR iterate
-#define kc_readdir_pos(filp, ctx) (ctx)->pos
-#define kc_dir_emit_dots(file, dirent, ctx) dir_emit_dots(file, ctx)
-#define kc_dir_emit(ctx, dirent, name, name_len, pos, ino, dt) \
-	dir_emit(ctx, name, name_len, ino, dt)
-#endif
-
-#ifndef KC_DIR_EMIT_DOTS
-/*
- * Kernels before ->iterate and don't have dir_emit_dots so we give them
- * one that works with the ->readdir() filldir() method.
- */
-static inline int dir_emit_dots(struct file *file, void *dirent,
-				filldir_t filldir)
-{
-	if (file->f_pos == 0) {
-		if (filldir(dirent, ".", 1, 1,
-			    file->f_path.dentry->d_inode->i_ino, DT_DIR))
-			return 0;
-		file->f_pos = 1;
-	}
-
-	if (file->f_pos == 1) {
-		if (filldir(dirent, "..", 2, 1,
-			    parent_ino(file->f_path.dentry), DT_DIR))
-			return 0;
-		file->f_pos = 2;
-	}
-
-	return 1;
-}
-#endif
-
 #ifdef KC_POSIX_ACL_VALID_USER_NS
 #define kc_posix_acl_valid(user_ns, acl) posix_acl_valid(user_ns, acl)
 #else
@@ -239,9 +195,11 @@ struct kc_shrinker_wrapper {
 #include <linux/inet.h>
 static inline int kc_kernel_getsockname(struct socket *sock, struct sockaddr *addr)
 {
-	int addrlen = sizeof(struct sockaddr_in);
+	int addrlen = sizeof(struct sockaddr_storage);
 	int ret = kernel_getsockname(sock, addr, &addrlen);
-	if (ret == 0 && addrlen != sizeof(struct sockaddr_in))
+	if (ret == 0 && (!(
+	    (addrlen == sizeof(struct sockaddr_in)) ||
+	    (addrlen == sizeof(struct sockaddr_in6)))))
 		return -EAFNOSUPPORT;
 	else if (ret < 0)
 		return ret;
@@ -250,9 +208,11 @@ static inline int kc_kernel_getsockname(struct socket *sock, struct sockaddr *ad
 }
 static inline int kc_kernel_getpeername(struct socket *sock, struct sockaddr *addr)
 {
-	int addrlen = sizeof(struct sockaddr_in);
+	int addrlen = sizeof(struct sockaddr_storage);
 	int ret = kernel_getpeername(sock, addr, &addrlen);
-	if (ret == 0 && addrlen != sizeof(struct sockaddr_in))
+	if (ret == 0 && (!(
+	    (addrlen == sizeof(struct sockaddr_in)) ||
+	    (addrlen == sizeof(struct sockaddr_in6)))))
 		return -EAFNOSUPPORT;
 	else if (ret < 0)
 		return ret;
@@ -307,6 +267,11 @@ typedef unsigned int blk_opf_t;
 #define kc__vmalloc __vmalloc
 #endif
 
+#ifdef KC_VFS_METHOD_MNT_IDMAP_ARG
+#define KC_VFS_NS_DEF struct mnt_idmap *mnt_idmap,
+#define KC_VFS_NS mnt_idmap,
+#define KC_VFS_INIT_NS &nop_mnt_idmap,
+#else
 #ifdef KC_VFS_METHOD_USER_NAMESPACE_ARG
 #define KC_VFS_NS_DEF struct user_namespace *mnt_user_ns,
 #define KC_VFS_NS mnt_user_ns,
@@ -316,6 +281,7 @@ typedef unsigned int blk_opf_t;
 #define KC_VFS_NS
 #define KC_VFS_INIT_NS
 #endif
+#endif /* KC_VFS_METHOD_MNT_IDMAP_ARG */
 
 #ifdef KC_BIO_ALLOC_DEV_OPF_ARGS
 #define kc_bio_alloc bio_alloc
@@ -438,4 +404,93 @@ static inline int kc_tcp_sock_set_nodelay(struct socket *sock)
 }
 #endif
 
+#ifdef KC_INODE_DIO_END
+#define kc_inode_dio_end inode_dio_end
+#else
+#define kc_inode_dio_end inode_dio_done
+#endif
+
+#ifndef KC_MM_VM_FAULT_T
+typedef unsigned int vm_fault_t;
+static inline vm_fault_t vmf_error(int err)
+{
+	if (err == -ENOMEM)
+		return VM_FAULT_OOM;
+	return VM_FAULT_SIGBUS;
+}
+#endif
+
+#include <linux/list_lru.h>
+
+#ifndef KC_LIST_LRU_SHRINK_COUNT_WALK
+/* we don't bother with sc->{nid,memcg} (which doesn't exist in oldest kernels) */
+static inline unsigned long list_lru_shrink_count(struct list_lru *lru,
+                                                  struct shrink_control *sc)
+{
+        return list_lru_count(lru);
+}
+static inline unsigned long
+list_lru_shrink_walk(struct list_lru *lru, struct shrink_control *sc,
+		     list_lru_walk_cb isolate, void *cb_arg)
+{
+	return list_lru_walk(lru, isolate, cb_arg, sc->nr_to_scan);
+}
+#endif
+
+#ifndef KC_LIST_LRU_ADD_OBJ
+#define list_lru_add_obj list_lru_add
+#define list_lru_del_obj list_lru_del
+#endif
+
+#if defined(KC_LIST_LRU_WALK_CB_LIST_LOCK) || defined(KC_LIST_LRU_WALK_CB_ITEM_LOCK)
+struct list_lru_one;
+typedef enum lru_status (*kc_list_lru_walk_cb_t)(struct list_head *item, struct list_lru_one *list,
+						 void *cb_arg);
+struct kc_isolate_args {
+	kc_list_lru_walk_cb_t isolate;
+	void *cb_arg;
+};
+unsigned long kc_list_lru_walk(struct list_lru *lru, kc_list_lru_walk_cb_t isolate, void *cb_arg,
+			       unsigned long nr_to_walk);
+unsigned long kc_list_lru_shrink_walk(struct list_lru *lru, struct shrink_control *sc,
+				      kc_list_lru_walk_cb_t isolate, void *cb_arg);
+#else
+#define kc_list_lru_shrink_walk list_lru_shrink_walk
+#endif
+
+#if defined(KC_LIST_LRU_WALK_CB_ITEM_LOCK)
+/* isolate moved by hand, nr_items updated in walk as _REMOVE returned */
+static inline void list_lru_isolate_move(struct list_lru_one *list, struct list_head *item,
+					 struct list_head *head)
+{
+        list_move(item, head);
+}
+#endif
+
+#ifndef KC_STACK_TRACE_SAVE
+#include <linux/stacktrace.h>
+static inline unsigned int stack_trace_save(unsigned long *store, unsigned int size,
+					    unsigned int skipnr)
+{
+        struct stack_trace trace = {
+                .entries        = store,
+                .max_entries    = size,
+                .skip           = skipnr,
+        };
+
+        save_stack_trace(&trace);
+        return trace.nr_entries;
+}
+
+static inline void stack_trace_print(unsigned long *entries, unsigned int nr_entries, int spaces)
+{
+        struct stack_trace trace = {
+                .entries        = entries,
+                .nr_entries     = nr_entries,
+        };
+
+	print_stack_trace(&trace, spaces);
+}
+#endif
+
 #endif

kmod/src/lock.c

@@ -53,8 +53,10 @@
  * all access to the lock (by revoking it down to a null mode) then the
  * lock is freed.
  *
- * Memory pressure on the client can cause the client to request a null
- * mode from the server so that once its granted the lock can be freed.
+ * Each client has a configurable number of locks that are allowed to
+ * remain idle after being granted, for use by future tasks.  Past the
+ * limit locks are freed by requesting a null mode from the server,
+ * governed by a LRU.
  *
  * So far we've only needed a minimal trylock.  We return -EAGAIN if a
  * lock attempt can't immediately match an existing granted lock.  This
@@ -79,14 +81,11 @@ struct lock_info {
 	bool unmounting;
 	struct rb_root lock_tree;
 	struct rb_root lock_range_tree;
-	KC_DEFINE_SHRINKER(shrinker);
+	u64 nr_locks;
 	struct list_head lru_list;
-	unsigned long long lru_nr;
 	struct workqueue_struct *workq;
 	struct work_struct inv_work;
 	struct list_head inv_list;
-	struct work_struct shrink_work;
-	struct list_head shrink_list;
 	atomic64_t next_refresh_gen;
 
 	struct dentry *tseq_dentry;
@@ -168,7 +167,6 @@ static int lock_invalidate(struct super_block *sb, struct scoutfs_lock *lock,
 			   enum scoutfs_lock_mode prev, enum scoutfs_lock_mode mode)
 {
 	struct scoutfs_lock_coverage *cov;
-	struct scoutfs_lock_coverage *tmp;
 	u64 ino, last;
 	int ret = 0;
 
@@ -192,19 +190,22 @@ static int lock_invalidate(struct super_block *sb, struct scoutfs_lock *lock,
 
 	/* have to invalidate if we're not in the only usable case */
 	if (!(prev == SCOUTFS_LOCK_WRITE && mode == SCOUTFS_LOCK_READ)) {
-retry:
-		/* remove cov items to tell users that their cache is stale */
+		/*
+		 * Remove cov items to tell users that their cache is
+		 * stale.  The unlock pattern comes from avoiding bad
+		 * sparse warnings when taking else in a failed trylock.
+		 */
 		spin_lock(&lock->cov_list_lock);
-		list_for_each_entry_safe(cov, tmp, &lock->cov_list, head) {
-			if (!spin_trylock(&cov->cov_lock)) {
-				spin_unlock(&lock->cov_list_lock);
-				cpu_relax();
-				goto retry;
+		while ((cov = list_first_entry_or_null(&lock->cov_list,
+						       struct scoutfs_lock_coverage, head))) {
+			if (spin_trylock(&cov->cov_lock)) {
+				list_del_init(&cov->head);
+				cov->lock = NULL;
+				spin_unlock(&cov->cov_lock);
+				scoutfs_inc_counter(sb, lock_invalidate_coverage);
 			}
-			list_del_init(&cov->head);
-			cov->lock = NULL;
-			spin_unlock(&cov->cov_lock);
-			scoutfs_inc_counter(sb, lock_invalidate_coverage);
+			spin_unlock(&lock->cov_list_lock);
+			spin_lock(&lock->cov_list_lock);
 		}
 		spin_unlock(&lock->cov_list_lock);
 
@@ -247,7 +248,6 @@ static void lock_free(struct lock_info *linfo, struct scoutfs_lock *lock)
 	BUG_ON(!RB_EMPTY_NODE(&lock->range_node));
 	BUG_ON(!list_empty(&lock->lru_head));
 	BUG_ON(!list_empty(&lock->inv_head));
-	BUG_ON(!list_empty(&lock->shrink_head));
 	BUG_ON(!list_empty(&lock->cov_list));
 
 	kfree(lock->inode_deletion_data);
@@ -275,7 +275,6 @@ static struct scoutfs_lock *lock_alloc(struct super_block *sb,
 	INIT_LIST_HEAD(&lock->lru_head);
 	INIT_LIST_HEAD(&lock->inv_head);
 	INIT_LIST_HEAD(&lock->inv_list);
-	INIT_LIST_HEAD(&lock->shrink_head);
 	spin_lock_init(&lock->cov_list_lock);
 	INIT_LIST_HEAD(&lock->cov_list);
 
@@ -302,6 +301,7 @@ static void lock_inc_count(unsigned int *counts, enum scoutfs_lock_mode mode)
 static void lock_dec_count(unsigned int *counts, enum scoutfs_lock_mode mode)
 {
 	BUG_ON(mode < 0 || mode >= SCOUTFS_LOCK_NR_MODES);
+	BUG_ON(counts[mode] == 0);
 	counts[mode]--;
 }
 
@@ -407,6 +407,7 @@ static bool lock_insert(struct super_block *sb, struct scoutfs_lock *ins)
 	rb_link_node(&ins->node, parent, node);
 	rb_insert_color(&ins->node, &linfo->lock_tree);
 
+	linfo->nr_locks++;
 	scoutfs_tseq_add(&linfo->tseq_tree, &ins->tseq_entry);
 
 	return true;
@@ -421,6 +422,7 @@ static void lock_remove(struct lock_info *linfo, struct scoutfs_lock *lock)
 	rb_erase(&lock->range_node, &linfo->lock_range_tree);
 	RB_CLEAR_NODE(&lock->range_node);
 
+	linfo->nr_locks--;
 	scoutfs_tseq_del(&linfo->tseq_tree, &lock->tseq_entry);
 }
 
@@ -460,10 +462,8 @@ static void __lock_del_lru(struct lock_info *linfo, struct scoutfs_lock *lock)
 {
 	assert_spin_locked(&linfo->lock);
 
-	if (!list_empty(&lock->lru_head)) {
+	if (!list_empty(&lock->lru_head))
 		list_del_init(&lock->lru_head);
-		linfo->lru_nr--;
-	}
 }
 
 /*
@@ -522,14 +522,16 @@ static struct scoutfs_lock *create_lock(struct super_block *sb,
  * indicate that the lock wasn't idle.  If it really is idle then we
  * either free it if it's null or put it back on the lru.
  */
-static void put_lock(struct lock_info *linfo,struct scoutfs_lock *lock)
+static void __put_lock(struct lock_info *linfo, struct scoutfs_lock *lock, bool tail)
 {
 	assert_spin_locked(&linfo->lock);
 
 	if (lock_idle(lock)) {
 		if (lock->mode != SCOUTFS_LOCK_NULL) {
-			list_add_tail(&lock->lru_head, &linfo->lru_list);
-			linfo->lru_nr++;
+			if (tail)
+				list_add_tail(&lock->lru_head, &linfo->lru_list);
+			else
+				list_add(&lock->lru_head, &linfo->lru_list);
 		} else {
 			lock_remove(linfo, lock);
 			lock_free(linfo, lock);
@@ -537,6 +539,11 @@ static void put_lock(struct lock_info *linfo,struct scoutfs_lock *lock)
 	}
 }
 
+static inline void put_lock(struct lock_info *linfo, struct scoutfs_lock *lock)
+{
+	__put_lock(linfo, lock, true);
+}
+
 /*
  * The caller has made a change (set a lock mode) which can let one of the
  * invalidating locks make forward progress.
@@ -710,14 +717,14 @@ static void lock_invalidate_worker(struct work_struct *work)
 		/* only lock protocol, inv can't call subsystems after shutdown */
 		if (!linfo->shutdown) {
 			ret = lock_invalidate(sb, lock, nl->old_mode, nl->new_mode);
-			BUG_ON(ret);
+			BUG_ON(ret < 0 && ret != -ENOLINK);
 		}
 
 		/* respond with the key and modes from the request, server might have died */
 		ret = scoutfs_client_lock_response(sb, ireq->net_id, nl);
 		if (ret == -ENOTCONN)
 			ret = 0;
-		BUG_ON(ret);
+		BUG_ON(ret < 0 && ret != -ENOLINK);
 
 		scoutfs_inc_counter(sb, lock_invalidate_response);
 	}
@@ -872,6 +879,69 @@ int scoutfs_lock_recover_request(struct super_block *sb, u64 net_id,
 	return ret;
 }
 
+/*
+ * This is called on every _lock call to try and keep the number of
+ * locks under the idle count.  We're intentionally trying to throttle
+ * shrinking bursts by tying its frequency to lock use.  It will only
+ * send requests to free unused locks, though, so it's always possible
+ * to exceed the high water mark under heavy load.
+ *
+ * We send a null request and the lock will be freed by the response
+ * once all users drain.  If this races with invalidation then the
+ * server will only send the grant response once the invalidation is
+ * finished.
+ */
+static bool try_shrink_lock(struct super_block *sb, struct lock_info *linfo, bool force)
+{
+	struct scoutfs_mount_options opts;
+	struct scoutfs_lock *lock = NULL;
+	struct scoutfs_net_lock nl;
+	int ret = 0;
+
+	scoutfs_options_read(sb, &opts);
+
+	/* avoiding lock contention with unsynchronized test, don't mind temp false results */
+	if (!force && (list_empty(&linfo->lru_list) ||
+	               READ_ONCE(linfo->nr_locks) <= opts.lock_idle_count))
+		return false;
+
+	spin_lock(&linfo->lock);
+
+	lock = list_first_entry_or_null(&linfo->lru_list, struct scoutfs_lock, lru_head);
+	if (lock && (force || (linfo->nr_locks > opts.lock_idle_count))) {
+		__lock_del_lru(linfo, lock);
+		lock->request_pending = 1;
+
+		nl.key = lock->start;
+		nl.old_mode = lock->mode;
+		nl.new_mode = SCOUTFS_LOCK_NULL;
+	} else {
+		lock = NULL;
+	}
+
+	spin_unlock(&linfo->lock);
+
+	if (lock) {
+		ret = scoutfs_client_lock_request(sb, &nl);
+		if (ret < 0) {
+			scoutfs_inc_counter(sb, lock_shrink_request_failed);
+
+			spin_lock(&linfo->lock);
+
+			lock->request_pending = 0;
+			wake_up(&lock->waitq);
+			__put_lock(linfo, lock, false);
+
+			spin_unlock(&linfo->lock);
+		} else {
+			scoutfs_inc_counter(sb, lock_shrink_attempted);
+			trace_scoutfs_lock_shrink(sb, lock);
+		}
+	}
+
+	return lock && ret == 0;
+}
+
 static bool lock_wait_cond(struct super_block *sb, struct scoutfs_lock *lock,
 			   enum scoutfs_lock_mode mode)
 {
@@ -934,6 +1004,8 @@ static int lock_key_range(struct super_block *sb, enum scoutfs_lock_mode mode, i
 	if (WARN_ON_ONCE(scoutfs_trans_held()))
 		return -EDEADLK;
 
+	try_shrink_lock(sb, linfo, false);
+
 	spin_lock(&linfo->lock);
 
 	/* drops and re-acquires lock if it allocates */
@@ -1377,134 +1449,12 @@ bool scoutfs_lock_protected(struct scoutfs_lock *lock, struct scoutfs_key *key,
 					  &lock->start, &lock->end) == 0;
 }
 
-/*
- * The shrink callback got the lock, marked it request_pending, and put
- * it on the shrink list.  We send a null request and the lock will be
- * freed by the response once all users drain.  If this races with
- * invalidation then the server will only send the grant response once
- * the invalidation is finished.
- */
-static void lock_shrink_worker(struct work_struct *work)
-{
-	struct lock_info *linfo = container_of(work, struct lock_info,
-					       shrink_work);
-	struct super_block *sb = linfo->sb;
-	struct scoutfs_net_lock nl;
-	struct scoutfs_lock *lock;
-	struct scoutfs_lock *tmp;
-	LIST_HEAD(list);
-	int ret;
-
-	scoutfs_inc_counter(sb, lock_shrink_work);
-
-	spin_lock(&linfo->lock);
-	list_splice_init(&linfo->shrink_list, &list);
-	spin_unlock(&linfo->lock);
-
-	list_for_each_entry_safe(lock, tmp, &list, shrink_head) {
-		list_del_init(&lock->shrink_head);
-
-		/* unlocked lock access, but should be stable since we queued */
-		nl.key = lock->start;
-		nl.old_mode = lock->mode;
-		nl.new_mode = SCOUTFS_LOCK_NULL;
-
-		ret = scoutfs_client_lock_request(sb, &nl);
-		if (ret) {
-			/* oh well, not freeing */
-			scoutfs_inc_counter(sb, lock_shrink_aborted);
-
-			spin_lock(&linfo->lock);
-
-			lock->request_pending = 0;
-			wake_up(&lock->waitq);
-			put_lock(linfo, lock);
-
-			spin_unlock(&linfo->lock);
-		}
-	}
-}
-
-static unsigned long lock_count_objects(struct shrinker *shrink,
-					struct shrink_control *sc)
-{
-	struct lock_info *linfo = KC_SHRINKER_CONTAINER_OF(shrink, struct lock_info);
-	struct super_block *sb = linfo->sb;
-
-	scoutfs_inc_counter(sb, lock_count_objects);
-
-	return shrinker_min_long(linfo->lru_nr);
-}
-
-/*
- * Start the shrinking process for locks on the lru.  If a lock is on
- * the lru then it can't have any active users.  We don't want to block
- * or allocate here so all we do is get the lock, mark it request
- * pending, and kick off the work.  The work sends a null request and
- * eventually the lock is freed by its response.
- *
- * Only a racing lock attempt that isn't matched can prevent the lock
- * from being freed.  It'll block waiting to send its request for its
- * mode which will prevent the lock from being freed when the null
- * response arrives.
- */
-static unsigned long lock_scan_objects(struct shrinker *shrink,
-				       struct shrink_control *sc)
-{
-	struct lock_info *linfo = KC_SHRINKER_CONTAINER_OF(shrink, struct lock_info);
-	struct super_block *sb = linfo->sb;
-	struct scoutfs_lock *lock;
-	struct scoutfs_lock *tmp;
-	unsigned long freed = 0;
-	unsigned long nr = sc->nr_to_scan;
-	bool added = false;
-
-	scoutfs_inc_counter(sb, lock_scan_objects);
-
-	spin_lock(&linfo->lock);
-
-restart:
-	list_for_each_entry_safe(lock, tmp, &linfo->lru_list, lru_head) {
-
-		BUG_ON(!lock_idle(lock));
-		BUG_ON(lock->mode == SCOUTFS_LOCK_NULL);
-		BUG_ON(!list_empty(&lock->shrink_head));
-
-		if (nr-- == 0)
-			break;
-
-		__lock_del_lru(linfo, lock);
-		lock->request_pending = 1;
-		list_add_tail(&lock->shrink_head, &linfo->shrink_list);
-		added = true;
-		freed++;
-
-		scoutfs_inc_counter(sb, lock_shrink_attempted);
-		trace_scoutfs_lock_shrink(sb, lock);
-
-		/* could have bazillions of idle locks */
-		if (cond_resched_lock(&linfo->lock))
-			goto restart;
-	}
-
-	spin_unlock(&linfo->lock);
-
-	if (added)
-		queue_work(linfo->workq, &linfo->shrink_work);
-
-	trace_scoutfs_lock_shrink_exit(sb, sc->nr_to_scan, freed);
-	return freed;
-}
-
 void scoutfs_free_unused_locks(struct super_block *sb)
 {
-	struct lock_info *linfo = SCOUTFS_SB(sb)->lock_info;
-	struct shrink_control sc = {
-		.gfp_mask = GFP_NOFS,
-		.nr_to_scan = INT_MAX,
-	};
+	DECLARE_LOCK_INFO(sb, linfo);
 
-	lock_scan_objects(KC_SHRINKER_FN(&linfo->shrinker), &sc);
+	while (try_shrink_lock(sb, linfo, true))
+		cond_resched();
 }
 
 static void lock_tseq_show(struct seq_file *m, struct scoutfs_tseq_entry *ent)
@@ -1587,10 +1537,10 @@ u64 scoutfs_lock_ino_refresh_gen(struct super_block *sb, u64 ino)
  * transitions and sending requests.   We set the shutdown flag to catch
  * anyone who breaks this rule.
  *
- * We unregister the shrinker so that we won't try and send null
- * requests in response to memory pressure.  The locks will all be
- * unceremoniously dropped once we get a farewell response from the
- * server which indicates that they destroyed our locking state.
+ * With no more lock callers, we'll no longer try to shrink the pool of
+ * granted locks.  We'll free all of them as _destroy() is called after
+ * the farewell response indicates that the server tore down all our
+ * lock state.
  *
  * We will still respond to invalidation requests that have to be
  * processed to let unmount in other mounts acquire locks and make
@@ -1610,10 +1560,6 @@ void scoutfs_lock_shutdown(struct super_block *sb)
 
 	trace_scoutfs_lock_shutdown(sb, linfo);
 
-	/* stop the shrinker from queueing work */
-	KC_UNREGISTER_SHRINKER(&linfo->shrinker);
-	flush_work(&linfo->shrink_work);
-
 	/* cause current and future lock calls to return errors */
 	spin_lock(&linfo->lock);
 	linfo->shutdown = true;
@@ -1704,8 +1650,6 @@ void scoutfs_lock_destroy(struct super_block *sb)
 			list_del_init(&lock->inv_head);
 			lock->invalidate_pending = 0;
 		}
-		if (!list_empty(&lock->shrink_head))
-			list_del_init(&lock->shrink_head);
 		lock_remove(linfo, lock);
 		lock_free(linfo, lock);
 	}
@@ -1730,14 +1674,9 @@ int scoutfs_lock_setup(struct super_block *sb)
 	spin_lock_init(&linfo->lock);
 	linfo->lock_tree = RB_ROOT;
 	linfo->lock_range_tree = RB_ROOT;
-	KC_INIT_SHRINKER_FUNCS(&linfo->shrinker, lock_count_objects,
-			       lock_scan_objects);
-	KC_REGISTER_SHRINKER(&linfo->shrinker, "scoutfs-lock:" SCSBF, SCSB_ARGS(sb));
 	INIT_LIST_HEAD(&linfo->lru_list);
 	INIT_WORK(&linfo->inv_work, lock_invalidate_worker);
 	INIT_LIST_HEAD(&linfo->inv_list);
-	INIT_WORK(&linfo->shrink_work, lock_shrink_worker);
-	INIT_LIST_HEAD(&linfo->shrink_list);
 	atomic64_set(&linfo->next_refresh_gen, 0);
 	scoutfs_tseq_tree_init(&linfo->tseq_tree, lock_tseq_show);
 

kmod/src/lock_server.c

@@ -202,21 +202,48 @@ static u8 invalidation_mode(u8 granted, u8 requested)
 
 /*
  * Return true of the client lock instances described by the entries can
- * be granted at the same time.  Typically this only means they're both
- * modes that are compatible between nodes. In addition there's the
- * special case where a read lock on a client is compatible with a write
- * lock on the same client because the client's cache covered by the
- * read lock is still valid if they get a write lock.
+ * be granted at the same time.  There's only three cases where this is
+ * true.
+ *
+ * First, the two locks are both of the same mode that allows full
+ * sharing -- read and write only.  The only point of these modes is
+ * that everyone can share them.
+ *
+ * Second, a write lock gives the client permission to read as well.
+ * This means that a client can upgrade its read lock to a write lock
+ * without having to invalidate the existing read and drop caches.
+ *
+ * Third, null locks are always compatible between clients.  It's as
+ * though the client with the null lock has no lock at all.  But it's
+ * never compatible with all locks on the client requesting null.
+ * Sending invalidations for existing locks on a client when we get a
+ * null request is how we resolve races in shrinking locks -- we turn it
+ * into the unsolicited remote invalidation case.
+ *
+ * All other mode and client combinations can not be shared, most
+ * typically a write lock invalidating all other non-write holders to
+ * drop caches and force a read after the write has completed.
  */
 static bool client_entries_compatible(struct client_lock_entry *granted,
 				      struct client_lock_entry *requested)
 {
-	return (granted->mode == requested->mode &&
-		(granted->mode == SCOUTFS_LOCK_READ ||
-		 granted->mode == SCOUTFS_LOCK_WRITE_ONLY)) ||
-	       (granted->rid == requested->rid &&
-		granted->mode == SCOUTFS_LOCK_READ &&
-		requested->mode == SCOUTFS_LOCK_WRITE);
+	/* only read and write_only can be full shared */
+	if ((granted->mode == requested->mode) &&
+	    (granted->mode == SCOUTFS_LOCK_READ || granted->mode == SCOUTFS_LOCK_WRITE_ONLY))
+		return true;
+
+	/* _write includes reading, so a client can upgrade its read to write */
+	if (granted->rid == requested->rid &&
+	    granted->mode == SCOUTFS_LOCK_READ &&
+	    requested->mode == SCOUTFS_LOCK_WRITE)
+		return true;
+
+	/* null is always compatible across clients, never within a client */
+	if ((granted->rid != requested->rid) &&
+	    (granted->mode == SCOUTFS_LOCK_NULL || requested->mode == SCOUTFS_LOCK_NULL))
+		return true;
+
+	return false;
 }
 
 /*
@@ -317,16 +344,18 @@ static void put_server_lock(struct lock_server_info *inf,
 
 	BUG_ON(!mutex_is_locked(&snode->mutex));
 
+	spin_lock(&inf->lock);
+
 	if (atomic_dec_and_test(&snode->refcount) &&
 	    list_empty(&snode->granted) &&
 	    list_empty(&snode->requested) &&
 	    list_empty(&snode->invalidated)) {
-		spin_lock(&inf->lock);
 		rb_erase(&snode->node, &inf->locks_root);
-		spin_unlock(&inf->lock);
 		should_free = true;
 	}
 
+	spin_unlock(&inf->lock);
+
 	mutex_unlock(&snode->mutex);
 
 	if (should_free) {
@@ -477,6 +506,19 @@ out:
  * because we don't know which locks they'll hold.  Once recover
  * finishes the server calls us to kick all the locks that were waiting
  * during recovery.
+ *
+ * The calling server shuts down if we return errors indicating that we
+ * weren't able to ensure forward progress in the lock state machine.
+ *
+ * Failure to send to a disconnected client is not a fatal error.
+ * During normal disconnection the client's state is removed before
+ * their connection is destroyed.  We can't use state to try and send to
+ * a non-existing connection.  But a client that fails to reconnect is
+ * disconnected before being fenced.  If we have multiple disconnected
+ * clients we can try to send to one while cleaning up another.  If
+ * they've uncleanly disconnected their locks are going to be removed
+ * and the lock can make forward progress again.  Or we'll shutdown for
+ * failure to fence.
  */
 static int process_waiting_requests(struct super_block *sb,
 				    struct server_lock_node *snode)
@@ -568,6 +610,10 @@ static int process_waiting_requests(struct super_block *sb,
 out:
 	put_server_lock(inf, snode);
 
+	/* disconnected clients will be fenced, trying to send to them isn't fatal */
+	if (ret == -ENOTCONN)
+		ret = 0;
+
 	return ret;
 }
 

kmod/src/msg.h

@@ -35,6 +35,12 @@ do {									\
 	}								\
 } while (0)								\
 
+#define scoutfs_bug_on_err(sb, err, fmt, args...) \
+do { \
+	__typeof__(err) _err = (err); \
+	scoutfs_bug_on(sb, _err < 0 && _err != -ENOLINK, fmt, ##args); \
+} while (0)
+
 /*
  * Each message is only generated once per volume.  Remounting resets
  * the messages.

kmod/src/net.h

@@ -1,10 +1,18 @@
 #ifndef _SCOUTFS_NET_H_
 #define _SCOUTFS_NET_H_
 
+#include <linux/spinlock.h>
+#include <linux/list.h>
 #include <linux/in.h>
 #include "endian_swap.h"
 #include "tseq.h"
 
+struct scoutfs_work_list {
+	struct work_struct work;
+	spinlock_t lock;
+	struct list_head list;
+};
+
 struct scoutfs_net_connection;
 
 /* These are called in their own blocking context */
@@ -41,15 +49,15 @@ struct scoutfs_net_connection {
 	unsigned long flags; /* CONN_FL_* bitmask */
 	unsigned long reconn_deadline;
 
-	struct sockaddr_in connect_sin;
+	struct sockaddr_storage connect_sin;
 	unsigned long connect_timeout_ms;
 
 	struct socket *sock;
 	u64 rid;
 	u64 greeting_id;
-	struct sockaddr_in sockname;
-	struct sockaddr_in peername;
-	struct sockaddr_in last_peername;
+	struct sockaddr_storage sockname;
+	struct sockaddr_storage peername;
+	struct sockaddr_storage last_peername;
 
 	struct list_head accepted_head;
 	struct scoutfs_net_connection *listening_conn;
@@ -59,8 +67,12 @@ struct scoutfs_net_connection {
 	u64 next_send_id;
 	struct list_head send_queue;
 	struct list_head resend_queue;
+	struct rb_root req_root;
+	struct rb_root resp_root;
 
 	atomic64_t recv_seq;
+	unsigned int ordered_proc_nr;
+	struct scoutfs_work_list *ordered_proc_wlists;
 
 	struct workqueue_struct *workq;
 	struct work_struct listen_work;
@@ -87,27 +99,44 @@ enum conn_flags {
 	CONN_FL_reconn_freeing = (1UL << 6), /* waiting done, setter frees */
 };
 
-#define SIN_FMT		"%pIS:%u"
-#define SIN_ARG(sin)	sin, be16_to_cpu((sin)->sin_port)
+#define SIN_FMT		"%pISpc"
+#define SIN_ARG(sin)	sin
 
-static inline void scoutfs_addr_to_sin(struct sockaddr_in *sin,
+static inline void scoutfs_addr_to_sin(struct sockaddr_storage *sin,
 				       union scoutfs_inet_addr *addr)
 {
-	BUG_ON(addr->v4.family != cpu_to_le16(SCOUTFS_AF_IPV4));
-
-	sin->sin_family = AF_INET;
-	sin->sin_addr.s_addr = cpu_to_be32(le32_to_cpu(addr->v4.addr));
-	sin->sin_port = cpu_to_be16(le16_to_cpu(addr->v4.port));
+	if (addr->v4.family == cpu_to_le16(SCOUTFS_AF_IPV4)) {
+		struct sockaddr_in *sin4 = (struct sockaddr_in *)sin;
+		memset(sin, 0, sizeof(struct sockaddr_storage));
+		sin4->sin_family = AF_INET;
+		sin4->sin_addr.s_addr = cpu_to_be32(le32_to_cpu(addr->v4.addr));
+		sin4->sin_port = cpu_to_be16(le16_to_cpu(addr->v4.port));
+	} else if (addr->v6.family == cpu_to_le16(SCOUTFS_AF_IPV6)) {
+		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sin;
+		memset(sin, 0, sizeof(struct sockaddr_storage));
+		sin6->sin6_family = AF_INET6;
+		memcpy(&sin6->sin6_addr.in6_u.u6_addr8, &addr->v6.addr, 16);
+		sin6->sin6_port = cpu_to_be16(le16_to_cpu(addr->v6.port));
+	} else
+		BUG();
 }
 
-static inline void scoutfs_sin_to_addr(union scoutfs_inet_addr *addr, struct sockaddr_in *sin)
+static inline void scoutfs_sin_to_addr(union scoutfs_inet_addr *addr, struct sockaddr_storage *sin)
 {
-	BUG_ON(sin->sin_family != AF_INET);
-
-	memset(addr, 0, sizeof(union scoutfs_inet_addr));
-	addr->v4.family = cpu_to_le16(SCOUTFS_AF_IPV4);
-	addr->v4.addr = be32_to_le32(sin->sin_addr.s_addr);
-	addr->v4.port = be16_to_le16(sin->sin_port);
+	if (sin->ss_family == AF_INET) {
+		struct sockaddr_in *sin4 = (struct sockaddr_in *)sin;
+		memset(addr, 0, sizeof(union scoutfs_inet_addr));
+		addr->v4.family = cpu_to_le16(SCOUTFS_AF_IPV4);
+		addr->v4.addr = be32_to_le32(sin4->sin_addr.s_addr);
+		addr->v4.port = be16_to_le16(sin4->sin_port);
+	} else if (sin->ss_family == AF_INET6) {
+		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sin;
+		memset(addr, 0, sizeof(union scoutfs_inet_addr));
+		addr->v6.family = cpu_to_le16(SCOUTFS_AF_IPV6);
+		memcpy(&addr->v6.addr, &sin6->sin6_addr.in6_u.u6_addr8, 16);
+		addr->v6.port = be16_to_le16(sin6->sin6_port);
+	} else
+		BUG();
 }
 
 struct scoutfs_net_connection *
@@ -118,10 +147,10 @@ scoutfs_net_alloc_conn(struct super_block *sb,
 u64 scoutfs_net_client_rid(struct scoutfs_net_connection *conn);
 int scoutfs_net_connect(struct super_block *sb,
 			struct scoutfs_net_connection *conn,
-			struct sockaddr_in *sin, unsigned long timeout_ms);
+			struct sockaddr_storage *sin, unsigned long timeout_ms);
 int scoutfs_net_bind(struct super_block *sb,
 		     struct scoutfs_net_connection *conn,
-		     struct sockaddr_in *sin);
+		     struct sockaddr_storage *sin);
 void scoutfs_net_listen(struct super_block *sb,
 			struct scoutfs_net_connection *conn);
 int scoutfs_net_submit_request(struct super_block *sb,

kmod/src/omap.c

@@ -592,7 +592,7 @@ static int handle_request(struct super_block *sb, struct omap_request *req)
 	ret = 0;
 out:
 	free_rids(&priv_rids);
-	if (ret < 0) {
+	if ((ret < 0) && (req != NULL)) {
 		ret = scoutfs_server_send_omap_response(sb, req->client_rid, req->client_id,
 							NULL, ret);
 		free_req(req);

kmod/src/options.c

@@ -33,12 +33,15 @@ enum {
 	Opt_acl,
 	Opt_data_prealloc_blocks,
 	Opt_data_prealloc_contig_only,
+	Opt_ino_alloc_per_lock,
+	Opt_lock_idle_count,
 	Opt_log_merge_wait_timeout_ms,
 	Opt_metadev_path,
 	Opt_noacl,
 	Opt_orphan_scan_delay_ms,
 	Opt_quorum_heartbeat_timeout_ms,
 	Opt_quorum_slot_nr,
+	Opt_tcp_keepalive_timeout_ms,
 	Opt_err,
 };
 
@@ -46,12 +49,15 @@ static const match_table_t tokens = {
 	{Opt_acl, "acl"},
 	{Opt_data_prealloc_blocks, "data_prealloc_blocks=%s"},
 	{Opt_data_prealloc_contig_only, "data_prealloc_contig_only=%s"},
+	{Opt_ino_alloc_per_lock, "ino_alloc_per_lock=%s"},
+	{Opt_lock_idle_count, "lock_idle_count=%s"},
 	{Opt_log_merge_wait_timeout_ms, "log_merge_wait_timeout_ms=%s"},
 	{Opt_metadev_path, "metadev_path=%s"},
 	{Opt_noacl, "noacl"},
 	{Opt_orphan_scan_delay_ms, "orphan_scan_delay_ms=%s"},
 	{Opt_quorum_heartbeat_timeout_ms, "quorum_heartbeat_timeout_ms=%s"},
 	{Opt_quorum_slot_nr, "quorum_slot_nr=%s"},
+	{Opt_tcp_keepalive_timeout_ms, "tcp_keepalive_timeout_ms=%s"},
 	{Opt_err, NULL}
 };
 
@@ -115,6 +121,10 @@ static void free_options(struct scoutfs_mount_options *opts)
 	kfree(opts->metadev_path);
 }
 
+#define MIN_LOCK_IDLE_COUNT	32
+#define DEFAULT_LOCK_IDLE_COUNT	(10 * 1000)
+#define MAX_LOCK_IDLE_COUNT	(100 * 1000)
+
 #define MIN_LOG_MERGE_WAIT_TIMEOUT_MS		100UL
 #define DEFAULT_LOG_MERGE_WAIT_TIMEOUT_MS	500
 #define MAX_LOG_MERGE_WAIT_TIMEOUT_MS		(60 * MSEC_PER_SEC)
@@ -126,16 +136,36 @@ static void free_options(struct scoutfs_mount_options *opts)
 #define MIN_DATA_PREALLOC_BLOCKS	1ULL
 #define MAX_DATA_PREALLOC_BLOCKS	((unsigned long long)SCOUTFS_BLOCK_SM_MAX)
 
+#define DEFAULT_TCP_KEEPALIVE_TIMEOUT_MS	(60 * MSEC_PER_SEC)
+
 static void init_default_options(struct scoutfs_mount_options *opts)
 {
 	memset(opts, 0, sizeof(*opts));
 
 	opts->data_prealloc_blocks = SCOUTFS_DATA_PREALLOC_DEFAULT_BLOCKS;
 	opts->data_prealloc_contig_only = 1;
+	opts->ino_alloc_per_lock = SCOUTFS_LOCK_INODE_GROUP_NR;
+	opts->lock_idle_count = DEFAULT_LOCK_IDLE_COUNT;
 	opts->log_merge_wait_timeout_ms = DEFAULT_LOG_MERGE_WAIT_TIMEOUT_MS;
 	opts->orphan_scan_delay_ms = -1;
 	opts->quorum_heartbeat_timeout_ms = SCOUTFS_QUORUM_DEF_HB_TIMEO_MS;
 	opts->quorum_slot_nr = -1;
+	opts->tcp_keepalive_timeout_ms = DEFAULT_TCP_KEEPALIVE_TIMEOUT_MS;
+}
+
+static int verify_lock_idle_count(struct super_block *sb, int ret, int val)
+{
+	if (ret < 0) {
+		scoutfs_err(sb, "failed to parse lock_idle_count value");
+		return -EINVAL;
+	}
+	if (val < MIN_LOCK_IDLE_COUNT || val > MAX_LOCK_IDLE_COUNT) {
+		scoutfs_err(sb, "invalid lock_idle_count value %d, must be between %u and %u",
+			    val, MIN_LOCK_IDLE_COUNT, MAX_LOCK_IDLE_COUNT);
+		return -EINVAL;
+	}
+
+	return 0;
 }
 
 static int verify_log_merge_wait_timeout_ms(struct super_block *sb, int ret, int val)
@@ -168,6 +198,21 @@ static int verify_quorum_heartbeat_timeout_ms(struct super_block *sb, int ret, u
 	return 0;
 }
 
+static int verify_tcp_keepalive_timeout_ms(struct super_block *sb, int ret, int val)
+{
+	if (ret < 0) {
+		scoutfs_err(sb, "failed to parse tcp_keepalive_timeout_ms value");
+		return -EINVAL;
+	}
+	if (val <= (UNRESPONSIVE_PROBES * MSEC_PER_SEC)) {
+		scoutfs_err(sb, "invalid tcp_keepalive_timeout_ms value %d, must be larger than %lu",
+			    val, (UNRESPONSIVE_PROBES * MSEC_PER_SEC));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 /*
  * Parse the option string into our options struct.   This can allocate
  * memory in the struct.  The caller is responsible for always calling
@@ -218,6 +263,34 @@ static int parse_options(struct super_block *sb, char *options, struct scoutfs_m
 			opts->data_prealloc_contig_only = nr;
 			break;
 
+		case Opt_ino_alloc_per_lock:
+			ret = match_int(args, &nr);
+			if (ret < 0 || nr < 1 || nr > SCOUTFS_LOCK_INODE_GROUP_NR) {
+				scoutfs_err(sb, "invalid ino_alloc_per_lock option, must be between 1 and %u",
+					    SCOUTFS_LOCK_INODE_GROUP_NR);
+				if (ret == 0)
+					ret = -EINVAL;
+				return ret;
+			}
+			opts->ino_alloc_per_lock = nr;
+			break;
+
+		case Opt_tcp_keepalive_timeout_ms:
+			ret = match_int(args, &nr);
+			ret = verify_tcp_keepalive_timeout_ms(sb, ret, nr);
+			if (ret < 0)
+				return ret;
+			opts->tcp_keepalive_timeout_ms = nr;
+			break;
+
+		case Opt_lock_idle_count:
+			ret = match_int(args, &nr);
+			ret = verify_lock_idle_count(sb, ret, nr);
+			if (ret < 0)
+				return ret;
+			opts->lock_idle_count = nr;
+			break;
+
 		case Opt_log_merge_wait_timeout_ms:
 			ret = match_int(args, &nr);
 			ret = verify_log_merge_wait_timeout_ms(sb, ret, nr);
@@ -365,12 +438,14 @@ int scoutfs_options_show(struct seq_file *seq, struct dentry *root)
 		seq_puts(seq, ",acl");
 	seq_printf(seq, ",data_prealloc_blocks=%llu", opts.data_prealloc_blocks);
 	seq_printf(seq, ",data_prealloc_contig_only=%u", opts.data_prealloc_contig_only);
+	seq_printf(seq, ",ino_alloc_per_lock=%u", opts.ino_alloc_per_lock);
 	seq_printf(seq, ",metadev_path=%s", opts.metadev_path);
 	if (!is_acl)
 		seq_puts(seq, ",noacl");
 	seq_printf(seq, ",orphan_scan_delay_ms=%u", opts.orphan_scan_delay_ms);
 	if (opts.quorum_slot_nr >= 0)
 		seq_printf(seq, ",quorum_slot_nr=%d", opts.quorum_slot_nr);
+	seq_printf(seq, ",tcp_keepalive_timeout_ms=%d", opts.tcp_keepalive_timeout_ms);
 
 	return 0;
 }
@@ -452,6 +527,82 @@ static ssize_t data_prealloc_contig_only_store(struct kobject *kobj, struct kobj
 }
 SCOUTFS_ATTR_RW(data_prealloc_contig_only);
 
+static ssize_t ino_alloc_per_lock_show(struct kobject *kobj, struct kobj_attribute *attr,
+					 char *buf)
+{
+	struct super_block *sb = SCOUTFS_SYSFS_ATTRS_SB(kobj);
+	struct scoutfs_mount_options opts;
+
+	scoutfs_options_read(sb, &opts);
+
+	return snprintf(buf, PAGE_SIZE, "%u", opts.ino_alloc_per_lock);
+}
+static ssize_t ino_alloc_per_lock_store(struct kobject *kobj, struct kobj_attribute *attr,
+					  const char *buf, size_t count)
+{
+	struct super_block *sb = SCOUTFS_SYSFS_ATTRS_SB(kobj);
+	DECLARE_OPTIONS_INFO(sb, optinf);
+	char nullterm[20]; /* more than enough for octal -U32_MAX */
+	long val;
+	int len;
+	int ret;
+
+	len = min(count, sizeof(nullterm) - 1);
+	memcpy(nullterm, buf, len);
+	nullterm[len] = '\0';
+
+	ret = kstrtol(nullterm, 0, &val);
+	if (ret < 0 || val < 1 || val > SCOUTFS_LOCK_INODE_GROUP_NR) {
+		scoutfs_err(sb, "invalid ino_alloc_per_lock option, must be between 1 and %u",
+			    SCOUTFS_LOCK_INODE_GROUP_NR);
+		return -EINVAL;
+	}
+
+	write_seqlock(&optinf->seqlock);
+	optinf->opts.ino_alloc_per_lock = val;
+	write_sequnlock(&optinf->seqlock);
+
+	return count;
+}
+SCOUTFS_ATTR_RW(ino_alloc_per_lock);
+
+static ssize_t lock_idle_count_show(struct kobject *kobj, struct kobj_attribute *attr,
+						char *buf)
+{
+	struct super_block *sb = SCOUTFS_SYSFS_ATTRS_SB(kobj);
+	struct scoutfs_mount_options opts;
+
+	scoutfs_options_read(sb, &opts);
+
+	return snprintf(buf, PAGE_SIZE, "%u", opts.lock_idle_count);
+}
+static ssize_t lock_idle_count_store(struct kobject *kobj, struct kobj_attribute *attr,
+						 const char *buf, size_t count)
+{
+	struct super_block *sb = SCOUTFS_SYSFS_ATTRS_SB(kobj);
+	DECLARE_OPTIONS_INFO(sb, optinf);
+	char nullterm[30]; /* more than enough for octal -U64_MAX */
+	int val;
+	int len;
+	int ret;
+
+	len = min(count, sizeof(nullterm) - 1);
+	memcpy(nullterm, buf, len);
+	nullterm[len] = '\0';
+
+	ret = kstrtoint(nullterm, 0, &val);
+	ret = verify_lock_idle_count(sb, ret, val);
+	if (ret == 0) {
+		write_seqlock(&optinf->seqlock);
+		optinf->opts.lock_idle_count = val;
+		write_sequnlock(&optinf->seqlock);
+		ret = count;
+	}
+
+	return ret;
+}
+SCOUTFS_ATTR_RW(lock_idle_count);
+
 static ssize_t log_merge_wait_timeout_ms_show(struct kobject *kobj, struct kobj_attribute *attr,
 						char *buf)
 {
@@ -592,6 +743,8 @@ SCOUTFS_ATTR_RO(quorum_slot_nr);
 static struct attribute *options_attrs[] = {
 	SCOUTFS_ATTR_PTR(data_prealloc_blocks),
 	SCOUTFS_ATTR_PTR(data_prealloc_contig_only),
+	SCOUTFS_ATTR_PTR(ino_alloc_per_lock),
+	SCOUTFS_ATTR_PTR(lock_idle_count),
 	SCOUTFS_ATTR_PTR(log_merge_wait_timeout_ms),
 	SCOUTFS_ATTR_PTR(metadev_path),
 	SCOUTFS_ATTR_PTR(orphan_scan_delay_ms),

kmod/src/options.h

@@ -8,13 +8,18 @@
 struct scoutfs_mount_options {
 	u64 data_prealloc_blocks;
 	bool data_prealloc_contig_only;
+	unsigned int ino_alloc_per_lock;
+	int lock_idle_count;
 	unsigned int log_merge_wait_timeout_ms;
 	char *metadev_path;
 	unsigned int orphan_scan_delay_ms;
 	int quorum_slot_nr;
 	u64 quorum_heartbeat_timeout_ms;
+	int tcp_keepalive_timeout_ms;
 };
 
+#define UNRESPONSIVE_PROBES	3
+
 void scoutfs_options_read(struct super_block *sb, struct scoutfs_mount_options *opts);
 int scoutfs_options_show(struct seq_file *seq, struct dentry *root);
 

kmod/src/quorum.c

@@ -145,14 +145,26 @@ struct quorum_info {
 #define DECLARE_QUORUM_INFO_KOBJ(kobj, name) \
 	DECLARE_QUORUM_INFO(SCOUTFS_SYSFS_ATTRS_SB(kobj), name)
 
-static bool quorum_slot_present(struct scoutfs_quorum_config *qconf, int i)
+static bool quorum_slot_ipv4(struct scoutfs_quorum_config *qconf, int i)
 {
 	BUG_ON(i < 0 || i > SCOUTFS_QUORUM_MAX_SLOTS);
 
 	return qconf->slots[i].addr.v4.family == cpu_to_le16(SCOUTFS_AF_IPV4);
 }
 
-static void quorum_slot_sin(struct scoutfs_quorum_config *qconf, int i, struct sockaddr_in *sin)
+static bool quorum_slot_ipv6(struct scoutfs_quorum_config *qconf, int i)
+{
+	BUG_ON(i < 0 || i > SCOUTFS_QUORUM_MAX_SLOTS);
+
+	return qconf->slots[i].addr.v6.family == cpu_to_le16(SCOUTFS_AF_IPV6);
+}
+
+static bool quorum_slot_present(struct scoutfs_quorum_config *qconf, int i)
+{
+	return quorum_slot_ipv4(qconf, i) || quorum_slot_ipv6(qconf, i);
+}
+
+static void quorum_slot_sin(struct scoutfs_quorum_config *qconf, int i, struct sockaddr_storage *sin)
 {
 	BUG_ON(i < 0 || i >= SCOUTFS_QUORUM_MAX_SLOTS);
 
@@ -179,11 +191,18 @@ static int create_socket(struct super_block *sb)
 {
 	DECLARE_QUORUM_INFO(sb, qinf);
 	struct socket *sock = NULL;
-	struct sockaddr_in sin;
+	struct sockaddr_storage sin;
+	struct scoutfs_quorum_slot slot = qinf->qconf.slots[qinf->our_quorum_slot_nr];
 	int addrlen;
 	int ret;
 
-	ret = kc_sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
+	if (le16_to_cpu(slot.addr.v4.family) == SCOUTFS_AF_IPV4)
+		ret = kc_sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
+	else if (le16_to_cpu(slot.addr.v6.family) == SCOUTFS_AF_IPV6)
+		ret = kc_sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
+	else
+		BUG();
+
 	if (ret) {
 		scoutfs_err(sb, "quorum couldn't create udp socket: %d", ret);
 		goto out;
@@ -192,9 +211,9 @@ static int create_socket(struct super_block *sb)
 	/* rather fail and retry than block waiting for free */
 	sock->sk->sk_allocation = GFP_ATOMIC;
 
+	addrlen = (le16_to_cpu(slot.addr.v4.family) == SCOUTFS_AF_IPV4) ?
+		  sizeof(struct sockaddr_in) : sizeof(struct sockaddr_in6);
 	quorum_slot_sin(&qinf->qconf, qinf->our_quorum_slot_nr, &sin);
-
-	addrlen = sizeof(sin);
 	ret = kernel_bind(sock, (struct sockaddr *)&sin, addrlen);
 	if (ret) {
 		scoutfs_err(sb, "quorum failed to bind udp socket to "SIN_FMT": %d",
@@ -241,12 +260,8 @@ static int send_msg_members(struct super_block *sb, int type, u64 term, int only
 		.iov_base = &qmes,
 		.iov_len = sizeof(qmes),
 	};
-	struct sockaddr_in sin;
+	struct sockaddr_storage sin;
 	struct msghdr mh = {
-#ifndef KC_MSGHDR_STRUCT_IOV_ITER
-		.msg_iov = (struct iovec *)&kv,
-		.msg_iovlen = 1,
-#endif
 		.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL,
 		.msg_name = &sin,
 		.msg_namelen = sizeof(sin),
@@ -268,9 +283,7 @@ static int send_msg_members(struct super_block *sb, int type, u64 term, int only
 
 		scoutfs_quorum_slot_sin(&qinf->qconf, i, &sin);
 		now = ktime_get();
-#ifdef KC_MSGHDR_STRUCT_IOV_ITER
-		iov_iter_init(&mh.msg_iter, WRITE, (struct iovec *)&kv, sizeof(qmes), 1);
-#endif
+
 		ret = kernel_sendmsg(qinf->sock, &mh, &kv, 1, kv.iov_len);
 		if (ret != kv.iov_len)
 			failed++;
@@ -312,10 +325,6 @@ static int recv_msg(struct super_block *sb, struct quorum_host_msg *msg,
 		.iov_len = sizeof(struct scoutfs_quorum_message),
 	};
 	struct msghdr mh = {
-#ifndef KC_MSGHDR_STRUCT_IOV_ITER
-		.msg_iov = (struct iovec *)&kv,
-		.msg_iovlen = 1,
-#endif
 		.msg_flags = MSG_NOSIGNAL,
 	};
 
@@ -333,9 +342,6 @@ static int recv_msg(struct super_block *sb, struct quorum_host_msg *msg,
 		ret = kc_tcp_sock_set_rcvtimeo(qinf->sock, rel_to);
 	}
 
-#ifdef KC_MSGHDR_STRUCT_IOV_ITER
-	iov_iter_init(&mh.msg_iter, READ, (struct iovec *)&kv, sizeof(struct scoutfs_quorum_message), 1);
-#endif
 	ret = kernel_recvmsg(qinf->sock, &mh, &kv, 1, kv.iov_len, mh.msg_flags);
 	if (ret < 0)
 		return ret;
@@ -520,10 +526,10 @@ static int update_quorum_block(struct super_block *sb, int event, u64 term, bool
 		set_quorum_block_event(sb, &blk, event, term);
 		ret = write_quorum_block(sb, blkno, &blk);
 		if (ret < 0)
-			scoutfs_err(sb, "error %d reading quorum block %llu to update event %d term %llu",
+			scoutfs_err(sb, "error %d writing quorum block %llu after updating event %d term %llu",
 				    ret, blkno, event, term);
 	} else {
-		scoutfs_err(sb, "error %d writing quorum block %llu after updating event %d term %llu",
+		scoutfs_err(sb, "error %d reading quorum block %llu to update event %d term %llu",
 			    ret, blkno, event, term);
 	}
 
@@ -555,10 +561,11 @@ int scoutfs_quorum_fence_leaders(struct super_block *sb, struct scoutfs_quorum_c
 				 u64 term)
 {
 #define NR_OLD 2
-	struct scoutfs_quorum_block_event old[SCOUTFS_QUORUM_MAX_SLOTS][NR_OLD] = {{{0,}}};
+	struct scoutfs_quorum_block_event (*old)[NR_OLD];
 	struct scoutfs_sb_info *sbi = SCOUTFS_SB(sb);
 	struct scoutfs_quorum_block blk;
-	struct sockaddr_in sin;
+	struct sockaddr_storage sin;
+	union scoutfs_inet_addr addr;
 	const __le64 lefsid = cpu_to_le64(sbi->fsid);
 	const u64 rid = sbi->rid;
 	bool fence_started = false;
@@ -571,13 +578,20 @@ int scoutfs_quorum_fence_leaders(struct super_block *sb, struct scoutfs_quorum_c
 
 	BUILD_BUG_ON(SCOUTFS_QUORUM_BLOCKS < SCOUTFS_QUORUM_MAX_SLOTS);
 
+	old = kmalloc(NR_OLD * SCOUTFS_QUORUM_MAX_SLOTS * sizeof(struct scoutfs_quorum_block_event), GFP_KERNEL);
+	if (!old) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	memset(old, 0, NR_OLD * SCOUTFS_QUORUM_MAX_SLOTS * sizeof(struct scoutfs_quorum_block_event));
+
 	for (i = 0; i < SCOUTFS_QUORUM_MAX_SLOTS; i++) {
 		if (!quorum_slot_present(qconf, i))
 			continue;
 
 		ret = read_quorum_block(sb, SCOUTFS_QUORUM_BLKNO + i, &blk, false);
 		if (ret < 0)
-			goto out;
+			goto out_free;
 
 		/* elected leader still running */
 		if (le64_to_cpu(blk.events[SCOUTFS_QUORUM_EVENT_ELECT].term) >
@@ -611,14 +625,17 @@ int scoutfs_quorum_fence_leaders(struct super_block *sb, struct scoutfs_quorum_c
 			scoutfs_info(sb, "fencing previous leader "SCSBF" at term %llu in slot %u with address "SIN_FMT,
 				     SCSB_LEFR_ARGS(lefsid, fence_rid),
 				     le64_to_cpu(old[i][j].term), i, SIN_ARG(&sin));
-			ret = scoutfs_fence_start(sb, le64_to_cpu(fence_rid), sin.sin_addr.s_addr,
+			scoutfs_sin_to_addr(&addr, &sin);
+			ret = scoutfs_fence_start(sb, le64_to_cpu(fence_rid), &addr,
 						  SCOUTFS_FENCE_QUORUM_BLOCK_LEADER);
 			if (ret < 0)
-				goto out;
+				goto out_free;
 			fence_started = true;
 		}
 	}
 
+out_free:
+	kfree(old);
 out:
 	err = scoutfs_fence_wait_fenced(sb, msecs_to_jiffies(SCOUTFS_QUORUM_FENCE_TO_MS));
 	if (ret == 0)
@@ -721,7 +738,7 @@ static void scoutfs_quorum_worker(struct work_struct *work)
 	struct quorum_info *qinf = container_of(work, struct quorum_info, work);
 	struct scoutfs_mount_options opts;
 	struct super_block *sb = qinf->sb;
-	struct sockaddr_in unused;
+	struct sockaddr_storage unused;
 	struct quorum_host_msg msg;
 	struct quorum_status qst = {0,};
 	struct hb_recording hbr;
@@ -822,6 +839,7 @@ static void scoutfs_quorum_worker(struct work_struct *work)
 
 		/* followers and candidates start new election on timeout */
 		if (qst.role != LEADER &&
+		    msg.type == SCOUTFS_QUORUM_MSG_INVALID &&
 		    ktime_after(ktime_get(), qst.timeout)) {
 			/* .. but only if their server has stopped */
 			if (!scoutfs_server_is_down(sb)) {
@@ -982,7 +1000,10 @@ static void scoutfs_quorum_worker(struct work_struct *work)
 	}
 
 	/* record that this slot no longer has an active quorum */
-	update_quorum_block(sb, SCOUTFS_QUORUM_EVENT_END, qst.term, true);
+	err = update_quorum_block(sb, SCOUTFS_QUORUM_EVENT_END, qst.term, true);
+	if (err < 0 && ret == 0)
+		ret = err;
+
 out:
 	if (ret < 0) {
 		scoutfs_err(sb, "quorum service saw error %d, shutting down.  This mount is no longer participating in quorum.  It should be remounted to restore service.",
@@ -999,7 +1020,7 @@ out:
  * leader with the greatest elected term.  If we get it wrong the
  * connection will timeout and the client will try again.
  */
-int scoutfs_quorum_server_sin(struct super_block *sb, struct sockaddr_in *sin)
+int scoutfs_quorum_server_sin(struct super_block *sb, struct sockaddr_storage *sin)
 {
 	struct scoutfs_super_block *super = NULL;
 	struct scoutfs_quorum_block blk;
@@ -1058,7 +1079,7 @@ u8 scoutfs_quorum_votes_needed(struct super_block *sb)
 	return qinf->votes_needed;
 }
 
-void scoutfs_quorum_slot_sin(struct scoutfs_quorum_config *qconf, int i, struct sockaddr_in *sin)
+void scoutfs_quorum_slot_sin(struct scoutfs_quorum_config *qconf, int i, struct sockaddr_storage *sin)
 {
 	return quorum_slot_sin(qconf, i, sin);
 }
@@ -1071,7 +1092,7 @@ static char *role_str(int role)
 		[LEADER] = "leader",
 	};
 
-	if (role < 0 || role > ARRAY_SIZE(roles) || !roles[role])
+	if (role < 0 || role >= ARRAY_SIZE(roles) || !roles[role])
 		return "invalid";
 
 	return roles[role];
@@ -1217,8 +1238,12 @@ static int verify_quorum_slots(struct super_block *sb, struct quorum_info *qinf,
 			       struct scoutfs_quorum_config *qconf)
 {
 	char slots[(SCOUTFS_QUORUM_MAX_SLOTS * 3) + 1];
-	struct sockaddr_in other;
-	struct sockaddr_in sin;
+	struct sockaddr_storage other;
+	struct sockaddr_storage sin;
+	struct sockaddr_in *sin4;
+	struct sockaddr_in *other4;
+	struct sockaddr_in6 *sin6;
+	struct sockaddr_in6 *other6;
 	int found = 0;
 	int ret;
 	int i;
@@ -1229,35 +1254,78 @@ static int verify_quorum_slots(struct super_block *sb, struct quorum_info *qinf,
 		if (!quorum_slot_present(qconf, i))
 			continue;
 
-		scoutfs_quorum_slot_sin(qconf, i, &sin);
+		if (quorum_slot_ipv4(qconf, i)) {
+			scoutfs_quorum_slot_sin(qconf, i, &sin);
+			sin4 = (struct sockaddr_in *)&sin;
 
-		if (!valid_ipv4_unicast(sin.sin_addr.s_addr)) {
-			scoutfs_err(sb, "quorum slot #%d has invalid ipv4 unicast address: "SIN_FMT,
-				    i,  SIN_ARG(&sin));
-			return -EINVAL;
-		}
-
-		if (!valid_ipv4_port(sin.sin_port)) {
-			scoutfs_err(sb, "quorum slot #%d has invalid ipv4 port number:"SIN_FMT,
-				    i,  SIN_ARG(&sin));
-			return -EINVAL;
-		}
-
-		for (j = i + 1; j < SCOUTFS_QUORUM_MAX_SLOTS; j++) {
-			if (!quorum_slot_present(qconf, j))
-				continue;
-
-			scoutfs_quorum_slot_sin(qconf, j, &other);
-
-			if (sin.sin_addr.s_addr == other.sin_addr.s_addr &&
-			    sin.sin_port == other.sin_port) {
-				scoutfs_err(sb, "quorum slots #%u and #%u have the same address: "SIN_FMT,
-					    i, j, SIN_ARG(&sin));
+			if (!valid_ipv4_unicast(sin4->sin_addr.s_addr)) {
+				scoutfs_err(sb, "quorum slot #%d has invalid ipv4 unicast address: "SIN_FMT,
+					    i, SIN_ARG(&sin));
 				return -EINVAL;
 			}
-		}
 
-		found++;
+			if (!valid_ipv4_port(sin4->sin_port)) {
+				scoutfs_err(sb, "quorum slot #%d has invalid ipv4 port number:"SIN_FMT,
+					    i, SIN_ARG(&sin));
+				return -EINVAL;
+			}
+
+			for (j = i + 1; j < SCOUTFS_QUORUM_MAX_SLOTS; j++) {
+				if (!quorum_slot_ipv4(qconf, j))
+					continue;
+
+				scoutfs_quorum_slot_sin(qconf, j, &other);
+				other4 = (struct sockaddr_in *)&other;
+
+				if (sin4->sin_addr.s_addr == other4->sin_addr.s_addr &&
+				    sin4->sin_port == other4->sin_port) {
+					scoutfs_err(sb, "quorum slots #%u and #%u have the same address: "SIN_FMT,
+						    i, j, SIN_ARG(&sin));
+					return -EINVAL;
+				}
+			}
+
+			found++;
+		} else if (quorum_slot_ipv6(qconf, i)) {
+			quorum_slot_sin(qconf, i, &sin);
+			sin6 = (struct sockaddr_in6 *)&sin;
+
+			if ((sin6->sin6_addr.in6_u.u6_addr32[0] == 0) && (sin6->sin6_addr.in6_u.u6_addr32[1] == 0) &&
+			    (sin6->sin6_addr.in6_u.u6_addr32[2] == 0) && (sin6->sin6_addr.in6_u.u6_addr32[3] == 0)) {
+				scoutfs_err(sb, "quorum slot #%d has unspecified ipv6 address:"SIN_FMT,
+					    i, SIN_ARG(&sin));
+				return -EINVAL;
+			}
+
+			if (sin6->sin6_addr.in6_u.u6_addr8[0] == 0xff) {
+				scoutfs_err(sb, "quorum slot #%d has multicast ipv6 address:"SIN_FMT,
+					    i, SIN_ARG(&sin));
+				return -EINVAL;
+			}
+
+			if (!valid_ipv4_port(sin6->sin6_port)) {
+				scoutfs_err(sb, "quorum slot #%d has invalid ipv6 port number:"SIN_FMT,
+					    i,  SIN_ARG(&sin));
+				return -EINVAL;
+			}
+
+			for (j = i + 1; j < SCOUTFS_QUORUM_MAX_SLOTS; j++) {
+				if (!quorum_slot_ipv6(qconf, j))
+					continue;
+
+				quorum_slot_sin(qconf, j, &other);
+				other6 = (struct sockaddr_in6 *)&other;
+
+				if ((ipv6_addr_equal(&sin6->sin6_addr, &other6->sin6_addr)) &&
+				    (sin6->sin6_port == other6->sin6_port)) {
+					scoutfs_err(sb, "quorum slots #%u and #%u have the same address: "SIN_FMT,
+						    i, j, SIN_ARG(&sin));
+					return -EINVAL;
+				}
+			}
+
+			found++;
+		}
 	}
 
 	if (found == 0)  {

kmod/src/quorum.h

@@ -1,11 +1,11 @@
 #ifndef _SCOUTFS_QUORUM_H_
 #define _SCOUTFS_QUORUM_H_
 
-int scoutfs_quorum_server_sin(struct super_block *sb, struct sockaddr_in *sin);
+int scoutfs_quorum_server_sin(struct super_block *sb, struct sockaddr_storage *sin);
 
 u8 scoutfs_quorum_votes_needed(struct super_block *sb);
 void scoutfs_quorum_slot_sin(struct scoutfs_quorum_config *qconf, int i,
-			     struct sockaddr_in *sin);
+			     struct sockaddr_storage *sin);
 
 int scoutfs_quorum_fence_leaders(struct super_block *sb, struct scoutfs_quorum_config *qconf,
 				 u64 term);

kmod/src/scoutfs_trace.h

@@ -286,6 +286,52 @@ TRACE_EVENT(scoutfs_data_alloc_block_enter,
 		  STE_ENTRY_ARGS(ext))
 );
 
+TRACE_EVENT(scoutfs_data_page_mkwrite,
+	TP_PROTO(struct super_block *sb, __u64 ino, __u64 pos, __u32 ret),
+
+	TP_ARGS(sb, ino, pos, ret),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+		__field(__u64, pos)
+		__field(__u32, ret)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+		__entry->pos = pos;
+		__entry->ret = ret;
+	),
+
+	TP_printk(SCSBF" ino %llu pos %llu ret %u ",
+		  SCSB_TRACE_ARGS, __entry->ino, __entry->pos, __entry->ret)
+);
+
+TRACE_EVENT(scoutfs_data_filemap_fault,
+	TP_PROTO(struct super_block *sb, __u64 ino, __u64 pos, __u32 ret),
+
+	TP_ARGS(sb, ino, pos, ret),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+		__field(__u64, pos)
+		__field(__u32, ret)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+		__entry->pos = pos;
+		__entry->ret = ret;
+	),
+
+	TP_printk(SCSBF" ino %llu pos %llu ret %u ",
+		  SCSB_TRACE_ARGS, __entry->ino, __entry->pos, __entry->ret)
+);
+
 DECLARE_EVENT_CLASS(scoutfs_data_file_extent_class,
 	TP_PROTO(struct super_block *sb, __u64 ino, struct scoutfs_extent *ext),
 
@@ -743,6 +789,80 @@ TRACE_EVENT(scoutfs_inode_walk_writeback,
 		  __entry->ino, __entry->write, __entry->ret)
 );
 
+TRACE_EVENT(scoutfs_orphan_scan_start,
+	TP_PROTO(struct super_block *sb),
+
+	TP_ARGS(sb),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+	),
+
+	TP_printk(SCSBF, SCSB_TRACE_ARGS)
+);
+
+TRACE_EVENT(scoutfs_orphan_scan_stop,
+	TP_PROTO(struct super_block *sb, bool work_todo),
+
+	TP_ARGS(sb, work_todo),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(bool, work_todo)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->work_todo = work_todo;
+	),
+
+	TP_printk(SCSBF" work_todo %d", SCSB_TRACE_ARGS, __entry->work_todo)
+);
+
+TRACE_EVENT(scoutfs_orphan_scan_work,
+	TP_PROTO(struct super_block *sb, __u64 ino),
+
+	TP_ARGS(sb, ino),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+	),
+
+	TP_printk(SCSBF" ino %llu", SCSB_TRACE_ARGS,
+		  __entry->ino)
+);
+
+TRACE_EVENT(scoutfs_orphan_scan_end,
+	TP_PROTO(struct super_block *sb, __u64 ino, int ret),
+
+	TP_ARGS(sb, ino, ret),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+		__field(int, ret)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+		__entry->ret = ret;
+	),
+
+	TP_printk(SCSBF" ino %llu ret %d", SCSB_TRACE_ARGS,
+		  __entry->ino, __entry->ret)
+);
+
 DECLARE_EVENT_CLASS(scoutfs_lock_info_class,
 	TP_PROTO(struct super_block *sb, struct lock_info *linfo),
 
@@ -777,13 +897,14 @@ DEFINE_EVENT(scoutfs_lock_info_class, scoutfs_lock_destroy,
 );
 
 TRACE_EVENT(scoutfs_xattr_set,
-	TP_PROTO(struct super_block *sb, size_t name_len, const void *value,
-		 size_t size, int flags),
+	TP_PROTO(struct super_block *sb, __u64 ino, size_t name_len,
+		 const void *value, size_t size, int flags),
 
-	TP_ARGS(sb, name_len, value, size, flags),
+	TP_ARGS(sb, ino, name_len, value, size, flags),
 
 	TP_STRUCT__entry(
 		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
 		__field(size_t, name_len)
 		__field(const void *, value)
 		__field(size_t, size)
@@ -792,15 +913,16 @@ TRACE_EVENT(scoutfs_xattr_set,
 
 	TP_fast_assign(
 		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
 		__entry->name_len = name_len;
 		__entry->value = value;
 		__entry->size = size;
 		__entry->flags = flags;
 	),
 
-	TP_printk(SCSBF" name_len %zu value %p size %zu flags 0x%x",
-		  SCSB_TRACE_ARGS, __entry->name_len, __entry->value,
-		  __entry->size, __entry->flags)
+	TP_printk(SCSBF" ino %llu name_len %zu value %p size %zu flags 0x%x",
+		  SCSB_TRACE_ARGS, __entry->ino,  __entry->name_len,
+		  __entry->value, __entry->size, __entry->flags)
 );
 
 TRACE_EVENT(scoutfs_advance_dirty_super,
@@ -988,6 +1110,82 @@ TRACE_EVENT(scoutfs_orphan_inode,
 		  MINOR(__entry->dev), __entry->ino)
 );
 
+DECLARE_EVENT_CLASS(scoutfs_try_delete_class,
+        TP_PROTO(struct super_block *sb, u64 ino),
+        TP_ARGS(sb, ino),
+        TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+        ),
+        TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+        ),
+	TP_printk(SCSBF" ino %llu", SCSB_TRACE_ARGS, __entry->ino)
+);
+
+DEFINE_EVENT(scoutfs_try_delete_class, scoutfs_try_delete,
+        TP_PROTO(struct super_block *sb, u64 ino),
+        TP_ARGS(sb, ino)
+);
+
+DEFINE_EVENT(scoutfs_try_delete_class, scoutfs_try_delete_local_busy,
+        TP_PROTO(struct super_block *sb, u64 ino),
+        TP_ARGS(sb, ino)
+);
+
+DEFINE_EVENT(scoutfs_try_delete_class, scoutfs_try_delete_cached,
+        TP_PROTO(struct super_block *sb, u64 ino),
+        TP_ARGS(sb, ino)
+);
+
+DEFINE_EVENT(scoutfs_try_delete_class, scoutfs_try_delete_no_item,
+        TP_PROTO(struct super_block *sb, u64 ino),
+        TP_ARGS(sb, ino)
+);
+
+TRACE_EVENT(scoutfs_try_delete_has_links,
+	TP_PROTO(struct super_block *sb, u64 ino, unsigned int nlink),
+
+	TP_ARGS(sb, ino, nlink),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+		__field(unsigned int, nlink)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+		__entry->nlink = nlink;
+	),
+
+	TP_printk(SCSBF" ino %llu nlink %u", SCSB_TRACE_ARGS, __entry->ino,
+		  __entry->nlink)
+);
+
+TRACE_EVENT(scoutfs_inode_orphan_delete,
+	TP_PROTO(struct super_block *sb, u64 ino, int ret),
+
+	TP_ARGS(sb, ino, ret),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, ino)
+		__field(int, ret)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+		__entry->ret = ret;
+	),
+
+	TP_printk(SCSBF" ino %llu ret %d", SCSB_TRACE_ARGS, __entry->ino,
+		__entry->ret)
+);
+
 TRACE_EVENT(scoutfs_delete_inode,
 	TP_PROTO(struct super_block *sb, u64 ino, umode_t mode, u64 size),
 
@@ -1012,6 +1210,32 @@ TRACE_EVENT(scoutfs_delete_inode,
 		  __entry->mode, __entry->size)
 );
 
+TRACE_EVENT(scoutfs_delete_inode_end,
+	TP_PROTO(struct super_block *sb, u64 ino, umode_t mode, u64 size, int ret),
+
+	TP_ARGS(sb, ino, mode, size, ret),
+
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(__u64, ino)
+		__field(umode_t, mode)
+		__field(__u64, size)
+		__field(int, ret)
+	),
+
+	TP_fast_assign(
+		__entry->dev = sb->s_dev;
+		__entry->ino = ino;
+		__entry->mode = mode;
+		__entry->size = size;
+		__entry->ret = ret;
+	),
+
+	TP_printk("dev %d,%d ino %llu, mode 0x%x size %llu, ret %d",
+		  MAJOR(__entry->dev), MINOR(__entry->dev), __entry->ino,
+		  __entry->mode, __entry->size, __entry->ret)
+);
+
 DECLARE_EVENT_CLASS(scoutfs_key_class,
         TP_PROTO(struct super_block *sb, struct scoutfs_key *key),
         TP_ARGS(sb, key),
@@ -1046,9 +1270,12 @@ DECLARE_EVENT_CLASS(scoutfs_lock_class,
 		sk_trace_define(start)
 		sk_trace_define(end)
 		__field(u64, refresh_gen)
+		__field(u64, write_seq)
+		__field(u64, dirty_trans_seq)
 		__field(unsigned char, request_pending)
 		__field(unsigned char, invalidate_pending)
 		__field(int, mode)
+		__field(int, invalidating_mode)
 		__field(unsigned int, waiters_cw)
 		__field(unsigned int, waiters_pr)
 		__field(unsigned int, waiters_ex)
@@ -1061,9 +1288,12 @@ DECLARE_EVENT_CLASS(scoutfs_lock_class,
 		sk_trace_assign(start, &lck->start);
 		sk_trace_assign(end, &lck->end);
 		__entry->refresh_gen = lck->refresh_gen;
+		__entry->write_seq = lck->write_seq;
+		__entry->dirty_trans_seq = lck->dirty_trans_seq;
 		__entry->request_pending = lck->request_pending;
 		__entry->invalidate_pending = lck->invalidate_pending;
 		__entry->mode = lck->mode;
+		__entry->invalidating_mode = lck->invalidating_mode;
 		__entry->waiters_pr = lck->waiters[SCOUTFS_LOCK_READ];
 		__entry->waiters_ex = lck->waiters[SCOUTFS_LOCK_WRITE];
 		__entry->waiters_cw = lck->waiters[SCOUTFS_LOCK_WRITE_ONLY];
@@ -1071,10 +1301,11 @@ DECLARE_EVENT_CLASS(scoutfs_lock_class,
 		__entry->users_ex = lck->users[SCOUTFS_LOCK_WRITE];
 		__entry->users_cw = lck->users[SCOUTFS_LOCK_WRITE_ONLY];
         ),
-        TP_printk(SCSBF" start "SK_FMT" end "SK_FMT" mode %u reqpnd %u invpnd %u rfrgen %llu waiters: pr %u ex %u cw %u users: pr %u ex %u cw %u",
+        TP_printk(SCSBF" start "SK_FMT" end "SK_FMT" mode %u invmd %u reqp %u invp %u refg %llu wris %llu dts %llu waiters: pr %u ex %u cw %u users: pr %u ex %u cw %u",
 		  SCSB_TRACE_ARGS, sk_trace_args(start), sk_trace_args(end),
-		  __entry->mode, __entry->request_pending,
-		  __entry->invalidate_pending, __entry->refresh_gen,
+		  __entry->mode, __entry->invalidating_mode, __entry->request_pending,
+		  __entry->invalidate_pending, __entry->refresh_gen, __entry->write_seq,
+		  __entry->dirty_trans_seq,
 		  __entry->waiters_pr, __entry->waiters_ex, __entry->waiters_cw,
 		  __entry->users_pr, __entry->users_ex, __entry->users_cw)
 );
@@ -1124,35 +1355,37 @@ DEFINE_EVENT(scoutfs_lock_class, scoutfs_lock_shrink,
 );
 
 DECLARE_EVENT_CLASS(scoutfs_net_class,
-        TP_PROTO(struct super_block *sb, struct sockaddr_in *name,
-		 struct sockaddr_in *peer, struct scoutfs_net_header *nh),
+        TP_PROTO(struct super_block *sb, struct sockaddr_storage *name,
+		 struct sockaddr_storage *peer, struct scoutfs_net_header *nh),
         TP_ARGS(sb, name, peer, nh),
         TP_STRUCT__entry(
 		SCSB_TRACE_FIELDS
-		si4_trace_define(name)
-		si4_trace_define(peer)
+		__field_struct(struct sockaddr_storage, name)
+		__field_struct(struct sockaddr_storage, peer)
 		snh_trace_define(nh)
         ),
         TP_fast_assign(
 		SCSB_TRACE_ASSIGN(sb);
-		si4_trace_assign(name, name);
-		si4_trace_assign(peer, peer);
+		memcpy(&__entry->name, name, sizeof(struct sockaddr_storage));
+		memcpy(&__entry->peer, peer, sizeof(struct sockaddr_storage));
 		snh_trace_assign(nh, nh);
         ),
-        TP_printk(SCSBF" name "SI4_FMT" peer "SI4_FMT" nh "SNH_FMT,
-		  SCSB_TRACE_ARGS, si4_trace_args(name), si4_trace_args(peer),
+        TP_printk(SCSBF" name "SIN_FMT" peer "SIN_FMT" nh "SNH_FMT,
+		  SCSB_TRACE_ARGS,
+		  &__entry->name,
+		  &__entry->peer,
 		  snh_trace_args(nh))
 );
 
 DEFINE_EVENT(scoutfs_net_class, scoutfs_net_send_message,
-        TP_PROTO(struct super_block *sb, struct sockaddr_in *name,
-		 struct sockaddr_in *peer, struct scoutfs_net_header *nh),
+        TP_PROTO(struct super_block *sb, struct sockaddr_storage *name,
+		 struct sockaddr_storage *peer, struct scoutfs_net_header *nh),
         TP_ARGS(sb, name, peer, nh)
 );
 
 DEFINE_EVENT(scoutfs_net_class, scoutfs_net_recv_message,
-        TP_PROTO(struct super_block *sb, struct sockaddr_in *name,
-		 struct sockaddr_in *peer, struct scoutfs_net_header *nh),
+        TP_PROTO(struct super_block *sb, struct sockaddr_storage *name,
+		 struct sockaddr_storage *peer, struct scoutfs_net_header *nh),
         TP_ARGS(sb, name, peer, nh)
 );
 
@@ -1185,8 +1418,8 @@ DECLARE_EVENT_CLASS(scoutfs_net_conn_class,
 		__field(void *, sock)
 		__field(__u64, c_rid)
 		__field(__u64, greeting_id)
-		si4_trace_define(sockname)
-		si4_trace_define(peername)
+		__field_struct(struct sockaddr_storage, sockname)
+		__field_struct(struct sockaddr_storage, peername)
 		__field(unsigned char, e_accepted_head)
 		__field(void *, listening_conn)
 		__field(unsigned char, e_accepted_list)
@@ -1204,8 +1437,8 @@ DECLARE_EVENT_CLASS(scoutfs_net_conn_class,
 		__entry->sock = conn->sock;
 		__entry->c_rid = conn->rid;
 		__entry->greeting_id = conn->greeting_id;
-		si4_trace_assign(sockname, &conn->sockname);
-		si4_trace_assign(peername, &conn->peername);
+		memcpy(&__entry->sockname, &conn->sockname, sizeof(struct sockaddr_storage));
+		memcpy(&__entry->peername, &conn->peername, sizeof(struct sockaddr_storage));
 		__entry->e_accepted_head = !!list_empty(&conn->accepted_head);
 		__entry->listening_conn = conn->listening_conn;
 		__entry->e_accepted_list = !!list_empty(&conn->accepted_list);
@@ -1215,7 +1448,7 @@ DECLARE_EVENT_CLASS(scoutfs_net_conn_class,
 		__entry->e_resend_queue = !!list_empty(&conn->resend_queue);
 		__entry->recv_seq = atomic64_read(&conn->recv_seq);
         ),
-        TP_printk(SCSBF" flags %s rc_dl %lu cto %lu sk %p rid %llu grid %llu sn "SI4_FMT" pn "SI4_FMT" eah %u lc %p eal %u nss %llu nsi %llu esq %u erq %u rs %llu",
+        TP_printk(SCSBF" flags %s rc_dl %lu cto %lu sk %p rid %llu grid %llu sn "SIN_FMT" pn "SIN_FMT" eah %u lc %p eal %u nss %llu nsi %llu esq %u erq %u rs %llu",
 		  SCSB_TRACE_ARGS,
 		  print_conn_flags(__entry->flags),
 		  __entry->reconn_deadline,
@@ -1223,8 +1456,8 @@ DECLARE_EVENT_CLASS(scoutfs_net_conn_class,
 		  __entry->sock,
 		  __entry->c_rid,
 		  __entry->greeting_id,
-		  si4_trace_args(sockname),
-		  si4_trace_args(peername),
+		  &__entry->sockname,
+		  &__entry->peername,
 		  __entry->e_accepted_head,
 		  __entry->listening_conn,
 		  __entry->e_accepted_list,
@@ -1388,28 +1621,6 @@ DEFINE_EVENT(scoutfs_work_class, scoutfs_data_return_server_extents_exit,
         TP_ARGS(sb, data, ret)
 );
 
-DECLARE_EVENT_CLASS(scoutfs_shrink_exit_class,
-        TP_PROTO(struct super_block *sb, unsigned long nr_to_scan, int ret),
-        TP_ARGS(sb, nr_to_scan, ret),
-        TP_STRUCT__entry(
-		__field(void *, sb)
-		__field(unsigned long, nr_to_scan)
-		__field(int, ret)
-        ),
-        TP_fast_assign(
-		__entry->sb = sb;
-		__entry->nr_to_scan = nr_to_scan;
-		__entry->ret = ret;
-        ),
-        TP_printk("sb %p nr_to_scan %lu ret %d",
-		  __entry->sb, __entry->nr_to_scan, __entry->ret)
-);
-
-DEFINE_EVENT(scoutfs_shrink_exit_class, scoutfs_lock_shrink_exit,
-        TP_PROTO(struct super_block *sb, unsigned long nr_to_scan, int ret),
-        TP_ARGS(sb, nr_to_scan, ret)
-);
-
 TRACE_EVENT(scoutfs_rename,
 	TP_PROTO(struct super_block *sb, struct inode *old_dir,
 		 struct dentry *old_dentry, struct inode *new_dir,
@@ -1913,15 +2124,17 @@ DEFINE_EVENT(scoutfs_server_client_count_class, scoutfs_server_client_down,
 );
 
 DECLARE_EVENT_CLASS(scoutfs_server_commit_users_class,
-        TP_PROTO(struct super_block *sb, int holding, int applying, int nr_holders,
-		 u32 avail_before, u32 freed_before, int committing, int exceeded),
-        TP_ARGS(sb, holding, applying, nr_holders, avail_before, freed_before, committing,
-		exceeded),
+        TP_PROTO(struct super_block *sb, int holding, int applying,
+		 int nr_holders, u32 budget,
+		 u32 avail_before, u32 freed_before,
+		 int committing, int exceeded),
+        TP_ARGS(sb, holding, applying, nr_holders, budget, avail_before, freed_before, committing, exceeded),
         TP_STRUCT__entry(
 		SCSB_TRACE_FIELDS
 		__field(int, holding)
 		__field(int, applying)
 		__field(int, nr_holders)
+		__field(u32, budget)
 		__field(__u32, avail_before)
 		__field(__u32, freed_before)
 		__field(int, committing)
@@ -1932,35 +2145,45 @@ DECLARE_EVENT_CLASS(scoutfs_server_commit_users_class,
 		__entry->holding = !!holding;
 		__entry->applying = !!applying;
 		__entry->nr_holders = nr_holders;
+		__entry->budget = budget;
 		__entry->avail_before = avail_before;
 		__entry->freed_before = freed_before;
 		__entry->committing = !!committing;
 		__entry->exceeded = !!exceeded;
         ),
-	TP_printk(SCSBF" holding %u applying %u nr %u avail_before %u freed_before %u committing %u exceeded %u",
-		  SCSB_TRACE_ARGS, __entry->holding, __entry->applying, __entry->nr_holders,
-		  __entry->avail_before, __entry->freed_before, __entry->committing,
-		  __entry->exceeded)
+	TP_printk(SCSBF" holding %u applying %u nr %u budget %u avail_before %u freed_before %u committing %u exceeded %u",
+		  SCSB_TRACE_ARGS, __entry->holding, __entry->applying,
+		  __entry->nr_holders, __entry->budget,
+		  __entry->avail_before, __entry->freed_before,
+		  __entry->committing, __entry->exceeded)
 );
 DEFINE_EVENT(scoutfs_server_commit_users_class, scoutfs_server_commit_hold,
-        TP_PROTO(struct super_block *sb, int holding, int applying, int nr_holders,
-		 u32 avail_before, u32 freed_before, int committing, int exceeded),
-        TP_ARGS(sb, holding, applying, nr_holders, avail_before, freed_before, committing, exceeded)
+        TP_PROTO(struct super_block *sb, int holding, int applying,
+		 int nr_holders, u32 budget,
+		 u32 avail_before, u32 freed_before,
+		 int committing, int exceeded),
+        TP_ARGS(sb, holding, applying, nr_holders, budget, avail_before, freed_before, committing, exceeded)
 );
 DEFINE_EVENT(scoutfs_server_commit_users_class, scoutfs_server_commit_apply,
-        TP_PROTO(struct super_block *sb, int holding, int applying, int nr_holders,
-		 u32 avail_before, u32 freed_before, int committing, int exceeded),
-        TP_ARGS(sb, holding, applying, nr_holders, avail_before, freed_before, committing, exceeded)
+        TP_PROTO(struct super_block *sb, int holding, int applying,
+		 int nr_holders, u32 budget,
+		 u32 avail_before, u32 freed_before,
+		 int committing, int exceeded),
+        TP_ARGS(sb, holding, applying, nr_holders, budget, avail_before, freed_before, committing, exceeded)
 );
 DEFINE_EVENT(scoutfs_server_commit_users_class, scoutfs_server_commit_start,
-        TP_PROTO(struct super_block *sb, int holding, int applying, int nr_holders,
-		 u32 avail_before, u32 freed_before, int committing, int exceeded),
-        TP_ARGS(sb, holding, applying, nr_holders, avail_before, freed_before, committing, exceeded)
+        TP_PROTO(struct super_block *sb, int holding, int applying,
+		 int nr_holders, u32 budget,
+		 u32 avail_before, u32 freed_before,
+		 int committing, int exceeded),
+        TP_ARGS(sb, holding, applying, nr_holders, budget, avail_before, freed_before, committing, exceeded)
 );
 DEFINE_EVENT(scoutfs_server_commit_users_class, scoutfs_server_commit_end,
-        TP_PROTO(struct super_block *sb, int holding, int applying, int nr_holders,
-		 u32 avail_before, u32 freed_before, int committing, int exceeded),
-        TP_ARGS(sb, holding, applying, nr_holders, avail_before, freed_before, committing, exceeded)
+        TP_PROTO(struct super_block *sb, int holding, int applying,
+		 int nr_holders, u32 budget,
+		 u32 avail_before, u32 freed_before,
+		 int committing, int exceeded),
+        TP_ARGS(sb, holding, applying, nr_holders, budget, avail_before, freed_before, committing, exceeded)
 );
 
 #define slt_symbolic(mode)						\
@@ -2398,6 +2621,27 @@ TRACE_EVENT(scoutfs_block_dirty_ref,
 		  __entry->block_blkno, __entry->block_seq)
 );
 
+TRACE_EVENT(scoutfs_get_file_block,
+	TP_PROTO(struct super_block *sb, u64 blkno, int flags),
+
+	TP_ARGS(sb, blkno, flags),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(__u64, blkno)
+		__field(int, flags)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->blkno = blkno;
+		__entry->flags = flags;
+	),
+
+	TP_printk(SCSBF" blkno %llu flags 0x%x",
+		  SCSB_TRACE_ARGS, __entry->blkno, __entry->flags)
+);
+
 TRACE_EVENT(scoutfs_block_stale,
 	TP_PROTO(struct super_block *sb, struct scoutfs_block_ref *ref,
 		 struct scoutfs_block_header *hdr, u32 magic, u32 crc),
@@ -2438,8 +2682,8 @@ TRACE_EVENT(scoutfs_block_stale,
 
 DECLARE_EVENT_CLASS(scoutfs_block_class,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno, int refcount, int io_count,
-		 unsigned long bits, __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed),
+		 unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits),
 	TP_STRUCT__entry(
 		SCSB_TRACE_FIELDS
 		__field(void *, bp)
@@ -2447,7 +2691,6 @@ DECLARE_EVENT_CLASS(scoutfs_block_class,
 		__field(int, refcount)
 		__field(int, io_count)
 		__field(long, bits)
-		__field(__u64, accessed)
 	),
 	TP_fast_assign(
 		SCSB_TRACE_ASSIGN(sb);
@@ -2456,71 +2699,65 @@ DECLARE_EVENT_CLASS(scoutfs_block_class,
 		__entry->refcount = refcount;
 		__entry->io_count = io_count;
 		__entry->bits = bits;
-		__entry->accessed = accessed;
 	),
-	TP_printk(SCSBF" bp %p blkno %llu refcount %d io_count %d bits 0x%lx accessed %llu",
+	TP_printk(SCSBF" bp %p blkno %llu refcount %x io_count %d bits 0x%lx",
 		  SCSB_TRACE_ARGS, __entry->bp, __entry->blkno, __entry->refcount,
-		  __entry->io_count, __entry->bits, __entry->accessed)
+		  __entry->io_count, __entry->bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_allocate,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_free,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_insert,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_remove,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_end_io,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_submit,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_invalidate,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_mark_dirty,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_forget,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 DEFINE_EVENT(scoutfs_block_class, scoutfs_block_shrink,
 	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
-		 int refcount, int io_count, unsigned long bits,
-		 __u64 accessed),
-	TP_ARGS(sb, bp, blkno, refcount, io_count, bits, accessed)
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
+);
+DEFINE_EVENT(scoutfs_block_class, scoutfs_block_isolate,
+	TP_PROTO(struct super_block *sb, void *bp, u64 blkno,
+		 int refcount, int io_count, unsigned long bits),
+	TP_ARGS(sb, bp, blkno, refcount, io_count, bits)
 );
 
 DECLARE_EVENT_CLASS(scoutfs_ext_next_class,
@@ -2995,6 +3232,45 @@ DEFINE_EVENT(scoutfs_srch_compact_class, scoutfs_srch_compact_client_recv,
 	TP_ARGS(sb, sc)
 );
 
+TRACE_EVENT(scoutfs_ioc_search_xattrs,
+	TP_PROTO(struct super_block *sb, u64 ino, u64 last_ino),
+
+	TP_ARGS(sb, ino, last_ino),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(u64, ino)
+		__field(u64, last_ino)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->ino = ino;
+		__entry->last_ino = last_ino;
+	),
+
+	TP_printk(SCSBF" ino %llu last_ino %llu", SCSB_TRACE_ARGS,
+		  __entry->ino, __entry->last_ino)
+);
+
+TRACE_EVENT(scoutfs_trigger_fired,
+	TP_PROTO(struct super_block *sb, const char *name),
+
+	TP_ARGS(sb, name),
+
+	TP_STRUCT__entry(
+		SCSB_TRACE_FIELDS
+		__field(const char *, name)
+	),
+
+	TP_fast_assign(
+		SCSB_TRACE_ASSIGN(sb);
+		__entry->name = name;
+	),
+
+	TP_printk(SCSBF" %s", SCSB_TRACE_ARGS, __entry->name)
+);
+
 #endif /* _TRACE_SCOUTFS_H */
 
 /* This part must be outside protection */

kmod/src/server.c

@@ -41,6 +41,7 @@
 #include "recov.h"
 #include "omap.h"
 #include "fence.h"
+#include "triggers.h"
 
 /*
  * Every active mount can act as the server that listens on a net
@@ -65,6 +66,7 @@ struct commit_users {
 	struct list_head holding;
 	struct list_head applying;
 	unsigned int nr_holders;
+	u32 budget;
 	u32 avail_before;
 	u32 freed_before;
 	bool committing;
@@ -84,8 +86,9 @@ static void init_commit_users(struct commit_users *cusers)
 do {												\
 	__typeof__(cusers) _cusers = (cusers);							\
 	trace_scoutfs_server_commit_##which(sb, !list_empty(&_cusers->holding),			\
-		!list_empty(&_cusers->applying), _cusers->nr_holders, _cusers->avail_before,	\
-		_cusers->freed_before, _cusers->committing, _cusers->exceeded);			\
+		!list_empty(&_cusers->applying), _cusers->nr_holders, _cusers->budget,		\
+		_cusers->avail_before, _cusers->freed_before, _cusers->committing,		\
+		_cusers->exceeded);								\
 } while (0)
 
 struct server_info {
@@ -303,7 +306,6 @@ static void check_holder_budget(struct super_block *sb, struct server_info *serv
 	u32 freed_used;
 	u32 avail_now;
 	u32 freed_now;
-	u32 budget;
 
 	assert_spin_locked(&cusers->lock);
 
@@ -318,15 +320,14 @@ static void check_holder_budget(struct super_block *sb, struct server_info *serv
 	else
 		freed_used = SCOUTFS_ALLOC_LIST_MAX_BLOCKS - freed_now;
 
-	budget = cusers->nr_holders * COMMIT_HOLD_ALLOC_BUDGET;
-	if (avail_used <= budget && freed_used <= budget)
+	if (avail_used <= cusers->budget && freed_used <= cusers->budget)
 		return;
 
 	exceeded_once = true;
 	cusers->exceeded = cusers->nr_holders;
 
-	scoutfs_err(sb, "%u holders exceeded alloc budget av: bef %u now %u, fr: bef %u now %u",
-		    cusers->nr_holders, cusers->avail_before, avail_now,
+	scoutfs_err(sb, "holders exceeded alloc budget %u av: bef %u now %u, fr: bef %u now %u",
+		    cusers->budget, cusers->avail_before, avail_now,
 		    cusers->freed_before, freed_now);
 
 	list_for_each_entry(hold, &cusers->holding, entry) {
@@ -349,7 +350,7 @@ static bool hold_commit(struct super_block *sb, struct server_info *server,
 {
 	bool has_room;
 	bool held;
-	u32 budget;
+	u32 new_budget;
 	u32 av;
 	u32 fr;
 
@@ -367,8 +368,8 @@ static bool hold_commit(struct super_block *sb, struct server_info *server,
 	}
 
 	/* +2 for our additional hold and then for the final commit work the server does */
-	budget = (cusers->nr_holders + 2) * COMMIT_HOLD_ALLOC_BUDGET;
-	has_room = av >= budget && fr >= budget;
+	new_budget = max(cusers->budget, (cusers->nr_holders + 2) * COMMIT_HOLD_ALLOC_BUDGET);
+	has_room = av >= new_budget && fr >= new_budget;
 	/* checking applying so holders drain once an apply caller starts waiting */
 	held = !cusers->committing && has_room && list_empty(&cusers->applying);
 
@@ -388,6 +389,7 @@ static bool hold_commit(struct super_block *sb, struct server_info *server,
 		list_add_tail(&hold->entry, &cusers->holding);
 
 		cusers->nr_holders++;
+		cusers->budget = new_budget;
 
 	} else if (!has_room && cusers->nr_holders == 0 && !cusers->committing) {
 		cusers->committing = true;
@@ -516,6 +518,7 @@ static void commit_end(struct super_block *sb, struct commit_users *cusers, int
 	list_for_each_entry_safe(hold, tmp, &cusers->applying, entry)
 		list_del_init(&hold->entry);
 	cusers->committing = false;
+	cusers->budget = 0;
 	spin_unlock(&cusers->lock);
 
 	wake_up(&cusers->waitq);
@@ -608,7 +611,7 @@ static void scoutfs_server_commit_func(struct work_struct *work)
 		goto out;
 
 	if (scoutfs_forcing_unmount(sb)) {
-		ret = -EIO;
+		ret = -ENOLINK;
 		goto out;
 	}
 
@@ -992,10 +995,11 @@ static int for_each_rid_last_lt(struct super_block *sb, struct scoutfs_btree_roo
 }
 
 /*
- * Log merge range items are stored at the starting fs key of the range.
- * The only fs key field that doesn't hold information is the zone, so
- * we use the zone to differentiate all types that we store in the log
- * merge tree.
+ * Log merge range items are stored at the starting fs key of the range
+ * with the zone overwritten to indicate the log merge item type.  This
+ * day0 mistake loses sorting information for items in the different
+ * zones in the fs root, so the range items aren't strictly sorted by
+ * the starting key of their range.
  */
 static void init_log_merge_key(struct scoutfs_key *key, u8 zone, u64 first,
 			       u64 second)
@@ -1027,6 +1031,51 @@ static int next_log_merge_item_key(struct super_block *sb, struct scoutfs_btree_
 	return ret;
 }
 
+/*
+ * The range items aren't sorted by their range.start because
+ * _RANGE_ZONE clobbers the range's zone.  We sweep all the items and
+ * find the range with the next least starting key that's greater than
+ * the caller's starting key.  We have to be careful to iterate over the
+ * log_merge tree keys because the ranges can overlap as they're mapped
+ * to the log_merge keys by clobbering their zone.
+ */
+static int next_log_merge_range(struct super_block *sb, struct scoutfs_btree_root *root,
+				struct scoutfs_key *start, struct scoutfs_log_merge_range *rng)
+{
+	struct scoutfs_log_merge_range *next;
+	SCOUTFS_BTREE_ITEM_REF(iref);
+	struct scoutfs_key key;
+	int ret;
+
+	key = *start;
+	key.sk_zone = SCOUTFS_LOG_MERGE_RANGE_ZONE;
+	scoutfs_key_set_ones(&rng->start);
+
+	do {
+		ret = scoutfs_btree_next(sb, root, &key, &iref);
+		if (ret == 0) {
+			if (iref.key->sk_zone != SCOUTFS_LOG_MERGE_RANGE_ZONE) {
+				ret = -ENOENT;
+			} else if (iref.val_len != sizeof(struct scoutfs_log_merge_range)) {
+				ret = -EIO;
+			} else {
+				next = iref.val;
+				if (scoutfs_key_compare(&next->start, &rng->start) < 0 &&
+				    scoutfs_key_compare(&next->start, start) >= 0)
+					*rng = *next;
+				key = *iref.key;
+				scoutfs_key_inc(&key);
+			}
+			scoutfs_btree_put_iref(&iref);
+		}
+	} while (ret == 0);
+
+	if (ret == -ENOENT && !scoutfs_key_is_ones(&rng->start))
+		ret = 0;
+
+	return ret;
+}
+
 static int next_log_merge_item(struct super_block *sb,
 			       struct scoutfs_btree_root *root,
 			       u8 zone, u64 first, u64 second,
@@ -1038,6 +1087,101 @@ static int next_log_merge_item(struct super_block *sb,
 	return next_log_merge_item_key(sb, root, zone, &key, val, val_len);
 }
 
+static int do_finalize_ours(struct super_block *sb,
+			    struct scoutfs_log_trees *lt,
+			    struct commit_hold *hold)
+{
+	struct server_info *server = SCOUTFS_SB(sb)->server_info;
+	struct scoutfs_super_block *super = DIRTY_SUPER_SB(sb);
+	struct scoutfs_key key;
+	char *err_str = NULL;
+	u64 rid = le64_to_cpu(lt->rid);
+	bool more;
+	int ret;
+	int err;
+
+	mutex_lock(&server->srch_mutex);
+	ret = scoutfs_srch_rotate_log(sb, &server->alloc, &server->wri,
+				      &super->srch_root, &lt->srch_file, true);
+	mutex_unlock(&server->srch_mutex);
+	if (ret < 0) {
+		scoutfs_err(sb, "error rotating srch log for rid %016llx: %d",
+			    rid, ret);
+		return ret;
+        }
+
+	do {
+		more = false;
+
+		/*
+		 * All of these can return errors, perhaps indicating successful
+		 * partial progress, after having modified the allocator trees.
+		 * We always have to update the roots in the log item.
+		 */
+		mutex_lock(&server->alloc_mutex);
+		ret = (err_str = "splice meta_freed to other_freed",
+				scoutfs_alloc_splice_list(sb, &server->alloc,
+					&server->wri, server->other_freed,
+					&lt->meta_freed)) ?:
+			(err_str = "splice meta_avail",
+			 scoutfs_alloc_splice_list(sb, &server->alloc,
+					&server->wri, server->other_freed,
+					&lt->meta_avail)) ?:
+			(err_str = "empty data_avail",
+			 alloc_move_empty(sb, &super->data_alloc,
+					  &lt->data_avail,
+					  COMMIT_HOLD_ALLOC_BUDGET / 2)) ?:
+			(err_str = "empty data_freed",
+			 alloc_move_empty(sb, &super->data_alloc,
+					  &lt->data_freed,
+					  COMMIT_HOLD_ALLOC_BUDGET / 2));
+		mutex_unlock(&server->alloc_mutex);
+
+		/*
+		 * only finalize, allowing merging, once the allocators are
+		 * fully freed
+		 */
+		if (ret == 0) {
+			/* the transaction is no longer open */
+			le64_add_cpu(&lt->flags, SCOUTFS_LOG_TREES_FINALIZED);
+			lt->finalize_seq = cpu_to_le64(scoutfs_server_next_seq(sb));
+		}
+
+		scoutfs_key_init_log_trees(&key, rid, le64_to_cpu(lt->nr));
+
+		err = scoutfs_btree_update(sb, &server->alloc, &server->wri,
+					   &super->logs_root, &key, lt,
+					   sizeof(*lt));
+		BUG_ON(err != 0); /* alloc, log, srch items out of sync */
+
+		if (ret == -EINPROGRESS) {
+			more = true;
+			mutex_unlock(&server->logs_mutex);
+			ret = server_apply_commit(sb, hold, 0);
+			if (ret < 0)
+				WARN_ON_ONCE(ret < 0);
+			server_hold_commit(sb, hold);
+			mutex_lock(&server->logs_mutex);
+		} else if (ret == 0) {
+			memset(&lt->item_root, 0, sizeof(lt->item_root));
+			memset(&lt->bloom_ref, 0, sizeof(lt->bloom_ref));
+			lt->inode_count_delta = 0;
+			lt->max_item_seq = 0;
+			lt->finalize_seq = 0;
+			le64_add_cpu(&lt->nr, 1);
+			lt->flags = 0;
+		}
+	} while (more);
+
+	if (ret < 0) {
+		scoutfs_err(sb,
+			    "error %d finalizing log trees for rid %016llx: %s",
+			    ret, rid, err_str);
+	}
+
+	return ret;
+}
+
 /*
  * Finalizing the log btrees for merging needs to be done carefully so
  * that items don't appear to go backwards in time.
@@ -1089,7 +1233,6 @@ static int finalize_and_start_log_merge(struct super_block *sb, struct scoutfs_l
 	struct scoutfs_log_merge_range rng;
 	struct scoutfs_mount_options opts;
 	struct scoutfs_log_trees each_lt;
-	struct scoutfs_log_trees fin;
 	unsigned int delay_ms;
 	unsigned long timeo;
 	bool saw_finalized;
@@ -1149,9 +1292,13 @@ static int finalize_and_start_log_merge(struct super_block *sb, struct scoutfs_l
 		 * meta was low so that deleted items are merged
 		 * promptly and freed blocks can bring the client out of
 		 * enospc.
+		 *
+		 * The trigger can be used to force a log merge in cases where
+		 * a test only generates small amounts of change.
 		 */
 		finalize_ours = (lt->item_root.height > 2) ||
-				(le32_to_cpu(lt->meta_avail.flags) & SCOUTFS_ALLOC_FLAG_LOW);
+				(le32_to_cpu(lt->meta_avail.flags) & SCOUTFS_ALLOC_FLAG_LOW) ||
+				scoutfs_trigger(sb, LOG_MERGE_FORCE_FINALIZE_OURS);
 
 		trace_scoutfs_server_finalize_decision(sb, rid, saw_finalized, others_active,
 						       ours_visible, finalize_ours, delay_ms,
@@ -1160,6 +1307,7 @@ static int finalize_and_start_log_merge(struct super_block *sb, struct scoutfs_l
 		/* done if we're not finalizing and there's no finalized */
 		if (!finalize_ours && !saw_finalized) {
 			ret = 0;
+			scoutfs_inc_counter(sb, log_merge_no_finalized);
 			break;
 		}
 
@@ -1194,32 +1342,11 @@ static int finalize_and_start_log_merge(struct super_block *sb, struct scoutfs_l
 
 		/* Finalize ours if it's visible to others */
 		if (ours_visible) {
-			fin = *lt;
-			memset(&fin.meta_avail, 0, sizeof(fin.meta_avail));
-			memset(&fin.meta_freed, 0, sizeof(fin.meta_freed));
-			memset(&fin.data_avail, 0, sizeof(fin.data_avail));
-			memset(&fin.data_freed, 0, sizeof(fin.data_freed));
-			memset(&fin.srch_file, 0, sizeof(fin.srch_file));
-			le64_add_cpu(&fin.flags, SCOUTFS_LOG_TREES_FINALIZED);
-			fin.finalize_seq = cpu_to_le64(scoutfs_server_next_seq(sb));
-
-			scoutfs_key_init_log_trees(&key, le64_to_cpu(fin.rid),
-						   le64_to_cpu(fin.nr));
-			ret = scoutfs_btree_update(sb, &server->alloc, &server->wri,
-						   &super->logs_root, &key, &fin,
-						   sizeof(fin));
+			ret = do_finalize_ours(sb, lt, hold);
 			if (ret < 0) {
-				err_str = "updating finalized log_trees";
+				err_str = "finalizing ours";
 				break;
 			}
-
-			memset(&lt->item_root, 0, sizeof(lt->item_root));
-			memset(&lt->bloom_ref, 0, sizeof(lt->bloom_ref));
-			lt->inode_count_delta = 0;
-			lt->max_item_seq = 0;
-			lt->finalize_seq = 0;
-			le64_add_cpu(&lt->nr, 1);
-			lt->flags = 0;
 		}
 
 		/* wait a bit for mounts to arrive */
@@ -1280,6 +1407,8 @@ static int finalize_and_start_log_merge(struct super_block *sb, struct scoutfs_l
 			BUG_ON(err); /* inconsistent */
 		}
 
+		scoutfs_inc_counter(sb, log_merge_start);
+
 		/* we're done, caller can make forward progress */
 		break;
 	}
@@ -1299,12 +1428,10 @@ static int finalize_and_start_log_merge(struct super_block *sb, struct scoutfs_l
  * is nested inside holding commits so we recheck the persistent item
  * each time we commit to make sure it's still what we think.   The
  * caller is still going to send the item to the client so we update the
- * caller's each time we make progress.  This is a best-effort attempt
- * to clean up and it's valid to leave extents in data_freed we don't
- * return errors to the caller.  The client will continue the work later
- * in get_log_trees or as the rid is reclaimed.
+ * caller's each time we make progress.  If we hit an error applying the
+ * changes we make then we can't send the log_trees to the client.
  */
-static void try_drain_data_freed(struct super_block *sb, struct scoutfs_log_trees *lt)
+static int try_drain_data_freed(struct super_block *sb, struct scoutfs_log_trees *lt)
 {
 	DECLARE_SERVER_INFO(sb, server);
 	struct scoutfs_super_block *super = DIRTY_SUPER_SB(sb);
@@ -1313,6 +1440,7 @@ static void try_drain_data_freed(struct super_block *sb, struct scoutfs_log_tree
 	struct scoutfs_log_trees drain;
 	struct scoutfs_key key;
 	COMMIT_HOLD(hold);
+	bool apply = false;
 	int ret = 0;
 	int err;
 
@@ -1321,22 +1449,27 @@ static void try_drain_data_freed(struct super_block *sb, struct scoutfs_log_tree
 	while (lt->data_freed.total_len != 0) {
 		server_hold_commit(sb, &hold);
 		mutex_lock(&server->logs_mutex);
+		apply = true;
 
 		ret = find_log_trees_item(sb, &super->logs_root, false, rid, U64_MAX, &drain);
-		if (ret < 0)
+		if (ret < 0) {
+			ret = 0;
 			break;
+		}
 
 		/* careful to only keep draining the caller's specific open trans */
 		if (drain.nr != lt->nr || drain.get_trans_seq != lt->get_trans_seq ||
 		    drain.commit_trans_seq != lt->commit_trans_seq || drain.flags != lt->flags) {
-			ret = -ENOENT;
+			ret = 0;
 			break;
 		}
 
 		ret = scoutfs_btree_dirty(sb, &server->alloc, &server->wri,
 					  &super->logs_root, &key);
-		if (ret < 0)
+		if (ret < 0) {
+			ret = 0;
 			break;
+		}
 
 		/* moving can modify and return errors, always update caller and item */
 		mutex_lock(&server->alloc_mutex);
@@ -1352,19 +1485,19 @@ static void try_drain_data_freed(struct super_block *sb, struct scoutfs_log_tree
 		BUG_ON(err < 0); /* dirtying must guarantee success */
 
 		mutex_unlock(&server->logs_mutex);
-
 		ret = server_apply_commit(sb, &hold, ret);
-		if (ret < 0) {
-			ret = 0; /* don't try to abort, ignoring ret */
+		apply = false;
+
+		if (ret < 0)
 			break;
-		}
 	}
 
-	/* try to cleanly abort and write any partial dirty btree blocks, but ignore result */
-	if (ret < 0) {
+	if (apply) {
 		mutex_unlock(&server->logs_mutex);
-		server_apply_commit(sb, &hold, 0);
+		server_apply_commit(sb, &hold, ret);
 	}
+
+	return ret;
 }
 
 /*
@@ -1492,7 +1625,8 @@ static int server_get_log_trees(struct super_block *sb,
 		goto update;
 	}
 
-	ret = alloc_move_empty(sb, &super->data_alloc, &lt.data_freed, 100);
+	ret = alloc_move_empty(sb, &super->data_alloc, &lt.data_freed,
+			       COMMIT_HOLD_ALLOC_BUDGET / 2);
 	if (ret == -EINPROGRESS)
 		ret = 0;
 	if (ret < 0) {
@@ -1572,9 +1706,9 @@ out:
 		scoutfs_err(sb, "error %d getting log trees for rid %016llx: %s",
 			    ret, rid, err_str);
 
-	/* try to drain excessive data_freed with additional commits, if needed, ignoring err */
+	/* try to drain excessive data_freed with additional commits, if needed */
 	if (ret == 0)
-		try_drain_data_freed(sb, &lt);
+		ret = try_drain_data_freed(sb, &lt);
 
 	return scoutfs_net_response(sb, conn, cmd, id, ret, &lt, sizeof(lt));
 }
@@ -1602,6 +1736,7 @@ static int server_commit_log_trees(struct super_block *sb,
 	int ret;
 
 	if (arg_len != sizeof(struct scoutfs_log_trees)) {
+		err_str = "invalid message log_trees size";
 		ret = -EINVAL;
 		goto out;
 	}
@@ -1665,7 +1800,7 @@ static int server_commit_log_trees(struct super_block *sb,
 
 	ret = scoutfs_btree_update(sb, &server->alloc, &server->wri,
 				   &super->logs_root, &key, &lt, sizeof(lt));
-	BUG_ON(ret < 0); /* dirtying should have guaranteed success */
+	BUG_ON(ret < 0); /* dirtying should have guaranteed success, srch item inconsistent */
 	if (ret < 0)
 		err_str = "updating log trees item";
 
@@ -1673,11 +1808,10 @@ unlock:
 	mutex_unlock(&server->logs_mutex);
 
 	ret = server_apply_commit(sb, &hold, ret);
-	if (ret < 0)
-		scoutfs_err(sb, "server error %d committing client logs for rid %016llx: %s",
-			    ret, rid, err_str);
 out:
-	WARN_ON_ONCE(ret < 0);
+	if (ret < 0)
+		scoutfs_err(sb, "server error %d committing client logs for rid %016llx, nr %llu: %s",
+			    ret, rid, le64_to_cpu(lt.nr), err_str);
 	return scoutfs_net_response(sb, conn, cmd, id, ret, NULL, 0);
 }
 
@@ -1787,9 +1921,11 @@ static int reclaim_open_log_tree(struct super_block *sb, u64 rid)
 	       scoutfs_alloc_splice_list(sb, &server->alloc, &server->wri, server->other_freed,
 					 &lt.meta_avail)) ?:
 	      (err_str = "empty data_avail",
-	       alloc_move_empty(sb, &super->data_alloc, &lt.data_avail, 100)) ?:
+	       alloc_move_empty(sb, &super->data_alloc, &lt.data_avail,
+				COMMIT_HOLD_ALLOC_BUDGET / 2)) ?:
 	      (err_str = "empty data_freed",
-	       alloc_move_empty(sb, &super->data_alloc, &lt.data_freed, 100));
+	       alloc_move_empty(sb, &super->data_alloc, &lt.data_freed,
+				COMMIT_HOLD_ALLOC_BUDGET / 2));
 	mutex_unlock(&server->alloc_mutex);
 
 	/* only finalize, allowing merging, once the allocators are fully freed */
@@ -1810,6 +1946,9 @@ static int reclaim_open_log_tree(struct super_block *sb, u64 rid)
 out:
 	mutex_unlock(&server->logs_mutex);
 
+	if (ret == 0)
+		scoutfs_inc_counter(sb, reclaimed_open_logs);
+
 	if (ret < 0 && ret != -EINPROGRESS)
 		scoutfs_err(sb, "server error %d reclaiming log trees for rid %016llx: %s",
 			    ret, rid, err_str);
@@ -2011,7 +2150,7 @@ static int server_srch_get_compact(struct super_block *sb,
 
 apply:
 	ret = server_apply_commit(sb, &hold, ret);
-	WARN_ON_ONCE(ret < 0 && ret != -ENOENT); /* XXX leaked busy item */
+	WARN_ON_ONCE(ret < 0 && ret != -ENOENT && ret != -ENOLINK); /* XXX leaked busy item */
 out:
 	ret = scoutfs_net_response(sb, conn, cmd, id, ret,
 				   sc, sizeof(struct scoutfs_srch_compact));
@@ -2051,7 +2190,7 @@ static int server_srch_commit_compact(struct super_block *sb,
 					  &super->srch_root, rid, sc,
 					  &av, &fr);
 	mutex_unlock(&server->srch_mutex);
-	if (ret < 0) /* XXX very bad, leaks allocators */
+	if (ret < 0)
 		goto apply;
 
 	/* reclaim allocators if they were set by _srch_commit_ */
@@ -2061,10 +2200,10 @@ static int server_srch_commit_compact(struct super_block *sb,
 	      scoutfs_alloc_splice_list(sb, &server->alloc, &server->wri,
 					server->other_freed, &fr);
 	mutex_unlock(&server->alloc_mutex);
+	WARN_ON(ret < 0); /* XXX leaks allocators */
 apply:
 	ret = server_apply_commit(sb, &hold, ret);
 out:
-	WARN_ON(ret < 0); /* XXX leaks allocators */
 	return scoutfs_net_response(sb, conn, cmd, id, ret, NULL, 0);
 }
 
@@ -2377,6 +2516,8 @@ static int splice_log_merge_completions(struct super_block *sb,
 		queue_work(server->wq, &server->log_merge_free_work);
 	else
 		err_str = "deleting merge status item";
+
+	scoutfs_inc_counter(sb, log_merge_complete);
 out:
 	if (upd_stat) {
 		init_log_merge_key(&key, SCOUTFS_LOG_MERGE_STATUS_ZONE, 0, 0);
@@ -2389,10 +2530,9 @@ out:
 		}
 	}
 
-	if (ret < 0)
-		scoutfs_err(sb, "server error %d splicing log merge completion: %s", ret, err_str);
-
-	BUG_ON(ret); /* inconsistent */
+	/* inconsistent */
+	scoutfs_bug_on_err(sb, ret,
+			   "server error %d splicing log merge completion: %s", ret, err_str);
 
 	return ret ?: einprogress;
 }
@@ -2527,7 +2667,7 @@ static void server_log_merge_free_work(struct work_struct *work)
 
 		ret = scoutfs_btree_free_blocks(sb, &server->alloc,
 						&server->wri, &fr.key,
-						&fr.root, COMMIT_HOLD_ALLOC_BUDGET / 2);
+						&fr.root, COMMIT_HOLD_ALLOC_BUDGET / 8);
 		if (ret < 0) {
 			err_str = "freeing log btree";
 			break;
@@ -2546,7 +2686,7 @@ static void server_log_merge_free_work(struct work_struct *work)
 		/* freed blocks are in allocator, we *have* to update fr */
 		BUG_ON(ret < 0);
 
-		if (server_hold_alloc_used_since(sb, &hold) >= COMMIT_HOLD_ALLOC_BUDGET / 2) {
+		if (server_hold_alloc_used_since(sb, &hold) >= (COMMIT_HOLD_ALLOC_BUDGET * 3) / 4) {
 			mutex_unlock(&server->logs_mutex);
 			ret = server_apply_commit(sb, &hold, ret);
 			commit = false;
@@ -2637,10 +2777,7 @@ restart:
 
 	/* find the next range, always checking for splicing */
 	for (;;) {
-		key = stat.next_range_key;
-		key.sk_zone = SCOUTFS_LOG_MERGE_RANGE_ZONE;
-		ret = next_log_merge_item_key(sb, &super->log_merge, SCOUTFS_LOG_MERGE_RANGE_ZONE,
-					      &key, &rng, sizeof(rng));
+		ret = next_log_merge_range(sb, &super->log_merge, &stat.next_range_key, &rng);
 		if (ret < 0 && ret != -ENOENT) {
 			err_str = "finding merge range item";
 			goto out;
@@ -2911,7 +3048,13 @@ static int server_commit_log_merge(struct super_block *sb,
 				  SCOUTFS_LOG_MERGE_STATUS_ZONE, 0, 0,
 				  &stat, sizeof(stat));
 	if (ret < 0) {
-		err_str = "getting merge status item";
+		/*
+		 * During a retransmission, it's possible that the server
+		 * already committed and resolved this log merge. ENOENT
+		 * is expected in that case.
+		 */
+		if (ret != -ENOENT)
+			err_str = "getting merge status item";
 		goto out;
 	}
 
@@ -3410,7 +3553,7 @@ static bool invalid_mounted_client_item(struct scoutfs_btree_item_ref *iref)
  * it's acceptable to see -EEXIST.
  */
 static int insert_mounted_client(struct super_block *sb, u64 rid, u64 gr_flags,
-				 struct sockaddr_in *sin)
+				 struct sockaddr_storage *sin)
 {
 	DECLARE_SERVER_INFO(sb, server);
 	struct scoutfs_super_block *super = DIRTY_SUPER_SB(sb);
@@ -4149,7 +4292,7 @@ static void fence_pending_recov_worker(struct work_struct *work)
 	struct server_info *server = container_of(work, struct server_info,
 						  fence_pending_recov_work);
 	struct super_block *sb = server->sb;
-	union scoutfs_inet_addr addr;
+	union scoutfs_inet_addr addr = {{0,}};
 	u64 rid = 0;
 	int ret = 0;
 
@@ -4163,7 +4306,7 @@ static void fence_pending_recov_worker(struct work_struct *work)
 			break;
 		}
 
-		ret = scoutfs_fence_start(sb, rid, le32_to_be32(addr.v4.addr),
+		ret = scoutfs_fence_start(sb, rid, &addr,
 					  SCOUTFS_FENCE_CLIENT_RECOVERY);
 		if (ret < 0) {
 			scoutfs_err(sb, "fence returned err %d, shutting down server", ret);
@@ -4314,7 +4457,7 @@ static void scoutfs_server_worker(struct work_struct *work)
 	struct scoutfs_net_connection *conn = NULL;
 	struct scoutfs_mount_options opts;
 	DECLARE_WAIT_QUEUE_HEAD(waitq);
-	struct sockaddr_in sin;
+	struct sockaddr_storage sin;
 	bool alloc_init = false;
 	u64 max_seq;
 	int ret;
@@ -4323,7 +4466,7 @@ static void scoutfs_server_worker(struct work_struct *work)
 
 	scoutfs_options_read(sb, &opts);
 	scoutfs_quorum_slot_sin(&server->qconf, opts.quorum_slot_nr, &sin);
-	scoutfs_info(sb, "server starting at "SIN_FMT, SIN_ARG(&sin));
+	scoutfs_info(sb, "server starting at "SIN_FMT, &sin);
 
 	scoutfs_block_writer_init(sb, &server->wri);
 	server->finalize_sent_seq = 0;

kmod/src/server.h

@@ -1,27 +1,6 @@
 #ifndef _SCOUTFS_SERVER_H_
 #define _SCOUTFS_SERVER_H_
 
-#define SI4_FMT		"%u.%u.%u.%u:%u"
-
-#define si4_trace_define(name)		\
-	__field(__u32, name##_addr)	\
-	__field(__u16, name##_port)
-
-#define si4_trace_assign(name, sin)					\
-do {									\
-	__typeof__(sin) _sin = (sin);					\
-									\
-	__entry->name##_addr = be32_to_cpu(_sin->sin_addr.s_addr);	\
-	__entry->name##_port = be16_to_cpu(_sin->sin_port);		\
-} while(0)
-
-#define si4_trace_args(name)			\
-	(__entry->name##_addr >> 24),		\
-	(__entry->name##_addr >> 16) & 255,	\
-	(__entry->name##_addr >> 8) & 255,	\
-	__entry->name##_addr & 255,		\
-	__entry->name##_port
-
 #define SNH_FMT \
 	"seq %llu recv_seq %llu id %llu data_len %u cmd %u flags 0x%x error %u"
 #define SNH_ARG(nh)							\

kmod/src/sparse-filtered.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+
+#
+# Unfortunately, kernels can ship which contain sparse errors that are
+# unrelated to us.
+#
+# The exit status of this filtering wrapper will indicate an error if
+# sparse wasn't found or if there were any unfiltered output lines.  It
+# can hide error exit status from sparse or grep if they don't produce
+# output that makes it past the filters.
+#
+
+# must have sparse.  Fail with error message, mask success path.
+which sparse > /dev/null || exit 1
+
+# initial unmatchable, additional added as RE+="|..."
+RE="$^"
+
+#
+# Darn.  sparse has multi-line error messages, and I'd rather not bother
+# with multi-line filters.  So we'll just drop this context.
+#
+# command-line: note: in included file (through include/linux/netlink.h, include/linux/ethtool.h, include/linux/netdevice.h, include/net/sock.h, /root/scoutfs/kmod/src/kernelcompat.h, builtin): 
+#         fprintf(stderr, "%s: note: in included file%s:\n",
+#
+RE+="|: note: in included file"
+
+# 3.10.0-1160.119.1.el7.x86_64.debug
+# include/linux/posix_acl.h:138:9: warning: incorrect type in assignment (different address spaces)
+# include/linux/posix_acl.h:138:9:    expected struct posix_acl *<noident>
+# include/linux/posix_acl.h:138:9:    got struct posix_acl [noderef] <asn:4>*<noident>
+RE+="|include/linux/posix_acl.h:"
+
+# 3.10.0-1160.119.1.el7.x86_64.debug
+#include/uapi/linux/perf_event.h:146:56: warning: cast truncates bits from constant value (8000000000000000 becomes 0)
+RE+="|include/uapi/linux/perf_event.h:"
+
+# 4.18.0-513.24.1.el8_9.x86_64+debug'
+#./include/linux/skbuff.h:824:1: warning: directive in macro's argument list
+RE+="|include/linux/skbuff.h:"
+
+sparse "$@" |& \
+	grep -E -v "($RE)" |& \
+	awk '{ print $0 } END { exit NR > 0 }'
+exit $?

kmod/src/srch.c

@@ -62,7 +62,7 @@
  * re-allocated and re-written.  Search can restart by checking the
  * btree for the current set of files.  Compaction reads log files which
  * are protected from other compactions by the persistent busy items
- * created by the server.  Compaction won't see it's blocks reused out
+ * created by the server.  Compaction won't see its blocks reused out
  * from under it, but it can encounter stale cached blocks that need to
  * be invalidated.
  */
@@ -442,6 +442,10 @@ out:
 	if (ret == 0 && (flags & GFB_INSERT) && blk >= le64_to_cpu(sfl->blocks))
 		sfl->blocks = cpu_to_le64(blk + 1);
 
+	if (bl) {
+		trace_scoutfs_get_file_block(sb, bl->blkno, flags);
+	}
+
 	*bl_ret = bl;
 	return ret;
 }
@@ -533,23 +537,35 @@ out:
  * the pairs cancel each other out by all readers (the second encoding
  * looks like deletion) so they aren't visible to the first/last bounds of
  * the block or file.
+ *
+ * We use the same entry repeatedly, so the diff between them will be empty.
+ * This lets us just emit the two-byte count word, leaving the other bytes
+ * as zero.
+ *
+ * Split the desired total len into two pieces, adding any remainder to the
+ * first four-bit value.
  */
-static int append_padded_entry(struct scoutfs_srch_file *sfl, u64 blk,
-			       struct scoutfs_srch_block *srb, struct scoutfs_srch_entry *sre)
+static void append_padded_entry(struct scoutfs_srch_file *sfl,
+				struct scoutfs_srch_block *srb,
+				int len)
 {
-	int ret;
+	int each;
+	int rem;
+	u16 lengths = 0;
+	u8 *buf = srb->entries + le32_to_cpu(srb->entry_bytes);
 
-	ret = encode_entry(srb->entries + le32_to_cpu(srb->entry_bytes),
-			   sre, &srb->tail);
-	if (ret > 0) {
-		srb->tail = *sre;
-		le32_add_cpu(&srb->entry_nr, 1);
-		le32_add_cpu(&srb->entry_bytes, ret);
-		le64_add_cpu(&sfl->entries, 1);
-		ret = 0;
-	}
+	each = (len - 2) >> 1;
+	rem = (len - 2) & 1;
 
-	return ret;
+	lengths |= each + rem;
+	lengths |= each << 4;
+
+	memset(buf, 0, len);
+	put_unaligned_le16(lengths, buf);
+
+	le32_add_cpu(&srb->entry_nr, 1);
+	le32_add_cpu(&srb->entry_bytes, len);
+	le64_add_cpu(&sfl->entries, 1);
 }
 
 /*
@@ -560,61 +576,41 @@ static int append_padded_entry(struct scoutfs_srch_file *sfl, u64 blk,
  * This is called when there is a single existing entry in the block.
  * We have the entire block to work with.  We encode pairs of matching
  * entries.  This hides them from readers (both searches and merging) as
- * they're interpreted as creation and deletion and are deleted.  We use
- * the existing hash value of the first entry in the block but then set
- * the inode to an impossibly large number so it doesn't interfere with
- * anything.
+ * they're interpreted as creation and deletion and are deleted.
  *
- * To hit the specific offset we very carefully manage the amount of
- * bytes of change between fields in the entry.  We know that if we
- * change all the byte of the ino and id we end up with a 20 byte
- * (2+8+8,2) encoding of the pair of entries.  To have the last entry
- * start at the _SAFE_POS offset we know that the final 20 byte pair
- * encoding needs to end at 2 bytes (second entry encoding) after the
- * _SAFE_POS offset.
+ * For simplicity and to maintain sort ordering within the block, we reuse
+ * the existing entry. This lets us skip the encoding step, because we know
+ * the diff will be zero. We can zero-pad the resulting entries to hit the
+ * target offset exactly.
  *
- * So as we encode pairs we watch the delta of our current offset from
- * that desired final offset of 2 past _SAFE_POS.  If we're a multiple
- * of 20 away then we encode the full 20 byte pairs.  If we're not, then
- * we drop a byte to encode 19 bytes.  That'll slowly change the offset
- * to be a multiple of 20 again while encoding large entries.
+ * Because we can't predict the exact number of entry_bytes when we start,
+ * we adjust the byte count of subsequent entries until we wind up at a
+ * multiple of 20 bytes away from our goal and then use that length for
+ * the remaining entries.
+ *
+ * We could just use a single pair of unnaturally large entries to consume
+ * the needed space, adjusting for an odd number of entry_bytes if necessary.
+ * The use of 19 or 20 bytes for the entry pair matches what we would see with
+ * real (non-zero) entries that vary from the existing entry.
  */
-static void pad_entries_at_safe(struct scoutfs_srch_file *sfl, u64 blk,
+static void pad_entries_at_safe(struct scoutfs_srch_file *sfl,
 				struct scoutfs_srch_block *srb)
 {
-	struct scoutfs_srch_entry sre;
 	u32 target;
 	s32 diff;
-	u64 hash;
-	u64 ino;
-	u64 id;
-	int ret;
-
-	hash = le64_to_cpu(srb->tail.hash);
-	ino = le64_to_cpu(srb->tail.ino) | (1ULL << 62);
-	id = le64_to_cpu(srb->tail.id);
 
 	target = SCOUTFS_SRCH_BLOCK_SAFE_BYTES + 2;
 
 	while ((diff = target - le32_to_cpu(srb->entry_bytes)) > 0) {
-		ino ^= 1ULL << (7 * 8);
+		append_padded_entry(sfl, srb, 10);
 		if (diff % 20 == 0) {
-			id ^= 1ULL << (7 * 8);
+			append_padded_entry(sfl, srb, 10);
 		} else {
-			id ^= 1ULL << (6 * 8);
+			append_padded_entry(sfl, srb, 9);
 		}
-
-		sre.hash = cpu_to_le64(hash);
-		sre.ino = cpu_to_le64(ino);
-		sre.id = cpu_to_le64(id);
-
-		ret = append_padded_entry(sfl, blk, srb, &sre);
-		if (ret == 0)
-			ret = append_padded_entry(sfl, blk, srb, &sre);
-		BUG_ON(ret != 0);
-
-		diff = target - le32_to_cpu(srb->entry_bytes);
 	}
+
+	WARN_ON_ONCE(diff != 0);
 }
 
 /*
@@ -749,14 +745,14 @@ static int search_log_file(struct super_block *sb,
 		for (i = 0; i < le32_to_cpu(srb->entry_nr); i++) {
 			if (pos > SCOUTFS_SRCH_BLOCK_SAFE_BYTES) {
 				/* can only be inconsistency :/ */
-				ret = EIO;
+				ret = -EIO;
 				break;
 			}
 
 			ret = decode_entry(srb->entries + pos, &sre, &prev);
 			if (ret <= 0) {
 				/* can only be inconsistency :/ */
-				ret = EIO;
+				ret = -EIO;
 				break;
 			}
 			pos += ret;
@@ -859,15 +855,15 @@ static int search_sorted_file(struct super_block *sb,
 
 		if (pos > SCOUTFS_SRCH_BLOCK_SAFE_BYTES) {
 			/* can only be inconsistency :/ */
-			ret = EIO;
-			break;
+			ret = -EIO;
+			goto out;
 		}
 
 		ret = decode_entry(srb->entries + pos, &sre, &prev);
 		if (ret <= 0) {
 			/* can only be inconsistency :/ */
-			ret = EIO;
-			break;
+			ret = -EIO;
+			goto out;
 		}
 		pos += ret;
 		prev = sre;
@@ -972,6 +968,8 @@ int scoutfs_srch_search_xattrs(struct super_block *sb,
 
 	scoutfs_inc_counter(sb, srch_search_xattrs);
 
+	trace_scoutfs_ioc_search_xattrs(sb, ino, last_ino);
+
 	*done = false;
 	srch_init_rb_root(sroot);
 
@@ -1408,7 +1406,7 @@ int scoutfs_srch_commit_compact(struct super_block *sb,
 			ret = -EIO;
 		scoutfs_btree_put_iref(&iref);
 	}
-	if (ret < 0) /* XXX leaks allocators */
+	if (ret < 0)
 		goto out;
 
 	/* restore busy to pending if the operation failed */
@@ -1428,10 +1426,8 @@ int scoutfs_srch_commit_compact(struct super_block *sb,
 	/* update file references if we finished compaction (!deleting) */
 	if (!(res->flags & SCOUTFS_SRCH_COMPACT_FLAG_DELETE)) {
 		ret = commit_files(sb, alloc, wri, root, res);
-		if (ret < 0) {
-			/* XXX we can't commit, shutdown? */
+		if (ret < 0)
 			goto out;
-		}
 
 		/* transition flags for deleting input files */
 		for (i = 0; i < res->nr; i++) {
@@ -1458,7 +1454,7 @@ update:
 			      le64_to_cpu(pending->id), 0);
 		ret = scoutfs_btree_insert(sb, alloc, wri, root, &key,
 					   pending, sizeof(*pending));
-		if (ret < 0)
+		if (WARN_ON_ONCE(ret < 0)) /* XXX inconsistency */
 			goto out;
 	}
 
@@ -1471,7 +1467,6 @@ update:
 		BUG_ON(err); /* both busy and pending present */
 	}
 out:
-	WARN_ON_ONCE(ret < 0); /* XXX inconsistency */
 	kfree(busy);
 	return ret;
 }
@@ -1669,7 +1664,7 @@ static int kway_merge(struct super_block *sb,
 			/* end sorted block on _SAFE offset for testing */
 			if (bl && le32_to_cpu(srb->entry_nr) == 1 && logs_input &&
 			    scoutfs_trigger(sb, SRCH_COMPACT_LOGS_PAD_SAFE)) {
-				pad_entries_at_safe(sfl, blk, srb);
+				pad_entries_at_safe(sfl, srb);
 				scoutfs_block_put(sb, bl);
 				bl = NULL;
 				blk++;
@@ -1802,7 +1797,7 @@ static void swap_page_sre(void *A, void *B, int size)
  * typically, ~10x worst case).
  *
  * Because we read and sort all the input files we must perform the full
- * compaction in one operation.  The server must have given us a
+ * compaction in one operation.  The server must have given us
  * sufficiently large avail/freed lists, otherwise we'll return ENOSPC.
  */
 static int compact_logs(struct super_block *sb,
@@ -1866,14 +1861,14 @@ static int compact_logs(struct super_block *sb,
 
 		if (pos > SCOUTFS_SRCH_BLOCK_SAFE_BYTES) {
 			/* can only be inconsistency :/ */
-			ret = EIO;
-			break;
+			ret = -EIO;
+			goto out;
 		}
 
 		ret = decode_entry(srb->entries + pos, sre, &prev);
 		if (ret <= 0) {
 			/* can only be inconsistency :/ */
-			ret = EIO;
+			ret = -EIO;
 			goto out;
 		}
 		prev = *sre;
@@ -2281,12 +2276,11 @@ static void scoutfs_srch_compact_worker(struct work_struct *work)
 	} else {
 		ret = -EINVAL;
 	}
-	if (ret < 0)
-		goto commit;
 
-	ret = scoutfs_alloc_prepare_commit(sb, &alloc, &wri) ?:
+	scoutfs_alloc_prepare_commit(sb, &alloc, &wri);
+	if (ret == 0)
 	      scoutfs_block_writer_write(sb, &wri);
-commit:
+
 	/* the server won't use our partial compact if _ERROR is set */
 	sc->meta_avail = alloc.avail;
 	sc->meta_freed = alloc.freed;
@@ -2303,7 +2297,7 @@ out:
 		scoutfs_inc_counter(sb, srch_compact_error);
 
 	scoutfs_block_writer_forget_all(sb, &wri);
-	queue_compact_work(srinf, sc->nr > 0 && ret == 0);
+	queue_compact_work(srinf, sc != NULL && sc->nr > 0 && ret == 0);
 
 	kfree(sc);
 }

kmod/src/super.c

@@ -160,11 +160,17 @@ static void scoutfs_metadev_close(struct super_block *sb)
 		 * from kill_sb->put_super.
 		 */
 		lockdep_off();
+
+#ifdef KC_BDEV_FILE_OPEN_BY_PATH
+		bdev_fput(sbi->meta_bdev_file);
+#else
 #ifdef KC_BLKDEV_PUT_HOLDER_ARG
 		blkdev_put(sbi->meta_bdev, sb);
 #else
 		blkdev_put(sbi->meta_bdev, SCOUTFS_META_BDEV_MODE);
 #endif
+#endif
+
 		lockdep_on();
 		sbi->meta_bdev = NULL;
 	}
@@ -481,7 +487,11 @@ out:
 static int scoutfs_fill_super(struct super_block *sb, void *data, int silent)
 {
 	struct scoutfs_mount_options opts;
+#ifdef KC_BDEV_FILE_OPEN_BY_PATH
+	struct file *meta_bdev_file;
+#else
 	struct block_device *meta_bdev;
+#endif
 	struct scoutfs_sb_info *sbi;
 	struct inode *inode;
 	int ret;
@@ -502,9 +512,9 @@ static int scoutfs_fill_super(struct super_block *sb, void *data, int silent)
 
 	sbi = kzalloc(sizeof(struct scoutfs_sb_info), GFP_KERNEL);
 	sb->s_fs_info = sbi;
-	sbi->sb = sb;
 	if (!sbi)
 		return -ENOMEM;
+	sbi->sb = sb;
 
 	ret = assign_random_id(sbi);
 	if (ret < 0)
@@ -527,6 +537,22 @@ static int scoutfs_fill_super(struct super_block *sb, void *data, int silent)
 		goto out;
 	}
 
+#ifdef KC_BDEV_FILE_OPEN_BY_PATH
+	/*
+	 * pass sbi as holder, since dev_mount already passes sb, which triggers a
+	 * WARN_ON because dev_mount also passes non-NULL hops. By passing sbi
+	 * here we just get a simple error in our test cases.
+	 */
+	meta_bdev_file = bdev_file_open_by_path(opts.metadev_path, SCOUTFS_META_BDEV_MODE, sbi, NULL);
+	if (IS_ERR(meta_bdev_file)) {
+		scoutfs_err(sb, "could not open metadev: error %ld",
+			    PTR_ERR(meta_bdev_file));
+		ret = PTR_ERR(meta_bdev_file);
+		goto out;
+	}
+	sbi->meta_bdev_file = meta_bdev_file;
+	sbi->meta_bdev = file_bdev(meta_bdev_file);
+#else
 #ifdef KC_BLKDEV_PUT_HOLDER_ARG
 	meta_bdev = blkdev_get_by_path(opts.metadev_path, SCOUTFS_META_BDEV_MODE, sb, NULL);
 #else
@@ -539,6 +565,8 @@ static int scoutfs_fill_super(struct super_block *sb, void *data, int silent)
 		goto out;
 	}
 	sbi->meta_bdev = meta_bdev;
+#endif
+
 	ret = set_blocksize(sbi->meta_bdev, SCOUTFS_BLOCK_SM_SIZE);
 	if (ret != 0) {
 		scoutfs_err(sb, "failed to set metadev blocksize, returned %d",

kmod/src/super.h

@@ -42,6 +42,9 @@ struct scoutfs_sb_info {
 	u64 fmt_vers;
 
 	struct block_device *meta_bdev;
+#ifdef KC_BDEV_FILE_OPEN_BY_PATH
+	struct file *meta_bdev_file;
+#endif
 
 	spinlock_t next_ino_lock;
 

kmod/src/trans.c

@@ -159,6 +159,58 @@ static bool drained_holders(struct trans_info *tri)
 	return holders == 0;
 }
 
+static int commit_current_log_trees(struct super_block *sb, char **str)
+{
+	DECLARE_TRANS_INFO(sb, tri);
+
+	return (*str = "data submit", scoutfs_inode_walk_writeback(sb, true)) ?:
+	       (*str = "item dirty", scoutfs_item_write_dirty(sb))  ?:
+	       (*str = "data prepare", scoutfs_data_prepare_commit(sb))  ?:
+	       (*str = "alloc prepare", scoutfs_alloc_prepare_commit(sb, &tri->alloc, &tri->wri)) ?:
+	       (*str = "meta write", scoutfs_block_writer_write(sb, &tri->wri))  ?:
+	       (*str = "data wait", scoutfs_inode_walk_writeback(sb, false)) ?:
+	       (*str = "commit log trees", commit_btrees(sb)) ?:
+	       scoutfs_item_write_done(sb);
+}
+
+static int get_next_log_trees(struct super_block *sb, char **str)
+{
+	return (*str = "get log trees", scoutfs_trans_get_log_trees(sb));
+}
+
+static int retry_forever(struct super_block *sb, int (*func)(struct super_block *sb, char **str))
+{
+	bool retrying = false;
+	char *str;
+	int ret;
+
+	do {
+		str = NULL;
+
+		ret = func(sb, &str);
+		if (ret < 0) {
+			if (!retrying) {
+				scoutfs_warn(sb, "critical transaction commit failure: %s = %d, retrying",
+					    str, ret);
+				retrying = true;
+			}
+
+			if (scoutfs_forcing_unmount(sb)) {
+				ret = -ENOLINK;
+				break;
+			}
+
+			msleep(2 * MSEC_PER_SEC);
+
+		} else if (retrying) {
+			scoutfs_info(sb, "retried transaction commit succeeded");
+		}
+
+	} while (ret < 0);
+
+	return ret;
+}
+
 /*
  * This work func is responsible for writing out all the dirty blocks
  * that make up the current dirty transaction.  It prevents writers from
@@ -184,8 +236,6 @@ void scoutfs_trans_write_func(struct work_struct *work)
 	struct trans_info *tri = container_of(work, struct trans_info, write_work.work);
 	struct super_block *sb = tri->sb;
 	struct scoutfs_sb_info *sbi = SCOUTFS_SB(sb);
-	bool retrying = false;
-	char *s = NULL;
 	int ret = 0;
 
 	tri->task = current;
@@ -202,7 +252,7 @@ void scoutfs_trans_write_func(struct work_struct *work)
 	}
 
 	if (scoutfs_forcing_unmount(sb)) {
-		ret = -EIO;
+		ret = -ENOLINK;
 		goto out;
 	}
 
@@ -214,37 +264,9 @@ void scoutfs_trans_write_func(struct work_struct *work)
 
 	scoutfs_inc_counter(sb, trans_commit_written);
 
-	do {
-		ret = (s = "data submit", scoutfs_inode_walk_writeback(sb, true)) ?:
-		      (s = "item dirty", scoutfs_item_write_dirty(sb))  ?:
-		      (s = "data prepare", scoutfs_data_prepare_commit(sb))  ?:
-		      (s = "alloc prepare", scoutfs_alloc_prepare_commit(sb, &tri->alloc,
-									 &tri->wri))  ?:
-		      (s = "meta write", scoutfs_block_writer_write(sb, &tri->wri))  ?:
-		      (s = "data wait", scoutfs_inode_walk_writeback(sb, false)) ?:
-		      (s = "commit log trees", commit_btrees(sb)) ?:
-		      scoutfs_item_write_done(sb) ?:
-		      (s = "get log trees", scoutfs_trans_get_log_trees(sb));
-		if (ret < 0) {
-			if (!retrying) {
-				scoutfs_warn(sb, "critical transaction commit failure: %s = %d, retrying",
-					    s, ret);
-				retrying = true;
-			}
-
-			if (scoutfs_forcing_unmount(sb)) {
-				ret = -EIO;
-				break;
-			}
-
-			msleep(2 * MSEC_PER_SEC);
-
-		} else if (retrying) {
-			scoutfs_info(sb, "retried transaction commit succeeded");
-		}
-
-	} while (ret < 0);
-
+	/* retry {commit,get}_log_trees until they succeeed, can only fail when forcing unmount */
+	ret = retry_forever(sb, commit_current_log_trees) ?:
+	      retry_forever(sb, get_next_log_trees);
 out:
 	spin_lock(&tri->write_lock);
 	tri->write_count++;

kmod/src/triggers.c

@@ -18,6 +18,7 @@
 
 #include "super.h"
 #include "triggers.h"
+#include "scoutfs_trace.h"
 
 /*
  * We have debugfs files we can write to which arm triggers which
@@ -39,6 +40,7 @@ struct scoutfs_triggers {
 
 static char *names[] = {
 	[SCOUTFS_TRIGGER_BLOCK_REMOVE_STALE] = "block_remove_stale",
+	[SCOUTFS_TRIGGER_LOG_MERGE_FORCE_FINALIZE_OURS] = "log_merge_force_finalize_ours",
 	[SCOUTFS_TRIGGER_SRCH_COMPACT_LOGS_PAD_SAFE] = "srch_compact_logs_pad_safe",
 	[SCOUTFS_TRIGGER_SRCH_FORCE_LOG_ROTATE] = "srch_force_log_rotate",
 	[SCOUTFS_TRIGGER_SRCH_MERGE_STOP_SAFE] = "srch_merge_stop_safe",
@@ -51,6 +53,7 @@ bool scoutfs_trigger_test_and_clear(struct super_block *sb, unsigned int t)
 	atomic_t *atom;
 	int old;
 	int mem;
+	bool fired;
 
 	BUG_ON(t >= SCOUTFS_TRIGGER_NR);
 	atom = &triggers->atomics[t];
@@ -64,7 +67,12 @@ bool scoutfs_trigger_test_and_clear(struct super_block *sb, unsigned int t)
 		mem = atomic_cmpxchg(atom, old, 0);
 	} while (mem && mem != old);
 
-	return !!mem;
+	fired = !!mem;
+
+	if (fired)
+		trace_scoutfs_trigger_fired(sb, names[t]);
+
+	return fired;
 }
 
 int scoutfs_setup_triggers(struct super_block *sb)

kmod/src/triggers.h

@@ -3,6 +3,7 @@
 
 enum scoutfs_trigger {
 	SCOUTFS_TRIGGER_BLOCK_REMOVE_STALE,
+	SCOUTFS_TRIGGER_LOG_MERGE_FORCE_FINALIZE_OURS,
 	SCOUTFS_TRIGGER_SRCH_COMPACT_LOGS_PAD_SAFE,
 	SCOUTFS_TRIGGER_SRCH_FORCE_LOG_ROTATE,
 	SCOUTFS_TRIGGER_SRCH_MERGE_STOP_SAFE,

kmod/src/xattr.c

@@ -742,7 +742,7 @@ int scoutfs_xattr_set_locked(struct inode *inode, const char *name, size_t name_
 	int ret;
 	int err;
 
-	trace_scoutfs_xattr_set(sb, name_len, value, size, flags);
+	trace_scoutfs_xattr_set(sb, ino, name_len, value, size, flags);
 
 	if (WARN_ON_ONCE(tgs->totl && tgs->indx) ||
 	    WARN_ON_ONCE((tgs->totl | tgs->indx) && !tag_lock))

restore/cmd/test_restore/test_restore.go

@@ -1,244 +0,0 @@
-package main
-
-import (
-	"flag"
-	"fmt"
-	"log"
-	"os"
-	"path/filepath"
-	"sync"
-	"syscall"
-
-	"restore/pkg/restore"
-)
-
-type options struct {
-	metaPath   string
-	sourceDir  string
-	numWorkers int
-}
-
-// hardlinkTracker keeps track of inodes we've already processed
-type hardlinkTracker struct {
-	sync.Mutex
-	seen map[uint64]bool
-}
-
-func newHardlinkTracker() *hardlinkTracker {
-	return &hardlinkTracker{
-		seen: make(map[uint64]bool),
-	}
-}
-
-func (h *hardlinkTracker) isNewInode(ino uint64, nlink bool) bool {
-	if !nlink {
-		return true
-	}
-
-	h.Lock()
-	defer h.Unlock()
-
-	if _, exists := h.seen[ino]; exists {
-		return false
-	}
-
-	h.seen[ino] = true
-	return true
-}
-
-// getFileInfo extracts file information from os.FileInfo
-func getFileInfo(info os.FileInfo) restore.FileInfo {
-	stat := info.Sys().(*syscall.Stat_t)
-
-	// Use target inode number if specified, otherwise use actual inode number
-	ino := uint64(stat.Ino)
-
-	return restore.FileInfo{
-		Ino:       ino,
-		Mode:      uint32(stat.Mode),
-		Uid:       uint32(stat.Uid),
-		Gid:       uint32(stat.Gid),
-		Size:      uint64(stat.Size),
-		Rdev:      uint64(stat.Rdev),
-		AtimeSec:  stat.Atim.Sec,
-		AtimeNsec: stat.Atim.Nsec,
-		MtimeSec:  stat.Mtim.Sec,
-		MtimeNsec: stat.Mtim.Nsec,
-		CtimeSec:  stat.Ctim.Sec,
-		CtimeNsec: stat.Ctim.Nsec,
-		IsDir:     info.IsDir(),
-		IsRegular: stat.Mode&syscall.S_IFMT == syscall.S_IFREG,
-	}
-}
-
-// getXAttrs gets extended attributes for a file/directory
-func getXAttrs(path string) ([]restore.XAttr, error) {
-	size, err := syscall.Listxattr(path, nil)
-	if err != nil || size == 0 {
-		return nil, err
-	}
-
-	buf := make([]byte, size)
-	size, err = syscall.Listxattr(path, buf)
-	if err != nil {
-		return nil, err
-	}
-
-	var xattrs []restore.XAttr
-	start := 0
-	for i := 0; i < size; i++ {
-		if buf[i] == 0 {
-			name := string(buf[start:i])
-			value, err := syscall.Getxattr(path, name, nil)
-			if err != nil {
-				continue
-			}
-
-			valueBuf := make([]byte, value)
-			_, err = syscall.Getxattr(path, name, valueBuf)
-			if err != nil {
-				continue
-			}
-
-			xattrs = append(xattrs, restore.XAttr{
-				Name:  name,
-				Value: valueBuf,
-			})
-			start = i + 1
-		}
-	}
-
-	return xattrs, nil
-}
-
-func restorePath(writer *restore.WorkerWriter, hlTracker *hardlinkTracker, path string, parentIno uint64) error {
-	entries, err := os.ReadDir(path)
-	if err != nil {
-		return fmt.Errorf("failed to read directory: %v", err)
-	}
-	log.Printf("Restoring path: %s", path)
-	var subdirs int
-	var nameBytes int
-
-	for pos, entry := range entries {
-		if entry.Name() == "." || entry.Name() == ".." {
-			continue
-		}
-
-		info, err := entry.Info()
-		if err != nil {
-			return fmt.Errorf("failed to get entry info: %v", err)
-		}
-
-		stat, ok := info.Sys().(*syscall.Stat_t)
-		if !ok {
-			return fmt.Errorf("failed to get stat_t")
-		}
-		nameBytes += len(entry.Name())
-		fullPath := filepath.Join(path, entry.Name())
-
-		// Recurse into directories
-		if info.IsDir() {
-			subdirs++
-
-			if err := restorePath(writer, hlTracker, fullPath, uint64(stat.Ino)); err != nil {
-				return err
-			}
-
-		}
-
-		err = writer.CreateEntry(parentIno, uint64(pos), uint64(stat.Ino), uint32(info.Mode()), entry.Name())
-		if err != nil {
-			return fmt.Errorf("failed to create entry: %v", err)
-		}
-
-		// Handle inode
-		isHardlink := stat.Nlink > 1
-		if !info.IsDir() && hlTracker.isNewInode(uint64(stat.Ino), isHardlink) {
-			fileInfo := getFileInfo(info)
-			err = writer.CreateInode(fileInfo)
-			if err != nil {
-				return fmt.Errorf("failed to create inode: %v", err)
-			}
-
-			// Handle xattrs
-			xattrs, err := getXAttrs(fullPath)
-			if err == nil {
-				for pos, xattr := range xattrs {
-					err = writer.CreateXAttr(uint64(stat.Ino), uint64(pos), xattr)
-					if err != nil {
-						return fmt.Errorf("failed to create xattr: %v", err)
-					}
-				}
-			}
-		}
-	}
-	// Get directory info
-	dirInfo, err := os.Stat(path)
-	if err != nil {
-		return fmt.Errorf("failed to stat directory: %v", err)
-	}
-
-	// Create directory inode
-	dirFileInfo := getFileInfo(dirInfo)
-	dirFileInfo.NrSubdirs = uint64(subdirs)
-	dirFileInfo.NameBytes = uint64(nameBytes)
-
-	return writer.CreateInode(dirFileInfo)
-}
-
-func main() {
-	opts := options{}
-	flag.StringVar(&opts.metaPath, "m", "", "path to metadata device")
-	flag.StringVar(&opts.sourceDir, "s", "", "path to source directory")
-	flag.IntVar(&opts.numWorkers, "w", 4, "number of worker threads")
-	flag.Parse()
-
-	if opts.metaPath == "" || opts.sourceDir == "" {
-		flag.Usage()
-		os.Exit(1)
-	}
-
-	// Create master and worker writers
-	master, workers, err := restore.NewWriters(opts.metaPath, opts.numWorkers)
-	if err != nil {
-		fmt.Fprintf(os.Stderr, "Failed to create writers: %v\n", err)
-		os.Exit(1)
-	}
-	defer master.Destroy()
-
-	// Create hardlink tracker
-	hlTracker := newHardlinkTracker()
-
-	// Start workers
-	var wg sync.WaitGroup
-	for i, worker := range workers {
-		wg.Add(1)
-		go func(w *restore.WorkerWriter, workerNum int) {
-			defer wg.Done()
-
-			// Each worker processes a subset of the directory tree
-			if err := restorePath(w, hlTracker, opts.sourceDir, 1); err != nil {
-				fmt.Fprintf(os.Stderr, "Worker %d failed: %v\n", workerNum, err)
-				os.Exit(1)
-			}
-			// Create root inode for source directory
-			rootInfo, err := os.Stat(opts.sourceDir)
-			if err != nil {
-				fmt.Fprintf(os.Stderr, "Failed to stat source directory: %v\n", err)
-				os.Exit(1)
-			}
-			w.CreateInode(getFileInfo(rootInfo))
-			err = w.Destroy()
-			if err != nil {
-				fmt.Fprintf(os.Stderr, "Failed to destroy worker: %v\n", err)
-				os.Exit(1)
-			}
-		}(worker, i)
-	}
-
-	// Wait for all workers to complete
-	wg.Wait()
-
-	fmt.Println("Restore completed successfully")
-}

restore/go.mod

@@ -1,3 +0,0 @@
-module restore
-
-go 1.21.11

restore/pkg/restore/restore.go

@@ -1,472 +0,0 @@
-package restore
-
-/*
-#cgo CFLAGS: -I${SRCDIR}/../../../utils/src -I${SRCDIR}/../../../kmod/src
-#cgo LDFLAGS: -L${SRCDIR}/../../../utils/src -l:scoutfs_parallel_restore.a -lm
-
-#include <stdlib.h>
-#include <linux/types.h>
-#include <stdbool.h>
-#include <math.h>
-#include "sparse.h"
-#include "util.h"
-#include "format.h"
-#include "parallel_restore.h"
-
-// If there are any type conflicts, you might need to add:
-// #include "kernel_types.h"
-*/
-import "C"
-import (
-    "errors"
-    "fmt"
-    "sync"
-    "syscall"
-    "unsafe"
-)
-
-const batchSize = 1000
-const bufSize = 2 * 1024 * 1024
-
-type WorkerWriter struct {
-    writer      *C.struct_scoutfs_parallel_restore_writer
-    progressCh  chan *ScoutfsParallelWriterProgress
-    fileCreated int64
-    devFd       int
-    buf         unsafe.Pointer
-    wg          *sync.WaitGroup
-}
-
-type MasterWriter struct {
-    writer     *C.struct_scoutfs_parallel_restore_writer
-    progressCh chan *ScoutfsParallelWriterProgress
-    workers    []*WorkerWriter
-    wg         sync.WaitGroup
-    slice      *C.struct_scoutfs_parallel_restore_slice // Add slice field
-    progressWg sync.WaitGroup
-    devFd      int
-    super      *C.struct_scoutfs_super_block
-}
-
-type ScoutfsParallelWriterProgress struct {
-    Progress *C.struct_scoutfs_parallel_restore_progress
-    Slice    *C.struct_scoutfs_parallel_restore_slice
-}
-
-func (m *MasterWriter) aggregateProgress() {
-    defer m.progressWg.Done()
-    for progress := range m.progressCh {
-        ret := C.scoutfs_parallel_restore_add_progress(m.writer, progress.Progress)
-        if ret != 0 {
-            // Handle error appropriately, e.g., log it
-            fmt.Printf("Failed to add progress, error code: %d\n", ret)
-        }
-        if progress.Slice != nil {
-            ret = C.scoutfs_parallel_restore_add_slice(m.writer, progress.Slice)
-            C.free(unsafe.Pointer(progress.Slice))
-            if ret != 0 {
-                // Handle error appropriately, e.g., log it
-                fmt.Printf("Failed to add slice, error code: %d\n", ret)
-            }
-        }
-        // Free the C-allocated progress structures
-        C.free(unsafe.Pointer(progress.Progress))
-    }
-}
-
-func (m *MasterWriter) Destroy() {
-    m.wg.Wait()
-    close(m.progressCh)
-    m.progressWg.Wait()
-
-    if m.slice != nil {
-        C.free(unsafe.Pointer(m.slice)) // Free slice on error
-    }
-    if m.super != nil {
-        C.free(unsafe.Pointer(m.super)) // Free superblock on error
-    }
-    if m.devFd != 0 {
-        syscall.Close(m.devFd)
-    }
-    // Destroy master writer
-    C.scoutfs_parallel_restore_destroy_writer(&m.writer)
-}
-
-func NewWriters(path string, numWriters int) (*MasterWriter, []*WorkerWriter, error) {
-    if numWriters <= 1 {
-        return nil, nil, errors.New("number of writers must be positive")
-    }
-
-    devFd, err := syscall.Open(path, syscall.O_DIRECT|syscall.O_RDWR|syscall.O_EXCL, 0)
-    if err != nil {
-        return nil, nil, fmt.Errorf("failed to open metadata device '%s': %v", path, err)
-    }
-
-    var masterWriter MasterWriter
-    masterWriter.progressCh = make(chan *ScoutfsParallelWriterProgress, numWriters*2)
-    masterWriter.workers = make([]*WorkerWriter, 0, numWriters-1)
-    masterWriter.devFd = devFd
-
-    var ret C.int
-    // Allocate aligned memory for superblock
-    var super unsafe.Pointer
-    ret = C.posix_memalign(&super, 4096, C.SCOUTFS_BLOCK_SM_SIZE)
-    if ret != 0 {
-        masterWriter.Destroy()
-        return nil, nil, fmt.Errorf("failed to allocate aligned memory for superblock: %d", ret)
-    }
-    masterWriter.super = (*C.struct_scoutfs_super_block)(super)
-
-    // Read the superblock from devFd
-    superOffset := C.SCOUTFS_SUPER_BLKNO << C.SCOUTFS_BLOCK_SM_SHIFT
-    count, err := syscall.Pread(devFd, (*[1 << 30]byte)(super)[:C.SCOUTFS_BLOCK_SM_SIZE], int64(superOffset))
-    if err != nil {
-        masterWriter.Destroy()
-        return nil, nil, fmt.Errorf("failed to read superblock: %v", err)
-    }
-    if count != int(C.SCOUTFS_BLOCK_SM_SIZE) {
-        masterWriter.Destroy()
-        return nil, nil, fmt.Errorf("failed to read superblock, bytes read: %d", count)
-    }
-
-    // Check if the superblock is valid.
-    if C.le64_to_cpu(masterWriter.super.flags)&C.SCOUTFS_FLAG_IS_META_BDEV == 0 {
-        masterWriter.Destroy()
-        return nil, nil, errors.New("superblock is not a metadata device")
-    }
-
-    // Create master writer
-    ret = C.scoutfs_parallel_restore_create_writer(&masterWriter.writer)
-    if ret != 0 {
-        masterWriter.Destroy()
-        return nil, nil, errors.New("failed to create master writer")
-    }
-
-    ret = C.scoutfs_parallel_restore_import_super(masterWriter.writer, masterWriter.super, C.int(devFd))
-    if ret != 0 {
-        masterWriter.Destroy()
-        return nil, nil, fmt.Errorf("failed to import superblock, error code: %d", ret)
-    }
-
-    // Initialize slices for each worker
-    masterWriter.slice = (*C.struct_scoutfs_parallel_restore_slice)(C.malloc(C.size_t(numWriters) *
-        C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_slice{}))))
-    if masterWriter.slice == nil {
-        masterWriter.Destroy()
-        return nil, nil, errors.New("failed to allocate slices")
-    }
-
-    ret = C.scoutfs_parallel_restore_init_slices(masterWriter.writer,
-        masterWriter.slice,
-        C.int(numWriters))
-    if ret != 0 {
-        masterWriter.Destroy()
-        return nil, nil, errors.New("failed to initialize slices")
-    }
-
-    ret = C.scoutfs_parallel_restore_add_slice(masterWriter.writer, masterWriter.slice)
-    if ret != 0 {
-        masterWriter.Destroy()
-        return nil, nil, errors.New("failed to add slice to master writer")
-    }
-
-    // Create worker writers
-    for i := 1; i < numWriters; i++ {
-        var bufPtr unsafe.Pointer
-        if ret := C.posix_memalign(&bufPtr, 4096, bufSize); ret != 0 {
-            masterWriter.Destroy()
-            return nil, nil, fmt.Errorf("failed to allocate aligned worker buffer: %d", ret)
-        }
-
-        worker := &WorkerWriter{
-            progressCh: masterWriter.progressCh,
-            buf:        bufPtr,
-            wg:         &masterWriter.wg,
-        }
-        ret = C.scoutfs_parallel_restore_create_writer(&worker.writer)
-        if ret != 0 {
-            masterWriter.Destroy()
-            return nil, nil, errors.New("failed to create worker writer")
-        }
-
-        masterWriter.wg.Add(1)
-
-        // Use each slice for the corresponding worker
-        slice := (*C.struct_scoutfs_parallel_restore_slice)(unsafe.Pointer(uintptr(unsafe.Pointer(masterWriter.slice)) +
-            uintptr(i)*unsafe.Sizeof(C.struct_scoutfs_parallel_restore_slice{})))
-        ret = C.scoutfs_parallel_restore_add_slice(worker.writer, slice)
-        if ret != 0 {
-            C.scoutfs_parallel_restore_destroy_writer(&worker.writer)
-            masterWriter.Destroy()
-            return nil, nil, errors.New("failed to add slice to worker writer")
-        }
-
-        masterWriter.workers = append(masterWriter.workers, worker)
-    }
-    masterWriter.progressWg.Add(1)
-    go masterWriter.aggregateProgress()
-
-    return &masterWriter, masterWriter.workers, nil
-
-}
-
-func (w *WorkerWriter) getProgress(withSlice bool) (*ScoutfsParallelWriterProgress, error) {
-    progress := (*C.struct_scoutfs_parallel_restore_progress)(
-        C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_progress{}))),
-    )
-    if progress == nil {
-        return nil, errors.New("failed to allocate memory for progress")
-    }
-
-    // Fetch the current progress from the C library
-    ret := C.scoutfs_parallel_restore_get_progress(w.writer, progress)
-    if ret != 0 {
-        C.free(unsafe.Pointer(progress))
-        return nil, fmt.Errorf("failed to get progress, error code: %d", ret)
-    }
-
-    var slice *C.struct_scoutfs_parallel_restore_slice
-    if withSlice {
-        slice = (*C.struct_scoutfs_parallel_restore_slice)(
-            C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_slice{}))),
-        )
-        if slice == nil {
-            C.free(unsafe.Pointer(progress))
-            return nil, errors.New("failed to allocate memory for slice")
-        }
-
-        // Optionally fetch the slice information
-        ret = C.scoutfs_parallel_restore_get_slice(w.writer, slice)
-        if ret != 0 {
-            C.free(unsafe.Pointer(progress))
-            C.free(unsafe.Pointer(slice))
-            return nil, fmt.Errorf("failed to get slice, error code: %d", ret)
-        }
-    }
-
-    return &ScoutfsParallelWriterProgress{
-        Progress: progress,
-        Slice:    slice,
-    }, nil
-}
-
-// writeBuffer writes data from the buffer to the device file descriptor.
-// It uses scoutfs_parallel_restore_write_buf to get data and pwrite to write it.
-func (w *WorkerWriter) writeBuffer() (int64, error) {
-    var totalWritten int64
-    var count int64
-    var off int64
-    var ret C.int
-
-    // Allocate memory for off and count
-    offPtr := (*C.off_t)(unsafe.Pointer(&off))
-    countPtr := (*C.size_t)(unsafe.Pointer(&count))
-
-    for {
-        ret = C.scoutfs_parallel_restore_write_buf(w.writer, w.buf,
-            C.size_t(bufSize), offPtr, countPtr)
-
-        if ret != 0 {
-            return totalWritten, fmt.Errorf("failed to write buffer: error code %d", ret)
-        }
-
-        if count > 0 {
-            n, err := syscall.Pwrite(w.devFd, unsafe.Slice((*byte)(w.buf), count), off)
-            if err != nil {
-                return totalWritten, fmt.Errorf("pwrite failed: %v", err)
-            }
-            if n != int(count) {
-                return totalWritten, fmt.Errorf("pwrite wrote %d bytes; expected %d", n, count)
-            }
-            totalWritten += int64(n)
-        }
-
-        if count == 0 {
-            break
-        }
-    }
-
-    return totalWritten, nil
-}
-
-func (w *WorkerWriter) InsertEntry(entry *C.struct_scoutfs_parallel_restore_entry) error {
-    // Add the entry using the C library
-    ret := C.scoutfs_parallel_restore_add_entry(w.writer, entry)
-    if ret != 0 {
-        return fmt.Errorf("failed to add entry, error code: %d", ret)
-    }
-
-    // Increment the fileCreated counter
-    w.fileCreated++
-    if w.fileCreated >= batchSize {
-        _, err := w.writeBuffer()
-        if err != nil {
-            return fmt.Errorf("error writing buffers: %v", err)
-        }
-        // Allocate memory for progress and slice structures
-        progress, err := w.getProgress(false)
-        if err != nil {
-            return err
-        }
-        // Send the progress update to the shared progress channel
-        w.progressCh <- progress
-        // Reset the fileCreated counter
-        w.fileCreated = 0
-    }
-
-    return nil
-}
-
-func (w *WorkerWriter) InsertXattr(xattr *C.struct_scoutfs_parallel_restore_xattr) error {
-    ret := C.scoutfs_parallel_restore_add_xattr(w.writer, xattr)
-    if ret != 0 {
-        return fmt.Errorf("failed to add xattr, error code: %d", ret)
-    }
-    return nil
-}
-
-func (w *WorkerWriter) InsertInode(inode *C.struct_scoutfs_parallel_restore_inode) error {
-    ret := C.scoutfs_parallel_restore_add_inode(w.writer, inode)
-    if ret != 0 {
-        return fmt.Errorf("failed to add inode, error code: %d", ret)
-    }
-    return nil
-}
-
-// should only be called once
-func (w *WorkerWriter) Destroy() error {
-    defer w.wg.Done()
-    // Send final progress if there are remaining entries
-    if w.fileCreated > 0 {
-        _, err := w.writeBuffer()
-        if err != nil {
-            return err
-        }
-        progress := &ScoutfsParallelWriterProgress{
-            Progress: (*C.struct_scoutfs_parallel_restore_progress)(C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_progress{})))),
-            Slice:    (*C.struct_scoutfs_parallel_restore_slice)(C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_slice{})))),
-        }
-        w.progressCh <- progress
-        w.fileCreated = 0
-    }
-
-    if w.buf != nil {
-        C.free(w.buf)
-        w.buf = nil
-    }
-
-    C.scoutfs_parallel_restore_destroy_writer(&w.writer)
-    return nil
-}
-
-// Add these new types and functions to the existing restore.go file
-
-type FileInfo struct {
-    Ino       uint64
-    Mode      uint32
-    Uid       uint32
-    Gid       uint32
-    Size      uint64
-    Rdev      uint64
-    AtimeSec  int64
-    AtimeNsec int64
-    MtimeSec  int64
-    MtimeNsec int64
-    CtimeSec  int64
-    CtimeNsec int64
-    NrSubdirs uint64
-    NameBytes uint64
-    IsDir     bool
-    IsRegular bool
-}
-
-type XAttr struct {
-    Name  string
-    Value []byte
-}
-
-// CreateInode creates a C inode structure from FileInfo
-func (w *WorkerWriter) CreateInode(info FileInfo) error {
-    inode := (*C.struct_scoutfs_parallel_restore_inode)(C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_inode{}))))
-    if inode == nil {
-        return fmt.Errorf("failed to allocate inode")
-    }
-    defer C.free(unsafe.Pointer(inode))
-
-    inode.ino = C.__u64(info.Ino)
-    inode.mode = C.__u32(info.Mode)
-    inode.uid = C.__u32(info.Uid)
-    inode.gid = C.__u32(info.Gid)
-    inode.size = C.__u64(info.Size)
-    inode.rdev = C.uint(info.Rdev)
-
-    inode.atime.tv_sec = C.__time_t(info.AtimeSec)
-    inode.atime.tv_nsec = C.long(info.AtimeNsec)
-    inode.mtime.tv_sec = C.__time_t(info.MtimeSec)
-    inode.mtime.tv_nsec = C.long(info.MtimeNsec)
-    inode.ctime.tv_sec = C.__time_t(info.CtimeSec)
-    inode.ctime.tv_nsec = C.long(info.CtimeNsec)
-    inode.crtime = inode.ctime
-
-    if info.IsRegular && info.Size > 0 {
-        inode.offline = C.bool(true)
-    }
-
-    if info.IsDir {
-        inode.nr_subdirs = C.__u64(info.NrSubdirs)
-        inode.total_entry_name_bytes = C.__u64(info.NameBytes)
-    }
-
-    return w.InsertInode(inode)
-}
-
-// CreateEntry creates a directory entry
-func (w *WorkerWriter) CreateEntry(dirIno uint64, pos uint64, ino uint64, mode uint32, name string) error {
-    entryC := (*C.struct_scoutfs_parallel_restore_entry)(C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_entry{})) + C.size_t(len(name))))
-
-    if entryC == nil {
-        return fmt.Errorf("failed to allocate entry")
-    }
-    defer C.free(unsafe.Pointer(entryC))
-
-    entryC.dir_ino = C.__u64(dirIno)
-    entryC.pos = C.__u64(pos)
-    entryC.ino = C.__u64(ino)
-    entryC.mode = C.__u32(mode)
-    entryC.name_len = C.uint(len(name))
-
-    entryC.name = (*C.char)(C.malloc(C.size_t(len(name))))
-    if entryC.name == nil {
-        return fmt.Errorf("failed to allocate entry name")
-    }
-    defer C.free(unsafe.Pointer(entryC.name))
-    copy((*[1 << 30]byte)(unsafe.Pointer(entryC.name))[:len(name)], []byte(name))
-
-    return w.InsertEntry(entryC)
-}
-
-// CreateXAttr creates an extended attribute
-func (w *WorkerWriter) CreateXAttr(ino uint64, pos uint64, xattr XAttr) error {
-    xattrC := (*C.struct_scoutfs_parallel_restore_xattr)(C.malloc(C.size_t(unsafe.Sizeof(C.struct_scoutfs_parallel_restore_xattr{})) + C.size_t(len(xattr.Name)) + C.size_t(len(xattr.Value))))
-    if xattrC == nil {
-        return fmt.Errorf("failed to allocate xattr")
-    }
-    defer C.free(unsafe.Pointer(xattrC))
-
-    xattrC.ino = C.__u64(ino)
-    xattrC.pos = C.__u64(pos)
-    xattrC.name_len = C.uint(len(xattr.Name))
-    xattrC.value_len = C.__u32(len(xattr.Value))
-
-    xattrC.name = (*C.char)(C.malloc(C.size_t(len(xattr.Name))))
-    if xattrC.name == nil {
-        return fmt.Errorf("failed to allocate xattr name")
-    }
-    defer C.free(unsafe.Pointer(xattrC.name))
-
-    copy((*[1 << 30]byte)(unsafe.Pointer(xattrC.name))[:len(xattr.Name)], []byte(xattr.Name))
-
-    xattrC.value = unsafe.Pointer(&xattr.Value[0])
-
-    return w.InsertXattr(xattrC)
-}

restore/pkg/restore/restore_test.go

@@ -1,10 +0,0 @@
-package restore
-
-import "testing"
-
-func TestNewWriters(t *testing.T) {
-	_, _, err := NewWriters("/tmp", 2)
-	if err != nil {
-		t.Fatalf("failed to create master writer: %v", err)
-	}
-}

tests/.gitignore

@@ -10,3 +10,5 @@ src/stage_tmpfile
 src/create_xattr_loop
 src/o_tmpfile_umask
 src/o_tmpfile_linkat
+src/mmap_stress
+src/mmap_validate

tests/Makefile

@@ -14,8 +14,8 @@ BIN := src/createmany			\
 	src/fragmented_data_extents	\
 	src/o_tmpfile_umask		\
 	src/o_tmpfile_linkat		\
-	src/parallel_restore		\
-	src/restore_copy
+	src/mmap_stress			\
+	src/mmap_validate
 
 DEPS := $(wildcard src/*.d)
 
@@ -25,12 +25,10 @@ ifneq ($(DEPS),)
 -include $(DEPS)
 endif
 
-src/parallel_restore_cflags := ../utils/src/scoutfs_parallel_restore.a -lm
-src/restore_copy_cflags := ../utils/src/scoutfs_parallel_restore.a -lm
+src/mmap_stress: LIBS+=-lpthread
 
 $(BIN): %: %.c Makefile
-	gcc $(CFLAGS) -MD -MP -MF $*.d $< -o $@ $($(@)_cflags)
-
+	gcc $(CFLAGS) -MD -MP -MF $*.d $< -o $@ $(LIBS)
 
 .PHONY: clean
 clean:

tests/README.md

@@ -117,6 +117,7 @@ used during the test.
 | T\_NR\_MOUNTS    | number of mounts     | -n              | 3                 |
 | T\_O[0-9]        | mount options        | created per run | -o server\_addr=  |
 | T\_QUORUM        | quorum count         | -q              | 2                 |
+| T\_EXTRA         | per-test file dir    | revision ctled  | tests/extra/t     |
 | T\_TMP           | per-test tmp prefix  | made for test   | results/tmp/t/tmp |
 | T\_TMPDIR        | per-test tmp dir dir | made for test   | results/tmp/t     |
 

tests/extra/xfstests/expected-results

@@ -0,0 +1,882 @@
+Ran:
+generic/001
+generic/002
+generic/004
+generic/005
+generic/006
+generic/007
+generic/008
+generic/009
+generic/011
+generic/012
+generic/013
+generic/014
+generic/015
+generic/016
+generic/018
+generic/020
+generic/021
+generic/022
+generic/023
+generic/024
+generic/025
+generic/026
+generic/028
+generic/029
+generic/030
+generic/031
+generic/032
+generic/033
+generic/034
+generic/035
+generic/037
+generic/039
+generic/040
+generic/041
+generic/050
+generic/052
+generic/053
+generic/056
+generic/057
+generic/058
+generic/059
+generic/060
+generic/061
+generic/062
+generic/063
+generic/064
+generic/065
+generic/066
+generic/067
+generic/069
+generic/070
+generic/071
+generic/073
+generic/076
+generic/078
+generic/079
+generic/080
+generic/081
+generic/082
+generic/084
+generic/086
+generic/087
+generic/088
+generic/090
+generic/091
+generic/092
+generic/094
+generic/096
+generic/097
+generic/098
+generic/099
+generic/101
+generic/104
+generic/105
+generic/106
+generic/107
+generic/110
+generic/111
+generic/113
+generic/114
+generic/115
+generic/116
+generic/117
+generic/118
+generic/119
+generic/120
+generic/121
+generic/122
+generic/123
+generic/124
+generic/126
+generic/128
+generic/129
+generic/130
+generic/131
+generic/134
+generic/135
+generic/136
+generic/138
+generic/139
+generic/140
+generic/141
+generic/142
+generic/143
+generic/144
+generic/145
+generic/146
+generic/147
+generic/148
+generic/149
+generic/150
+generic/151
+generic/152
+generic/153
+generic/154
+generic/155
+generic/156
+generic/157
+generic/158
+generic/159
+generic/160
+generic/161
+generic/162
+generic/163
+generic/169
+generic/171
+generic/172
+generic/173
+generic/174
+generic/177
+generic/178
+generic/179
+generic/180
+generic/181
+generic/182
+generic/183
+generic/184
+generic/185
+generic/188
+generic/189
+generic/190
+generic/191
+generic/193
+generic/194
+generic/195
+generic/196
+generic/197
+generic/198
+generic/199
+generic/200
+generic/201
+generic/202
+generic/203
+generic/205
+generic/206
+generic/207
+generic/210
+generic/211
+generic/212
+generic/214
+generic/215
+generic/216
+generic/217
+generic/218
+generic/219
+generic/220
+generic/221
+generic/222
+generic/223
+generic/225
+generic/227
+generic/228
+generic/229
+generic/230
+generic/235
+generic/236
+generic/237
+generic/238
+generic/240
+generic/244
+generic/245
+generic/246
+generic/247
+generic/248
+generic/249
+generic/250
+generic/252
+generic/253
+generic/254
+generic/255
+generic/256
+generic/257
+generic/258
+generic/259
+generic/260
+generic/261
+generic/262
+generic/263
+generic/264
+generic/265
+generic/266
+generic/267
+generic/268
+generic/271
+generic/272
+generic/276
+generic/277
+generic/278
+generic/279
+generic/281
+generic/282
+generic/283
+generic/284
+generic/286
+generic/287
+generic/288
+generic/289
+generic/290
+generic/291
+generic/292
+generic/293
+generic/294
+generic/295
+generic/296
+generic/301
+generic/302
+generic/303
+generic/304
+generic/305
+generic/306
+generic/307
+generic/308
+generic/309
+generic/312
+generic/313
+generic/314
+generic/315
+generic/316
+generic/317
+generic/319
+generic/322
+generic/324
+generic/325
+generic/326
+generic/327
+generic/328
+generic/329
+generic/330
+generic/331
+generic/332
+generic/335
+generic/336
+generic/337
+generic/341
+generic/342
+generic/343
+generic/346
+generic/348
+generic/353
+generic/355
+generic/358
+generic/359
+generic/360
+generic/361
+generic/362
+generic/363
+generic/364
+generic/365
+generic/366
+generic/367
+generic/368
+generic/369
+generic/370
+generic/371
+generic/372
+generic/373
+generic/374
+generic/375
+generic/376
+generic/377
+generic/378
+generic/379
+generic/380
+generic/381
+generic/382
+generic/383
+generic/384
+generic/385
+generic/386
+generic/389
+generic/391
+generic/392
+generic/393
+generic/394
+generic/395
+generic/396
+generic/397
+generic/398
+generic/400
+generic/401
+generic/402
+generic/403
+generic/404
+generic/406
+generic/407
+generic/408
+generic/412
+generic/413
+generic/414
+generic/417
+generic/419
+generic/420
+generic/421
+generic/422
+generic/424
+generic/425
+generic/426
+generic/427
+generic/428
+generic/436
+generic/437
+generic/439
+generic/440
+generic/443
+generic/445
+generic/446
+generic/448
+generic/449
+generic/450
+generic/451
+generic/452
+generic/453
+generic/454
+generic/456
+generic/458
+generic/460
+generic/462
+generic/463
+generic/465
+generic/466
+generic/468
+generic/469
+generic/470
+generic/471
+generic/474
+generic/477
+generic/478
+generic/479
+generic/480
+generic/481
+generic/483
+generic/485
+generic/486
+generic/487
+generic/488
+generic/489
+generic/490
+generic/491
+generic/492
+generic/498
+generic/499
+generic/501
+generic/502
+generic/503
+generic/504
+generic/505
+generic/506
+generic/507
+generic/508
+generic/509
+generic/510
+generic/511
+generic/512
+generic/513
+generic/514
+generic/515
+generic/516
+generic/517
+generic/518
+generic/519
+generic/520
+generic/523
+generic/524
+generic/525
+generic/526
+generic/527
+generic/528
+generic/529
+generic/530
+generic/531
+generic/533
+generic/534
+generic/535
+generic/536
+generic/537
+generic/538
+generic/539
+generic/540
+generic/541
+generic/542
+generic/543
+generic/544
+generic/545
+generic/546
+generic/547
+generic/548
+generic/549
+generic/550
+generic/552
+generic/553
+generic/555
+generic/556
+generic/557
+generic/566
+generic/567
+generic/571
+generic/572
+generic/573
+generic/574
+generic/575
+generic/576
+generic/577
+generic/578
+generic/580
+generic/581
+generic/582
+generic/583
+generic/584
+generic/586
+generic/587
+generic/588
+generic/591
+generic/592
+generic/593
+generic/594
+generic/595
+generic/596
+generic/597
+generic/598
+generic/599
+generic/600
+generic/601
+generic/602
+generic/603
+generic/604
+generic/605
+generic/606
+generic/607
+generic/608
+generic/609
+generic/610
+generic/611
+generic/612
+generic/613
+generic/614
+generic/618
+generic/621
+generic/623
+generic/624
+generic/625
+generic/626
+generic/628
+generic/629
+generic/630
+generic/632
+generic/634
+generic/635
+generic/637
+generic/638
+generic/639
+generic/640
+generic/644
+generic/645
+generic/646
+generic/647
+generic/651
+generic/652
+generic/653
+generic/654
+generic/655
+generic/657
+generic/658
+generic/659
+generic/660
+generic/661
+generic/662
+generic/663
+generic/664
+generic/665
+generic/666
+generic/667
+generic/668
+generic/669
+generic/673
+generic/674
+generic/675
+generic/676
+generic/677
+generic/678
+generic/679
+generic/680
+generic/681
+generic/682
+generic/683
+generic/684
+generic/685
+generic/686
+generic/687
+generic/688
+generic/689
+shared/002
+shared/032
+Not
+run:
+generic/008
+generic/009
+generic/012
+generic/015
+generic/016
+generic/018
+generic/021
+generic/022
+generic/025
+generic/026
+generic/031
+generic/033
+generic/050
+generic/052
+generic/058
+generic/059
+generic/060
+generic/061
+generic/063
+generic/064
+generic/078
+generic/079
+generic/081
+generic/082
+generic/091
+generic/094
+generic/096
+generic/110
+generic/111
+generic/113
+generic/114
+generic/115
+generic/116
+generic/118
+generic/119
+generic/121
+generic/122
+generic/123
+generic/128
+generic/130
+generic/134
+generic/135
+generic/136
+generic/138
+generic/139
+generic/140
+generic/142
+generic/143
+generic/144
+generic/145
+generic/146
+generic/147
+generic/148
+generic/149
+generic/150
+generic/151
+generic/152
+generic/153
+generic/154
+generic/155
+generic/156
+generic/157
+generic/158
+generic/159
+generic/160
+generic/161
+generic/162
+generic/163
+generic/171
+generic/172
+generic/173
+generic/174
+generic/177
+generic/178
+generic/179
+generic/180
+generic/181
+generic/182
+generic/183
+generic/185
+generic/188
+generic/189
+generic/190
+generic/191
+generic/193
+generic/194
+generic/195
+generic/196
+generic/197
+generic/198
+generic/199
+generic/200
+generic/201
+generic/202
+generic/203
+generic/205
+generic/206
+generic/207
+generic/210
+generic/211
+generic/212
+generic/214
+generic/216
+generic/217
+generic/218
+generic/219
+generic/220
+generic/222
+generic/223
+generic/225
+generic/227
+generic/229
+generic/230
+generic/235
+generic/238
+generic/240
+generic/244
+generic/250
+generic/252
+generic/253
+generic/254
+generic/255
+generic/256
+generic/259
+generic/260
+generic/261
+generic/262
+generic/263
+generic/264
+generic/265
+generic/266
+generic/267
+generic/268
+generic/271
+generic/272
+generic/276
+generic/277
+generic/278
+generic/279
+generic/281
+generic/282
+generic/283
+generic/284
+generic/287
+generic/288
+generic/289
+generic/290
+generic/291
+generic/292
+generic/293
+generic/295
+generic/296
+generic/301
+generic/302
+generic/303
+generic/304
+generic/305
+generic/312
+generic/314
+generic/316
+generic/317
+generic/324
+generic/326
+generic/327
+generic/328
+generic/329
+generic/330
+generic/331
+generic/332
+generic/353
+generic/355
+generic/358
+generic/359
+generic/361
+generic/362
+generic/363
+generic/364
+generic/365
+generic/366
+generic/367
+generic/368
+generic/369
+generic/370
+generic/371
+generic/372
+generic/373
+generic/374
+generic/378
+generic/379
+generic/380
+generic/381
+generic/382
+generic/383
+generic/384
+generic/385
+generic/386
+generic/391
+generic/392
+generic/395
+generic/396
+generic/397
+generic/398
+generic/400
+generic/402
+generic/404
+generic/406
+generic/407
+generic/408
+generic/412
+generic/413
+generic/414
+generic/417
+generic/419
+generic/420
+generic/421
+generic/422
+generic/424
+generic/425
+generic/427
+generic/439
+generic/440
+generic/446
+generic/449
+generic/450
+generic/451
+generic/453
+generic/454
+generic/456
+generic/458
+generic/462
+generic/463
+generic/465
+generic/466
+generic/468
+generic/469
+generic/470
+generic/471
+generic/474
+generic/485
+generic/487
+generic/488
+generic/491
+generic/492
+generic/499
+generic/501
+generic/503
+generic/505
+generic/506
+generic/507
+generic/508
+generic/511
+generic/513
+generic/514
+generic/515
+generic/516
+generic/517
+generic/518
+generic/519
+generic/520
+generic/528
+generic/530
+generic/536
+generic/537
+generic/538
+generic/539
+generic/540
+generic/541
+generic/542
+generic/543
+generic/544
+generic/545
+generic/546
+generic/548
+generic/549
+generic/550
+generic/552
+generic/553
+generic/555
+generic/556
+generic/566
+generic/567
+generic/572
+generic/573
+generic/574
+generic/575
+generic/576
+generic/577
+generic/578
+generic/580
+generic/581
+generic/582
+generic/583
+generic/584
+generic/586
+generic/587
+generic/588
+generic/591
+generic/592
+generic/593
+generic/594
+generic/595
+generic/596
+generic/597
+generic/598
+generic/599
+generic/600
+generic/601
+generic/602
+generic/603
+generic/605
+generic/606
+generic/607
+generic/608
+generic/609
+generic/610
+generic/612
+generic/613
+generic/621
+generic/623
+generic/624
+generic/625
+generic/626
+generic/628
+generic/629
+generic/630
+generic/635
+generic/644
+generic/645
+generic/646
+generic/647
+generic/651
+generic/652
+generic/653
+generic/654
+generic/655
+generic/657
+generic/658
+generic/659
+generic/660
+generic/661
+generic/662
+generic/663
+generic/664
+generic/665
+generic/666
+generic/667
+generic/668
+generic/669
+generic/673
+generic/674
+generic/675
+generic/677
+generic/678
+generic/679
+generic/680
+generic/681
+generic/682
+generic/683
+generic/684
+generic/685
+generic/686
+generic/687
+generic/688
+generic/689
+shared/002
+shared/032
+Passed all 512 tests

tests/extra/xfstests/local.exclude

@@ -0,0 +1,44 @@
+generic/003	# missing atime update in buffered read
+generic/075	# file content mismatch failures (fds, etc)
+generic/103	# enospc causes trans commit failures
+generic/108	# mount fails on failing device?
+generic/112	# file content mismatch failures (fds, etc)
+generic/213	# enospc causes trans commit failures
+generic/318	# can't support user namespaces until v5.11
+generic/321	# requires selinux enabled for '+' in ls?
+generic/338	# BUG_ON update inode error handling
+generic/347	# _dmthin_mount doesn't work?
+generic/356	# swap
+generic/357	# swap
+generic/409	# bind mounts not scripted yet
+generic/410	# bind mounts not scripted yet
+generic/411	# bind mounts not scripted yet
+generic/423	# symlink inode size is strlen() + 1 on scoutfs
+generic/430	# xfs_io copy_range missing in el7
+generic/431	# xfs_io copy_range missing in el7
+generic/432	# xfs_io copy_range missing in el7
+generic/433	# xfs_io copy_range missing in el7
+generic/434	# xfs_io copy_range missing in el7
+generic/441	# dm-mapper
+generic/444	# el9's posix_acl_update_mode is buggy ?
+generic/467	# open_by_handle ESTALE
+generic/472	# swap
+generic/484	# dm-mapper
+generic/493	# swap
+generic/494	# swap
+generic/495	# swap
+generic/496	# swap
+generic/497	# swap
+generic/532	# xfs_io statx attrib_mask missing in el7
+generic/554	# swap
+generic/563	# cgroup+loopdev
+generic/564	# xfs_io copy_range missing in el7
+generic/565	# xfs_io copy_range missing in el7
+generic/568	# falloc not resulting in block count increase
+generic/569	# swap
+generic/570	# swap
+generic/620	# dm-hugedisk
+generic/633	# id-mapped mounts missing in el7
+generic/636	# swap
+generic/641	# swap
+generic/643	# swap

tests/fenced-local-force-unmount.sh

@@ -8,36 +8,33 @@
 
 echo "$0 running rid '$SCOUTFS_FENCED_REQ_RID' ip '$SCOUTFS_FENCED_REQ_IP' args '$@'"
 
-log() {
-	echo "$@" > /dev/stderr
+echo_fail() {
+	echo "$@" >&2
 	exit 1
 }
 
-echo_fail() {
-	echo "$@" > /dev/stderr
-	exit 1
+# silence error messages
+quiet_cat()
+{
+	cat "$@" 2>/dev/null
 }
 
 rid="$SCOUTFS_FENCED_REQ_RID"
 
+shopt -s nullglob
 for fs in /sys/fs/scoutfs/*; do
-	[ ! -d "$fs" ] && continue
+	fs_rid="$(quiet_cat $fs/rid)"
+	nr="$(quiet_cat $fs/data_device_maj_min)"
+	[ ! -d "$fs" -o "$fs_rid" != "$rid" ] && continue
 
-	fs_rid="$(cat $fs/rid)" || \
-		echo_fail "failed to get rid in $fs"
-	if [ "$fs_rid" != "$rid" ]; then
-		continue
+	mnt=$(findmnt -l -n -t scoutfs -o TARGET -S $nr)
+	[ -z "$mnt" ] && continue
+
+	if ! umount -qf "$mnt"; then
+		if [ -d "$fs" ]; then
+			echo_fail "umount -qf $mnt failed"
+		fi
 	fi
-
-	nr="$(cat $fs/data_device_maj_min)" || \
-		echo_fail "failed to get data device major:minor in $fs"
-
-	mnts=$(findmnt -l -n -t scoutfs -o TARGET -S $nr) || \
-		echo_fail "findmnt -t scoutfs -S $nr failed"
-	for mnt in $mnts; do
-		umount -f "$mnt" || \
-			echo_fail "umout -f $mnt failed"
-	done
 done
 
 exit 0

tests/funcs/exec.sh

@@ -64,19 +64,37 @@ t_rc()
 }
 
 #
-# redirect test output back to the output of the invoking script intead
-# of the compared output.
+# As run, stdout/err are redirected to a file that will be compared with
+# the stored expected golden output of the test.  This redirects
+# stdout/err in the script to stdout of the invoking run-test.  It's
+# intended to give visible output of tests without being included in the
+# golden output.
 #
-t_restore_output()
+# (see the goofy "exec" fd manipulation in the main run-tests as it runs
+# each test)
+#
+t_stdout_invoked()
 {
 	exec >&6 2>&1
 }
 
 #
-# redirect a command's output back to the compared output after the
-# test has restored its output
+# This undoes t_stdout_invokved, returning the test's stdout/err to the
+# output file as it was when it was launched.
 #
-t_compare_output()
+t_stdout_compare()
 {
-	"$@" >&7 2>&1
+	exec >&7 2>&1
+}
+
+#
+# usually bash prints an annoying output message when jobs
+# are killed.  We can avoid that by redirecting stderr for
+# the bash process when it reaps the jobs that are killed.
+#
+t_silent_kill() {
+	exec {ERR}>&2 2>/dev/null
+	kill "$@"
+	wait "$@"
+	exec 2>&$ERR {ERR}>&-
 }

tests/funcs/filter.sh

@@ -121,6 +121,7 @@ t_filter_dmesg()
 
 	# in debugging kernels we can slow things down a bit
 	re="$re|hrtimer: interrupt took .*"
+	re="$re|clocksource: Long readout interval"
 
 	# fencing tests force unmounts and trigger timeouts
 	re="$re|scoutfs .* forcing unmount"
@@ -140,6 +141,9 @@ t_filter_dmesg()
 	re="$re|scoutfs .* error.*server failed to bind to.*"
 	re="$re|scoutfs .* critical transaction commit failure.*"
 
+	# ENOLINK (-67) indicates an expected forced unmount error
+	re="$re|scoutfs .* error -67 .*"
+
 	# change-devices causes loop device resizing
 	re="$re|loop: module loaded"
 	re="$re|loop[0-9].* detected capacity change from.*"
@@ -160,6 +164,15 @@ t_filter_dmesg()
 	re="$re|Pipe handler or fully qualified core dump path required.*"
 	re="$re|Set kernel.core_pattern before fs.suid_dumpable.*"
 
+	# perf warning that it adjusted sample rate
+	re="$re|perf: interrupt took too long.*lowering kernel.perf_event_max_sample_rate.*"
+
+	# some ci test guests are unresponsive
+	re="$re|longest quorum heartbeat .* delay"
+
+	# creating block devices may trigger this
+	re="$re|block device autoloading is deprecated and will be removed."
+
 	egrep -v "($re)" | \
 		ignore_harmless_unwind_kasan_stack_oob
 }

tests/funcs/fs.sh

@@ -283,6 +283,30 @@ t_reinsert_remount_all()
 	t_quiet t_mount_all || t_fail "mounting all failed"
 }
 
+#
+# scratch helpers
+#
+t_scratch_mkfs()
+{
+	scoutfs mkfs -f -Q 0,127.0.0.1,$T_SCRATCH_PORT "$T_EX_META_DEV" "$T_EX_DATA_DEV" "$@" > $T_TMP.mkfs.out 2>&1 || \
+		t_fail "scratch mkfs failed"
+}
+
+t_scratch_mount()
+{
+	mkdir -p "$T_MSCR"
+	mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$@" "$T_EX_DATA_DEV" "$T_MSCR" || \
+		t_fail "scratch mount failed"
+}
+
+t_scratch_umount()
+{
+	umount "$T_MSCR" || \
+		t_fail "scratch umount failed"
+	rmdir "$T_MSCR"
+}
+
+
 t_trigger_path() {
 	local nr="$1"
 
@@ -498,3 +522,121 @@ t_restore_all_sysfs_mount_options() {
 		t_set_sysfs_mount_option $i $name "${_saved_opts[$ind]}"
 	done
 }
+
+t_force_log_merge() {
+	local sv=$(t_server_nr)
+	local merges_started
+	local last_merges_started
+	local merges_completed
+	local last_merges_completed
+
+	while true; do
+		last_merges_started=$(t_counter log_merge_start $sv)
+		last_merges_completed=$(t_counter log_merge_complete $sv)
+
+		t_trigger_arm_silent log_merge_force_finalize_ours $sv
+
+		t_sync_seq_index
+
+		while test "$(t_trigger_get log_merge_force_finalize_ours $sv)" == "1"; do
+			sleep .5
+		done
+
+		merges_started=$(t_counter log_merge_start $sv)
+
+		if (( merges_started > last_merges_started )); then
+			merges_completed=$(t_counter log_merge_complete $sv)
+
+			while (( merges_completed == last_merges_completed )); do
+				sleep .5
+				merges_completed=$(t_counter log_merge_complete $sv)
+			done
+			break
+		fi
+	done
+}
+
+declare -A _last_scan
+t_get_orphan_scan_runs() {
+	local i
+
+	for i in $(t_fs_nrs); do
+		_last_scan[$i]=$(t_counter orphan_scan $i)
+	done
+}
+
+t_wait_for_orphan_scan_runs() {
+	local i
+	local scan
+
+	t_get_orphan_scan_runs
+
+	for i in $(t_fs_nrs); do
+		while true; do
+			scan=$(t_counter orphan_scan $i)
+			if (( scan != _last_scan[$i] )); then
+				break
+			fi
+			sleep .5
+		done
+	done
+}
+
+declare -A _last_empty
+t_get_orphan_scan_empty() {
+	local i
+
+	for i in $(t_fs_nrs); do
+		_last_empty[$i]=$(t_counter orphan_scan_empty $i)
+	done
+}
+
+t_wait_for_no_orphans() {
+	local i;
+	local working;
+	local empty;
+
+	t_get_orphan_scan_empty
+
+	while true; do
+		working=0
+
+		t_wait_for_orphan_scan_runs
+
+		for i in $(t_fs_nrs); do
+			empty=$(t_counter orphan_scan_empty $i)
+			if (( empty == _last_empty[$i] )); then
+				(( working++ ))
+			else
+				(( _last_empty[$i] = empty ))
+			fi
+		done
+
+		if (( working == 0 )); then
+			break
+		fi
+
+		sleep 1
+	done
+}
+
+#
+# Repeatedly run the arguments as a command, sleeping in between, until
+# it returns success.  The first argument is a relative timeout in
+# seconds.  The remaining arguments are the command and its arguments.
+#
+# If the timeout expires without the command returning 0 then the test
+# fails.
+#
+t_wait_until_timeout() {
+	local relative="$1"
+	local expire="$((SECONDS + relative))"
+	shift
+
+	while (( SECONDS < expire )); do
+		"$@" && return
+		sleep 1
+	done
+
+	t_fail "command failed for $relative sec: $@"
+}

tests/funcs/tap.sh

@@ -0,0 +1,101 @@
+
+#
+# Generate TAP format test results
+#
+
+t_tap_header()
+{
+	local runid=$1
+	local sequence=( $(echo $tests) )
+	local count=${#sequence[@]}
+
+	# avoid recreating the same TAP result over again - harness sets this
+	[[ -z "$runid" ]] && runid="*test*"
+
+	cat > $T_RESULTS/scoutfs.tap <<TAPEOF
+TAP version 14
+1..${count}
+#
+# TAP results for run ${runid}
+#
+# host/run info:
+#
+#   hostname: ${HOSTNAME}
+#   test start time: $(date --utc)
+#   uname -r: $(uname -r)
+#   scoutfs commit id: $(git describe --tags)
+#
+# sequence for this run:
+#
+TAPEOF
+
+	# Sequence
+	for t in ${tests}; do
+		 echo ${t/.sh/}
+	done | cat -n | expand | column -c 120 | expand | sed 's/^ /#/' >> $T_RESULTS/scoutfs.tap
+	echo "#" >> $T_RESULTS/scoutfs.tap
+}
+
+t_tap_progress()
+{
+(
+	local i=$(( testcount + 1 ))
+	local testname=$1
+	local result=$2
+
+	local stmsg=""
+	local diff=""
+	local dmsg=""
+
+	if [[ -s $T_RESULTS/tmp/${testname}/status.msg ]]; then
+		stmsg="1"
+	fi
+
+	if [[ -s "$T_RESULTS/tmp/${testname}/dmesg.new" ]]; then
+		dmsg="1"
+	fi
+
+	if ! cmp -s golden/${testname} $T_RESULTS/output/${testname}; then
+		diff="1"
+	fi
+
+	if [[ "${result}" == "100" ]] && [[ -z "${dmsg}" ]] && [[ -z "${diff}" ]]; then
+		echo "ok ${i} - ${testname}"
+	elif [[ "${result}" == "103" ]]; then
+		echo "ok ${i} - ${testname}"
+		echo "# ${testname} ** skipped - permitted **"
+	else
+		echo "not ok ${i} - ${testname}"
+
+		case ${result} in
+		101)
+			echo "# ${testname} ** skipped **"
+			;;
+		102)
+			echo "# ${testname} ** failed **"
+			;;
+		esac
+
+		if [[ -n "${stmsg}" ]]; then
+			echo "#"
+			echo "# status:"
+			echo "#"
+			cat $T_RESULTS/tmp/${testname}/status.msg | sed 's/^/# - /'
+		fi
+
+		if [[ -n "${diff}" ]]; then
+			echo "#"
+			echo "# diff:"
+			echo "#"
+			diff -u golden/${testname} $T_RESULTS/output/${testname} | expand | sed 's/^/#   /'
+		fi
+
+		if [[ -n "${dmsg}" ]]; then
+			echo "#"
+			echo "# dmesg:"
+			echo "#"
+			cat "$T_RESULTS/tmp/${testname}/dmesg.new" | sed 's/^/#   /'
+		fi
+	fi
+) >> $T_RESULTS/scoutfs.tap
+}

tests/golden/basic-acl-consistency

@@ -0,0 +1,6 @@
+== make scratch fs
+== create uid/gids
+== set acls and permissions
+== compare output
+== drop caches and compare again
+== cleanup scratch fs

tests/golden/inode-deletion

@@ -17,7 +17,7 @@ ino not found in dseq index
 mount 0 contents after mount 1 rm: contents
 ino found in dseq index
 ino found in dseq index
-stat: cannot stat '/mnt/test/test/inode-deletion/file': No such file or directory
+stat: cannot stat '/mnt/test/test/inode-deletion/badfile': No such file or directory
 ino not found in dseq index
 ino not found in dseq index
 == lots of deletions use one open map

tests/golden/large-fragmented-free

@@ -1,4 +1,3 @@
-== setting longer hung task timeout
 == creating fragmented extents
 == unlink file with moved extents to free extents per block
 == cleanup

tests/golden/lock-shrink-read-race

@@ -0,0 +1,2 @@
+=== setup
+=== spin reading and shrinking

tests/golden/mmap

@@ -0,0 +1,27 @@
+== mmap_stress
+thread 0 complete
+thread 1 complete
+thread 2 complete
+thread 3 complete
+thread 4 complete
+== basic mmap/read/write consistency checks
+== mmap read from offline extent
+0: offset: 0 length: 2 flags: O.L
+extents: 1
+1
+00000200:  ea ea ea ea ea ea ea ea ea ea ea ea ea ea ea ea  ................
+0
+0: offset: 0 length: 2 flags: ..L
+extents: 1
+== mmap write to an offline extent
+0: offset: 0 length: 2 flags: O.L
+extents: 1
+1
+0
+0: offset: 0 length: 2 flags: ..L
+extents: 1
+00000000  ea ea ea ea ea ea ea ea  ea ea ea ea ea ea ea ea  |................|
+00000010  11 11 11 11 11 11 11 11  11 11 11 11 11 11 11 11  |................|
+00000020  ea ea ea ea ea ea ea ea  ea ea ea ea ea ea ea ea  |................|
+00000030
+== done

tests/golden/offline-extent-waiting

@@ -49,7 +49,7 @@ offline wating should be empty:
 0
 == truncating does wait
 truncate should be waiting for first block:
-trunate should no longer be waiting:
+truncate should no longer be waiting:
 0
 == writing waits
 should be waiting for write

tests/golden/parallel_restore

@@ -1,28 +0,0 @@
-== simple mkfs/restore/mount
-committed_seq     1120
-total_meta_blocks 163840
-total_data_blocks 15728640
-   1440    1440   57120
-     80      80     400
-0: offset: 0 length: 1 flags: O.L
-extents: 1
-0: offset: 0 length: 1 flags: O.L
-extents: 1
-0: offset: 0 length: 1 flags: O.L
-extents: 1
-0: offset: 0 length: 1 flags: O.L
-extents: 1
-    Type  Size     Total   Used      Free  Use%  
-MetaData  64KB    163840  34722    129118    21  
-    Data   4KB  15728640     64  15728576     0  
-  7 13,L,- 15,L,- 17,L,- I 33 -
-== just under ENOSPC
-    Type  Size     Total    Used      Free  Use%  
-MetaData  64KB    163840  155666      8174    95  
-    Data   4KB  15728640      64  15728576     0  
-== just over ENOSPC
-== ENOSPC
-== attempt to restore data device
-== attempt format_v1 restore
-== test if previously mounted
-== cleanup

tests/golden/simple-readdir

@@ -0,0 +1,97 @@
+== create content
+== readdir all
+00000000: d_off: 0x00000001 d_reclen: 0x18 d_type: DT_DIR d_name: .
+00000001: d_off: 0x00000002 d_reclen: 0x18 d_type: DT_DIR d_name: ..
+00000002: d_off: 0x00000003 d_reclen: 0x18 d_type: DT_REG d_name: a
+00000003: d_off: 0x00000004 d_reclen: 0x20 d_type: DT_REG d_name: aaaaaaaa
+00000004: d_off: 0x00000005 d_reclen: 0x28 d_type: DT_REG d_name: aaaaaaaaaaaaaaa
+00000005: d_off: 0x00000006 d_reclen: 0x30 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaa
+00000006: d_off: 0x00000007 d_reclen: 0x38 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000007: d_off: 0x00000008 d_reclen: 0x38 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000008: d_off: 0x00000009 d_reclen: 0x40 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000009: d_off: 0x0000000a d_reclen: 0x48 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000000a: d_off: 0x0000000b d_reclen: 0x50 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000000b: d_off: 0x0000000c d_reclen: 0x58 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000000c: d_off: 0x0000000d d_reclen: 0x60 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000000d: d_off: 0x0000000e d_reclen: 0x68 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000000e: d_off: 0x0000000f d_reclen: 0x70 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000000f: d_off: 0x00000010 d_reclen: 0x70 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000010: d_off: 0x00000011 d_reclen: 0x78 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000011: d_off: 0x00000012 d_reclen: 0x80 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000012: d_off: 0x00000013 d_reclen: 0x88 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000013: d_off: 0x00000014 d_reclen: 0x90 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000014: d_off: 0x00000015 d_reclen: 0x98 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000015: d_off: 0x00000016 d_reclen: 0xa0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000016: d_off: 0x00000017 d_reclen: 0xa8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000017: d_off: 0x00000018 d_reclen: 0xa8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000018: d_off: 0x00000019 d_reclen: 0xb0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000019: d_off: 0x0000001a d_reclen: 0xb8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001a: d_off: 0x0000001b d_reclen: 0xc0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001b: d_off: 0x0000001c d_reclen: 0xc8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001c: d_off: 0x0000001d d_reclen: 0xd0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001d: d_off: 0x0000001e d_reclen: 0xd8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001e: d_off: 0x0000001f d_reclen: 0xe0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001f: d_off: 0x00000020 d_reclen: 0xe0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000020: d_off: 0x00000021 d_reclen: 0xe8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000021: d_off: 0x00000022 d_reclen: 0xf0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000022: d_off: 0x00000023 d_reclen: 0xf8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000023: d_off: 0x00000024 d_reclen: 0x100 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000024: d_off: 0x00000025 d_reclen: 0x108 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000025: d_off: 0x00000026 d_reclen: 0x110 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+== readdir offset
+00000014: d_off: 0x00000015 d_reclen: 0x98 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000015: d_off: 0x00000016 d_reclen: 0xa0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000016: d_off: 0x00000017 d_reclen: 0xa8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000017: d_off: 0x00000018 d_reclen: 0xa8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000018: d_off: 0x00000019 d_reclen: 0xb0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000019: d_off: 0x0000001a d_reclen: 0xb8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001a: d_off: 0x0000001b d_reclen: 0xc0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001b: d_off: 0x0000001c d_reclen: 0xc8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001c: d_off: 0x0000001d d_reclen: 0xd0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001d: d_off: 0x0000001e d_reclen: 0xd8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001e: d_off: 0x0000001f d_reclen: 0xe0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001f: d_off: 0x00000020 d_reclen: 0xe0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000020: d_off: 0x00000021 d_reclen: 0xe8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000021: d_off: 0x00000022 d_reclen: 0xf0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000022: d_off: 0x00000023 d_reclen: 0xf8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000023: d_off: 0x00000024 d_reclen: 0x100 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000024: d_off: 0x00000025 d_reclen: 0x108 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000025: d_off: 0x00000026 d_reclen: 0x110 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+== readdir len (bytes)
+00000000: d_off: 0x00000001 d_reclen: 0x18 d_type: DT_DIR d_name: .
+00000001: d_off: 0x00000002 d_reclen: 0x18 d_type: DT_DIR d_name: ..
+00000002: d_off: 0x00000003 d_reclen: 0x18 d_type: DT_REG d_name: a
+00000003: d_off: 0x00000004 d_reclen: 0x20 d_type: DT_REG d_name: aaaaaaaa
+00000004: d_off: 0x00000005 d_reclen: 0x28 d_type: DT_REG d_name: aaaaaaaaaaaaaaa
+00000005: d_off: 0x00000006 d_reclen: 0x30 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaa
+00000006: d_off: 0x00000007 d_reclen: 0x38 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+== introduce gap
+00000000: d_off: 0x00000001 d_reclen: 0x18 d_type: DT_DIR d_name: .
+00000001: d_off: 0x00000002 d_reclen: 0x18 d_type: DT_DIR d_name: ..
+00000002: d_off: 0x00000003 d_reclen: 0x18 d_type: DT_REG d_name: a
+00000003: d_off: 0x00000004 d_reclen: 0x20 d_type: DT_REG d_name: aaaaaaaa
+00000004: d_off: 0x00000005 d_reclen: 0x28 d_type: DT_REG d_name: aaaaaaaaaaaaaaa
+00000005: d_off: 0x00000006 d_reclen: 0x30 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaa
+00000006: d_off: 0x00000007 d_reclen: 0x38 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000007: d_off: 0x00000008 d_reclen: 0x38 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000008: d_off: 0x00000009 d_reclen: 0x40 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000009: d_off: 0x00000014 d_reclen: 0x48 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000014: d_off: 0x00000015 d_reclen: 0x98 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000015: d_off: 0x00000016 d_reclen: 0xa0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000016: d_off: 0x00000017 d_reclen: 0xa8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000017: d_off: 0x00000018 d_reclen: 0xa8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000018: d_off: 0x00000019 d_reclen: 0xb0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000019: d_off: 0x0000001a d_reclen: 0xb8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001a: d_off: 0x0000001b d_reclen: 0xc0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001b: d_off: 0x0000001c d_reclen: 0xc8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001c: d_off: 0x0000001d d_reclen: 0xd0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001d: d_off: 0x0000001e d_reclen: 0xd8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001e: d_off: 0x0000001f d_reclen: 0xe0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+0000001f: d_off: 0x00000020 d_reclen: 0xe0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000020: d_off: 0x00000021 d_reclen: 0xe8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000021: d_off: 0x00000022 d_reclen: 0xf0 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000022: d_off: 0x00000023 d_reclen: 0xf8 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000023: d_off: 0x00000024 d_reclen: 0x100 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000024: d_off: 0x00000025 d_reclen: 0x108 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+00000025: d_off: 0x00000026 d_reclen: 0x110 d_type: DT_REG d_name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+== cleanup

tests/golden/xfstests

@@ -1,865 +0,0 @@
-Ran:
-generic/001
-generic/002
-generic/004
-generic/005
-generic/006
-generic/007
-generic/008
-generic/009
-generic/011
-generic/012
-generic/013
-generic/014
-generic/015
-generic/016
-generic/018
-generic/020
-generic/021
-generic/022
-generic/023
-generic/024
-generic/025
-generic/026
-generic/028
-generic/031
-generic/032
-generic/033
-generic/034
-generic/035
-generic/037
-generic/039
-generic/040
-generic/041
-generic/050
-generic/052
-generic/053
-generic/056
-generic/057
-generic/058
-generic/059
-generic/060
-generic/061
-generic/062
-generic/063
-generic/064
-generic/065
-generic/066
-generic/067
-generic/069
-generic/070
-generic/071
-generic/073
-generic/076
-generic/078
-generic/079
-generic/081
-generic/082
-generic/084
-generic/086
-generic/087
-generic/088
-generic/090
-generic/091
-generic/092
-generic/094
-generic/096
-generic/097
-generic/098
-generic/099
-generic/101
-generic/104
-generic/105
-generic/106
-generic/107
-generic/110
-generic/111
-generic/113
-generic/114
-generic/115
-generic/116
-generic/117
-generic/118
-generic/119
-generic/121
-generic/122
-generic/123
-generic/124
-generic/128
-generic/129
-generic/130
-generic/131
-generic/134
-generic/135
-generic/136
-generic/138
-generic/139
-generic/140
-generic/142
-generic/143
-generic/144
-generic/145
-generic/146
-generic/147
-generic/148
-generic/149
-generic/150
-generic/151
-generic/152
-generic/153
-generic/154
-generic/155
-generic/156
-generic/157
-generic/158
-generic/159
-generic/160
-generic/161
-generic/162
-generic/163
-generic/169
-generic/171
-generic/172
-generic/173
-generic/174
-generic/177
-generic/178
-generic/179
-generic/180
-generic/181
-generic/182
-generic/183
-generic/184
-generic/185
-generic/188
-generic/189
-generic/190
-generic/191
-generic/193
-generic/194
-generic/195
-generic/196
-generic/197
-generic/198
-generic/199
-generic/200
-generic/201
-generic/202
-generic/203
-generic/205
-generic/206
-generic/207
-generic/210
-generic/211
-generic/212
-generic/214
-generic/216
-generic/217
-generic/218
-generic/219
-generic/220
-generic/221
-generic/222
-generic/223
-generic/225
-generic/227
-generic/228
-generic/229
-generic/230
-generic/235
-generic/236
-generic/237
-generic/238
-generic/240
-generic/244
-generic/245
-generic/249
-generic/250
-generic/252
-generic/253
-generic/254
-generic/255
-generic/256
-generic/257
-generic/258
-generic/259
-generic/260
-generic/261
-generic/262
-generic/263
-generic/264
-generic/265
-generic/266
-generic/267
-generic/268
-generic/271
-generic/272
-generic/276
-generic/277
-generic/278
-generic/279
-generic/281
-generic/282
-generic/283
-generic/284
-generic/286
-generic/287
-generic/288
-generic/289
-generic/290
-generic/291
-generic/292
-generic/293
-generic/294
-generic/295
-generic/296
-generic/301
-generic/302
-generic/303
-generic/304
-generic/305
-generic/306
-generic/307
-generic/308
-generic/309
-generic/312
-generic/313
-generic/314
-generic/315
-generic/316
-generic/317
-generic/319
-generic/322
-generic/324
-generic/326
-generic/327
-generic/328
-generic/329
-generic/330
-generic/331
-generic/332
-generic/335
-generic/336
-generic/337
-generic/341
-generic/342
-generic/343
-generic/348
-generic/353
-generic/355
-generic/358
-generic/359
-generic/360
-generic/361
-generic/362
-generic/363
-generic/364
-generic/365
-generic/366
-generic/367
-generic/368
-generic/369
-generic/370
-generic/371
-generic/372
-generic/373
-generic/374
-generic/375
-generic/376
-generic/377
-generic/378
-generic/379
-generic/380
-generic/381
-generic/382
-generic/383
-generic/384
-generic/385
-generic/386
-generic/389
-generic/391
-generic/392
-generic/393
-generic/394
-generic/395
-generic/396
-generic/397
-generic/398
-generic/400
-generic/401
-generic/402
-generic/403
-generic/404
-generic/406
-generic/407
-generic/408
-generic/412
-generic/413
-generic/414
-generic/417
-generic/419
-generic/420
-generic/421
-generic/422
-generic/424
-generic/425
-generic/426
-generic/427
-generic/436
-generic/439
-generic/440
-generic/443
-generic/445
-generic/446
-generic/448
-generic/449
-generic/450
-generic/451
-generic/453
-generic/454
-generic/456
-generic/458
-generic/460
-generic/462
-generic/463
-generic/465
-generic/466
-generic/468
-generic/469
-generic/470
-generic/471
-generic/474
-generic/477
-generic/478
-generic/479
-generic/480
-generic/481
-generic/483
-generic/485
-generic/486
-generic/487
-generic/488
-generic/489
-generic/490
-generic/491
-generic/492
-generic/498
-generic/499
-generic/501
-generic/502
-generic/503
-generic/504
-generic/505
-generic/506
-generic/507
-generic/508
-generic/509
-generic/510
-generic/511
-generic/512
-generic/513
-generic/514
-generic/515
-generic/516
-generic/517
-generic/518
-generic/519
-generic/520
-generic/523
-generic/524
-generic/525
-generic/526
-generic/527
-generic/528
-generic/529
-generic/530
-generic/531
-generic/533
-generic/534
-generic/535
-generic/536
-generic/537
-generic/538
-generic/539
-generic/540
-generic/541
-generic/542
-generic/543
-generic/544
-generic/545
-generic/546
-generic/547
-generic/548
-generic/549
-generic/550
-generic/552
-generic/553
-generic/555
-generic/556
-generic/557
-generic/566
-generic/567
-generic/571
-generic/572
-generic/573
-generic/574
-generic/575
-generic/576
-generic/577
-generic/578
-generic/580
-generic/581
-generic/582
-generic/583
-generic/584
-generic/586
-generic/587
-generic/588
-generic/591
-generic/592
-generic/593
-generic/594
-generic/595
-generic/596
-generic/597
-generic/598
-generic/599
-generic/600
-generic/601
-generic/602
-generic/603
-generic/604
-generic/605
-generic/606
-generic/607
-generic/608
-generic/609
-generic/610
-generic/611
-generic/612
-generic/613
-generic/618
-generic/621
-generic/623
-generic/624
-generic/625
-generic/626
-generic/628
-generic/629
-generic/630
-generic/632
-generic/634
-generic/635
-generic/637
-generic/639
-generic/640
-generic/644
-generic/645
-generic/646
-generic/647
-generic/651
-generic/652
-generic/653
-generic/654
-generic/655
-generic/657
-generic/658
-generic/659
-generic/660
-generic/661
-generic/662
-generic/663
-generic/664
-generic/665
-generic/666
-generic/667
-generic/668
-generic/669
-generic/673
-generic/674
-generic/675
-generic/676
-generic/677
-generic/678
-generic/679
-generic/680
-generic/681
-generic/682
-generic/683
-generic/684
-generic/685
-generic/686
-generic/687
-generic/688
-generic/689
-shared/002
-shared/032
-Not
-run:
-generic/008
-generic/009
-generic/012
-generic/015
-generic/016
-generic/018
-generic/021
-generic/022
-generic/025
-generic/026
-generic/031
-generic/033
-generic/050
-generic/052
-generic/058
-generic/059
-generic/060
-generic/061
-generic/063
-generic/064
-generic/078
-generic/079
-generic/081
-generic/082
-generic/091
-generic/094
-generic/096
-generic/110
-generic/111
-generic/113
-generic/114
-generic/115
-generic/116
-generic/118
-generic/119
-generic/121
-generic/122
-generic/123
-generic/128
-generic/130
-generic/134
-generic/135
-generic/136
-generic/138
-generic/139
-generic/140
-generic/142
-generic/143
-generic/144
-generic/145
-generic/146
-generic/147
-generic/148
-generic/149
-generic/150
-generic/151
-generic/152
-generic/153
-generic/154
-generic/155
-generic/156
-generic/157
-generic/158
-generic/159
-generic/160
-generic/161
-generic/162
-generic/163
-generic/171
-generic/172
-generic/173
-generic/174
-generic/177
-generic/178
-generic/179
-generic/180
-generic/181
-generic/182
-generic/183
-generic/185
-generic/188
-generic/189
-generic/190
-generic/191
-generic/193
-generic/194
-generic/195
-generic/196
-generic/197
-generic/198
-generic/199
-generic/200
-generic/201
-generic/202
-generic/203
-generic/205
-generic/206
-generic/207
-generic/210
-generic/211
-generic/212
-generic/214
-generic/216
-generic/217
-generic/218
-generic/219
-generic/220
-generic/222
-generic/223
-generic/225
-generic/227
-generic/229
-generic/230
-generic/235
-generic/238
-generic/240
-generic/244
-generic/250
-generic/252
-generic/253
-generic/254
-generic/255
-generic/256
-generic/259
-generic/260
-generic/261
-generic/262
-generic/263
-generic/264
-generic/265
-generic/266
-generic/267
-generic/268
-generic/271
-generic/272
-generic/276
-generic/277
-generic/278
-generic/279
-generic/281
-generic/282
-generic/283
-generic/284
-generic/287
-generic/288
-generic/289
-generic/290
-generic/291
-generic/292
-generic/293
-generic/295
-generic/296
-generic/301
-generic/302
-generic/303
-generic/304
-generic/305
-generic/312
-generic/314
-generic/316
-generic/317
-generic/324
-generic/326
-generic/327
-generic/328
-generic/329
-generic/330
-generic/331
-generic/332
-generic/353
-generic/355
-generic/358
-generic/359
-generic/361
-generic/362
-generic/363
-generic/364
-generic/365
-generic/366
-generic/367
-generic/368
-generic/369
-generic/370
-generic/371
-generic/372
-generic/373
-generic/374
-generic/378
-generic/379
-generic/380
-generic/381
-generic/382
-generic/383
-generic/384
-generic/385
-generic/386
-generic/391
-generic/392
-generic/395
-generic/396
-generic/397
-generic/398
-generic/400
-generic/402
-generic/404
-generic/406
-generic/407
-generic/408
-generic/412
-generic/413
-generic/414
-generic/417
-generic/419
-generic/420
-generic/421
-generic/422
-generic/424
-generic/425
-generic/427
-generic/439
-generic/440
-generic/446
-generic/449
-generic/450
-generic/451
-generic/453
-generic/454
-generic/456
-generic/458
-generic/462
-generic/463
-generic/465
-generic/466
-generic/468
-generic/469
-generic/470
-generic/471
-generic/474
-generic/485
-generic/487
-generic/488
-generic/491
-generic/492
-generic/499
-generic/501
-generic/503
-generic/505
-generic/506
-generic/507
-generic/508
-generic/511
-generic/513
-generic/514
-generic/515
-generic/516
-generic/517
-generic/518
-generic/519
-generic/520
-generic/528
-generic/530
-generic/536
-generic/537
-generic/538
-generic/539
-generic/540
-generic/541
-generic/542
-generic/543
-generic/544
-generic/545
-generic/546
-generic/548
-generic/549
-generic/550
-generic/552
-generic/553
-generic/555
-generic/556
-generic/566
-generic/567
-generic/572
-generic/573
-generic/574
-generic/575
-generic/576
-generic/577
-generic/578
-generic/580
-generic/581
-generic/582
-generic/583
-generic/584
-generic/586
-generic/587
-generic/588
-generic/591
-generic/592
-generic/593
-generic/594
-generic/595
-generic/596
-generic/597
-generic/598
-generic/599
-generic/600
-generic/601
-generic/602
-generic/603
-generic/605
-generic/606
-generic/607
-generic/608
-generic/609
-generic/610
-generic/612
-generic/613
-generic/621
-generic/623
-generic/624
-generic/625
-generic/626
-generic/628
-generic/629
-generic/630
-generic/635
-generic/644
-generic/645
-generic/646
-generic/647
-generic/651
-generic/652
-generic/653
-generic/654
-generic/655
-generic/657
-generic/658
-generic/659
-generic/660
-generic/661
-generic/662
-generic/663
-generic/664
-generic/665
-generic/666
-generic/667
-generic/668
-generic/669
-generic/673
-generic/674
-generic/675
-generic/677
-generic/678
-generic/679
-generic/680
-generic/681
-generic/682
-generic/683
-generic/684
-generic/685
-generic/686
-generic/687
-generic/688
-generic/689
-shared/002
-shared/032
-Passed all 495 tests

tests/run-tests.sh

@@ -56,6 +56,7 @@ $(basename $0) options:
               | only tests matching will be run.  Can be provided multiple
               | times
     -i        | Force removing and inserting the built scoutfs.ko module.
+    -l <nr>   | Loop each test <nr> times while passing, last run counts.
     -M <file> | Specify the filesystem's meta data device path that contains
               | the file system to be tested.  Will be clobbered by -m mkfs.
     -m        | Run mkfs on the device before mounting and running
@@ -69,6 +70,7 @@ $(basename $0) options:
     -r <dir>  | Specify the directory in which to store results of
               | test runs.  The directory will be created if it doesn't
               | exist.  Previous results will be deleted as each test runs.
+    -R        | shuffle the test order randomly using shuf
     -s        | Skip git repo checkouts.
     -t        | Enabled trace events that match the given glob argument.
               | Multiple options enable multiple globbed events.
@@ -89,6 +91,8 @@ done
 # set some T_ defaults
 T_TRACE_DUMP="0"
 T_TRACE_PRINTK="0"
+T_PORT_START="19700"
+T_LOOP_ITER="1"
 
 # array declarations to be able to use array ops
 declare -a T_TRACE_GLOB
@@ -129,6 +133,12 @@ while true; do
 	-i)
 		T_INSMOD="1"
 		;;
+	-l)
+	        test -n "$2" || die "-l must have a nr iterations argument"
+		test "$2" -eq "$2" 2>/dev/null || die "-l <nr> argument must be an integer"
+		T_LOOP_ITER="$2"
+		shift
+		;;
 	-M)
 	        test -n "$2" || die "-z must have meta device file argument"
 	        T_META_DEVICE="$2"
@@ -164,6 +174,9 @@ while true; do
 		T_RESULTS="$2"
 		shift
 		;;
+	-R)
+		T_SHUF="1"
+		;;
 	-s)
 	        T_SKIP_CHECKOUT="1"
 		;;
@@ -261,13 +274,37 @@ for e in T_META_DEVICE T_DATA_DEVICE T_EX_META_DEV T_EX_DATA_DEV T_KMOD T_RESULT
 	eval $e=\"$(readlink -f "${!e}")\"
 done
 
+# try and check ports, but not necessary
+T_TEST_PORT="$T_PORT_START"
+T_SCRATCH_PORT="$((T_PORT_START + 100))"
+T_DEV_PORT="$((T_PORT_START + 200))"
+read local_start local_end < /proc/sys/net/ipv4/ip_local_port_range
+if [ -n "$local_start" -a -n "$local_end" -a "$local_start" -lt "$local_end" ]; then
+	if [ ! "$T_DEV_PORT" -lt "$local_start" -a ! "$T_TEST_PORT" -gt "$local_end" ]; then
+		die "listening port range $T_TEST_PORT - $T_DEV_PORT is within local dynamic port range $local_start - $local_end in /proc/sys/net/ipv4/ip_local_port_range"
+	fi
+fi
+
+# permute sequence?
+T_SEQUENCE=sequence
+if [ -n "$T_SHUF" ]; then
+	msg "shuffling test order"
+	shuf sequence -o sequence.shuf
+	# keep xfstests at the end
+	if grep -q 'xfstests.sh' sequence.shuf ; then
+		sed -i '/xfstests.sh/d' sequence.shuf
+		echo "xfstests.sh" >> sequence.shuf
+	fi
+	T_SEQUENCE=sequence.shuf
+fi
+
 # include everything by default
 test -z "$T_INCLUDE" && T_INCLUDE="-e '.*'"
 # (quickly) exclude nothing by default
 test -z "$T_EXCLUDE" && T_EXCLUDE="-e '\Zx'"
 
 # eval to strip re ticks but not expand
-tests=$(grep -v "^#" sequence |
+tests=$(grep -v "^#" $T_SEQUENCE |
 	eval grep "$T_INCLUDE" | eval grep -v "$T_EXCLUDE")
 test -z "$tests" && \
 	die "no tests found by including $T_INCLUDE and excluding $T_EXCLUDE"
@@ -346,7 +383,7 @@ fi
 quo=""
 if [ -n "$T_MKFS" ]; then
 	for i in $(seq -0 $((T_QUORUM - 1))); do
-		quo="$quo -Q $i,127.0.0.1,$((42000 + i))"
+		quo="$quo -Q $i,::1,$((T_TEST_PORT + i))"
 	done
 
 	msg "making new filesystem with $T_QUORUM quorum members"
@@ -363,7 +400,8 @@ if [ -n "$T_INSMOD" ]; then
 fi
 
 if [ -n "$T_TRACE_MULT" ]; then
-	orig_trace_size=$(cat /sys/kernel/debug/tracing/buffer_size_kb)
+#	orig_trace_size=$(cat /sys/kernel/debug/tracing/buffer_size_kb)
+	orig_trace_size=1408
 	mult_trace_size=$((orig_trace_size * T_TRACE_MULT))
 	msg "increasing trace buffer size from $orig_trace_size KiB to $mult_trace_size KiB"
 	echo $mult_trace_size > /sys/kernel/debug/tracing/buffer_size_kb
@@ -401,6 +439,30 @@ cmd grep .  /sys/kernel/debug/tracing/options/trace_printk \
 	    /sys/kernel/debug/tracing/buffer_size_kb \
 	    /proc/sys/kernel/ftrace_dump_on_oops
 
+# we can record pids to kill as we exit, we kill in reverse added order
+atexit_kill_pids=""
+add_atexit_kill_pid()
+{
+	atexit_kill_pids="$1 $atexit_kill_pids"
+}
+atexit_kill()
+{
+	local pid
+
+	# suppress bg function exited messages
+	exec {ERR}>&2 2>/dev/null
+
+	for pid in $atexit_kill_pids; do
+		if test -e "/proc/$pid/status" ; then
+			kill "$pid"
+			wait "$pid"
+		fi
+	done
+
+	exec 2>&$ERR {ERR}>&-
+}
+trap atexit_kill EXIT
+
 #
 # Build a fenced config that runs scripts out of the repository rather
 # than the default system directory
@@ -414,26 +476,46 @@ EOF
 export SCOUTFS_FENCED_CONFIG_FILE="$conf"
 T_FENCED_LOG="$T_RESULTS/fenced.log"
 
-#
-# Run the agent in the background, log its output, an kill it if we
-# exit
-#
-fenced_log()
-{
-	echo "[$(timestamp)] $*" >> "$T_FENCED_LOG"
-}
-fenced_pid=""
-kill_fenced()
-{
-	if test -n "$fenced_pid" -a -d "/proc/$fenced_pid" ; then
-		fenced_log "killing fenced pid $fenced_pid"
-		kill "$fenced_pid"
-	fi
-}
-trap kill_fenced EXIT
 $T_UTILS/fenced/scoutfs-fenced > "$T_FENCED_LOG" 2>&1 &
 fenced_pid=$!
-fenced_log "started fenced pid $fenced_pid in the background"
+add_atexit_kill_pid $fenced_pid
+
+#
+# some critical failures will cause fs operations to hang.  We can watch
+# for evidence of them and cause the system to crash, at least.
+#
+crash_monitor()
+{
+	local bad=0
+
+	while sleep 1; do
+		if dmesg | grep -q "inserting extent.*overlaps existing"; then
+			echo "run-tests monitor saw overlapping extent message"
+			bad=1
+		fi
+
+		if dmesg | grep -q "error indicated by fence action" ; then
+			echo "run-tests monitor saw fence agent error message"
+			bad=1
+		fi
+
+		if [ ! -e "/proc/${fenced_pid}/status" ]; then
+			echo "run-tests monitor didn't see fenced pid $fenced_pid /proc dir"
+			bad=1
+		fi
+
+		if [ "$bad" != 0 ]; then
+			echo "run-tests monitor syncing and triggering crash"
+			# hail mary, the sync could well hang
+			(echo s > /proc/sysrq-trigger) &
+			sleep 5
+			echo c > /proc/sysrq-trigger
+			exit 1
+		fi
+	done
+}
+crash_monitor &
+add_atexit_kill_pid $!
 
 # setup dm tables
 echo "0 $(blockdev --getsz $T_META_DEVICE) linear $T_META_DEVICE 0" > \
@@ -506,12 +588,17 @@ fi
 . funcs/filter.sh
 
 # give tests access to built binaries in src/, prefer over installed
-PATH="$PWD/src:$PATH"
+export PATH="$PWD/src:$PATH"
 
 msg "running tests"
 > "$T_RESULTS/skip.log"
 > "$T_RESULTS/fail.log"
 
+# generate a test ID to make sure we can de-duplicate TAP results in aggregation
+. funcs/tap.sh
+t_tap_header $(uuidgen)
+
+testcount=0
 passed=0
 skipped=0
 failed=0
@@ -521,98 +608,113 @@ for t in $tests; do
 	t="tests/$t"
 	test_name=$(basename "$t" | sed -e 's/.sh$//')
 
-	# create a temporary dir and file path for the test
-	T_TMPDIR="$T_RESULTS/tmp/$test_name"
-	T_TMP="$T_TMPDIR/tmp"
-	cmd rm -rf "$T_TMPDIR"
-	cmd mkdir -p "$T_TMPDIR"
-
-	# create a test name dir in the fs
-	T_DS=""
-	for i in $(seq 0 $((T_NR_MOUNTS - 1))); do
-		dir="${T_M[$i]}/test/$test_name"
-
-		test $i == 0 && cmd mkdir -p "$dir"
-
-		eval T_D$i=$dir
-		T_D[$i]=$dir
-		T_DS+="$dir "
-	done
-
-	# export all our T_ variables
-	for v in ${!T_*}; do
-		eval export $v
-	done
-	export PATH # give test access to scoutfs binary
-
-	# prepare to compare output to golden output
-	test -e "$T_RESULTS/output" || cmd mkdir -p "$T_RESULTS/output"
-	out="$T_RESULTS/output/$test_name"
-	> "$T_TMPDIR/status.msg"
-	golden="golden/$test_name"
-
 	# get stats from previous pass
 	last="$T_RESULTS/last-passed-test-stats"
 	stats=$(grep -s "^$test_name " "$last" | cut -d " " -f 2-)
 	test -n "$stats" && stats="last: $stats"
-
 	printf "  %-30s $stats" "$test_name"
 
 	# mark in dmesg as to what test we are running
 	echo "run scoutfs test $test_name" > /dev/kmsg
 
-	# record dmesg before
-	dmesg | t_filter_dmesg > "$T_TMPDIR/dmesg.before"
+	# let the test get at its extra files
+	T_EXTRA="$T_TESTS/extra/$test_name"
 
-	# give tests stdout and compared output on specific fds
-	exec 6>&1
-	exec 7>$out
+	for iter in $(seq 1 $T_LOOP_ITER); do
 
-	# run the test with access to our functions
-	start_secs=$SECONDS
-	bash -c "for f in funcs/*.sh; do . \$f; done; . $t" >&7 2>&1
-	sts="$?"
-	log "test $t exited with status $sts"
-	stats="$((SECONDS - start_secs))s"
+		# create a temporary dir and file path for the test
+		T_TMPDIR="$T_RESULTS/tmp/$test_name"
+		T_TMP="$T_TMPDIR/tmp"
+		cmd rm -rf "$T_TMPDIR"
+		cmd mkdir -p "$T_TMPDIR"
 
-	# close our weird descriptors
-	exec 6>&-
-	exec 7>&-
+		# assign scratch mount point in temporary dir
+		T_MSCR="$T_TMPDIR/scratch"
 
-	# compare output if the test returned passed status
-	if [ "$sts" == "$T_PASS_STATUS" ]; then
-		if [ ! -e "$golden" ]; then
-			message="no golden output"
-			sts=$T_FAIL_STATUS
-		elif ! cmp -s "$golden" "$out"; then 
-			message="output differs"
-			sts=$T_FAIL_STATUS
-			diff -u "$golden" "$out" >> "$T_RESULTS/fail.log"
+		# create a test name dir in the fs, clean up old data as needed
+		T_DS=""
+		for i in $(seq 0 $((T_NR_MOUNTS - 1))); do
+			dir="${T_M[$i]}/test/$test_name"
+
+			test $i == 0 && (
+				test -d "$dir" && cmd rm -rf "$dir"
+				cmd mkdir -p "$dir"
+			)
+
+			eval T_D$i=$dir
+			T_D[$i]=$dir
+			T_DS+="$dir "
+		done
+
+		# export all our T_ variables
+		for v in ${!T_*}; do
+			eval export $v
+		done
+
+		# prepare to compare output to golden output
+		test -e "$T_RESULTS/output" || cmd mkdir -p "$T_RESULTS/output"
+		out="$T_RESULTS/output/$test_name"
+		> "$T_TMPDIR/status.msg"
+		golden="golden/$test_name"
+
+		# record dmesg before
+		dmesg | t_filter_dmesg > "$T_TMPDIR/dmesg.before"
+
+		# give tests stdout and compared output on specific fds
+		exec 6>&1
+		exec 7>$out
+
+		# run the test with access to our functions
+		start_secs=$SECONDS
+		bash -c "for f in funcs/*.sh; do . \$f; done; . $t" >&7 2>&1
+		sts="$?"
+		log "test $t exited with status $sts"
+		stats="$((SECONDS - start_secs))s"
+
+		# close our weird descriptors
+		exec 6>&-
+		exec 7>&-
+
+		# compare output if the test returned passed status
+		if [ "$sts" == "$T_PASS_STATUS" ]; then
+			if [ ! -e "$golden" ]; then
+				message="no golden output"
+				sts=$T_FAIL_STATUS
+			elif ! cmp -s "$golden" "$out"; then 
+				message="output differs"
+				sts=$T_FAIL_STATUS
+				diff -u "$golden" "$out" >> "$T_RESULTS/fail.log"
+			fi
+		else
+			# get message from t_*() functions
+			message=$(cat "$T_TMPDIR/status.msg")
 		fi
-	else
-		# get message from t_*() functions
-		message=$(cat "$T_TMPDIR/status.msg")
-	fi
 
-	# see if anything unexpected was added to dmesg
-	if [ "$sts" == "$T_PASS_STATUS" ]; then
-		dmesg | t_filter_dmesg > "$T_TMPDIR/dmesg.after"
-		diff --old-line-format="" --unchanged-line-format="" \
-			"$T_TMPDIR/dmesg.before" "$T_TMPDIR/dmesg.after" > \
-			"$T_TMPDIR/dmesg.new"
+		# see if anything unexpected was added to dmesg
+		if [ "$sts" == "$T_PASS_STATUS" ]; then
+			dmesg | t_filter_dmesg > "$T_TMPDIR/dmesg.after"
+			diff --old-line-format="" --unchanged-line-format="" \
+				"$T_TMPDIR/dmesg.before" "$T_TMPDIR/dmesg.after" > \
+				"$T_TMPDIR/dmesg.new"
 
-		if [ -s "$T_TMPDIR/dmesg.new" ]; then
-			message="unexpected messages in dmesg"
-			sts=$T_FAIL_STATUS
-			cat "$T_TMPDIR/dmesg.new" >> "$T_RESULTS/fail.log"
+			if [ -s "$T_TMPDIR/dmesg.new" ]; then
+				message="unexpected messages in dmesg"
+				sts=$T_FAIL_STATUS
+				cat "$T_TMPDIR/dmesg.new" >> "$T_RESULTS/fail.log"
+			fi
 		fi
-	fi
 
-	# record unknown exit status
-	if [ "$sts" -lt "$T_FIRST_STATUS" -o "$sts" -gt "$T_LAST_STATUS" ]; then
-		message="unknown status: $sts"
-		sts=$T_FAIL_STATUS
-	fi
+		# record unknown exit status
+		if [ "$sts" -lt "$T_FIRST_STATUS" -o "$sts" -gt "$T_LAST_STATUS" ]; then
+			message="unknown status: $sts"
+			sts=$T_FAIL_STATUS
+		fi
+
+		# stop looping if we didn't pass
+		if [ "$sts" != "$T_PASS_STATUS" ]; then
+			break;
+		fi
+	done
 
 	# show and record the result of the test
 	if [ "$sts" == "$T_PASS_STATUS" ]; then
@@ -637,6 +739,11 @@ for t in $tests; do
 
 		test -n "$T_ABORT" && die "aborting after first failure"
 	fi
+
+	# record results for TAP format output
+	t_tap_progress $test_name $sts
+	((testcount++))
+
 done
 
 msg "all tests run: $passed passed, $skipped skipped, $skipped_permitted skipped (permitted), $failed failed"

tests/sequence

@@ -2,10 +2,12 @@ export-get-name-parent.sh
 basic-block-counts.sh
 basic-bad-mounts.sh
 basic-posix-acl.sh
+basic-acl-consistency.sh
 inode-items-updated.sh
 simple-inode-index.sh
 simple-staging.sh
 simple-release-extents.sh
+simple-readdir.sh
 get-referring-entries.sh
 fallocate.sh
 basic-truncate.sh
@@ -17,6 +19,7 @@ projects.sh
 large-fragmented-free.sh
 format-version-forward-back.sh
 enospc.sh
+mmap.sh
 srch-safe-merge-pos.sh
 srch-basic-functionality.sh
 simple-xattr-unit.sh
@@ -25,6 +28,7 @@ totl-xattr-tag.sh
 quota.sh
 lock-refleak.sh
 lock-shrink-consistency.sh
+lock-shrink-read-race.sh
 lock-pr-cw-conflict.sh
 lock-revoke-getcwd.sh
 lock-recover-invalidate.sh
@@ -54,5 +58,4 @@ archive-light-cycle.sh
 block-stale-reads.sh
 inode-deletion.sh
 renameat2-noreplace.sh
-parallel_restore.sh
 xfstests.sh

tests/src/mmap_stress.c

@@ -0,0 +1,192 @@
+#define _GNU_SOURCE
+/*
+ * mmap() stress test for scoutfs
+ *
+ * This test exercises the scoutfs kernel module's locking by
+ * repeatedly reading/writing using mmap and pread/write calls
+ * across 5 clients (mounts).
+ *
+ * Each thread operates on a single thread/client, and performs
+ * operations in a random order on the file.
+ *
+ * The goal is to assure that locking between _page_mkwrite vfs
+ * calls and the normal read/write paths do not cause deadlocks.
+ *
+ * There is no content validation performed. All that is done is
+ * assure that the programs continues without errors.
+ */
+
+#include <sys/types.h>
+#include <stdio.h>
+#include <sys/stat.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdbool.h>
+#include <sys/mman.h>
+#include <pthread.h>
+#include <errno.h>
+
+static int size = 0;
+static int duration = 0;
+
+struct thread_info {
+	int nr;
+	int fd;
+};
+
+static void *run_test_func(void *ptr)
+{
+	void *buf = NULL;
+	char *addr = NULL;
+	struct thread_info *tinfo = ptr;
+	uint64_t seconds = 0;
+	struct timespec ts;
+	int c = 0;
+	int fd;
+	ssize_t read, written, ret;
+	int preads = 0, pwrites = 0, mreads = 0, mwrites = 0;
+
+	fd = tinfo->fd;
+
+	if (posix_memalign(&buf, 4096, size) != 0) {
+		perror("calloc");
+		exit(-1);
+	}
+
+	addr = mmap(NULL, size, PROT_WRITE | PROT_READ, MAP_SHARED, fd, 0);
+	if (addr == MAP_FAILED) {
+		perror("mmap");
+		exit(-1);
+	}
+
+	usleep(100000); /* 0.1sec to allow all threads to start roughly at the same time */
+
+	clock_gettime(CLOCK_REALTIME, &ts); /* record start time */
+	seconds = ts.tv_sec + duration;
+
+	for (;;) {
+		if (++c % 16 == 0) {
+			clock_gettime(CLOCK_REALTIME, &ts);
+			if (ts.tv_sec >= seconds)
+				break;
+		}
+
+		switch (rand() % 4) {
+		case 0: /* pread */
+			preads++;
+			for (read = 0; read < size;) {
+				ret = pread(fd, buf, size - read, read);
+				if (ret < 0) {
+					perror("pwrite");
+					exit(-1);
+				}
+				read += ret;
+			}
+			break;
+		case 1: /* pwrite */
+			pwrites++;
+			memset(buf, (char)(c & 0xff), size);
+			for (written = 0; written < size;) {
+				ret = pwrite(fd, buf, size - written, written);
+				if (ret < 0) {
+					perror("pwrite");
+					exit(-1);
+				}
+				written += ret;
+			}
+			break;
+		case 2: /* mmap read */
+			mreads++;
+			memcpy(buf, addr, size); /* noerr */
+			break;
+		case 3: /* mmap write */
+			mwrites++;
+			memset(buf, (char)(c & 0xff), size);
+			memcpy(addr, buf, size); /* noerr */
+			break;
+		}
+
+		usleep(10000);
+	}
+
+	munmap(addr, size);
+
+	free(buf);
+
+	printf("thread %u complete: preads %u pwrites %u mreads %u mwrites %u\n", tinfo->nr,
+		mreads, mwrites, preads, pwrites);
+
+	return NULL;
+}
+
+int main(int argc, char **argv)
+{
+	pthread_t thread[5];
+	struct thread_info tinfo[5];
+	int fd[5];
+	int ret;
+	int i;
+
+	if (argc != 8) {
+		fprintf(stderr, "%s requires 7 arguments - size duration file1 file2 file3 file4 file5\n", argv[0]);
+		exit(-1);
+	}
+
+	size = atoi(argv[1]);
+	if (size <= 0) {
+		fprintf(stderr, "invalid size, must be greater than 0\n");
+		exit(-1);
+	}
+
+	duration = atoi(argv[2]);
+	if (duration < 0) {
+		fprintf(stderr, "invalid duration, must be greater than or equal to 0\n");
+		exit(-1);
+	}
+
+	/* create and truncate one fd */
+	fd[0] = open(argv[3], O_RDWR | O_CREAT | O_TRUNC, 00644);
+	if (fd[0] < 0) {
+		perror("open");
+		exit(-1);
+	}
+
+	/* make it the test size */
+	if (posix_fallocate(fd[0], 0, size) != 0) {
+		perror("fallocate");
+		exit(-1);
+	}
+
+	/* now open the rest of the fds */
+	for (i = 1; i < 5; i++) {
+		fd[i] = open(argv[3+i], O_RDWR);
+		if (fd[i] < 0) {
+			perror("open");
+			exit(-1);
+		}
+	}
+
+	/* start threads */
+	for (i = 0; i < 5; i++) {
+		tinfo[i].fd = fd[i];
+		tinfo[i].nr = i;
+		ret = pthread_create(&thread[i], NULL, run_test_func, (void*)&tinfo[i]);
+
+		if (ret) {
+			perror("pthread_create");
+			exit(-1);
+		}
+	}
+
+	/* wait for complete */
+	for (i = 0; i < 5; i++)
+		pthread_join(thread[i], NULL);
+
+	for (i = 0; i < 5; i++)
+		close(fd[i]);
+
+	exit(0);
+}

tests/src/mmap_validate.c

@@ -0,0 +1,159 @@
+#define _GNU_SOURCE
+/*
+ * mmap() content consistency checking for scoutfs
+ *
+ * This test program validates that content from memory mappings
+ * are consistent across clients, whether written/read with mmap or
+ * normal writes/reads.
+ *
+ * One side of (read/write) will always be memory mapped. It may
+ * be that both sides do memory mapped (33% of the time).
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <fcntl.h>
+#include <errno.h>
+
+static int count = 0;
+static int size = 0;
+
+static void run_test_func(int fd1, int fd2)
+{
+	void *buf1 = NULL;
+	void *buf2 = NULL;
+	char *addr1 = NULL;
+	char *addr2 = NULL;
+	int c = 0;
+	ssize_t read, written, ret;
+
+	/* buffers for both sides to compare */
+	if (posix_memalign(&buf1, 4096, size) != 0) {
+		perror("calloc1");
+		exit(-1);
+	}
+
+	if (posix_memalign(&buf2, 4096, size) != 0) {
+		perror("calloc1");
+		exit(-1);
+	}
+
+	/* memory maps for both sides */
+	addr1 = mmap(NULL, size, PROT_WRITE | PROT_READ, MAP_SHARED, fd1, 0);
+	if (addr1 == MAP_FAILED) {
+		perror("mmap1");
+		exit(-1);
+	}
+
+	addr2 = mmap(NULL, size, PROT_WRITE | PROT_READ, MAP_SHARED, fd2, 0);
+	if (addr2 == MAP_FAILED) {
+		perror("mmap2");
+		exit(-1);
+	}
+
+	for (;;) {
+		if (++c > count) /* 10k iterations */
+			break;
+
+		/* put a pattern in buf1 */
+		memset(buf1, c & 0xff, size);
+
+		/* pwrite or mmap write from buf1 */
+		switch (c % 3) {
+		case 0:	/* pwrite */
+			for (written = 0; written < size;) {
+				ret = pwrite(fd1, buf1, size - written, written);
+				if (ret < 0) {
+					perror("pwrite");
+					exit(-1);
+				}
+				written += ret;
+			}
+			break;
+		default: /* mmap write */
+			memcpy(addr1, buf1, size);
+			break;
+		}
+
+		/* pread or mmap read to buf2 */
+		switch (c % 3) {
+		case 2: /* pread */
+			for (read = 0; read < size;) {
+				ret = pread(fd2, buf2, size - read, read);
+				if (ret < 0) {
+					perror("pwrite");
+					exit(-1);
+				}
+				read += ret;
+			}
+			break;
+		default: /* mmap read */
+			memcpy(buf2, addr2, size);
+			break;
+		}
+
+		/* compare bufs */
+		if (memcmp(buf1, buf2, size) != 0) {
+			fprintf(stderr, "memcmp() failed\n");
+			exit(-1);
+		}
+	}
+
+	munmap(addr1, size);
+	munmap(addr2, size);
+
+	free(buf1);
+	free(buf2);
+}
+
+int main(int argc, char **argv)
+{
+	int fd[1];
+
+	if (argc != 5) {
+		fprintf(stderr, "%s requires 4 arguments - size count file1 file2\n", argv[0]);
+		exit(-1);
+	}
+
+	size = atoi(argv[1]);
+	if (size <= 0) {
+		fprintf(stderr, "invalid size, must be greater than 0\n");
+		exit(-1);
+	}
+
+	count = atoi(argv[2]);
+	if (count < 3) {
+		fprintf(stderr, "invalid count, must be greater than 3\n");
+		exit(-1);
+	}
+
+	/* create and truncate one fd */
+	fd[0] = open(argv[3], O_RDWR | O_CREAT | O_TRUNC, 00644);
+	if (fd[0] < 0) {
+		perror("open");
+		exit(-1);
+	}
+
+	fd[1] = open(argv[4], O_RDWR , 00644);
+	if (fd[1] < 0) {
+		perror("open");
+		exit(-1);
+	}
+
+	/* make it the test size */
+	if (posix_fallocate(fd[0], 0, size) != 0) {
+		perror("fallocate");
+		exit(-1);
+	}
+
+	/* run the test function */
+	run_test_func(fd[0], fd[1]);
+
+	close(fd[0]);
+	close(fd[1]);
+
+	exit(0);
+}

tests/src/parallel_restore.c

@@ -1,838 +0,0 @@
-#define _GNU_SOURCE /* O_DIRECT */
-#include <stdlib.h>
-#include <stdio.h>
-#include <stdbool.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/xattr.h>
-#include <ctype.h>
-#include <string.h>
-#include <errno.h>
-#include <limits.h>
-#include <time.h>
-#include <sys/prctl.h>
-#include <signal.h>
-#include <sys/socket.h>
-
-#include "../../utils/src/sparse.h"
-#include "../../utils/src/util.h"
-#include "../../utils/src/list.h"
-#include "../../utils/src/parse.h"
-#include "../../kmod/src/format.h"
-#include "../../utils/src/parallel_restore.h"
-
-/*
- * XXX:
- *  - add a nice description of what's going on
- *  - mention allocator contention
- *  - test child process dying handling
- *  - root dir entry name length is wrong
- */
-
-#define ERRF " errno %d (%s)"
-#define ERRA errno, strerror(errno)
-
-#define error_exit(cond, fmt, args...)			\
-do {							\
-	if (cond) {					\
-		printf("error: "fmt"\n", ##args);	\
-		exit(1);				\
-	}						\
-} while (0)
-
-#define dprintf(fmt, args...)		\
-do {					\
-	if (0)				\
-		printf(fmt, ##args);	\
-} while (0)
-
-#define REG_MODE (S_IFREG | 0644)
-#define DIR_MODE (S_IFDIR | 0755)
-
-struct opts {
-	unsigned long long buf_size;
-
-	unsigned long long write_batch;
-	unsigned long long low_dirs;
-	unsigned long long high_dirs;
-	unsigned long long low_files;
-	unsigned long long high_files;
-	char *meta_path;
-	unsigned long long total_files;
-	bool read_only;
-	unsigned long long seed;
-	unsigned long long nr_writers;
-};
-
-static void usage(void)
-{
-	printf("usage:\n"
-	       " -b NR       | threads write blocks in batches files (100000)\n"
-	       " -d LOW:HIGH | range of subdirs per directory (5:10)\n"
-	       " -f LOW:HIGH | range of files per directory (10:20)\n"
-	       " -m PATH     | path to metadata device\n"
-	       " -n NR       | total number of files to create (100)\n"
-	       " -r          | read-only, all work except writing, measure cpu cost\n"
-	       " -s NR       | randomization seed (random)\n"
-	       " -w NR       | number of writing processes to fork (online cpus)\n"
-	       );
-}
-
-static size_t write_bufs(struct opts *opts, struct scoutfs_parallel_restore_writer *wri,
-			 void *buf, size_t buf_size, int dev_fd)
-{
-	size_t total = 0;
-	size_t count;
-	off_t off;
-	int ret;
-
-	do {
-		ret = scoutfs_parallel_restore_write_buf(wri, buf, buf_size, &off, &count);
-		error_exit(ret, "write buf %d", ret);
-
-		if (count > 0) {
-			if (!opts->read_only)
-				ret = pwrite(dev_fd, buf, count, off);
-			else
-				ret = count;
-			error_exit(ret != count, "pwrite count %zu ret %d", count, ret);
-			total += ret;
-		}
-	} while (count > 0);
-
-	return total;
-}
-
-struct gen_inode {
-	struct scoutfs_parallel_restore_inode inode;
-	struct scoutfs_parallel_restore_xattr **xattrs;
-	u64 nr_xattrs;
-	struct scoutfs_parallel_restore_entry **entries;
-	u64 nr_files;
-	u64 nr_entries;
-};
-
-static void free_gino(struct gen_inode *gino)
-{
-	u64 i;
-
-	if (gino) {
-		if (gino->entries) {
-			for (i = 0; i < gino->nr_entries; i++)
-				free(gino->entries[i]);
-			free(gino->entries);
-		}
-		if (gino->xattrs) {
-			for (i = 0; i < gino->nr_xattrs; i++)
-				free(gino->xattrs[i]);
-			free(gino->xattrs);
-		}
-		free(gino);
-	}
-}
-
-static struct scoutfs_parallel_restore_xattr *
-generate_xattr(struct opts *opts, u64 ino, u64 pos, char *name, int name_len, void *value,
-		int value_len)
-{
-	struct scoutfs_parallel_restore_xattr *xattr;
-
-	xattr = malloc(sizeof(struct scoutfs_parallel_restore_xattr) + name_len + value_len);
-	error_exit(!xattr, "error allocating generated xattr");
-
-	*xattr = (struct scoutfs_parallel_restore_xattr) {
-		.ino = ino,
-		.pos = pos,
-		.name_len = name_len,
-		.value_len = value_len,
-	};
-
-	xattr->name = (void *)(xattr + 1);
-	xattr->value = (void *)(xattr->name + name_len);
-
-	memcpy(xattr->name, name, name_len);
-	if (value_len)
-		memcpy(xattr->value, value, value_len);
-
-	return xattr;
-}
-
-static struct gen_inode *generate_inode(struct opts *opts, u64 ino, mode_t mode)
-{
-	struct gen_inode *gino;
-	struct timespec now;
-
-	clock_gettime(CLOCK_REALTIME, &now);
-
-	gino = calloc(1, sizeof(struct gen_inode));
-	error_exit(!gino, "failure allocating generated inode");
-
-	gino->inode = (struct scoutfs_parallel_restore_inode) {
-		.ino = ino,
-		.meta_seq = ino,
-		.data_seq = 0,
-		.mode = mode,
-		.atime = now,
-		.ctime = now,
-		.mtime = now,
-		.crtime = now,
-	};
-
-	/*
-	 * hacky creation of a bunch of xattrs for now.
-	 */
-	if ((mode & S_IFMT) == S_IFREG) {
-		#define NV(n, v) { n, sizeof(n) - 1, v, sizeof(v) - 1, }
-		struct name_val {
-			char *name;
-			int len;
-			char *value;
-			int value_len;
-		} nv[] = {
-			NV("scoutfs.hide.totl.acct.8314611887310466424.2.0", "1"),
-			NV("scoutfs.hide.srch.sam_vol_E01001L6_4", ""),
-			NV("scoutfs.hide.sam_reqcopies", ""),
-			NV("scoutfs.hide.sam_copy_2", ""),
-			NV("scoutfs.hide.totl.acct.F01030L6.8314611887310466424.7.30", "1"),
-			NV("scoutfs.hide.sam_copy_1", ""),
-			NV("scoutfs.hide.srch.sam_vol_F01030L6_4", ""),
-			NV("scoutfs.hide.srch.sam_release_cand", ""),
-			NV("scoutfs.hide.sam_restime", ""),
-			NV("scoutfs.hide.sam_uuid", ""),
-			NV("scoutfs.hide.totl.acct.8314611887310466424.3.0", "1"),
-			NV("scoutfs.hide.srch.sam_vol_F01030L6", ""),
-			NV("scoutfs.hide.srch.sam_uuid_865939b7-24d6-472f-b85c-7ce7afeb813a", ""),
-			NV("scoutfs.hide.srch.sam_vol_E01001L6", ""),
-			NV("scoutfs.hide.totl.acct.E01001L6.8314611887310466424.7.1", "1"),
-			NV("scoutfs.hide.totl.acct.8314611887310466424.4.0", "1"),
-			NV("scoutfs.hide.totl.acct.8314611887310466424.11.0", "1"),
-			NV("scoutfs.hide.totl.acct.8314611887310466424.1.0", "1"),
-		};
-		unsigned int nr = array_size(nv);
-		int i;
-
-		gino->xattrs = calloc(nr, sizeof(struct scoutfs_parallel_restore_xattr *));
-
-		for (i = 0; i < nr; i++)
-			gino->xattrs[i] = generate_xattr(opts, ino, i, nv[i].name, nv[i].len,
-							 nv[i].value, nv[i].value_len);
-
-		gino->nr_xattrs = nr;
-		gino->inode.nr_xattrs = nr;
-
-		gino->inode.size = 4096;
-		gino->inode.offline = true;
-	}
-
-	return gino;
-}
-
-static struct scoutfs_parallel_restore_entry *
-generate_entry(struct opts *opts, char *prefix, u64 nr, u64 dir_ino, u64 pos, u64 ino, mode_t mode)
-{
-	struct scoutfs_parallel_restore_entry *entry;
-	char buf[PATH_MAX];
-	int bytes;
-
-	bytes = snprintf(buf, sizeof(buf), "%s-%llu", prefix, nr);
-
-	entry = malloc(sizeof(struct scoutfs_parallel_restore_entry) + bytes);
-	error_exit(!entry, "error allocating generated entry");
-
-	*entry = (struct scoutfs_parallel_restore_entry) {
-		.dir_ino = dir_ino,
-		.pos = pos,
-		.ino = ino,
-		.mode = mode,
-		.name = (void *)(entry + 1),
-		.name_len = bytes,
-	};
-
-	memcpy(entry->name, buf, bytes);
-
-	return entry;
-}
-
-/*
- * since the _parallel_restore_quota_rule mimics the squota_rule found in the
- * kernel we can also mimic its rule_to_irule function
- */
-
-#define TEST_RULE_STR "7 13,L,- 15,L,- 17,L,- I 33 -"
-
-static struct scoutfs_parallel_restore_quota_rule *
-generate_quota(struct opts *opts)
-{
-	struct scoutfs_parallel_restore_quota_rule *prule;
-	int err;
-
-	prule = calloc(1, sizeof(struct scoutfs_parallel_restore_quota_rule));
-	error_exit(!prule, "Quota rule alloc failed");
-
-	err = sscanf(TEST_RULE_STR, " %hhu %llu,%c,%c %llu,%c,%c %llu,%c,%c %c %llu %c",
-		     &prule->prio,
-			 &prule->names[0].val, &prule->names[0].source, &prule->names[0].flags,
-		     &prule->names[1].val, &prule->names[1].source, &prule->names[1].flags,
-			 &prule->names[2].val, &prule->names[2].source, &prule->names[2].flags,
-			 &prule->op, &prule->limit, &prule->rule_flags);
-	error_exit(err != 13, "invalid quota rule, missing fields. nr fields: %d rule str: %s\n", err, TEST_RULE_STR);
-
-	return prule;
-}
-
-static u64 random64(void)
-{
-	return ((u64)lrand48() << 32) | lrand48();
-}
-
-static u64 random_range(u64 low, u64 high)
-{
-	return low + (random64() % (high - low + 1));
-}
-
-static struct gen_inode *generate_dir(struct opts *opts, u64 dir_ino, u64 ino_start, u64 ino_len,
-				      bool no_dirs)
-{
-	struct scoutfs_parallel_restore_entry *entry;
-	struct gen_inode *gino;
-	u64 nr_entries;
-	u64 nr_files;
-	u64 nr_dirs;
-	u64 ino;
-	char *prefix;
-	mode_t mode;
-	u64 i;
-
-	nr_dirs = no_dirs ? 0 : random_range(opts->low_dirs, opts->high_dirs);
-	nr_files = random_range(opts->low_files, opts->high_files);
-
-	if (1 + nr_dirs + nr_files > ino_len) {
-		nr_dirs = no_dirs ? 0 : (ino_len - 1) / 2;
-		nr_files = (ino_len - 1) - nr_dirs;
-	}
-
-	nr_entries = nr_dirs + nr_files;
-
-	gino = generate_inode(opts, dir_ino, DIR_MODE);
-	error_exit(!gino, "error allocating generated inode");
-
-	gino->inode.nr_subdirs = nr_dirs;
-	gino->nr_files = nr_files;
-
-	if (nr_entries) {
-		gino->entries = calloc(nr_entries, sizeof(struct scoutfs_parallel_restore_entry *));
-		error_exit(!gino->entries, "error allocating generated inode entries");
-
-		gino->nr_entries = nr_entries;
-	}
-
-	mode = DIR_MODE;
-	prefix = "dir";
-	for (i = 0; i < nr_entries; i++) {
-		if (i == nr_dirs) {
-			mode = REG_MODE;
-			prefix = "file";
-		}
-
-		ino = ino_start + i;
-		entry = generate_entry(opts, prefix, ino, gino->inode.ino,
-				       SCOUTFS_DIRENT_FIRST_POS + i, ino, mode);
-
-		gino->entries[i] = entry;
-		gino->inode.total_entry_name_bytes += entry->name_len;
-	}
-
-	return gino;
-}
-
-/*
- * Restore a generated inode.  If it's a directory then we also restore
- * all its entries.  The caller is going to descend into subdir entries and generate
- * those dir inodes.  We have to generate and restore all non-dir inodes referenced
- * by this inode's entries.
- */
-static void restore_inode(struct opts *opts, struct scoutfs_parallel_restore_writer *wri,
-			  struct gen_inode *gino)
-{
-	struct gen_inode *nondir;
-	int ret;
-	u64 i;
-
-	ret = scoutfs_parallel_restore_add_inode(wri, &gino->inode);
-	error_exit(ret, "thread add root inode %d", ret);
-
-	for (i = 0; i < gino->nr_entries; i++) {
-		ret = scoutfs_parallel_restore_add_entry(wri, gino->entries[i]);
-		error_exit(ret, "thread add entry %d", ret);
-
-		/* caller only needs subdir entries, generate and free others */
-		if ((gino->entries[i]->mode & S_IFMT) != S_IFDIR) {
-
-			nondir = generate_inode(opts, gino->entries[i]->ino,
-						gino->entries[i]->mode);
-			restore_inode(opts, wri, nondir);
-			free_gino(nondir);
-
-			free(gino->entries[i]);
-			if (i != gino->nr_entries - 1)
-				gino->entries[i] = gino->entries[gino->nr_entries - 1];
-			gino->nr_entries--;
-			gino->nr_files--;
-			i--;
-		}
-	}
-
-	for (i = 0; i < gino->nr_xattrs; i++) {
-		ret = scoutfs_parallel_restore_add_xattr(wri, gino->xattrs[i]);
-		error_exit(ret, "thread add xattr %d", ret);
-	}
-}
-
-struct writer_args {
-	struct list_head head;
-
-	int dev_fd;
-	int pair_fd;
-
-	struct scoutfs_parallel_restore_slice slice;
-	u64 writer_nr;
-	u64 dir_height;
-	u64 ino_start;
-	u64 ino_len;
-};
-
-struct write_result {
-	struct scoutfs_parallel_restore_progress prog;
-	struct scoutfs_parallel_restore_slice slice;
-	__le64 files_created;
-	__le64 bytes_written;
-};
-
-static void write_bufs_and_send(struct opts *opts, struct scoutfs_parallel_restore_writer *wri,
-				  void *buf, size_t buf_size, int dev_fd,
-				  struct write_result *res, bool get_slice, int pair_fd)
-{
-	size_t total;
-	int ret;
-
-	total = write_bufs(opts, wri, buf, buf_size, dev_fd);
-	le64_add_cpu(&res->bytes_written, total);
-
-	ret = scoutfs_parallel_restore_get_progress(wri, &res->prog);
-	error_exit(ret, "get prog %d", ret);
-
-	if (get_slice) {
-		ret = scoutfs_parallel_restore_get_slice(wri, &res->slice);
-		error_exit(ret, "thread get slice %d", ret);
-	}
-
-	ret = write(pair_fd, res, sizeof(struct write_result));
-	error_exit(ret != sizeof(struct write_result), "result send error");
-
-	memset(res, 0, sizeof(struct write_result));
-}
-
-/*
- * Calculate the number of bytes in toplevel "dir-%llu" entry names for the given
- * number of writers.
- */
-static u64 topdir_entry_bytes(u64 nr_writers)
-{
-	u64 bytes = (3 + 1) * nr_writers;
-	u64 limit;
-	u64 done;
-	u64 wid;
-	u64 nr;
-
-	for (done = 0, wid = 1, limit = 10; done < nr_writers; done += nr, wid++, limit *= 10) {
-		nr = min(limit - done, nr_writers - done);
-		bytes += nr * wid;
-	}
-
-	return bytes;
-}
-
-struct dir_pos {
-	struct gen_inode *gino;
-	u64 pos;
-};
-
-static void writer_proc(struct opts *opts, struct writer_args *args)
-{
-	struct scoutfs_parallel_restore_writer *wri = NULL;
-	struct scoutfs_parallel_restore_entry *entry;
-	struct dir_pos *dirs = NULL;
-	struct write_result res;
-	struct gen_inode *gino;
-	void *buf = NULL;
-	u64 level;
-	u64 ino;
-	int ret;
-
-	memset(&res, 0, sizeof(res));
-
-	dirs = calloc(args->dir_height, sizeof(struct dir_pos));
-	error_exit(errno, "error allocating parent dirs "ERRF, ERRA);
-
-	errno = posix_memalign((void **)&buf, 4096, opts->buf_size);
-	error_exit(errno, "error allocating block buf "ERRF, ERRA);
-
-	ret = scoutfs_parallel_restore_create_writer(&wri);
-	error_exit(ret, "create writer %d", ret);
-
-	ret = scoutfs_parallel_restore_add_slice(wri, &args->slice);
-	error_exit(ret, "add slice %d", ret);
-
-	/* writer 0 creates the root dir */
-	if (args->writer_nr == 0) {
-		gino = generate_inode(opts, SCOUTFS_ROOT_INO, DIR_MODE);
-		gino->inode.nr_subdirs = opts->nr_writers;
-		gino->inode.total_entry_name_bytes = topdir_entry_bytes(opts->nr_writers);
-
-		ret = scoutfs_parallel_restore_add_inode(wri, &gino->inode);
-		error_exit(ret, "thread add root inode %d", ret);
-		free_gino(gino);
-	}
-
-	/* create root entry for our top level dir */
-	ino = args->ino_start++;
-	args->ino_len--;
-
-	entry = generate_entry(opts, "top", args->writer_nr,
-			       SCOUTFS_ROOT_INO, SCOUTFS_DIRENT_FIRST_POS + args->writer_nr,
-			       ino, DIR_MODE);
-
-	ret = scoutfs_parallel_restore_add_entry(wri, entry);
-	error_exit(ret, "thread top entry %d", ret);
-	free(entry);
-
-	level = args->dir_height - 1;
-
-	while (args->ino_len > 0 && level < args->dir_height) {
-		gino = dirs[level].gino;
-
-		/* generate and restore if we follow entries */
-		if (!gino) {
-			gino = generate_dir(opts, ino, args->ino_start, args->ino_len, level == 0);
-			args->ino_start += gino->nr_entries;
-			args->ino_len -= gino->nr_entries;
-			le64_add_cpu(&res.files_created, gino->nr_files);
-
-			restore_inode(opts, wri, gino);
-			dirs[level].gino = gino;
-		}
-
-		if (dirs[level].pos == gino->nr_entries) {
-			/* ascend if we're done with this dir */
-			dirs[level].gino = NULL;
-			dirs[level].pos = 0;
-			free_gino(gino);
-			level++;
-
-		} else {
-			/* otherwise descend into subdir entry */
-			ino = gino->entries[dirs[level].pos]->ino;
-			dirs[level].pos++;
-			level--;
-		}
-
-		/* do a partial write at batch intervals when there's still more to do */
-		if (le64_to_cpu(res.files_created) >= opts->write_batch && args->ino_len > 0)
-			write_bufs_and_send(opts, wri, buf, opts->buf_size, args->dev_fd,
-					    &res, false, args->pair_fd);
-	}
-
-	write_bufs_and_send(opts, wri, buf, opts->buf_size, args->dev_fd,
-			    &res, true, args->pair_fd);
-
-	scoutfs_parallel_restore_destroy_writer(&wri);
-
-	free(dirs);
-	free(buf);
-}
-
-/*
- * If any of our children exited with an error code, we hard exit.
- * The child processes should themselves report out any errors
- * encountered. Any remaining children will receive SIGHUP and
- * terminate.
- */
-static void sigchld_handler(int signo, siginfo_t *info, void *context)
-{
-	if (info->si_status)
-		exit(EXIT_FAILURE);
-}
-
-static void fork_writer(struct opts *opts, struct writer_args *args)
-{
-	pid_t parent = getpid();
-	pid_t pid;
-	int ret;
-
-	pid = fork();
-	error_exit(pid == -1, "fork error");
-
-	if (pid != 0)
-		return;
-
-	ret = prctl(PR_SET_PDEATHSIG, SIGHUP);
-	error_exit(ret < 0, "failed to set parent death sig");
-
-	printf("pid %u getpid() %u parent %u getppid() %u\n",
-		pid, getpid(), parent, getppid());
-	error_exit(getppid() != parent, "child parent already changed");
-
-	writer_proc(opts, args);
-	exit(0);
-}
-
-static int do_restore(struct opts *opts)
-{
-	struct scoutfs_parallel_restore_writer *wri = NULL;
-	struct scoutfs_parallel_restore_slice *slices = NULL;
-	struct scoutfs_parallel_restore_quota_rule *rule = NULL;
-	struct scoutfs_super_block *super = NULL;
-	struct write_result res;
-	struct writer_args *args;
-	struct timespec begin;
-	struct timespec end;
-	LIST_HEAD(writers);
-	u64 next_ino;
-	u64 ino_per;
-	u64 avg_dirs;
-	u64 avg_files;
-	u64 dir_height;
-	u64 tot_files;
-	u64 tot_bytes;
-	int pair[2] = {-1, -1};
-	float secs;
-	void *buf = NULL;
-	int dev_fd = -1;
-	int ret;
-	int i;
-
-	ret = socketpair(PF_LOCAL, SOCK_STREAM, 0, pair);
-	error_exit(ret, "socketpair error "ERRF, ERRA);
-
-	dev_fd = open(opts->meta_path, O_DIRECT | (opts->read_only ? O_RDONLY : (O_RDWR|O_EXCL)));
-	error_exit(dev_fd < 0, "error opening '%s': "ERRF, opts->meta_path, ERRA);
-
-	errno = posix_memalign((void **)&super, 4096, SCOUTFS_BLOCK_SM_SIZE) ?:
-		posix_memalign((void **)&buf, 4096, opts->buf_size);
-	error_exit(errno, "error allocating block bufs "ERRF, ERRA);
-
-	ret = pread(dev_fd, super, SCOUTFS_BLOCK_SM_SIZE,
-		    SCOUTFS_SUPER_BLKNO << SCOUTFS_BLOCK_SM_SHIFT);
-	error_exit(ret != SCOUTFS_BLOCK_SM_SIZE, "error reading super, ret %d", ret);
-
-	ret = scoutfs_parallel_restore_create_writer(&wri);
-	error_exit(ret, "create writer %d", ret);
-
-	ret = scoutfs_parallel_restore_import_super(wri, super, dev_fd);
-	error_exit(ret, "import super %d", ret);
-
-	rule = generate_quota(opts);
-	ret = scoutfs_parallel_restore_add_quota_rule(wri, rule);
-	free(rule);
-	error_exit(ret, "add quotas %d", ret);
-
-	slices = calloc(1 + opts->nr_writers, sizeof(struct scoutfs_parallel_restore_slice));
-	error_exit(!slices, "alloc slices");
-
-	scoutfs_parallel_restore_init_slices(wri, slices, 1 + opts->nr_writers);
-
-	ret = scoutfs_parallel_restore_add_slice(wri, &slices[0]);
-	error_exit(ret, "add slices[0] %d", ret);
-
-	next_ino = (SCOUTFS_ROOT_INO | SCOUTFS_LOCK_INODE_GROUP_MASK) + 1;
-	ino_per = opts->total_files / opts->nr_writers;
-	avg_dirs = (opts->low_dirs + opts->high_dirs) / 2;
-	avg_files = (opts->low_files + opts->high_files) / 2;
-
-	dir_height = 1;
-	tot_files = avg_files * opts->nr_writers;
-
-	while (tot_files < opts->total_files) {
-		dir_height++;
-		tot_files *= avg_dirs;
-	}
-
-	dprintf("height %llu tot %llu total %llu\n", dir_height, tot_files, opts->total_files);
-
-	clock_gettime(CLOCK_MONOTONIC_RAW, &begin);
-
-	/* start each writing process */
-	for (i = 0; i < opts->nr_writers; i++) {
-		args = calloc(1, sizeof(struct writer_args));
-		error_exit(!args, "alloc writer args");
-
-		args->dev_fd = dev_fd;
-		args->pair_fd = pair[1];
-		args->slice = slices[1 + i];
-		args->writer_nr = i;
-		args->dir_height = dir_height;
-		args->ino_start = next_ino;
-		args->ino_len = ino_per;
-
-		list_add_tail(&args->head, &writers);
-		next_ino += ino_per;
-
-		fork_writer(opts, args);
-	}
-
-	/* read results and watch for writers to finish */
-	tot_files = 0;
-	tot_bytes = 0;
-	i = 0;
-	while (i < opts->nr_writers) {
-		ret = read(pair[0], &res, sizeof(struct write_result));
-		error_exit(ret != sizeof(struct write_result), "result read error %d", ret);
-
-		ret = scoutfs_parallel_restore_add_progress(wri, &res.prog);
-		error_exit(ret, "add thr prog %d", ret);
-
-		if (res.slice.meta_len != 0) {
-			ret = scoutfs_parallel_restore_add_slice(wri, &res.slice);
-			error_exit(ret, "add thr slice %d", ret);
-			i++;
-		}
-
-		tot_files += le64_to_cpu(res.files_created);
-		tot_bytes += le64_to_cpu(res.bytes_written);
-	}
-
-	tot_bytes += write_bufs(opts, wri, buf, opts->buf_size, dev_fd);
-
-	ret = scoutfs_parallel_restore_export_super(wri, super);
-	error_exit(ret, "update super %d", ret);
-
-	if (!opts->read_only) {
-		ret = pwrite(dev_fd, super, SCOUTFS_BLOCK_SM_SIZE,
-			     SCOUTFS_SUPER_BLKNO << SCOUTFS_BLOCK_SM_SHIFT);
-		error_exit(ret != SCOUTFS_BLOCK_SM_SIZE, "error writing super, ret %d", ret);
-	}
-
-	clock_gettime(CLOCK_MONOTONIC_RAW, &end);
-
-	scoutfs_parallel_restore_destroy_writer(&wri);
-
-	secs = ((float)end.tv_sec + ((float)end.tv_nsec/NSEC_PER_SEC)) -
-	       ((float)begin.tv_sec + ((float)begin.tv_nsec/NSEC_PER_SEC));
-	printf("created %llu files in %llu bytes and %f secs => %f bytes/file, %f files/sec\n",
-		tot_files, tot_bytes, secs,
-		(float)tot_bytes / tot_files, (float)tot_files / secs);
-
-	if (dev_fd >= 0)
-		close(dev_fd);
-	if (pair[0] >= 0)
-		close(pair[0]);
-	if (pair[1] >= 0)
-		close(pair[1]);
-	free(super);
-	free(slices);
-	free(buf);
-
-	return 0;
-}
-
-static int parse_low_high(char *str, u64 *low_ret, u64 *high_ret)
-{
-	char *sep;
-	int ret = 0;
-
-	sep = index(str, ':');
-	if (sep) {
-		*sep = '\0';
-		ret = parse_u64(sep + 1, high_ret);
-	}
-
-	if (ret == 0)
-		ret = parse_u64(str, low_ret);
-
-	if (sep)
-		*sep = ':';
-
-	return ret;
-}
-
-int main(int argc, char **argv)
-{
-	struct opts opts = {
-		.buf_size = (32 * 1024 * 1024),
-
-		.write_batch = 1000000,
-		.low_dirs = 5,
-		.high_dirs = 10,
-		.low_files = 10,
-		.high_files = 20,
-		.total_files = 100,
-	};
-	struct sigaction act = { 0 };
-	int ret;
-	int c;
-
-	opts.seed = random64();
-	opts.nr_writers = sysconf(_SC_NPROCESSORS_ONLN);
-
-        while ((c = getopt(argc, argv, "b:d:f:m:n:rs:w:")) != -1) {
-                switch(c) {
-                case 'b':
-			ret = parse_u64(optarg, &opts.write_batch);
-			error_exit(ret, "error parsing -b '%s'\n", optarg);
-			error_exit(opts.write_batch == 0, "-b can't be 0");
-                        break;
-                case 'd':
-			ret = parse_low_high(optarg, &opts.low_dirs, &opts.high_dirs);
-			error_exit(ret, "error parsing -d '%s'\n", optarg);
-                        break;
-                case 'f':
-			ret = parse_low_high(optarg, &opts.low_files, &opts.high_files);
-			error_exit(ret, "error parsing -f '%s'\n", optarg);
-                        break;
-                case 'm':
-                        opts.meta_path = strdup(optarg);
-                        break;
-                case 'n':
-			ret = parse_u64(optarg, &opts.total_files);
-			error_exit(ret, "error parsing -n '%s'\n", optarg);
-                        break;
-                case 'r':
-			opts.read_only = true;
-			break;
-                case 's':
-			ret = parse_u64(optarg, &opts.seed);
-			error_exit(ret, "error parsing -s '%s'\n", optarg);
-                        break;
-                case 'w':
-			ret = parse_u64(optarg, &opts.nr_writers);
-			error_exit(ret, "error parsing -w '%s'\n", optarg);
-                        break;
-                case '?':
-                        printf("Unknown option '%c'\n", optopt);
-                        usage();
-			exit(1);
-                }
-        }
-
-	error_exit(opts.low_dirs > opts.high_dirs, "LOW > HIGH in -d %llu:%llu",
-		   opts.low_dirs, opts.high_dirs);
-	error_exit(opts.low_files > opts.high_files, "LOW > HIGH in -f %llu:%llu",
-		   opts.low_files, opts.high_files);
-	error_exit(!opts.meta_path, "must specify metadata device path with -m");
-
-	printf("recreate with: -d %llu:%llu -f %llu:%llu -n %llu -s %llu -w %llu\n",
-		opts.low_dirs, opts.high_dirs, opts.low_files, opts.high_files,
-		opts.total_files, opts.seed, opts.nr_writers);
-
-	act.sa_flags = SA_SIGINFO | SA_RESTART;
-	act.sa_sigaction = &sigchld_handler;
-	if (sigaction(SIGCHLD, &act, NULL) == -1)
-		error_exit(ret, "error setting up signal handler\n");
-
-	ret = do_restore(&opts);
-
-	free(opts.meta_path);
-
-	return ret == 0 ? 0 : 1;
-}

tests/src/restore_copy.c

@@ -1,817 +0,0 @@
-#define _GNU_SOURCE /* O_DIRECT */
-#include <stdlib.h>
-#include <stdio.h>
-#include <stdbool.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/types.h>
-#include <sys/xattr.h>
-#include <sys/stat.h>
-#include <sys/ioctl.h>
-#include <ctype.h>
-#include <string.h>
-#include <errno.h>
-#include <limits.h>
-#include <time.h>
-#include <sys/prctl.h>
-#include <sys/socket.h>
-#include <sys/signal.h>
-#include <sys/statfs.h>
-#include <dirent.h>
-
-#include "../../utils/src/sparse.h"
-#include "../../utils/src/util.h"
-#include "../../utils/src/list.h"
-#include "../../utils/src/parse.h"
-#include "../../kmod/src/format.h"
-#include "../../kmod/src/ioctl.h"
-#include "../../utils/src/parallel_restore.h"
-
-/*
- * XXX:
- */
-
-#define ERRF " errno %d (%s)"
-#define ERRA errno, strerror(errno)
-
-#define error_exit(cond, fmt, args...)			\
-do {							\
-	if (cond) {					\
-		printf("error: "fmt"\n", ##args);	\
-		exit(1);				\
-	}						\
-} while (0)
-
-#define REG_MODE (S_IFREG | 0644)
-#define DIR_MODE (S_IFDIR | 0755)
-#define LNK_MODE (S_IFLNK | 0777)
-
-/*
- * At about 1k files we seem to be writing about 1MB of data, so
- * set buffer sizes adequately above that.
- */
-#define BATCH_FILES 1024
-#define BUF_SIZ 2 * 1024 * 1024
-
-/*
- * We can't make duplicate inodes for hardlinked files, so we
- * will need to track these as we generate them. Not too costly
- * to do, since it's just an integer, and sorting shouldn't matter
- * until we get into the millions of entries, hopefully.
- */
-static struct list_head hardlinks;
-struct hardlink_head {
-	struct list_head head;
-	u64 ino;
-};
-
-struct opts {
-	char *meta_path;
-	char *source_dir;
-};
-
-static bool warn_scoutfs = false;
-
-static void usage(void)
-{
-	printf("usage:\n"
-	       " -m PATH     | path to metadata device\n"
-	       " -s PATH     | path to source directory\n"
-	       );
-}
-
-static size_t write_bufs(struct scoutfs_parallel_restore_writer *wri,
-			 void *buf, int dev_fd)
-{
-	size_t total = 0;
-	size_t count;
-	off_t off;
-	int ret;
-
-	do {
-		ret = scoutfs_parallel_restore_write_buf(wri, buf, BUF_SIZ, &off, &count);
-		error_exit(ret, "write buf %d", ret);
-
-		if (count > 0) {
-			ret = pwrite(dev_fd, buf, count, off);
-			error_exit(ret != count, "pwrite count %zu ret %d", count, ret);
-			total += ret;
-		}
-	} while (count > 0);
-
-	return total;
-}
-
-struct write_result {
-	struct scoutfs_parallel_restore_progress prog;
-	struct scoutfs_parallel_restore_slice slice;
-	__le64 files_created;
-	__le64 dirs_created;
-	__le64 bytes_written;
-	bool complete;
-};
-
-static void write_bufs_and_send(struct scoutfs_parallel_restore_writer *wri,
-				void *buf, int dev_fd,
-				struct write_result *res, bool get_slice, int pair_fd)
-{
-	size_t total;
-	int ret;
-
-	total = write_bufs(wri, buf, dev_fd);
-	le64_add_cpu(&res->bytes_written, total);
-
-	ret = scoutfs_parallel_restore_get_progress(wri, &res->prog);
-	error_exit(ret, "get prog %d", ret);
-
-	if (get_slice) {
-		ret = scoutfs_parallel_restore_get_slice(wri, &res->slice);
-		error_exit(ret, "thread get slice %d", ret);
-	}
-
-	ret = write(pair_fd, res, sizeof(struct write_result));
-	error_exit(ret != sizeof(struct write_result), "result send error");
-
-	memset(res, 0, sizeof(struct write_result));
-}
-
-/*
- * Adding xattrs is supported for files and directories only.
- *
- * If the filesystem on which the path resides isn't scoutfs, we omit the
- * scoutfs specific ioctl to fetch hidden xattrs.
- *
- * Untested if the hidden xattr ioctl works on directories or symlinks.
- */
-static void add_xattrs(struct scoutfs_parallel_restore_writer *wri, char *path, u64 ino, bool is_scoutfs)
-{
-	struct scoutfs_ioctl_listxattr_hidden lxh;
-	struct scoutfs_parallel_restore_xattr *xattr;
-	char *buf = NULL;
-	char *name = NULL;
-	int fd = -1;
-	int bytes;
-	int len;
-	int value_len;
-	int ret;
-	int pos = 0;
-
-	if (!is_scoutfs)
-		goto normal_xattrs;
-
-	fd = open(path, O_RDONLY);
-	error_exit(fd < 0, "open"ERRF, ERRA);
-
-	memset(&lxh, 0, sizeof(lxh));
-	lxh.id_pos = 0;
-	lxh.hash_pos = 0;
-	lxh.buf_bytes = 256 * 1024;
-
-	buf = malloc(lxh.buf_bytes);
-	error_exit(!buf, "alloc xattr_hidden buf");
-	lxh.buf_ptr = (unsigned long)buf;
-
-	/* hidden */
-	for (;;) {
-		ret = ioctl(fd, SCOUTFS_IOC_LISTXATTR_HIDDEN, &lxh);
-		if (ret == 0) /* done */
-			break;
-		error_exit(ret < 0, "listxattr_hidden"ERRF, ERRA);
-		bytes = ret;
-		error_exit(bytes > lxh.buf_bytes, "listxattr_hidden overflow");
-		error_exit(buf[bytes - 1] != '\0', "listxattr_hidden didn't term");
-
-		name = buf;
-
-		do {
-			len = strlen(name);
-			error_exit(len == 0, "listxattr_hidden empty name");
-			error_exit(len > SCOUTFS_XATTR_MAX_NAME_LEN, "listxattr_hidden long name");
-
-			/* get value len */
-			value_len = fgetxattr(fd, name, NULL, 0);
-			error_exit(value_len < 0, "malloc value hidden"ERRF, ERRA);
-
-			/* allocate everything at once */
-			xattr = malloc(sizeof(struct scoutfs_parallel_restore_xattr) + len + value_len);
-			error_exit(!xattr, "error allocating generated xattr");
-
-			*xattr = (struct scoutfs_parallel_restore_xattr) {
-				.ino = ino,
-				.pos = pos++,
-				.name_len = len,
-				.value_len = value_len,
-			};
-			xattr->name = (void *)(xattr + 1);
-			xattr->value = (void *)(xattr->name + len);
-
-			/* get value into xattr directly */
-			ret = fgetxattr(fd, name, (void *)(xattr->name + len), value_len);
-			error_exit(ret != value_len, "fgetxattr value"ERRF, ERRA);
-
-			memcpy(xattr->name, name, len);
-
-			ret = scoutfs_parallel_restore_add_xattr(wri, xattr);
-			error_exit(ret, "add hidden xattr %d", ret);
-
-			free(xattr);
-
-			name += len + 1;
-			bytes -= len + 1;
-		} while (bytes > 0);
-	}
-
-	free(buf);
-	close(fd);
-
-normal_xattrs:
-	value_len = listxattr(path, NULL, 0);
-	error_exit(value_len < 0, "hidden listxattr "ERRF, ERRA);
-	if (value_len == 0)
-		return;
-
-	buf = calloc(1, value_len);
-	error_exit(!buf, "malloc value"ERRF, ERRA);
-
-	ret = listxattr(path, buf, value_len);
-	error_exit(ret < 0, "hidden listxattr %d", ret);
-
-	name = buf;
-	bytes = ret;
-	do {
-		len = strlen(name);
-
-		error_exit(len == 0, "listxattr_hidden empty name");
-		error_exit(len > SCOUTFS_XATTR_MAX_NAME_LEN, "listxattr_hidden long name");
-
-		value_len = getxattr(path, name, NULL, 0);
-		error_exit(value_len < 0, "value "ERRF, ERRA);
-
-		xattr = malloc(sizeof(struct scoutfs_parallel_restore_xattr) + len + value_len);
-		error_exit(!xattr, "error allocating generated xattr");
-
-		*xattr = (struct scoutfs_parallel_restore_xattr) {
-			.ino = ino,
-			.pos = pos++,
-			.name_len = len,
-			.value_len = value_len,
-		};
-		xattr->name = (void *)(xattr + 1);
-		xattr->value = (void *)(xattr->name + len);
-
-		ret = getxattr(path, name, (void *)(xattr->name + len), value_len);
-		error_exit(ret != value_len, "fgetxattr value"ERRF, ERRA);
-
-		memcpy(xattr->name, name, len);
-
-		ret = scoutfs_parallel_restore_add_xattr(wri, xattr);
-		error_exit(ret, "add xattr %d", ret);
-
-		free(xattr);
-
-		name += len + 1;
-		bytes -= len + 1;
-	} while (bytes > 0);
-
-	free(buf);
-}
-
-/*
- * We can't store the same inode multiple times, so we need to make
- * sure to account for hardlinks. Maintain a LL that stores the first
- * hardlink inode we encounter, and every subsequent hardlink to this
- * inode will omit inserting an inode, and just adds another entry
- */
-static bool is_new_inode_item(bool nlink, u64 ino)
-{
-	struct hardlink_head *hh_tmp;
-	struct hardlink_head *hh;
-
-	if (!nlink)
-		return true;
-
-	/* lineair search, pretty awful, should be a binary tree */
-	list_for_each_entry_safe(hh, hh_tmp, &hardlinks, head) {
-		if (hh->ino == ino)
-			return false;
-	}
-
-	/* insert item */
-	hh = malloc(sizeof(struct hardlink_head));
-	error_exit(!hh, "malloc");
-	hh->ino = ino;
-	list_add_tail(&hh->head, &hardlinks);
-
-	/*
-	 *  XXX
-	 *
-	 * We can be confident that if we don't traverse filesystems
-	 * that once we've created N entries of an N-linked inode, that
-	 * it can be removed from the LL. This would significantly
-	 * improve the manageability of the list.
-	 *
-	 * All we'd need to do is add a counter and compare it to the nr_links
-	 * field of the inode.
-	 */
-
-	return true;
-}
-
-/*
- * create the inode data for a given path as best as possible
- * duplicating the exact data from the source path
- */
-static struct scoutfs_parallel_restore_inode *read_inode_data(char *path, u64 ino, bool *nlink, bool is_scoutfs)
-{
-	struct scoutfs_parallel_restore_inode *inode = NULL;
-	struct scoutfs_ioctl_stat_more stm;
-	struct stat st;
-	int ret;
-	int fd;
-
-	inode = calloc(1, sizeof(struct scoutfs_parallel_restore_inode));
-	error_exit(!inode, "failure allocating inode");
-
-	ret = lstat(path, &st);
-	error_exit(ret, "failure stat inode");
-
-	/* use exact inode numbers from path, except for root ino */
-	if (ino != SCOUTFS_ROOT_INO)
-		inode->ino = st.st_ino;
-	else
-		inode->ino = SCOUTFS_ROOT_INO;
-
-	inode->mode = st.st_mode;
-	inode->uid = st.st_uid;
-	inode->gid = st.st_gid;
-	inode->atime = st.st_atim;
-	inode->ctime = st.st_ctim;
-	inode->mtime = st.st_mtim;
-	inode->size = st.st_size;
-
-	inode->rdev = st.st_rdev;
-
-	/* scoutfs specific */
-	inode->meta_seq = 0;
-	inode->data_seq = 0;
-	inode->crtime = st.st_ctim;
-
-	if (S_ISREG(inode->mode)) {
-		if (inode->size > 0)
-			inode->offline = true;
-
-		if (is_scoutfs) {
-			fd = open(path, O_RDONLY);
-			error_exit(!fd, "open failure"ERRF, ERRA);
-
-			ret = ioctl(fd, SCOUTFS_IOC_STAT_MORE, &stm);
-			error_exit(ret, "failure SCOUTFS_IOC_STAT_MORE inode");
-
-			inode->meta_seq = stm.meta_seq;
-			inode->data_seq = stm.data_seq;
-			inode->crtime = (struct timespec){.tv_sec = stm.crtime_sec, .tv_nsec = stm.crtime_nsec};
-
-			close(fd);
-		}
-
-	}
-
-	/* pass whether item is hardlinked or not */
-	*nlink = (st.st_nlink > 1);
-
-	return inode;
-}
-
-struct writer_args {
-	struct list_head head;
-
-	int dev_fd;
-	int pair_fd;
-
-	struct scoutfs_parallel_restore_slice slice;
-};
-
-static void restore_path(struct scoutfs_parallel_restore_writer *wri, struct writer_args *args, struct write_result *res, void *buf, char *path, u64 ino)
-{
-	struct scoutfs_parallel_restore_inode *inode;
-	struct scoutfs_parallel_restore_entry *entry;
-	DIR *dirp = NULL;
-	char *subdir = NULL;
-	char link[PATH_MAX + 1];
-	struct dirent *ent;
-	struct statfs stf;
-	int ret = 0;
-	int subdir_count = 0, file_count = 0;
-	size_t ent_len = 0;
-	size_t pos = 0;
-	bool nlink = false;
-	char ind = '?';
-	u64 mode;
-	bool is_scoutfs = false;
-
-	/* get fs info once per path */
-	ret = statfs(path, &stf);
-	error_exit(ret != 0, "statfs"ERRF, ERRA);
-	is_scoutfs = (stf.f_type == 0x554f4353);
-
-	if (!is_scoutfs && !warn_scoutfs) {
-		warn_scoutfs = true;
-		fprintf(stderr, "Non-scoutfs source path detected: scoutfs specific features disabled\n");
-	}
-
-	/* traverse the entire tree */
-	dirp = opendir(path);
-	errno = 0;
-	while ((ent = readdir(dirp))) {
-		if (ent->d_type == DT_DIR) {
-			if ((strcmp(ent->d_name, ".") == 0) ||
-			    (strcmp(ent->d_name, "..") == 0)) {
-				/* position still matters */
-				pos++;
-				continue;
-			}
-
-			/* recurse into subdir */
-			ret = asprintf(&subdir, "%s/%s", path, ent->d_name);
-			error_exit(ret == -1, "asprintf subdir"ERRF, ERRA);
-			restore_path(wri, args, res, buf, subdir, ent->d_ino);
-
-			subdir_count++;
-
-			ent_len += strlen(ent->d_name);
-
-			entry = malloc(sizeof(struct scoutfs_parallel_restore_entry) + strlen(ent->d_name));
-			error_exit(!entry, "error allocating generated entry");
-
-			*entry = (struct scoutfs_parallel_restore_entry) {
-				.dir_ino = ino,
-				.pos = pos++,
-				.ino = ent->d_ino,
-				.mode = DIR_MODE,
-				.name = (void *)(entry + 1),
-				.name_len = strlen(ent->d_name),
-			};
-
-			memcpy(entry->name, ent->d_name, strlen(ent->d_name));
-			ret = scoutfs_parallel_restore_add_entry(wri, entry);
-			error_exit(ret, "add entry %d", ret);
-			free(entry);
-
-			add_xattrs(wri, subdir, ent->d_ino, is_scoutfs);
-
-			free(subdir);
-
-			le64_add_cpu(&res->dirs_created, 1);
-		} else if (ent->d_type == DT_REG) {
-
-			file_count++;
-
-			ent_len += strlen(ent->d_name);
-
-			entry = malloc(sizeof(struct scoutfs_parallel_restore_entry) + strlen(ent->d_name));
-			error_exit(!entry, "error allocating generated entry");
-
-			*entry = (struct scoutfs_parallel_restore_entry) {
-				.dir_ino = ino,
-				.pos = pos++,
-				.ino = ent->d_ino,
-				.mode = REG_MODE,
-				.name = (void *)(entry + 1),
-				.name_len = strlen(ent->d_name),
-			};
-
-			memcpy(entry->name, ent->d_name, strlen(ent->d_name));
-			ret = scoutfs_parallel_restore_add_entry(wri, entry);
-			error_exit(ret, "add entry %d", ret);
-			free(entry);
-
-			ret = asprintf(&subdir, "%s/%s", path, ent->d_name);
-			error_exit(ret == -1, "asprintf subdir"ERRF, ERRA);
-
-			/* file inode */
-			inode = read_inode_data(subdir, ent->d_ino, &nlink, is_scoutfs);
-			fprintf(stdout, "f %s/%s\n", path, ent->d_name);
-			if (is_new_inode_item(nlink, ent->d_ino)) {
-				ret = scoutfs_parallel_restore_add_inode(wri, inode);
-				error_exit(ret, "add reg file inode %d", ret);
-
-				/* xattrs */
-				add_xattrs(wri, subdir, ent->d_ino, is_scoutfs);
-			}
-			free(inode);
-
-			free(subdir);
-
-			le64_add_cpu(&res->files_created, 1);
-		} else if (ent->d_type == DT_LNK) {
-			/* readlink */
-
-			ret = asprintf(&subdir, "%s/%s", path, ent->d_name);
-			error_exit(ret == -1, "asprintf subdir"ERRF, ERRA);
-
-			ent_len += strlen(ent->d_name);
-
-			ret = readlink(subdir, link, PATH_MAX);
-			error_exit(ret < 0, "readlink %d", ret);
-			/* must 0-terminate if we want to print it */
-			link[ret] = 0;
-
-			entry = malloc(sizeof(struct scoutfs_parallel_restore_entry) + strlen(ent->d_name));
-			error_exit(!entry, "error allocating generated entry");
-
-			*entry = (struct scoutfs_parallel_restore_entry) {
-				.dir_ino = ino,
-				.pos = pos++,
-				.ino = ent->d_ino,
-				.mode = LNK_MODE,
-				.name = (void *)(entry + 1),
-				.name_len = strlen(ent->d_name),
-			};
-
-			memcpy(entry->name, ent->d_name, strlen(ent->d_name));
-			ret = scoutfs_parallel_restore_add_entry(wri, entry);
-			error_exit(ret, "add symlink entry %d", ret);
-
-			/* link inode */
-			inode = read_inode_data(subdir, ent->d_ino, &nlink, is_scoutfs);
-
-			fprintf(stdout, "l %s/%s -> %s\n", path, ent->d_name, link);
-
-			inode->mode = LNK_MODE;
-			inode->target = link;
-			inode->target_len = strlen(link) + 1; /* scoutfs null terminates symlinks */
-
-			ret = scoutfs_parallel_restore_add_inode(wri, inode);
-			error_exit(ret, "add syml inode %d", ret);
-
-			free(inode);
-			free(subdir);
-
-			le64_add_cpu(&res->files_created, 1);
-		} else {
-			/* odd stuff */
-			switch(ent->d_type) {
-			case DT_CHR:
-				ind = 'c';
-				mode = S_IFCHR;
-				break;
-			case DT_BLK:
-				ind = 'b';
-				mode = S_IFBLK;
-				break;
-			case DT_FIFO:
-				ind = 'p';
-				mode = S_IFIFO;
-				break;
-			case DT_SOCK:
-				ind = 's';
-				mode = S_IFSOCK;
-				break;
-			default:
-				error_exit(true, "Unknown readdir entry type");
-				;;
-			}
-
-			file_count++;
-
-			ent_len += strlen(ent->d_name);
-
-			entry = malloc(sizeof(struct scoutfs_parallel_restore_entry) + strlen(ent->d_name));
-			error_exit(!entry, "error allocating generated entry");
-
-			*entry = (struct scoutfs_parallel_restore_entry) {
-				.dir_ino = ino,
-				.pos = pos++,
-				.ino = ent->d_ino,
-				.mode = mode,
-				.name = (void *)(entry + 1),
-				.name_len = strlen(ent->d_name),
-			};
-
-			memcpy(entry->name, ent->d_name, strlen(ent->d_name));
-			ret = scoutfs_parallel_restore_add_entry(wri, entry);
-			error_exit(ret, "add entry %d", ret);
-
-			free(entry);
-
-			ret = asprintf(&subdir, "%s/%s", path, ent->d_name);
-			error_exit(ret == -1, "asprintf subdir"ERRF, ERRA);
-
-			/* file inode */
-			inode = read_inode_data(subdir, ent->d_ino, &nlink, is_scoutfs);
-			fprintf(stdout, "%c %lld %s/%s\n", ind, inode->ino, path, ent->d_name);
-			if (is_new_inode_item(nlink, ent->d_ino)) {
-				ret = scoutfs_parallel_restore_add_inode(wri, inode);
-				error_exit(ret, "add reg file inode %d", ret);
-			}
-			free(inode);
-
-			free(subdir);
-
-			le64_add_cpu(&res->files_created, 1);
-		}
-
-		/* batch out changes, will be about 1M */
-		if (le64_to_cpu(res->files_created) > BATCH_FILES) {
-			write_bufs_and_send(wri, buf, args->dev_fd, res, false, args->pair_fd);
-		}
-
-	}
-	if (ent != NULL)
-		error_exit(errno, "readdir"ERRF, ERRA);
-	closedir(dirp);
-
-	/* create the dir itself */
-	inode = read_inode_data(path, ino, &nlink, is_scoutfs);
-	inode->nr_subdirs = subdir_count;
-	inode->total_entry_name_bytes = ent_len;
-	fprintf(stdout, "d %s\n", path);
-
-	ret = scoutfs_parallel_restore_add_inode(wri, inode);
-	error_exit(ret, "add dir inode %d", ret);
-
-	free(inode);
-
-	/* No need to send, we'll send final after last directory is complete */
-}
-
-static int do_restore(struct opts *opts)
-{
-	struct scoutfs_parallel_restore_writer *pwri, *wri = NULL;
-	struct scoutfs_parallel_restore_slice *slices = NULL;
-	struct scoutfs_super_block *super = NULL;
-	struct writer_args *args;
-	struct write_result res;
-	int pair[2] = {-1, -1};
-	LIST_HEAD(writers);
-	void *buf = NULL;
-	void *bufp = NULL;
-	int dev_fd = -1;
-	pid_t pid;
-	int ret;
-	u64 tot_bytes;
-	u64 tot_dirs;
-	u64 tot_files;
-
-	ret = socketpair(PF_LOCAL, SOCK_STREAM, 0, pair);
-	error_exit(ret, "socketpair error "ERRF, ERRA);
-
-	dev_fd = open(opts->meta_path, O_DIRECT | (O_RDWR|O_EXCL));
-	error_exit(dev_fd < 0, "error opening '%s': "ERRF, opts->meta_path, ERRA);
-
-	errno = posix_memalign((void **)&super, 4096, SCOUTFS_BLOCK_SM_SIZE) ?:
-		posix_memalign((void **)&buf, 4096, BUF_SIZ);
-	error_exit(errno, "error allocating block bufs "ERRF, ERRA);
-
-	ret = pread(dev_fd, super, SCOUTFS_BLOCK_SM_SIZE,
-		    SCOUTFS_SUPER_BLKNO << SCOUTFS_BLOCK_SM_SHIFT);
-	error_exit(ret != SCOUTFS_BLOCK_SM_SIZE, "error reading super, ret %d", ret);
-
-	error_exit((super->flags & SCOUTFS_FLAG_IS_META_BDEV) == 0, "super block is not meta dev");
-
-	ret = scoutfs_parallel_restore_create_writer(&wri);
-	error_exit(ret, "create writer %d", ret);
-
-	ret = scoutfs_parallel_restore_import_super(wri, super, dev_fd);
-	error_exit(ret, "import super %d", ret);
-
-	slices = calloc(2, sizeof(struct scoutfs_parallel_restore_slice));
-	error_exit(!slices, "alloc slices");
-
-	scoutfs_parallel_restore_init_slices(wri, slices, 2);
-
-	ret = scoutfs_parallel_restore_add_slice(wri, &slices[0]);
-	error_exit(ret, "add slices[0] %d", ret);
-
-	args = calloc(1, sizeof(struct writer_args));
-	error_exit(!args, "alloc writer args");
-
-	args->dev_fd = dev_fd;
-	args->slice = slices[1];
-	args->pair_fd = pair[1];
-	list_add_tail(&args->head, &writers);
-
-	/* fork writer process */
-	pid = fork();
-	error_exit(pid == -1, "fork error");
-
-	if (pid == 0) {
-		ret = prctl(PR_SET_PDEATHSIG, SIGHUP);
-		error_exit(ret < 0, "failed to set parent death sig");
-
-		errno = posix_memalign((void **)&bufp, 4096, BUF_SIZ);
-		error_exit(errno, "error allocating block bufp "ERRF, ERRA);
-
-		ret = scoutfs_parallel_restore_create_writer(&pwri);
-		error_exit(ret, "create pwriter %d", ret);
-
-		ret = scoutfs_parallel_restore_add_slice(pwri, &args->slice);
-		error_exit(ret, "add pslice %d", ret);
-
-		memset(&res, 0, sizeof(res));
-
-		restore_path(pwri, args, &res, bufp, opts->source_dir, SCOUTFS_ROOT_INO);
-
-		res.complete = true;
-
-		write_bufs_and_send(pwri, buf, args->dev_fd, &res, true, args->pair_fd);
-
-		scoutfs_parallel_restore_destroy_writer(&pwri);
-		free(bufp);
-
-		exit(0);
-	};
-
-	/* read results and wait for writer to finish */
-	tot_bytes = 0;
-	tot_dirs = 1;
-	tot_files = 0;
-	for (;;) {
-		ret = read(pair[0], &res, sizeof(struct write_result));
-		error_exit(ret != sizeof(struct write_result), "result read error %d", ret);
-
-		ret = scoutfs_parallel_restore_add_progress(wri, &res.prog);
-		error_exit(ret, "add thr prog %d", ret);
-
-		if (res.slice.meta_len != 0) {
-			ret = scoutfs_parallel_restore_add_slice(wri, &res.slice);
-			error_exit(ret, "add thr slice %d", ret);
-
-			if (res.complete)
-				break;
-		}
-
-		tot_bytes += le64_to_cpu(res.bytes_written);
-		tot_files += le64_to_cpu(res.files_created);
-		tot_dirs += le64_to_cpu(res.dirs_created);
-	}
-
-	tot_bytes += write_bufs(wri, buf, args->dev_fd);
-
-	fprintf(stdout, "Wrote %lld directories, %lld files, %lld bytes total\n",
-		tot_dirs, tot_files, tot_bytes);
-
-	/* write super to finalize */
-	ret = scoutfs_parallel_restore_export_super(wri, super);
-	error_exit(ret, "update super %d", ret);
-
-	ret = pwrite(dev_fd, super, SCOUTFS_BLOCK_SM_SIZE,
-		     SCOUTFS_SUPER_BLKNO << SCOUTFS_BLOCK_SM_SHIFT);
-	error_exit(ret != SCOUTFS_BLOCK_SM_SIZE, "error writing super, ret %d", ret);
-
-	scoutfs_parallel_restore_destroy_writer(&wri);
-
-	if (dev_fd >= 0)
-		close(dev_fd);
-	if (pair[0] > 0)
-		close(pair[0]);
-	if (pair[1] > 0)
-		close(pair[1]);
-	free(super);
-	free(args);
-	free(slices);
-	free(buf);
-
-	return 0;
-}
-
-int main(int argc, char **argv)
-{
-	struct opts opts = (struct opts){ 0 };
-	struct hardlink_head *hh_tmp;
-	struct hardlink_head *hh;
-	int ret;
-	int c;
-
-	INIT_LIST_HEAD(&hardlinks);
-
-        while ((c = getopt(argc, argv, "b:m:s:")) != -1) {
-                switch(c) {
-                case 'm':
-                        opts.meta_path = strdup(optarg);
-                        break;
-		case 's':
-			opts.source_dir = strdup(optarg);
-			break;
-                case '?':
-                        printf("Unknown option '%c'\n", optopt);
-                        usage();
-			exit(1);
-                }
-        }
-
-	error_exit(!opts.meta_path, "must specify metadata device path with -m");
-	error_exit(!opts.source_dir, "must specify source directory path with -s");
-
-	ret = do_restore(&opts);
-
-	free(opts.meta_path);
-	free(opts.source_dir);
-
-	list_for_each_entry_safe(hh, hh_tmp, &hardlinks, head) {
-		list_del_init(&hh->head);
-		free(hh);
-	}
-
-	return ret == 0 ? 0 : 1;
-}

tests/tests/basic-acl-consistency.sh

@@ -0,0 +1,117 @@
+
+#
+# Test basic clustered posix acl consistency.
+#
+
+t_require_commands getfacl setfacl
+
+GETFACL="getfacl --absolute-names"
+
+filter_scratch() {
+	sed "s@$T_MSCR@t_mscr@g"
+}
+
+acl_compare()
+{
+	diff -u - <($GETFACL $T_MSCR/data/dir_a/dir_b | filter_scratch) <<EOF1
+# file: t_mscr/data/dir_a/dir_b
+# owner: t_usr_3
+# group: t_grp_3
+# flags: -s-
+user::rwx
+group::rwx
+group:t_grp_2:r-x
+mask::rwx
+other::---
+default:user::rwx
+default:group::rwx
+default:group:t_grp_2:r-x
+default:group:t_grp_3:rwx
+default:mask::rwx
+default:other::---
+
+EOF1
+
+	test $? -eq 0 || t_fail "dir_b differs"
+
+	diff -u - <($GETFACL -p $T_MSCR/data/dir_a/dir_b/dir_c/dir_d | filter_scratch) <<EOF3
+# file: t_mscr/data/dir_a/dir_b/dir_c/dir_d
+# owner: t_usr_1
+# group: t_grp_1
+# flags: -s-
+user::rwx
+group::rwx
+group:t_grp_2:r-x
+mask::rwx
+other::---
+default:user::rwx
+default:group::rwx
+default:group:t_grp_2:r-x
+default:group:t_grp_3:rwx
+default:mask::rwx
+default:other::---
+
+EOF3
+	test $? -eq 0 || t_fail "dir_d differs"
+
+	diff -u - <($GETFACL $T_MSCR/data/dir_a/dir_b/dir_c | filter_scratch) <<EOF2
+# file: t_mscr/data/dir_a/dir_b/dir_c
+# owner: t_usr_3
+# group: t_grp_2
+# flags: -s-
+user::rwx
+group::rwx
+group:t_grp_2:r-x
+mask::rwx
+other::---
+default:user::rwx
+default:group::rwx
+default:group:t_grp_2:r-x
+default:group:t_grp_3:rwx
+default:mask::rwx
+default:other::---
+
+EOF2
+	test $? -eq 0 || t_fail "dir_c differs"
+}
+echo "== make scratch fs"
+t_scratch_mkfs
+t_scratch_mount
+
+rm -rf $T_MSCR/data
+
+echo "== create uid/gids"
+groupadd -g 7101 t_grp_1 > /dev/null 2>&1
+useradd -g 7101 -u 7101 t_usr_1 > /dev/null 2>&1
+groupadd -g 7102 t_grp_2 > /dev/null 2>&1
+groupadd -g 7103 t_grp_3 > /dev/null 2>&1
+useradd -g 7103 -u 7103 t_usr_3 > /dev/null 2>&1
+
+echo "== set acls and permissions"
+mkdir -p $T_MSCR/data/dir_a/dir_b
+chown t_usr_3:t_grp_3 $T_MSCR/data/dir_a/dir_b
+chmod 2770 $T_MSCR/data/dir_a/dir_b
+setfacl -m g:t_grp_2:rx $T_MSCR/data/dir_a/dir_b
+setfacl -m d:g:t_grp_2:rx $T_MSCR/data/dir_a/dir_b
+setfacl -m d:g:t_grp_3:rwx $T_MSCR/data/dir_a/dir_b
+
+mkdir -p $T_MSCR/data/dir_a/dir_b/dir_c
+chown t_usr_3:t_grp_2 $T_MSCR/data/dir_a/dir_b/dir_c
+setfacl -x g:t_grp_3 $T_MSCR/data/dir_a/dir_b/dir_c
+
+mkdir -p $T_MSCR/data/dir_a/dir_b/dir_c/dir_d
+chown t_usr_1:t_grp_1 $T_MSCR/data/dir_a/dir_b/dir_c/dir_d
+setfacl -x g:t_grp_3 $T_MSCR/data/dir_a/dir_b/dir_c/dir_d
+
+echo "== compare output"
+acl_compare
+
+echo "== drop caches and compare again"
+sync
+echo 3 > /proc/sys/vm/drop_caches
+acl_compare
+
+echo "== cleanup scratch fs"
+t_scratch_umount
+
+t_pass

tests/tests/basic-bad-mounts.sh

@@ -12,25 +12,22 @@ mount_fail()
 }
 
 echo "== prepare devices, mount point, and logs"
-SCR="$T_TMPDIR/mnt.scratch"
-mkdir -p "$SCR"
+t_scratch_mkfs
 > $T_TMP.mount.out
-scoutfs mkfs -f -Q 0,127.0.0.1,53000 "$T_EX_META_DEV" "$T_EX_DATA_DEV" > $T_TMP.mkfs.out 2>&1 \
-	|| t_fail "mkfs failed"
 
 echo "== bad devices, bad options"
-mount_fail -o _bad /dev/null /dev/null "$SCR"
+mount_fail -o _bad /dev/null /dev/null "$T_MSCR"
 
 echo "== swapped devices"
-mount_fail -o metadev_path=$T_EX_DATA_DEV,quorum_slot_nr=0 "$T_EX_META_DEV" "$SCR"
+mount_fail -o metadev_path=$T_EX_DATA_DEV,quorum_slot_nr=0 "$T_EX_META_DEV" "$T_MSCR"
 
 echo "== both meta devices"
-mount_fail -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_META_DEV" "$SCR"
+mount_fail -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_META_DEV" "$T_MSCR"
 
 echo "== both data devices"
-mount_fail -o metadev_path=$T_EX_DATA_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$SCR"
+mount_fail -o metadev_path=$T_EX_DATA_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$T_MSCR"
 
 echo "== good volume, bad option and good options"
-mount_fail -o _bad,metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$SCR" 
+mount_fail -o _bad,metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$T_MSCR"
 
 t_pass

tests/tests/basic-truncate.sh

@@ -11,7 +11,7 @@ FILE="$T_D0/file"
 # final block as we truncated past it.
 #
 echo "== truncate writes zeroed partial end of file block"
-yes | dd of="$FILE" bs=8K count=1 status=none iflag=fullblock
+yes 2>/dev/null | dd of="$FILE" bs=8K count=1 status=none iflag=fullblock
 sync
 
 # not passing iflag=fullblock causes the file occasionally to just be

tests/tests/change-devices.sh

@@ -11,9 +11,8 @@ truncate -s $sz "$T_TMP.equal"
 truncate -s $large_sz "$T_TMP.large"
 
 echo "== make scratch fs"
-t_quiet scoutfs mkfs -f -Q 0,127.0.0.1,53000 "$T_EX_META_DEV" "$T_EX_DATA_DEV"
-SCR="$T_TMPDIR/mnt.scratch"
-mkdir -p "$SCR"
+t_scratch_mkfs
+mkdir -p "$T_MSCR"
 
 echo "== small new data device fails"
 t_rc scoutfs prepare-empty-data-device "$T_EX_META_DEV" "$T_TMP.small"
@@ -23,13 +22,13 @@ t_rc scoutfs prepare-empty-data-device --check "$T_EX_META_DEV" "$T_TMP.small"
 t_rc scoutfs prepare-empty-data-device --check "$T_EX_META_DEV"
 
 echo "== preparing while mounted fails"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$SCR"
+mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$T_MSCR"
 t_rc scoutfs prepare-empty-data-device "$T_EX_META_DEV" "$T_TMP.equal"
-umount "$SCR"
+umount "$T_MSCR"
 
 echo "== preparing without recovery fails"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$SCR"
-umount -f "$SCR"
+mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$T_MSCR"
+umount -f "$T_MSCR"
 t_rc scoutfs prepare-empty-data-device "$T_EX_META_DEV" "$T_TMP.equal"
 
 echo "== check sees metadata errors"
@@ -37,16 +36,16 @@ t_rc scoutfs prepare-empty-data-device --check "$T_EX_META_DEV"
 t_rc scoutfs prepare-empty-data-device --check "$T_EX_META_DEV" "$T_TMP.equal"
 
 echo "== preparing with file data fails"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$SCR"
-echo hi > "$SCR"/file
-umount "$SCR"
+mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$T_MSCR"
+echo hi > "$T_MSCR"/file
+umount "$T_MSCR"
 scoutfs print "$T_EX_META_DEV" > "$T_TMP.print"
 t_rc scoutfs prepare-empty-data-device "$T_EX_META_DEV" "$T_TMP.equal"
 
 echo "== preparing after emptied"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$SCR"
-rm -f "$SCR"/file
-umount "$SCR"
+mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$T_EX_DATA_DEV" "$T_MSCR"
+rm -f "$T_MSCR"/file
+umount "$T_MSCR"
 t_rc scoutfs prepare-empty-data-device "$T_EX_META_DEV" "$T_TMP.equal"
 
 echo "== checks pass"
@@ -55,22 +54,22 @@ t_rc scoutfs prepare-empty-data-device --check "$T_EX_META_DEV" "$T_TMP.equal"
 
 echo "== using prepared"
 scr_loop=$(losetup --find --show "$T_TMP.equal")
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$scr_loop" "$SCR"
-touch "$SCR"/equal_prepared
-equal_tot=$(scoutfs statfs -s total_data_blocks -p "$SCR")
-umount "$SCR"
+mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$scr_loop" "$T_MSCR"
+touch "$T_MSCR"/equal_prepared
+equal_tot=$(scoutfs statfs -s total_data_blocks -p "$T_MSCR")
+umount "$T_MSCR"
 losetup -d "$scr_loop"
 
 echo "== preparing larger and resizing"
 t_rc scoutfs prepare-empty-data-device "$T_EX_META_DEV" "$T_TMP.large"
 scr_loop=$(losetup --find --show "$T_TMP.large")
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$scr_loop" "$SCR"
-touch "$SCR"/large_prepared
-ls "$SCR"
-scoutfs resize-devices -p "$SCR" -d $large_sz
-large_tot=$(scoutfs statfs -s total_data_blocks -p "$SCR")
+mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 "$scr_loop" "$T_MSCR"
+touch "$T_MSCR"/large_prepared
+ls "$T_MSCR"
+scoutfs resize-devices -p "$T_MSCR" -d $large_sz
+large_tot=$(scoutfs statfs -s total_data_blocks -p "$T_MSCR")
 test "$large_tot" -gt "$equal_tot" ; echo "resized larger test rc: $?"
-umount "$SCR"
+umount "$T_MSCR"
 losetup -d "$scr_loop"
 
 echo "== cleanup"

tests/tests/enospc.sh

@@ -54,21 +54,16 @@ after=$(free_blocks Data "$T_M0")
 test "$before" == "$after" || \
 	t_fail "$after free data blocks after rm, expected $before"
 
-# XXX this is all pretty manual, would be nice to have helpers
 echo "== make small meta fs"
 # meta device just big enough for reserves and the metadata we'll fill
-scoutfs mkfs -A -f -Q 0,127.0.0.1,53000 -m 10G "$T_EX_META_DEV" "$T_EX_DATA_DEV" > $T_TMP.mkfs.out 2>&1 || \
-	t_fail "mkfs failed"
-SCR="$T_TMPDIR/mnt.scratch"
-mkdir -p "$SCR"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 \
-	"$T_EX_DATA_DEV" "$SCR"
+t_scratch_mkfs -A -m 10G
+t_scratch_mount
 
 echo "== create large xattrs until we fill up metadata"
-mkdir -p "$SCR/xattrs"
+mkdir -p "$T_MSCR/xattrs"
 
 for f in $(seq 1 100000); do
-	file="$SCR/xattrs/file-$f"
+	file="$T_MSCR/xattrs/file-$f"
 	touch "$file"
 
 	LC_ALL=C create_xattr_loop -c 1000 -n user.scoutfs-enospc -p "$file" -s 65535 > $T_TMP.cxl 2>&1
@@ -84,17 +79,21 @@ for f in $(seq 1 100000); do
 done
 
 echo "== remove files with xattrs after enospc"
-rm -rf "$SCR/xattrs"
+rm -rf "$T_MSCR/xattrs"
 
 echo "== make sure we can create again"
-file="$SCR/file-after"
+file="$T_MSCR/file-after"
+C=120
+while (( C-- )); do
+	touch $file 2> /dev/null && break
+	sleep 1
+done
 touch $file
 setfattr -n user.scoutfs-enospc -v 1 "$file"
 sync
 rm -f "$file"
 
 echo "== cleanup small meta fs"
-umount "$SCR"
-rmdir "$SCR"
+t_scratch_umount
 
 t_pass

tests/tests/fence-and-reclaim.sh

@@ -5,6 +5,9 @@
 t_require_commands sleep touch grep sync scoutfs
 t_require_mounts 2
 
+# regularly see ~20/~30s
+VERIFY_TIMEOUT_SECS=90
+
 #
 # Make sure that all mounts can read the results of a write from each
 # mount.
@@ -40,8 +43,10 @@ verify_fenced_run()
 
 	for rid in $rids; do
 		grep -q ".* running rid '$rid'.* args 'ignored run args'" "$T_FENCED_LOG" || \
-			t_fail "fenced didn't execute RUN script for rid $rid"
+			return 1
 	done
+
+	return 0
 }
 
 echo "== make sure all mounts can see each other"
@@ -54,14 +59,7 @@ rid=$(t_mount_rid $cl)
 echo "cl $cl sv $sv rid $rid" >> "$T_TMP.log"
 sync
 t_force_umount $cl
-# wait for client reconnection to timeout
-while grep -q $rid $(t_debugfs_path $sv)/connections; do
-	sleep .5
-done
-while t_rid_is_fencing $rid; do
-	sleep .5
-done
-verify_fenced_run $rid
+t_wait_until_timeout $VERIFY_TIMEOUT_SECS verify_fenced_run $rid
 t_mount $cl
 check_read_write
 
@@ -83,15 +81,7 @@ for cl in $(t_fs_nrs); do
 	t_force_umount $cl
 done
 
-# wait for all client reconnections to timeout
-while egrep -q "($pattern)" $(t_debugfs_path $sv)/connections; do
-	sleep .5
-done
-# wait for all fence requests to complete
-while test -d $(echo /sys/fs/scoutfs/*/fence/* | cut -d " " -f 1); do
-	sleep .5
-done
-verify_fenced_run $rids
+t_wait_until_timeout $VERIFY_TIMEOUT_SECS verify_fenced_run $rids
 # remount all the clients
 for cl in $(t_fs_nrs); do
 	if [ $cl == $sv ]; then
@@ -107,12 +97,7 @@ rid=$(t_mount_rid $sv)
 echo "sv $sv rid $rid" >> "$T_TMP.log"
 sync
 t_force_umount $sv
-t_wait_for_leader
-# wait until new server is done fencing unmounted leader rid
-while t_rid_is_fencing $rid; do
-	sleep .5
-done
-verify_fenced_run $rid
+t_wait_until_timeout $VERIFY_TIMEOUT_SECS verify_fenced_run $rid
 t_mount $sv
 check_read_write
 
@@ -127,11 +112,7 @@ for nr in $(t_fs_nrs); do
 	t_force_umount $nr
 done
 t_mount_all
-# wait for all fence requests to complete
-while test -d $(echo /sys/fs/scoutfs/*/fence/* | cut -d " " -f 1); do
-	sleep .5
-done
-verify_fenced_run $rids
+t_wait_until_timeout $VERIFY_TIMEOUT_SECS verify_fenced_run $rids
 check_read_write
 
 t_pass

tests/tests/format-version-forward-back.sh

@@ -11,8 +11,8 @@
 # format version.
 #
 
-# not supported on el9!
-if [ $(source /etc/os-release ; echo ${VERSION_ID:0:1}) -gt 8 ]; then
+# not supported on el8 or higher
+if [ $(source /etc/os-release ; echo ${VERSION_ID:0:1}) -gt 7 ]; then
 	t_skip_permitted "Unsupported OS version"
 fi
 
@@ -89,7 +89,7 @@ for vers in $(seq $MIN $((MAX - 1))); do
 	old_module="$builds/$vers/scoutfs.ko"
 
 	echo "mkfs $vers" >> "$T_TMP.log"
-	t_quiet $old_scoutfs mkfs -f -Q 0,127.0.0.1,53000 "$T_EX_META_DEV" "$T_EX_DATA_DEV" \
+	t_quiet $old_scoutfs mkfs -f -Q 0,127.0.0.1,$T_SCRATCH_PORT "$T_EX_META_DEV" "$T_EX_DATA_DEV" \
 		|| t_fail "mkfs $vers failed"
 
 	echo "mount $vers with $vers" >> "$T_TMP.log"

tests/tests/get-referring-entries.sh

@@ -72,7 +72,7 @@ touch $T_D0/dir/file
 mkdir $T_D0/dir/dir
 ln -s $T_D0/dir/file $T_D0/dir/symlink
 mknod $T_D0/dir/char c 1 3 # null
-mknod $T_D0/dir/block b 7 0 # loop0
+mknod $T_D0/dir/block b 42 0 # SAMPLE block dev - nonexistant/demo use only number
 for name in $(ls -UA $T_D0/dir | sort); do
 	ino=$(stat -c '%i' $T_D0/dir/$name)
 	$GRE $ino | filter_types

tests/tests/inode-deletion.sh

@@ -61,18 +61,28 @@ rm -f "$T_D1/file"
 check_ino_index "$ino" "$dseq" "$T_M0"
 check_ino_index "$ino" "$dseq" "$T_M1"
 
+# Hurry along the orphan scanners. If any are currently asleep, we will
+# have to wait at least their current scan interval before they wake up,
+# run, and notice their new interval.
+t_save_all_sysfs_mount_options orphan_scan_delay_ms
+t_set_all_sysfs_mount_options orphan_scan_delay_ms 500
+t_wait_for_orphan_scan_runs
+
 echo "== unlink wait for open on other mount"
-echo "contents" > "$T_D0/file"
-ino=$(stat -c "%i" "$T_D0/file")
-dseq=$(scoutfs stat -s data_seq "$T_D0/file")
-exec {FD}<"$T_D0/file"
-rm -f "$T_D1/file"
+echo "contents" > "$T_D0/badfile"
+ino=$(stat -c "%i" "$T_D0/badfile")
+dseq=$(scoutfs stat -s data_seq "$T_D0/badfile")
+exec {FD}<"$T_D0/badfile"
+rm -f "$T_D1/badfile"
 echo "mount 0 contents after mount 1 rm: $(cat <&$FD)"
 check_ino_index "$ino" "$dseq" "$T_M0"
 check_ino_index "$ino" "$dseq" "$T_M1"
 exec {FD}>&-  # close
 # we know that revalidating will unhash the remote dentry
-stat "$T_D0/file" 2>&1 | sed 's/cannot statx/cannot stat/' | t_filter_fs
+stat "$T_D0/badfile" 2>&1 | sed 's/cannot statx/cannot stat/' | t_filter_fs
+t_force_log_merge
+# wait for orphan scanners to pick up the unlinked inode and become idle
+t_wait_for_no_orphans
 check_ino_index "$ino" "$dseq" "$T_M0"
 check_ino_index "$ino" "$dseq" "$T_M1"
 
@@ -83,16 +93,20 @@ rm -f "$T_D0/dir"/files-*
 rmdir "$T_D0/dir"
 
 echo "== open files survive remote scanning orphans"
-echo "contents" > "$T_D0/file"
-ino=$(stat -c "%i" "$T_D0/file")
-dseq=$(scoutfs stat -s data_seq "$T_D0/file")
-exec {FD}<"$T_D0/file"
-rm -f "$T_D0/file"
+echo "contents" > "$T_D0/lastfile"
+ino=$(stat -c "%i" "$T_D0/lastfile")
+dseq=$(scoutfs stat -s data_seq "$T_D0/lastfile")
+exec {FD}<"$T_D0/lastfile"
+rm -f "$T_D0/lastfile"
 t_umount 1
 t_mount 1
 echo "mount 0 contents after mount 1 remounted: $(cat <&$FD)"
 exec {FD}>&-  # close
+t_force_log_merge
+t_wait_for_no_orphans
 check_ino_index "$ino" "$dseq" "$T_M0"
 check_ino_index "$ino" "$dseq" "$T_M1"
 
+t_restore_all_sysfs_mount_options orphan_scan_delay_ms
+
 t_pass

tests/tests/large-fragmented-free.sh

@@ -10,30 +10,6 @@ EXTENTS_PER_BTREE_BLOCK=600
 EXTENTS_PER_LIST_BLOCK=8192
 FREED_EXTENTS=$((EXTENTS_PER_BTREE_BLOCK * EXTENTS_PER_LIST_BLOCK))
 
-#
-# This test specifically creates a pathologically sparse file that will
-# be as expensive as possible to free.  This is usually fine on
-# dedicated or reasonable hardware, but trying to run this in
-# virtualized debug kernels can take a very long time.  This test is
-# about making sure that the server doesn't fail, not that the platform
-# can handle the scale of work that our btree formats happen to require
-# while execution is bogged down with use-after-free memory reference
-# tracking.  So we give the test a lot more breathing room before
-# deciding that its hung.
-#
-echo "== setting longer hung task timeout"
-if [ -w /proc/sys/kernel/hung_task_timeout_secs ]; then
-	secs=$(cat /proc/sys/kernel/hung_task_timeout_secs)
-	test "$secs" -gt 0 || \
-		t_fail "confusing value '$secs' from /proc/sys/kernel/hung_task_timeout_secs"
-	restore_hung_task_timeout()
-	{
-		echo "$secs" > /proc/sys/kernel/hung_task_timeout_secs
-	}
-	trap restore_hung_task_timeout EXIT
-	echo "$((secs * 5))" > /proc/sys/kernel/hung_task_timeout_secs
-fi
-
 echo "== creating fragmented extents"
 fragmented_data_extents $FREED_EXTENTS $EXTENTS_PER_BTREE_BLOCK "$T_D0/alloc" "$T_D0/move"
 

tests/tests/lock-recover-invalidate.sh

@@ -38,6 +38,6 @@ while [ "$SECONDS" -lt "$END" ]; do
 done
 
 echo "== stopping background load"
-kill $load_pids
+t_silent_kill $load_pids
 
 t_pass

tests/tests/lock-shrink-read-race.sh

@@ -0,0 +1,40 @@
+#
+# We had a lock server refcounting bug that could let one thread get a
+# reference on a lock struct that was being freed by another thread.  We
+# were able to reproduce this by having all clients try and produce a
+# lot of read and null requests.
+#
+# This will manfiest as a hung lock and timed out test runs, probably
+# with hung task messages on the console.  Depending on how the race
+# turns out, it can trigger KASAN warnings in
+# process_waiting_requests().
+#
+
+READERS_PER=3
+SECS=30
+
+echo "=== setup"
+touch "$T_D0/file"
+
+echo "=== spin reading and shrinking"
+END=$((SECONDS + SECS))
+for m in $(t_fs_nrs); do
+	eval file="\$T_D${m}/file"
+
+	# lots of tasks reading as fast as they can
+	for t in $(seq 1 $READERS_PER); do
+		(while [ $SECONDS -lt $END ]; do
+			stat $file > /dev/null
+		 done) &
+	done
+	# one task shrinking (triggering null requests) and reading
+	(while [ $SECONDS -lt $END ]; do
+		stat $file > /dev/null
+		t_trigger_arm_silent statfs_lock_purge $m
+		stat -f "$file" > /dev/null
+	 done) &
+done
+
+wait
+
+t_pass

tests/tests/mmap.sh

@@ -0,0 +1,54 @@
+#
+# test mmap() and normal read/write consistency between different nodes
+#
+
+t_require_commands mmap_stress mmap_validate scoutfs xfs_io
+
+echo "== mmap_stress"
+mmap_stress 8192 30 "$T_D0/mmap_stress" "$T_D0/mmap_stress" "$T_D0/mmap_stress" "$T_D3/mmap_stress" "$T_D3/mmap_stress" | sed 's/:.*//g' | sort
+
+echo "== basic mmap/read/write consistency checks"
+mmap_validate 256 1000 "$T_D0/mmap_val1" "$T_D1/mmap_val1"
+mmap_validate 8192 1000 "$T_D0/mmap_val2" "$T_D1/mmap_val2"
+mmap_validate 88400 1000 "$T_D0/mmap_val3" "$T_D1/mmap_val3"
+
+echo "== mmap read from offline extent"
+F="$T_D0/mmap-offline"
+touch "$F"
+xfs_io -c "pwrite -S 0xEA 0 8192" "$F" > /dev/null
+cp "$F" "${F}-stage"
+vers=$(scoutfs stat -s data_version "$F")
+scoutfs release "$F" -V "$vers" -o 0 -l 8192
+scoutfs get-fiemap -L "$F"
+xfs_io -c "mmap -rwx 0 8192" \
+	-c "mread -v 512 16" "$F" &
+sleep 1
+# should be 1 - data waiting
+jobs | wc -l
+scoutfs stage "${F}-stage" "$F" -V "$vers" -o 0 -l 8192
+# xfs_io thread <here> will output 16 bytes of read data
+sleep 1
+# should be 0 - no more waiting jobs, xfs_io should have exited
+jobs | wc -l
+scoutfs get-fiemap -L "$F"
+
+echo "== mmap write to an offline extent"
+# reuse the same file
+scoutfs release "$F" -V "$vers" -o 0 -l 8192
+scoutfs get-fiemap -L "$F"
+xfs_io -c "mmap -rwx 0 8192" \
+	-c "mwrite -S 0x11 528 16" "$F" &
+sleep 1
+# should be 1 job waiting
+jobs | wc -l
+scoutfs stage "${F}-stage" "$F" -V "$vers" -o 0 -l 8192
+# no output here from write
+sleep 1
+# should be 0 - no more waiting jobs, xfs_io should have exited
+jobs | wc -l
+scoutfs get-fiemap -L "$F"
+# read back contents to assure write changed the file
+dd status=none if="$F" bs=1 count=48 skip=512 | hexdump -C
+
+echo "== done"
+t_pass

tests/tests/offline-extent-waiting.sh

@@ -157,7 +157,7 @@ echo "truncate should be waiting for first block:"
 expect_wait "$DIR/file" "change_size" $ino 0
 scoutfs stage "$DIR/golden" "$DIR/file" -V "$vers" -o 0 -l $BYTES
 sleep .1
-echo "trunate should no longer be waiting:"
+echo "truncate should no longer be waiting:"
 scoutfs data-waiting -B 0 -I 0 -p "$DIR" | wc -l
 cat "$DIR/golden" > "$DIR/file"
 vers=$(scoutfs stat -s data_version "$DIR/file")
@@ -168,10 +168,13 @@ scoutfs release "$DIR/file" -V "$vers" -o 0 -l $BYTES
 # overwrite, not truncate+write
 dd if="$DIR/other" of="$DIR/file" \
 	bs=$BS count=$BLOCKS conv=notrunc status=none &
+pid="$!"
 sleep .1
 echo "should be waiting for write"
 expect_wait "$DIR/file" "write" $ino 0
 scoutfs stage "$DIR/golden" "$DIR/file" -V "$vers" -o 0 -l $BYTES
+# wait for the background dd to complete
+wait "$pid" 2> /dev/null
 cmp "$DIR/file" "$DIR/other"
 
 echo "== cleanup"

tests/tests/orphan-inodes.sh

@@ -5,18 +5,6 @@
 t_require_commands sleep touch sync stat handle_cat kill rm
 t_require_mounts 2
 
-#
-# usually bash prints an annoying output message when jobs
-# are killed.  We can avoid that by redirecting stderr for
-# the bash process when it reaps the jobs that are killed.
-#
-silent_kill() {
-	exec {ERR}>&2 2>/dev/null
-	kill "$@"
-	wait "$@"
-	exec 2>&$ERR {ERR}>&-
-}
-
 #
 # We don't have a great way to test that inode items still exist.   We
 # don't prevent opening handles with nlink 0 today, so we'll use that.
@@ -52,7 +40,7 @@ inode_exists $ino || echo "$ino didn't exist"
 
 echo "== orphan from failed evict deletion is picked up"
 # pending kill signal stops evict from getting locks and deleting
-silent_kill $pid
+t_silent_kill $pid
 t_set_sysfs_mount_option 0 orphan_scan_delay_ms 1000
 sleep 5
 inode_exists $ino && echo "$ino still exists"
@@ -70,7 +58,7 @@ for nr in $(t_fs_nrs); do
 	rm -f "$path"
 done
 sync
-silent_kill $pids
+t_silent_kill $pids
 for nr in $(t_fs_nrs); do
 	t_force_umount $nr
 done
@@ -79,10 +67,49 @@ t_mount_all
 while test -d $(echo /sys/fs/scoutfs/*/fence/* | cut -d " " -f 1); do
 	sleep .5
 done
+
+
+sv=$(t_server_nr)
+
+# wait for reclaim_open_log_tree() to complete for each mount
+while [ $(t_counter reclaimed_open_logs $sv) -lt $T_NR_MOUNTS ]; do
+	sleep 1
+done
+
+# wait for finalize_and_start_log_merge() to find no active merges in flight
+# and not find any finalized trees
+while [ $(t_counter log_merge_no_finalized $sv) -lt 1 ]; do
+	sleep 1
+done
+
 # wait for orphan scans to run
 t_set_all_sysfs_mount_options orphan_scan_delay_ms 1000
-# also have to wait for delayed log merge work from mount
-sleep 15
+# wait until we see two consecutive orphan scan attempts without
+# any inode deletion forward progress in each mount
+for nr in $(t_fs_nrs); do
+	C=0
+	LOSA=$(t_counter orphan_scan_attempts $nr)
+	LDOP=$(t_counter inode_deleted $nr)
+
+	while [ $C -lt 2 ]; do
+		sleep 1
+
+		OSA=$(t_counter orphan_scan_attempts $nr)
+		DOP=$(t_counter inode_deleted $nr)
+
+		if [ $OSA != $LOSA ]; then
+			if [ $DOP == $LDOP ]; then
+				(( C++ ))
+			else
+				C=0
+			fi
+		fi
+
+		LOSA=$OSA
+		LDOP=$DOP
+	done
+done
+
 for ino in $inos; do
 	inode_exists $ino && echo "$ino still exists"
 done
@@ -131,7 +158,7 @@ while [ $SECONDS -lt $END ]; do
 	done
 
 	# trigger eviction deletion of each file in each mount
-	silent_kill $pids
+	t_silent_kill $pids
 
 	wait || t_fail "handle_fsetxattr failed"
 

tests/tests/parallel_restore.sh

@@ -1,78 +0,0 @@
-#
-# validate parallel restore library
-#
-
-t_require_commands scoutfs parallel_restore find xargs
-
-SCR="$T_TMPDIR/mnt.scratch"
-mkdir -p "$SCR"
-
-scratch_mkfs() {
-	scoutfs mkfs $@ \
-		-A -f -Q 0,127.0.0.1,53000 $T_EX_META_DEV $T_EX_DATA_DEV
-}
-
-scratch_check() {
-	# give ample time for writes to commit
-	sleep 1
-	sync
-	scoutfs check -d ${T_TMPDIR}/check.debug $T_EX_META_DEV $T_EX_DATA_DEV
-}
-
-scratch_mount() {
-	mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 $T_EX_DATA_DEV $SCR
-}
-
-
-echo "== simple mkfs/restore/mount"
-# meta device just big enough for reserves and the metadata we'll fill
-scratch_mkfs -V 2 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-parallel_restore -m "$T_EX_META_DEV" > /dev/null || t_fail "parallel_restore"
-scratch_check || t_fail "check failed"
-scratch_mount
-
-scoutfs statfs -p "$SCR" | grep -v -e 'fsid' -e 'rid'
-find "$SCR" -exec scoutfs list-hidden-xattrs {} \; | wc
-scoutfs search-xattrs -p "$SCR" scoutfs.hide.srch.sam_vol_F01030L6 -p "$SCR" | wc
-find "$SCR" -type f -name "file-*" | head -n 4 | xargs -n 1 scoutfs get-fiemap -L
-scoutfs df -p "$SCR"
-scoutfs quota-list -p "$SCR"
-umount "$SCR"
-scratch_check || t_fail "check after mount failed"
-
-echo "== just under ENOSPC"
-scratch_mkfs -V 2 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-parallel_restore -m "$T_EX_META_DEV" -n 3000000 > /dev/null || t_fail "parallel_restore"
-scratch_check || t_fail "check failed"
-scratch_mount
-scoutfs df -p "$SCR"
-umount "$SCR"
-scratch_check || t_fail "check after mount failed"
-
-echo "== just over ENOSPC"
-scratch_mkfs -V 2 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-parallel_restore -m "$T_EX_META_DEV" -n 3500000 | grep died 2>&1 && t_fail "parallel_restore"
-scratch_check || t_fail "check failed"
-
-echo "== ENOSPC"
-scratch_mkfs -V 2 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-parallel_restore -m "$T_EX_META_DEV" -d 600:1000 -f 600:1000 -n 4000000 | grep died 2>&1 && t_fail "parallel_restore"
-
-echo "== attempt to restore data device"
-scratch_mkfs -V 2 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-parallel_restore -m "$T_EX_DATA_DEV" | grep died 2>&1 && t_fail "parallel_restore"
-
-echo "== attempt format_v1 restore"
-scratch_mkfs -V 1 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-parallel_restore -m "$T_EX_META_DEV" | grep died 2>&1 && t_fail "parallel_restore"
-
-echo "== test if previously mounted"
-scratch_mkfs -V 2 -m 10G -d 60G > $T_TMP.mkfs.out 2>&1 || t_fail "mkfs failed"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 \
-	"$T_EX_DATA_DEV" "$SCR"
-umount "$SCR"
-parallel_restore -m "$T_EX_META_DEV" | grep died 2>&1 && t_fail "parallel_restore"
-
-echo "== cleanup"
-rmdir "$SCR"
-t_pass

tests/tests/quorum-heartbeat-timeout.sh

@@ -62,7 +62,7 @@ test_timeout()
 	sleep 1
 
 	# tear down the current server/leader
-	t_force_umount $sv
+	t_force_umount $sv &
 
 	# see how long it takes for the next leader to start
 	start=$(time_ms)
@@ -73,6 +73,7 @@ test_timeout()
 	echo "to $to delay $delay" >> $T_TMP.delay
 
 	# restore the mount that we tore down
+	wait
 	t_mount $sv
 
 	# make sure the new leader delay was reasonable, allowing for some slack

tests/tests/renameat2-noreplace.sh

@@ -8,19 +8,19 @@ t_require_mounts 2
 echo "=== renameat2 noreplace flag test"
 
 # give each mount their own dir (lock group) to minimize create contention
-mkdir $T_M0/dir0
-mkdir $T_M1/dir1
+mkdir $T_D0/dir0
+mkdir $T_D1/dir1
 
 echo "=== run two asynchronous calls to renameat2 NOREPLACE"
 for i in $(seq 0 100); do
         # prepare inputs in isolation
-        touch "$T_M0/dir0/old0"
-        touch "$T_M1/dir1/old1"
+        touch "$T_D0/dir0/old0"
+        touch "$T_D1/dir1/old1"
 
         # race doing noreplace renames, both can't succeed
-        dumb_renameat2 -n "$T_M0/dir0/old0" "$T_M0/dir0/sharednew" 2> /dev/null &
+        dumb_renameat2 -n "$T_D0/dir0/old0" "$T_D0/dir0/sharednew" 2> /dev/null &
         pid0=$!
-        dumb_renameat2 -n "$T_M1/dir1/old1" "$T_M1/dir0/sharednew" 2> /dev/null &
+        dumb_renameat2 -n "$T_D1/dir1/old1" "$T_D1/dir0/sharednew" 2> /dev/null &
         pid1=$!
 
         wait $pid0
@@ -31,7 +31,7 @@ for i in $(seq 0 100); do
         test "$rc0" == 0 -a "$rc1" == 0 && t_fail "both renames succeeded"
 
         # blow away possible files for either race outcome
-        rm -f "$T_M0/dir0/old0" "$T_M1/dir1/old1" "$T_M0/dir0/sharednew" "$T_M1/dir1/sharednew"
+        rm -f "$T_D0/dir0/old0" "$T_D1/dir1/old1" "$T_D0/dir0/sharednew" "$T_D1/dir1/sharednew"
 done
 
 t_pass

tests/tests/resize-devices.sh

@@ -19,8 +19,8 @@ df_free() {
 }
 
 same_totals() {
-	cur_meta_tot=$(statfs_total meta "$SCR")
-	cur_data_tot=$(statfs_total data "$SCR")
+	cur_meta_tot=$(statfs_total meta "$T_MSCR")
+	cur_data_tot=$(statfs_total data "$T_MSCR")
 
 	test "$cur_meta_tot" == "$exp_meta_tot" || \
 		t_fail "cur total_meta_blocks $cur_meta_tot != expected $exp_meta_tot"
@@ -34,10 +34,10 @@ same_totals() {
 # some slop to account for reserved blocks and concurrent allocation.
 #
 devices_grew() {
-	cur_meta_tot=$(statfs_total meta "$SCR")
-	cur_data_tot=$(statfs_total data "$SCR")
-	cur_meta_df=$(df_free MetaData "$SCR")
-	cur_data_df=$(df_free Data "$SCR")
+	cur_meta_tot=$(statfs_total meta "$T_MSCR")
+	cur_data_tot=$(statfs_total data "$T_MSCR")
+	cur_meta_df=$(df_free MetaData "$T_MSCR")
+	cur_data_df=$(df_free Data "$T_MSCR")
 
 	local grow_meta_tot=$(echo "$exp_meta_tot * 2" | bc)
 	local grow_data_tot=$(echo "$exp_data_tot * 2" | bc)
@@ -70,19 +70,13 @@ size_data=$(blockdev --getsize64 "$T_EX_DATA_DEV")
 quarter_meta=$(echo "$size_meta / 4" | bc)
 quarter_data=$(echo "$size_data / 4" | bc)
 
-# XXX this is all pretty manual, would be nice to have helpers
 echo "== make initial small fs"
-scoutfs mkfs -A -f -Q 0,127.0.0.1,53000 -m $quarter_meta -d $quarter_data \
-	"$T_EX_META_DEV" "$T_EX_DATA_DEV" > $T_TMP.mkfs.out 2>&1 || \
-		t_fail "mkfs failed"
-SCR="$T_TMPDIR/mnt.scratch"
-mkdir -p "$SCR"
-mount -t scoutfs -o metadev_path=$T_EX_META_DEV,quorum_slot_nr=0 \
-	"$T_EX_DATA_DEV" "$SCR"
+t_scratch_mkfs -A -m $quarter_meta -d $quarter_data
+t_scratch_mount
 
 # then calculate sizes based on blocks that mkfs used
-quarter_meta=$(echo "$(statfs_total meta "$SCR") * 64 * 1024" | bc)
-quarter_data=$(echo "$(statfs_total data "$SCR") * 4 * 1024" | bc)
+quarter_meta=$(echo "$(statfs_total meta "$T_MSCR") * 64 * 1024" | bc)
+quarter_data=$(echo "$(statfs_total data "$T_MSCR") * 4 * 1024" | bc)
 whole_meta=$(echo "$quarter_meta * 4" | bc)
 whole_data=$(echo "$quarter_data * 4" | bc)
 outsize_meta=$(echo "$whole_meta * 2" | bc)
@@ -93,59 +87,58 @@ shrink_meta=$(echo "$quarter_meta / 2" | bc)
 shrink_data=$(echo "$quarter_data / 2" | bc)
 
 # and save expected values for checks
-exp_meta_tot=$(statfs_total meta "$SCR")
-exp_meta_df=$(df_free MetaData "$SCR")
-exp_data_tot=$(statfs_total data "$SCR")
-exp_data_df=$(df_free Data "$SCR")
+exp_meta_tot=$(statfs_total meta "$T_MSCR")
+exp_meta_df=$(df_free MetaData "$T_MSCR")
+exp_data_tot=$(statfs_total data "$T_MSCR")
+exp_data_df=$(df_free Data "$T_MSCR")
 
 echo "== 0s do nothing"
-scoutfs resize-devices -p "$SCR" 
-scoutfs resize-devices -p "$SCR" -m 0
-scoutfs resize-devices -p "$SCR" -d 0
-scoutfs resize-devices -p "$SCR" -m 0 -d 0
+scoutfs resize-devices -p "$T_MSCR"
+scoutfs resize-devices -p "$T_MSCR" -m 0
+scoutfs resize-devices -p "$T_MSCR" -d 0
+scoutfs resize-devices -p "$T_MSCR" -m 0 -d 0
 
 echo "== shrinking fails"
-scoutfs resize-devices -p "$SCR" -m $shrink_meta
-scoutfs resize-devices -p "$SCR" -d $shrink_data
-scoutfs resize-devices -p "$SCR" -m $shrink_meta -d $shrink_data
+scoutfs resize-devices -p "$T_MSCR" -m $shrink_meta
+scoutfs resize-devices -p "$T_MSCR" -d $shrink_data
+scoutfs resize-devices -p "$T_MSCR" -m $shrink_meta -d $shrink_data
 same_totals
 
 echo "== existing sizes do nothing"
-scoutfs resize-devices -p "$SCR" -m $quarter_meta
-scoutfs resize-devices -p "$SCR" -d $quarter_data
-scoutfs resize-devices -p "$SCR" -m $quarter_meta -d $quarter_data
+scoutfs resize-devices -p "$T_MSCR" -m $quarter_meta
+scoutfs resize-devices -p "$T_MSCR" -d $quarter_data
+scoutfs resize-devices -p "$T_MSCR" -m $quarter_meta -d $quarter_data
 same_totals
 
 echo "== growing outside device fails"
-scoutfs resize-devices -p "$SCR" -m $outsize_meta
-scoutfs resize-devices -p "$SCR" -d $outsize_data
-scoutfs resize-devices -p "$SCR" -m $outsize_meta -d $outsize_data
+scoutfs resize-devices -p "$T_MSCR" -m $outsize_meta
+scoutfs resize-devices -p "$T_MSCR" -d $outsize_data
+scoutfs resize-devices -p "$T_MSCR" -m $outsize_meta -d $outsize_data
 same_totals
 
 echo "== resizing meta works"
-scoutfs resize-devices -p "$SCR" -m $half_meta
+scoutfs resize-devices -p "$T_MSCR" -m $half_meta
 devices_grew meta
 
 echo "== resizing data works"
-scoutfs resize-devices -p "$SCR" -d $half_data
+scoutfs resize-devices -p "$T_MSCR" -d $half_data
 devices_grew data
 
 echo "== shrinking back fails"
-scoutfs resize-devices -p "$SCR" -m $quarter_meta
-scoutfs resize-devices -p "$SCR" -m $quarter_data
+scoutfs resize-devices -p "$T_MSCR" -m $quarter_meta
+scoutfs resize-devices -p "$T_MSCR" -m $quarter_data
 same_totals
 
 echo "== resizing again does nothing"
-scoutfs resize-devices -p "$SCR" -m $half_meta
-scoutfs resize-devices -p "$SCR" -m $half_data
+scoutfs resize-devices -p "$T_MSCR" -m $half_meta
+scoutfs resize-devices -p "$T_MSCR" -m $half_data
 same_totals
 
 echo "== resizing to full works"
-scoutfs resize-devices -p "$SCR" -m $whole_meta -d $whole_data
+scoutfs resize-devices -p "$T_MSCR" -m $whole_meta -d $whole_data
 devices_grew meta data
 
 echo "== cleanup extra fs"
-umount "$SCR"
-rmdir "$SCR"
+t_scratch_umount
 
 t_pass

tests/tests/simple-readdir.sh

@@ -0,0 +1,37 @@
+#
+# verify d_off output of xfs_io is consistent.
+#
+
+t_require_commands xfs_io
+
+filt()
+{
+	grep d_off | cut -d ' ' -f 1,4-
+}
+
+echo "== create content"
+for s in $(seq 1 7 250); do
+	f=$(printf '%*s' $s | tr ' ' 'a')
+	touch ${T_D0}/$f
+done
+
+echo "== readdir all"
+xfs_io -c "readdir -v" $T_D0 | filt
+
+echo "== readdir offset"
+xfs_io -c "readdir -v -o 20" $T_D0 | filt
+
+echo "== readdir len (bytes)"
+xfs_io -c "readdir -v -l 193" $T_D0 | filt
+
+echo "== introduce gap"
+for s in $(seq 57 7 120); do
+	f=$(printf '%*s' $s | tr ' ' 'a')
+	rm -f ${T_D0}/$f
+done
+xfs_io -c "readdir -v" $T_D0 | filt
+
+echo "== cleanup"
+rm -rf $T_D0
+
+t_pass

tests/tests/xfstests.sh

@@ -50,9 +50,9 @@ t_quiet sync
 cat << EOF > local.config
 export FSTYP=scoutfs
 export MKFS_OPTIONS="-f"
-export MKFS_TEST_OPTIONS="-Q 0,127.0.0.1,42000"
-export MKFS_SCRATCH_OPTIONS="-Q 0,127.0.0.1,43000"
-export MKFS_DEV_OPTIONS="-Q 0,127.0.0.1,44000"
+export MKFS_TEST_OPTIONS="-Q 0,127.0.0.1,$T_TEST_PORT"
+export MKFS_SCRATCH_OPTIONS="-Q 0,127.0.0.1,$T_SCRATCH_PORT"
+export MKFS_DEV_OPTIONS="-Q 0,127.0.0.1,$T_DEV_PORT"
 export TEST_DEV=$T_DB0
 export TEST_DIR=$T_M0
 export SCRATCH_META_DEV=$T_EX_META_DEV
@@ -63,90 +63,47 @@ export MOUNT_OPTIONS="-o quorum_slot_nr=0,metadev_path=$T_MB0"
 export TEST_FS_MOUNT_OPTS="-o quorum_slot_nr=0,metadev_path=$T_MB0"
 EOF
 
-cat << EOF > local.exclude
-generic/003	# missing atime update in buffered read
-generic/029	# mmap missing
-generic/030	# mmap missing
-generic/075	# file content mismatch failures (fds, etc)
-generic/080	# mmap missing
-generic/103	# enospc causes trans commit failures
-generic/108	# mount fails on failing device?
-generic/112	# file content mismatch failures (fds, etc)
-generic/120	# (can't exec 'cause no mmap)
-generic/126	# (can't exec 'cause no mmap)
-generic/141	# mmap missing
-generic/213	# enospc causes trans commit failures
-generic/215	# mmap missing
-generic/246	# mmap missing
-generic/247	# mmap missing
-generic/248	# mmap missing
-generic/318	# can't support user namespaces until v5.11
-generic/321	# requires selinux enabled for '+' in ls?
-generic/325	# mmap missing
-generic/338	# BUG_ON update inode error handling
-generic/346	# mmap missing
-generic/347	# _dmthin_mount doesn't work?
-generic/356	# swap
-generic/357	# swap
-generic/409	# bind mounts not scripted yet
-generic/410	# bind mounts not scripted yet
-generic/411	# bind mounts not scripted yet
-generic/423	# symlink inode size is strlen() + 1 on scoutfs
-generic/428	# mmap missing
-generic/430	# xfs_io copy_range missing in el7
-generic/431	# xfs_io copy_range missing in el7
-generic/432	# xfs_io copy_range missing in el7
-generic/433	# xfs_io copy_range missing in el7
-generic/434	# xfs_io copy_range missing in el7
-generic/437	# mmap missing
-generic/441	# dm-mapper
-generic/444	# el9's posix_acl_update_mode is buggy ?
-generic/452	# exec test - no mmap
-generic/467	# open_by_handle ESTALE
-generic/472	# swap
-generic/484	# dm-mapper
-generic/493	# swap
-generic/494	# swap
-generic/495	# swap
-generic/496	# swap
-generic/497	# swap
-generic/532	# xfs_io statx attrib_mask missing in el7
-generic/554	# swap
-generic/563	# cgroup+loopdev
-generic/564	# xfs_io copy_range missing in el7
-generic/565	# xfs_io copy_range missing in el7
-generic/568	# falloc not resulting in block count increase
-generic/569	# swap
-generic/570	# swap
-generic/614	# mmap missing
-generic/620	# dm-hugedisk
-generic/633	# mmap, id-mapped mounts missing in el7
-generic/636	# swap
-generic/638	# mmap missing
-generic/641	# swap
-generic/643	# swap
-EOF
+cp "$T_EXTRA/local.exclude" local.exclude
 
-t_restore_output
+t_stdout_invoked
 echo "  (showing output of xfstests)"
 
 args="-E local.exclude ${T_XFSTESTS_ARGS:--g quick}"
 ./check $args
 # the fs is unmounted when check finishes
 
+t_stdout_compare
+
 #
-# ./check writes the results of the run to check.log.  It lists
-# the tests it ran, skipped, or failed.  Then it writes a line saying
-# everything passed or some failed.  We scrape the most recent run and
-# use it as the output to compare to make sure that we run the right
-# tests and get the right results.
+# ./check writes the results of the run to check.log.  It lists the
+# tests it ran, skipped, or failed.  Then it writes a line saying
+# everything passed or some failed.
+#
+
+#
+# If XFSTESTS_ARGS were specified then we just pass/fail to match the
+# check run.
+#
+if [ -n "$T_XFSTESTS_ARGS" ]; then
+	if tail -1 results/check.log | grep -q "Failed"; then
+		t_fail
+	else
+		t_pass
+	fi
+fi
+
+#
+# Otherwise, typically, when there were no args then we scrape the most
+# recent run and use it as the output to compare to make sure that we
+# run the right tests and get the right results.
 #
 awk '
 	/^(Ran|Not run|Failures):.*/ {
 		if (pf) {
 			res=""
 			pf=""
-		} res = res "\n" $0
+		}
+		res = res "\n" $0
 	}
 	/^(Passed|Failed).*tests$/ {
 		pf=$0
@@ -156,10 +113,14 @@ awk '
 	}' < results/check.log  > "$T_TMPDIR/results"
 
 # put a test per line so diff shows tests that differ
-egrep "^(Ran|Not run|Failures):" "$T_TMPDIR/results" | \
-	fmt -w 1 > "$T_TMPDIR/results.fmt"
-egrep "^(Passed|Failed).*tests$" "$T_TMPDIR/results" >> "$T_TMPDIR/results.fmt"
+grep -E "^(Ran|Not run|Failures):" "$T_TMPDIR/results" | fmt -w 1 > "$T_TMPDIR/results.fmt"
+grep -E "^(Passed|Failed).*tests$" "$T_TMPDIR/results" >> "$T_TMPDIR/results.fmt"
 
-t_compare_output cat "$T_TMPDIR/results.fmt"
+diff -u "$T_EXTRA/expected-results" "$T_TMPDIR/results.fmt" > "$T_TMPDIR/results.diff"
+if [ -s "$T_TMPDIR/results.diff" ]; then
+	echo "tests that were skipped/run differed from expected:"
+	cat "$T_TMPDIR/results.diff"
+	t_fail
+fi
 
 t_pass

utils/Makefile

@@ -7,7 +7,7 @@ FMTIOC_H := format.h ioctl.h
 FMTIOC_KMOD := $(addprefix ../kmod/src/,$(FMTIOC_H))
 
 CFLAGS := -Wall -O2 -Werror -D_FILE_OFFSET_BITS=64 -g -msse4.2 \
-	-I src/ -fno-strict-aliasing \
+	-fno-strict-aliasing \
 	-DSCOUTFS_FORMAT_HASH=0x$(SCOUTFS_FORMAT_HASH)LLU
 
 ifneq ($(wildcard $(firstword $(FMTIOC_KMOD))),)
@@ -15,13 +15,10 @@ CFLAGS += -I../kmod/src
 endif
 
 BIN := src/scoutfs
-OBJ_DIRS := src src/check
-OBJ := $(foreach dir,$(OBJ_DIRS),$(patsubst %.c,%.o,$(wildcard $(dir)/*.c)))
-DEPS := $(foreach dir,$(OBJ_DIRS),$(wildcard $(dir)/*.d))
+OBJ := $(patsubst %.c,%.o,$(wildcard src/*.c))
+DEPS := $(wildcard */*.d)
 
-AR := src/scoutfs_parallel_restore.a
-
-all: $(BIN) $(AR)
+all: $(BIN)
 
 ifneq ($(DEPS),)
 -include $(DEPS)
@@ -39,10 +36,6 @@ $(BIN): $(OBJ)
 	$(QU)  [BIN $@]
 	$(VE)gcc -o $@ $^ -luuid -lm -lcrypto -lblkid
 
-$(AR): $(OBJ)
-	$(QU)  [AR $@]
-	$(VE)ar rcs $@ $^
-
 %.o %.d: %.c Makefile sparse.sh
 	$(QU)  [CC $<]
 	$(VE)gcc $(CFLAGS) -MD -MP -MF $*.d -c $< -o $*.o

utils/fenced/scoutfs-fenced

@@ -62,31 +62,27 @@ test -x "$SCOUTFS_FENCED_RUN" || \
 # files disappear.
 #
 
-# generate failure messages to stderr while still echoing 0 for the caller
-careful_cat()
+# silence error messages
+quiet_cat()
 {
-	local path="$@"
-
-	cat "$@" || echo 0
+	cat "$@" 2>/dev/null
 }
 
 while sleep $SCOUTFS_FENCED_DELAY; do
+	shopt -s nullglob
 	for fence in /sys/fs/scoutfs/*/fence/*; do
-		# catches unmatched regex when no dirs
-		if [ ! -d "$fence" ]; then
-			continue
-		fi
-
-		# skip requests that have been handled
-		if [ "$(careful_cat $fence/fenced)" == 1 -o \
-		     "$(careful_cat $fence/error)" == 1 ]; then
-			continue
-		fi
 
 		srv=$(basename $(dirname $(dirname $fence)))
-		rid="$(cat $fence/rid)"
-		ip="$(cat $fence/ipv4_addr)"
-		reason="$(cat $fence/reason)"
+		fenced="$(quiet_cat $fence/fenced)"
+		error="$(quiet_cat $fence/error)"
+		rid="$(quiet_cat $fence/rid)"
+		ip="$(quiet_cat $fence/ipv4_addr)"
+		reason="$(quiet_cat $fence/reason)"
+
+		# request dirs can linger then disappear after fenced/error is set
+		if [ ! -d "$fence" -o "$fenced" == "1" -o "$error" == "1" ]; then
+			continue
+		fi
 
 		log_message "server $srv fencing rid $rid at IP $ip for $reason"
 

utils/man/scoutfs.5

@@ -55,6 +55,30 @@ with initial sparse regions (perhaps by multiple threads writing to
 different regions) and wasted space isn't an issue (perhaps because the
 file population contains few small files).
 .TP
+.B ino_alloc_per_lock=<number>
+This option determines how many inode numbers are allocated in the same
+cluster lock.  The default, and maximum, is 1024.  The minimum is 1.
+Allocating fewer inodes per lock can allow more parallelism between
+mounts because there are more locks that cover the same number of
+created files.  This can be helpful when working with smaller numbers of
+large files.
+.TP
+.B lock_idle_count=<number>
+This option sets the number of locks that the client will allow to
+remain idle after being granted.  If the number of locks exceeds this
+count then the client will try to free the oldest locks.  This setting
+is per-mount and only changes the behavior of that mount.
+.sp
+Idle locks are not reclaimed by memory pressure so this option
+determines the limit of how much memory is likely to be pinned by
+allocated idle locks.  Setting this too low can increase latency of
+operations as repeated use of a working set of locks has to request the
+locks from the network rather than using granted idle locks.
+.sp
+The count is not strictly enforced.  Operations are allowed to use locks
+while over the limit to avoid deadlocks under heavy concurrent load.
+Exceeding the count only attempts freeing of idle locks.
+.TP
 .B log_merge_wait_timeout_ms=<number>
 This option sets the amount of time, in milliseconds, that log merge
 creation can wait before timing out.  This setting is per-mount, only
@@ -130,6 +154,23 @@ the server for the filesystem if it is elected leader.
 The assigned number must match one of the slots defined with \-Q options
 when the filesystem was created with mkfs.  If the number assigned
 doesn't match a number created during mkfs then the mount will fail.
+.TP
+.B tcp_keepalive_timeout_ms=<number>
+This option sets the amount of time, in milliseconds, that a client
+connection will wait for active TCP packets, before deciding that
+the connection is dead. This setting is per-mount and only changes
+the behavior of that mount.
+.sp
+The default value of this setting is 60000msec (60s). Any precision
+beyond a whole second is likely unrealistic due to the nature of
+TCP keepalive mechanisms in the Linux kernel. Valid values are any
+value higher than 3000 (3s).
+.sp
+The TCP keepalive mechanism is complex and observing a lost connection
+quickly is important to maintain cluster stability. If the local
+network suffers from intermittent outages this option may provide
+some respite to overcome these outages without the cluster becoming
+desynchronized.
 .SH VOLUME OPTIONS
 Volume options are persistent options which are stored in the super
 block in the metadata device and which apply to all mounts of the volume.

utils/man/scoutfs.8

@@ -76,41 +76,6 @@ run when the file system will not be mounted.
 .RE
 .PD
 
-.TP
-.BI "check META-DEVICE DATA-DEVICE [-d|--debug FILE]"
-.sp
-Performs an offline file system check. The program iterates through all the
-data structures on disk directly - the filesystem must not be mounted while
-this operation is running.
-.RS 1.0i
-.PD 0
-.sp
-.TP
-.B "-d, --debug FILE"
-An output file where the program can output debug information about the
-state of the filesystem as it performs the check. If
-.B FILE
-is "-", the debug output is written to the Standard Error output.
-.TP
-.RE
-.sp
-.B RETURN VALUE
-The check function can return the following exit codes:
-.RS
-.TP
-\fB 0 \fR - no filesystem issues detected
-.TP
-\fB 1 \fR - file system issues were detected
-.TP
-\fB 8 \fR - operational error
-.TP
-\fB 16 \fR - usage error
-.TP
-\fB 32 \fR - cancelled by user (SIGINT)
-.TP
-.RE
-.PD
-
 .TP
 .BI "counters [-t|--table] SYSFS-DIR"
 .sp

utils/scoutfs-utils.spec.in

@@ -54,8 +54,6 @@ cp man/*.8.gz $RPM_BUILD_ROOT%{_mandir}/man8/.
 install -m 755 -D src/scoutfs $RPM_BUILD_ROOT%{_sbindir}/scoutfs
 install -m 644 -D src/ioctl.h $RPM_BUILD_ROOT%{_includedir}/scoutfs/ioctl.h
 install -m 644 -D src/format.h $RPM_BUILD_ROOT%{_includedir}/scoutfs/format.h
-install -m 644 -D src/parallel_restore.h $RPM_BUILD_ROOT%{_includedir}/scoutfs/parallel_restore.h
-install -m 644 -D src/scoutfs_parallel_restore.a $RPM_BUILD_ROOT%{_libdir}/scoutfs/libscoutfs_parallel_restore.a
 install -m 755 -D fenced/scoutfs-fenced $RPM_BUILD_ROOT%{_libexecdir}/scoutfs-fenced/scoutfs-fenced
 install -m 644 -D fenced/scoutfs-fenced.service $RPM_BUILD_ROOT%{_unitdir}/scoutfs-fenced.service
 install -m 644 -D fenced/scoutfs-fenced.conf.example $RPM_BUILD_ROOT%{_sysconfdir}/scoutfs/scoutfs-fenced.conf.example
@@ -72,7 +70,6 @@ install -m 644 -D fenced/scoutfs-fenced.conf.example $RPM_BUILD_ROOT%{_sysconfdi
 %files -n scoutfs-devel
 %defattr(644,root,root,755)
 %{_includedir}/scoutfs
-%{_libdir}/scoutfs
 
 %clean
 rm -rf %{buildroot}

utils/sparse.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 
-# can we find sparse?  If not, we're done.
-which sparse > /dev/null 2>&1 || exit 0
+# must have sparse.  Fail with error message, mask success path.
+which sparse > /dev/null || exit 1
 
 # 
 # one of the problems with using sparse in userspace is that it picks up
@@ -22,6 +22,11 @@ RE="$RE|warning: memset with byte count of 4194304"
 # some sparse versions don't know about some builtins
 RE="$RE|error: undefined identifier '__builtin_fpclassify'"
 
+# on el8, sparse can't handle __has_include for some reason when _GNU_SOURCE
+# is defined, and we need that for O_DIRECT.
+RE="$RE|note: in included file .through /usr/include/sys/stat.h.:"
+RE="$RE|/usr/include/bits/statx.h:30:6: error: "
+
 #
 # don't filter out 'too many errors' here, it can signify that
 # sparse doesn't understand something and is throwing a *ton*

utils/src/check/alloc.c

@@ -1,166 +0,0 @@
-#include <unistd.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdbool.h>
-#include <sys/mman.h>
-#include <errno.h>
-
-#include "sparse.h"
-#include "util.h"
-#include "format.h"
-#include "bitmap.h"
-#include "key.h"
-
-#include "alloc.h"
-#include "block.h"
-#include "btree.h"
-#include "extent.h"
-#include "iter.h"
-#include "sns.h"
-
-/*
- * We check the list blocks serially.
- *
- * XXX:
- *  - compare ref seqs
- *  - detect cycles?
- */
-int alloc_list_meta_iter(struct scoutfs_alloc_list_head *lhead, extent_cb_t cb, void *cb_arg)
-{
-	struct scoutfs_alloc_list_block *lblk;
-	struct scoutfs_block_ref ref;
-	struct block *blk = NULL;
-	u64 blkno;
-	int ret;
-
-	ref = lhead->ref;
-
-	while (ref.blkno) {
-		blkno = le64_to_cpu(ref.blkno);
-
-		ret = cb(blkno, 1, cb_arg);
-		if (ret < 0) {
-			ret = xlate_iter_errno(ret);
-			goto out;
-		}
-
-		ret = block_get(&blk, blkno, 0);
-		if (ret < 0)
-			goto out;
-
-		lblk = block_buf(blk);
-		/* XXX verify block */
-		ret = block_hdr_valid(blk, blkno, 0, SCOUTFS_BLOCK_MAGIC_ALLOC_LIST);
-		if (ret < 0)
-			goto out;
-
-		/* XXX sort?   maybe */
-
-		ref = lblk->next;
-
-		block_put(&blk);
-	}
-
-	ret = 0;
-out:
-	return ret;
-}
-
-int alloc_root_meta_iter(struct scoutfs_alloc_root *root, extent_cb_t cb, void *cb_arg)
-{
-	return btree_meta_iter(&root->root, cb, cb_arg);
-}
-
-int alloc_list_extent_iter(struct scoutfs_alloc_list_head *lhead, extent_cb_t cb, void *cb_arg)
-{
-	struct scoutfs_alloc_list_block *lblk;
-	struct scoutfs_block_ref ref;
-	struct block *blk = NULL;
-	u64 blkno;
-	int ret;
-	int i;
-
-	ref = lhead->ref;
-
-	while (ref.blkno) {
-		blkno = le64_to_cpu(ref.blkno);
-
-		ret = block_get(&blk, blkno, 0);
-		if (ret < 0)
-			goto out;
-
-		sns_push("alloc_list_block", blkno, 0);
-
-		lblk = block_buf(blk);
-		/* XXX verify block */
-		ret = block_hdr_valid(blk, blkno, 0, SCOUTFS_BLOCK_MAGIC_ALLOC_LIST);
-		if (ret < 0)
-			goto out;
-		/* XXX sort?   maybe */
-
-		ret = 0;
-		for (i = 0; i < le32_to_cpu(lblk->nr); i++) {
-			blkno = le64_to_cpu(lblk->blknos[le32_to_cpu(lblk->start) + i]);
-
-			ret = cb(blkno, 1, cb_arg);
-			if (ret < 0)
-				break;
-		}
-
-		ref = lblk->next;
-
-		block_put(&blk);
-		sns_pop();
-		if (ret < 0) {
-			ret = xlate_iter_errno(ret);
-			goto out;
-		}
-	}
-
-	ret = 0;
-out:
-	return ret;
-}
-
-static bool valid_free_extent_key(struct scoutfs_key *key)
-{
-	return (key->sk_zone == SCOUTFS_FREE_EXTENT_BLKNO_ZONE ||
-	        key->sk_zone == SCOUTFS_FREE_EXTENT_ORDER_ZONE) &&
-	       (!key->_sk_fourth && !key->sk_type &&
-		(key->sk_zone == SCOUTFS_FREE_EXTENT_ORDER_ZONE || !key->_sk_third));
-}
-
-static int free_item_cb(struct scoutfs_key *key, void *val, u16 val_len, void *cb_arg)
-{
-	struct extent_cb_arg_t *ecba = cb_arg;
-	u64 start;
-	u64 len;
-
-	/* XXX not sure these eios are what we want */
-
-	if (val_len != 0)
-		return -EIO;
-
-	if (!valid_free_extent_key(key))
-		return -EIO;
-
-	if (key->sk_zone == SCOUTFS_FREE_EXTENT_ORDER_ZONE)
-		return -ECHECK_ITER_DONE;
-
-	start = le64_to_cpu(key->skfb_end) - le64_to_cpu(key->skfb_len) + 1;
-	len = le64_to_cpu(key->skfb_len);
-
-	return ecba->cb(start, len, ecba->cb_arg);
-}
-
-/*
- * Call the callback with each of the primary BLKNO free extents stored
- * in item in the given alloc root.  It doesn't visit the secondary
- * ORDER extents.
- */
-int alloc_root_extent_iter(struct scoutfs_alloc_root *root, extent_cb_t cb, void *cb_arg)
-{
-	struct extent_cb_arg_t ecba = { .cb = cb, .cb_arg = cb_arg };
-
-	return btree_item_iter(&root->root, free_item_cb, &ecba);
-}

utils/src/check/alloc.h

@@ -1,12 +0,0 @@
-#ifndef _SCOUTFS_UTILS_CHECK_ALLOC_H
-#define _SCOUTFS_UTILS_CHECK_ALLOC_H
-
-#include "extent.h"
-
-int alloc_list_meta_iter(struct scoutfs_alloc_list_head *lhead, extent_cb_t cb, void *cb_arg);
-int alloc_root_meta_iter(struct scoutfs_alloc_root *root, extent_cb_t cb, void *cb_arg);
-
-int alloc_list_extent_iter(struct scoutfs_alloc_list_head *lhead, extent_cb_t cb, void *cb_arg);
-int alloc_root_extent_iter(struct scoutfs_alloc_root *root, extent_cb_t cb, void *cb_arg);
-
-#endif