Swap meta allocators with much more reserve.

We increase the reserve from 2x to 3x the minimum number of blocks
needed for our reserve, and change the algorithm that determines when
to swap them.

The old algorithm swaps them if _avail is just larger than _freed. While
the simplest algorithm, it suffers from the problem that in practice,
when we hit ENOSPC conditions, it will almost always swap on every
iteration when we hit our low water mark.

In our testing, we regularly see failures because what is effectively
happening is that we starve both allocators by slowly draining them
block by block, trying to do work. The work of course requires us to
drain more blocks to commit changes. This cycle doesn't end until both
allocators are almost completely drained and not enough blocks remain to
do any real work anymore.

The new algorithm will not swap allocators unless _freed is 50%
larger than _avail.

The outcome is that, during meta space pressure, we're allowing _avail to
slowly drain down to the same levels as before, effectively. However, we
then swap to _freed which is now 1.5x larger.

This results in us being able to do a whole chunk of work without
needing to swap. While draining _avail for longer, we allow work to
commit and recycle blocks back into _freed muich more effectively.

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

ReleaseNotes.md

@@ -1,6 +1,28 @@
 Versity ScoutFS Release Notes
 =============================
 
+---
+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/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

@@ -192,7 +192,7 @@ endif
 #
 # 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
 
@@ -413,3 +413,21 @@ 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

kmod/src/ioctl.c

@@ -524,7 +524,9 @@ 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 +539,9 @@ 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);

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) {

kmod/src/net.c

@@ -502,12 +502,12 @@ static void scoutfs_net_proc_worker(struct work_struct *work)
  * Free live responses up to and including the seq by marking them dead
  * and moving them to the send queue to be freed.
  */
-static int move_acked_responses(struct scoutfs_net_connection *conn,
-				struct list_head *list, u64 seq)
+static bool move_acked_responses(struct scoutfs_net_connection *conn,
+				 struct list_head *list, u64 seq)
 {
 	struct message_send *msend;
 	struct message_send *tmp;
-	int ret = 0;
+	bool moved = false;
 
 	assert_spin_locked(&conn->lock);
 
@@ -519,20 +519,20 @@ static int move_acked_responses(struct scoutfs_net_connection *conn,
 
 		msend->dead = 1;
 		list_move(&msend->head, &conn->send_queue);
-		ret = 1;
+		moved = true;
 	}
 
-	return ret;
+	return moved;
 }
 
 /* acks are processed inline in the recv worker */
 static void free_acked_responses(struct scoutfs_net_connection *conn, u64 seq)
 {
-	int moved;
+	bool moved;
 
 	spin_lock(&conn->lock);
 
-	moved = move_acked_responses(conn, &conn->send_queue, seq) +
+	moved = move_acked_responses(conn, &conn->send_queue, seq) |
 		move_acked_responses(conn, &conn->resend_queue, seq);
 
 	spin_unlock(&conn->lock);

kmod/src/scoutfs_trace.h

@@ -1046,9 +1046,12 @@ 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 +1064,12 @@ 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 +1077,11 @@ 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)
 );

kmod/src/server.c

@@ -668,14 +668,16 @@ static void scoutfs_server_commit_func(struct work_struct *work)
 	 * the reserved blocks after having filled the log trees's avail
 	 * allocator during its transaction.  To avoid prematurely
 	 * setting the low flag and causing enospc we make sure that the
-	 * next transaction's meta_avail has 2x the reserved blocks so
+	 * next transaction's meta_avail has 3x the reserved blocks so
 	 * that it can consume a full reserved amount and still have
 	 * enough to avoid enospc.  We swap to freed if avail is under
-	 * the buffer and freed is larger.
+	 * the buffer and freed is larger by 50%. This results in much less
+	 * swapping overall and allows the pools to refill naturally.
 	 */
 	if ((le64_to_cpu(server->meta_avail->total_len) <
-	     (scoutfs_server_reserved_meta_blocks(sb) * 2)) &&
-	    (le64_to_cpu(server->meta_freed->total_len) >
+	     (scoutfs_server_reserved_meta_blocks(sb) * 3)) &&
+	    ((le64_to_cpu(server->meta_freed->total_len) +
+	     (le64_to_cpu(server->meta_freed->total_len) >> 1)) >
 	     le64_to_cpu(server->meta_avail->total_len)))
 		swap(server->meta_avail, server->meta_freed);
 

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;
@@ -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;
 

tests/golden/lock-shrink-read-race

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

tests/sequence

@@ -25,6 +25,7 @@ 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

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

utils/src/bitmap.c

@@ -10,6 +10,11 @@
  * Just a quick simple native bitmap.
  */
 
+int test_bit(unsigned long *bits, u64 nr)
+{
+	return !!(bits[nr / BITS_PER_LONG] & (1UL << (nr & (BITS_PER_LONG - 1))));
+}
+
 void set_bit(unsigned long *bits, u64 nr)
 {
 	bits[nr / BITS_PER_LONG] |= 1UL << (nr & (BITS_PER_LONG - 1));

utils/src/bitmap.h

@@ -1,6 +1,7 @@
 #ifndef _BITMAP_H_
 #define _BITMAP_H_
 
+int test_bit(unsigned long *bits, u64 nr);
 void set_bit(unsigned long *bits, u64 nr);
 void clear_bit(unsigned long *bits, u64 nr);
 u64 find_next_set_bit(unsigned long *start, u64 from, u64 total);

utils/src/bloom.c

@@ -0,0 +1,20 @@
+#include <errno.h>
+
+#include "sparse.h"
+#include "util.h"
+#include "format.h"
+#include "hash.h"
+#include "bloom.h"
+
+void calc_bloom_nrs(struct scoutfs_key *key, unsigned int *nrs)
+{
+	u64 hash;
+	int i;
+
+	hash = scoutfs_hash64(key, sizeof(struct scoutfs_key));
+
+	for (i = 0; i < SCOUTFS_FOREST_BLOOM_NRS; i++) {
+		nrs[i] = (u32)hash % SCOUTFS_FOREST_BLOOM_BITS;
+		hash >>= SCOUTFS_FOREST_BLOOM_FUNC_BITS;
+	}
+}

utils/src/bloom.h

@@ -0,0 +1,6 @@
+#ifndef _BLOOM_H_
+#define _BLOOM_H_
+
+void calc_bloom_nrs(struct scoutfs_key *key, unsigned int *nrs);
+
+#endif

utils/src/btree.c

@@ -8,7 +8,7 @@
 #include "leaf_item_hash.h"
 #include "btree.h"
 
-static void init_block(struct scoutfs_btree_block *bt, int level)
+void btree_init_block(struct scoutfs_btree_block *bt, int level)
 {
 	int free;
 
@@ -33,7 +33,7 @@ void btree_init_root_single(struct scoutfs_btree_root *root,
 
 	memset(bt, 0, SCOUTFS_BLOCK_LG_SIZE);
 
-	init_block(bt, 0);
+	btree_init_block(bt, 0);
 }
 
 static void *alloc_val(struct scoutfs_btree_block *bt, int len)

utils/src/btree.h

@@ -1,6 +1,7 @@
 #ifndef _BTREE_H_
 #define _BTREE_H_
 
+void btree_init_block(struct scoutfs_btree_block *bt, int level);
 void btree_init_root_single(struct scoutfs_btree_root *root,
 			    struct scoutfs_btree_block *bt,
 			    u64 seq, u64 blkno);

utils/src/list.h

@@ -156,6 +156,16 @@ static inline void list_move_tail(struct list_head *list,
         list_add_tail(list, head);
 }
 
+/**
+ * list_is_head - tests whether @list is the list @head
+ * @list: the entry to test
+ * @head: the head of the list
+ */
+static inline int list_is_head(const struct list_head *list, const struct list_head *head)
+{
+	return list == head;
+}
+
 /**
  * list_empty - tests whether a list is empty
  * @head: the list to test.
@@ -242,6 +252,15 @@ static inline void list_splice_init(struct list_head *list,
         for (pos = (head)->next, n = pos->next; pos != (head); \
                 pos = n, n = pos->next)
 
+/**
+ * list_entry_is_head - test if the entry points to the head of the list
+ * @pos:	the type * to cursor
+ * @head:	the head for your list.
+ * @member:	the name of the list_head within the struct.
+ */
+#define list_entry_is_head(pos, head, member)				\
+	(&pos->member == (head))
+
 /**
  * list_for_each_entry        -        iterate over list of given type
  * @pos:        the type * to use as a loop counter.
@@ -307,4 +326,28 @@ static inline void list_splice_init(struct list_head *list,
 #define list_next_entry(pos, member) \
         list_entry((pos)->member.next, typeof(*(pos)), member)
 
+/**
+ * list_prev_entry - get the prev element in list
+ * @pos:	the type * to cursor
+ * @member:	the name of the list_head within the struct.
+ */
+#define list_prev_entry(pos, member) \
+	list_entry((pos)->member.prev, typeof(*(pos)), member)
+
+/**
+ * list_for_each_entry_safe_reverse - iterate backwards over list safe against removal
+ * @pos:	the type * to use as a loop cursor.
+ * @n:		another type * to use as temporary storage
+ * @head:	the head for your list.
+ * @member:	the name of the list_head within the struct.
+ *
+ * Iterate backwards over list of given type, safe against removal
+ * of list entry.
+ */
+#define list_for_each_entry_safe_reverse(pos, n, head, member)		\
+	for (pos = list_last_entry(head, typeof(*pos), member),		\
+		n = list_prev_entry(pos, member);			\
+	     !list_entry_is_head(pos, head, member); 			\
+	     pos = n, n = list_prev_entry(n, member))
+
 #endif

utils/src/lk_rbtree_wrapper.h

@@ -0,0 +1,24 @@
+#ifndef _LK_RBTREE_WRAPPER_H_
+#define _LK_RBTREE_WRAPPER_H_
+
+/*
+ * We're using this lame hack to build and use the kernel's rbtree in
+ * userspace.  We drop the kernel's rbtree*[ch] implementation in and
+ * use them with this wrapper.  We only have to remove the kernel
+ * includes from the imported files.
+ */
+
+#include <stdbool.h>
+#include "util.h"
+
+#define rcu_assign_pointer(a, b)	do { a = b; } while (0)
+#define READ_ONCE(a)			({ a; })
+#define WRITE_ONCE(a, b)		do { a = b; } while (0)
+#define unlikely(a)			({ a; })
+#define EXPORT_SYMBOL(a)		/* nop */
+
+#include "rbtree_types.h"
+#include "rbtree.h"
+#include "rbtree_augmented.h"
+
+#endif

utils/src/mode_types.c

@@ -0,0 +1,24 @@
+#include <unistd.h>
+#include <sys/stat.h>
+
+#include "sparse.h"
+#include "util.h"
+#include "format.h"
+#include "mode_types.h"
+
+unsigned int mode_to_type(mode_t mode)
+{
+#define S_SHIFT 12
+	static unsigned char mode_types[S_IFMT >> S_SHIFT] = {
+		[S_IFIFO >> S_SHIFT]	= SCOUTFS_DT_FIFO,
+		[S_IFCHR >> S_SHIFT]	= SCOUTFS_DT_CHR,
+		[S_IFDIR >> S_SHIFT]	= SCOUTFS_DT_DIR,
+		[S_IFBLK >> S_SHIFT]	= SCOUTFS_DT_BLK,
+		[S_IFREG >> S_SHIFT]	= SCOUTFS_DT_REG,
+		[S_IFLNK >> S_SHIFT]	= SCOUTFS_DT_LNK,
+		[S_IFSOCK >> S_SHIFT]	= SCOUTFS_DT_SOCK,
+	};
+
+	return mode_types[(mode & S_IFMT) >> S_SHIFT];
+#undef S_SHIFT
+}

utils/src/mode_types.h

@@ -0,0 +1,6 @@
+#ifndef _MODE_TYPES_H_
+#define _MODE_TYPES_H_
+
+unsigned int mode_to_type(mode_t mode);
+
+#endif

utils/src/name_hash.h

@@ -0,0 +1,46 @@
+#ifndef _SCOUTFS_NAME_HASH_H_
+#define _SCOUTFS_NAME_HASH_H_
+
+#include "hash.h"
+
+/*
+ * Test a bit number as though an array of bytes is a large len-bit
+ * big-endian value.  nr 0 is the LSB of the final byte, nr (len - 1) is
+ * the MSB of the first byte.
+ */
+static int test_be_bytes_bit(int nr, const char *bytes, int len)
+{
+	return bytes[(len - 1 - nr) >> 3] & (1 << (nr & 7));
+}
+
+/*
+ * Generate a 32bit "fingerprint" of the name by extracting 32 evenly
+ * distributed bits from the name.  The intent is to have the sort order
+ * of the fingerprints reflect the memcmp() sort order of the names
+ * while mapping large names down to small fs keys.
+ *
+ * Names that are smaller than 32bits are biased towards the high bits
+ * of the fingerprint so that most significant bits of the fingerprints
+ * consistently reflect the initial characters of the names.
+ */
+static inline u32 dirent_name_fingerprint(const char *name, unsigned int name_len)
+{
+	int name_bits = name_len * 8;
+	int skip = max(name_bits / 32, 1);
+	u32 fp = 0;
+	int f;
+	int n;
+
+	for (f = 31, n = name_bits - 1; f >= 0 && n >= 0; f--, n -= skip)
+		fp |= !!test_be_bytes_bit(n, name, name_bits) << f;
+
+	return fp;
+}
+
+static inline u64 dirent_name_hash(const char *name, unsigned int name_len)
+{
+       return scoutfs_hash32(name, name_len) |
+              ((u64)dirent_name_fingerprint(name, name_len) << 32);
+}
+
+#endif

utils/src/print.c

@@ -645,6 +645,8 @@ static int print_alloc_list_block(int fd, char *str, struct scoutfs_block_ref *r
 	u64 blkno;
 	u64 start;
 	u64 len;
+	u64 st;
+	u64 nr;
 	int wid;
 	int ret;
 	int i;
@@ -663,27 +665,37 @@ static int print_alloc_list_block(int fd, char *str, struct scoutfs_block_ref *r
 	       AL_REF_A(&lblk->next), le32_to_cpu(lblk->start),
 	       le32_to_cpu(lblk->nr));
 
-	if (lblk->nr) {
-		wid = printf("  exts: ");
-		start = 0;
-		len = 0;
-		for (i = 0; i < le32_to_cpu(lblk->nr); i++) {
-			if (len == 0)
-				start = le64_to_cpu(lblk->blknos[i]);
-			len++;
-
-			if (i == (le32_to_cpu(lblk->nr) - 1) ||
-			    start + len != le64_to_cpu(lblk->blknos[i + 1])) {
-				if (wid >= 72)
-					wid = printf("\n        ");
-
-				wid += printf("%llu,%llu ", start, len);
-				len = 0;
-			}
-		}
-		printf("\n");
+	st = le32_to_cpu(lblk->start);
+	nr = le32_to_cpu(lblk->nr);
+	if (st >= SCOUTFS_ALLOC_LIST_MAX_BLOCKS ||
+	    nr > SCOUTFS_ALLOC_LIST_MAX_BLOCKS ||
+	    (st + nr) > SCOUTFS_ALLOC_LIST_MAX_BLOCKS) {
+		printf("  (invalid start and nr fields)\n");
+		goto out;
 	}
 
+	if (lblk->nr == 0)
+		goto out;
+
+	wid = printf("  exts: ");
+	start = 0;
+	len = 0;
+	for (i = 0; i < nr; i++) {
+		if (len == 0)
+			start = le64_to_cpu(lblk->blknos[st + i]);
+		len++;
+
+		if (i == (nr - 1) || (start + len) != le64_to_cpu(lblk->blknos[st + i + 1])) {
+			if (wid >= 72)
+				wid = printf("\n        ");
+
+			wid += printf("%llu,%llu ", start, len);
+			len = 0;
+		}
+	}
+	printf("\n");
+
+out:
 	next = lblk->next;
 	free(lblk);
 	return print_alloc_list_block(fd, str, &next);

utils/src/rbtree.c

@@ -0,0 +1,629 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+  Red Black Trees
+  (C) 1999  Andrea Arcangeli <andrea@suse.de>
+  (C) 2002  David Woodhouse <dwmw2@infradead.org>
+  (C) 2012  Michel Lespinasse <walken@google.com>
+
+
+  linux/lib/rbtree.c
+*/
+
+#include "lk_rbtree_wrapper.h"
+
+/*
+ * red-black trees properties:  https://en.wikipedia.org/wiki/Rbtree
+ *
+ *  1) A node is either red or black
+ *  2) The root is black
+ *  3) All leaves (NULL) are black
+ *  4) Both children of every red node are black
+ *  5) Every simple path from root to leaves contains the same number
+ *     of black nodes.
+ *
+ *  4 and 5 give the O(log n) guarantee, since 4 implies you cannot have two
+ *  consecutive red nodes in a path and every red node is therefore followed by
+ *  a black. So if B is the number of black nodes on every simple path (as per
+ *  5), then the longest possible path due to 4 is 2B.
+ *
+ *  We shall indicate color with case, where black nodes are uppercase and red
+ *  nodes will be lowercase. Unknown color nodes shall be drawn as red within
+ *  parentheses and have some accompanying text comment.
+ */
+
+/*
+ * Notes on lockless lookups:
+ *
+ * All stores to the tree structure (rb_left and rb_right) must be done using
+ * WRITE_ONCE(). And we must not inadvertently cause (temporary) loops in the
+ * tree structure as seen in program order.
+ *
+ * These two requirements will allow lockless iteration of the tree -- not
+ * correct iteration mind you, tree rotations are not atomic so a lookup might
+ * miss entire subtrees.
+ *
+ * But they do guarantee that any such traversal will only see valid elements
+ * and that it will indeed complete -- does not get stuck in a loop.
+ *
+ * It also guarantees that if the lookup returns an element it is the 'correct'
+ * one. But not returning an element does _NOT_ mean it's not present.
+ *
+ * NOTE:
+ *
+ * Stores to __rb_parent_color are not important for simple lookups so those
+ * are left undone as of now. Nor did I check for loops involving parent
+ * pointers.
+ */
+
+static inline void rb_set_black(struct rb_node *rb)
+{
+	rb->__rb_parent_color |= RB_BLACK;
+}
+
+static inline struct rb_node *rb_red_parent(struct rb_node *red)
+{
+	return (struct rb_node *)red->__rb_parent_color;
+}
+
+/*
+ * Helper function for rotations:
+ * - old's parent and color get assigned to new
+ * - old gets assigned new as a parent and 'color' as a color.
+ */
+static inline void
+__rb_rotate_set_parents(struct rb_node *old, struct rb_node *new,
+			struct rb_root *root, int color)
+{
+	struct rb_node *parent = rb_parent(old);
+	new->__rb_parent_color = old->__rb_parent_color;
+	rb_set_parent_color(old, new, color);
+	__rb_change_child(old, new, parent, root);
+}
+
+static __always_inline void
+__rb_insert(struct rb_node *node, struct rb_root *root,
+	    void (*augment_rotate)(struct rb_node *old, struct rb_node *new))
+{
+	struct rb_node *parent = rb_red_parent(node), *gparent, *tmp;
+
+	while (true) {
+		/*
+		 * Loop invariant: node is red.
+		 */
+		if (unlikely(!parent)) {
+			/*
+			 * The inserted node is root. Either this is the
+			 * first node, or we recursed at Case 1 below and
+			 * are no longer violating 4).
+			 */
+			rb_set_parent_color(node, NULL, RB_BLACK);
+			break;
+		}
+
+		/*
+		 * If there is a black parent, we are done.
+		 * Otherwise, take some corrective action as,
+		 * per 4), we don't want a red root or two
+		 * consecutive red nodes.
+		 */
+		if(rb_is_black(parent))
+			break;
+
+		gparent = rb_red_parent(parent);
+
+		tmp = gparent->rb_right;
+		if (parent != tmp) {	/* parent == gparent->rb_left */
+			if (tmp && rb_is_red(tmp)) {
+				/*
+				 * Case 1 - node's uncle is red (color flips).
+				 *
+				 *       G            g
+				 *      / \          / \
+				 *     p   u  -->   P   U
+				 *    /            /
+				 *   n            n
+				 *
+				 * However, since g's parent might be red, and
+				 * 4) does not allow this, we need to recurse
+				 * at g.
+				 */
+				rb_set_parent_color(tmp, gparent, RB_BLACK);
+				rb_set_parent_color(parent, gparent, RB_BLACK);
+				node = gparent;
+				parent = rb_parent(node);
+				rb_set_parent_color(node, parent, RB_RED);
+				continue;
+			}
+
+			tmp = parent->rb_right;
+			if (node == tmp) {
+				/*
+				 * Case 2 - node's uncle is black and node is
+				 * the parent's right child (left rotate at parent).
+				 *
+				 *      G             G
+				 *     / \           / \
+				 *    p   U  -->    n   U
+				 *     \           /
+				 *      n         p
+				 *
+				 * This still leaves us in violation of 4), the
+				 * continuation into Case 3 will fix that.
+				 */
+				tmp = node->rb_left;
+				WRITE_ONCE(parent->rb_right, tmp);
+				WRITE_ONCE(node->rb_left, parent);
+				if (tmp)
+					rb_set_parent_color(tmp, parent,
+							    RB_BLACK);
+				rb_set_parent_color(parent, node, RB_RED);
+				augment_rotate(parent, node);
+				parent = node;
+				tmp = node->rb_right;
+			}
+
+			/*
+			 * Case 3 - node's uncle is black and node is
+			 * the parent's left child (right rotate at gparent).
+			 *
+			 *        G           P
+			 *       / \         / \
+			 *      p   U  -->  n   g
+			 *     /                 \
+			 *    n                   U
+			 */
+			WRITE_ONCE(gparent->rb_left, tmp); /* == parent->rb_right */
+			WRITE_ONCE(parent->rb_right, gparent);
+			if (tmp)
+				rb_set_parent_color(tmp, gparent, RB_BLACK);
+			__rb_rotate_set_parents(gparent, parent, root, RB_RED);
+			augment_rotate(gparent, parent);
+			break;
+		} else {
+			tmp = gparent->rb_left;
+			if (tmp && rb_is_red(tmp)) {
+				/* Case 1 - color flips */
+				rb_set_parent_color(tmp, gparent, RB_BLACK);
+				rb_set_parent_color(parent, gparent, RB_BLACK);
+				node = gparent;
+				parent = rb_parent(node);
+				rb_set_parent_color(node, parent, RB_RED);
+				continue;
+			}
+
+			tmp = parent->rb_left;
+			if (node == tmp) {
+				/* Case 2 - right rotate at parent */
+				tmp = node->rb_right;
+				WRITE_ONCE(parent->rb_left, tmp);
+				WRITE_ONCE(node->rb_right, parent);
+				if (tmp)
+					rb_set_parent_color(tmp, parent,
+							    RB_BLACK);
+				rb_set_parent_color(parent, node, RB_RED);
+				augment_rotate(parent, node);
+				parent = node;
+				tmp = node->rb_left;
+			}
+
+			/* Case 3 - left rotate at gparent */
+			WRITE_ONCE(gparent->rb_right, tmp); /* == parent->rb_left */
+			WRITE_ONCE(parent->rb_left, gparent);
+			if (tmp)
+				rb_set_parent_color(tmp, gparent, RB_BLACK);
+			__rb_rotate_set_parents(gparent, parent, root, RB_RED);
+			augment_rotate(gparent, parent);
+			break;
+		}
+	}
+}
+
+/*
+ * Inline version for rb_erase() use - we want to be able to inline
+ * and eliminate the dummy_rotate callback there
+ */
+static __always_inline void
+____rb_erase_color(struct rb_node *parent, struct rb_root *root,
+	void (*augment_rotate)(struct rb_node *old, struct rb_node *new))
+{
+	struct rb_node *node = NULL, *sibling, *tmp1, *tmp2;
+
+	while (true) {
+		/*
+		 * Loop invariants:
+		 * - node is black (or NULL on first iteration)
+		 * - node is not the root (parent is not NULL)
+		 * - All leaf paths going through parent and node have a
+		 *   black node count that is 1 lower than other leaf paths.
+		 */
+		sibling = parent->rb_right;
+		if (node != sibling) {	/* node == parent->rb_left */
+			if (rb_is_red(sibling)) {
+				/*
+				 * Case 1 - left rotate at parent
+				 *
+				 *     P               S
+				 *    / \             / \
+				 *   N   s    -->    p   Sr
+				 *      / \         / \
+				 *     Sl  Sr      N   Sl
+				 */
+				tmp1 = sibling->rb_left;
+				WRITE_ONCE(parent->rb_right, tmp1);
+				WRITE_ONCE(sibling->rb_left, parent);
+				rb_set_parent_color(tmp1, parent, RB_BLACK);
+				__rb_rotate_set_parents(parent, sibling, root,
+							RB_RED);
+				augment_rotate(parent, sibling);
+				sibling = tmp1;
+			}
+			tmp1 = sibling->rb_right;
+			if (!tmp1 || rb_is_black(tmp1)) {
+				tmp2 = sibling->rb_left;
+				if (!tmp2 || rb_is_black(tmp2)) {
+					/*
+					 * Case 2 - sibling color flip
+					 * (p could be either color here)
+					 *
+					 *    (p)           (p)
+					 *    / \           / \
+					 *   N   S    -->  N   s
+					 *      / \           / \
+					 *     Sl  Sr        Sl  Sr
+					 *
+					 * This leaves us violating 5) which
+					 * can be fixed by flipping p to black
+					 * if it was red, or by recursing at p.
+					 * p is red when coming from Case 1.
+					 */
+					rb_set_parent_color(sibling, parent,
+							    RB_RED);
+					if (rb_is_red(parent))
+						rb_set_black(parent);
+					else {
+						node = parent;
+						parent = rb_parent(node);
+						if (parent)
+							continue;
+					}
+					break;
+				}
+				/*
+				 * Case 3 - right rotate at sibling
+				 * (p could be either color here)
+				 *
+				 *   (p)           (p)
+				 *   / \           / \
+				 *  N   S    -->  N   sl
+				 *     / \             \
+				 *    sl  Sr            S
+				 *                       \
+				 *                        Sr
+				 *
+				 * Note: p might be red, and then both
+				 * p and sl are red after rotation(which
+				 * breaks property 4). This is fixed in
+				 * Case 4 (in __rb_rotate_set_parents()
+				 *         which set sl the color of p
+				 *         and set p RB_BLACK)
+				 *
+				 *   (p)            (sl)
+				 *   / \            /  \
+				 *  N   sl   -->   P    S
+				 *       \        /      \
+				 *        S      N        Sr
+				 *         \
+				 *          Sr
+				 */
+				tmp1 = tmp2->rb_right;
+				WRITE_ONCE(sibling->rb_left, tmp1);
+				WRITE_ONCE(tmp2->rb_right, sibling);
+				WRITE_ONCE(parent->rb_right, tmp2);
+				if (tmp1)
+					rb_set_parent_color(tmp1, sibling,
+							    RB_BLACK);
+				augment_rotate(sibling, tmp2);
+				tmp1 = sibling;
+				sibling = tmp2;
+			}
+			/*
+			 * Case 4 - left rotate at parent + color flips
+			 * (p and sl could be either color here.
+			 *  After rotation, p becomes black, s acquires
+			 *  p's color, and sl keeps its color)
+			 *
+			 *      (p)             (s)
+			 *      / \             / \
+			 *     N   S     -->   P   Sr
+			 *        / \         / \
+			 *      (sl) sr      N  (sl)
+			 */
+			tmp2 = sibling->rb_left;
+			WRITE_ONCE(parent->rb_right, tmp2);
+			WRITE_ONCE(sibling->rb_left, parent);
+			rb_set_parent_color(tmp1, sibling, RB_BLACK);
+			if (tmp2)
+				rb_set_parent(tmp2, parent);
+			__rb_rotate_set_parents(parent, sibling, root,
+						RB_BLACK);
+			augment_rotate(parent, sibling);
+			break;
+		} else {
+			sibling = parent->rb_left;
+			if (rb_is_red(sibling)) {
+				/* Case 1 - right rotate at parent */
+				tmp1 = sibling->rb_right;
+				WRITE_ONCE(parent->rb_left, tmp1);
+				WRITE_ONCE(sibling->rb_right, parent);
+				rb_set_parent_color(tmp1, parent, RB_BLACK);
+				__rb_rotate_set_parents(parent, sibling, root,
+							RB_RED);
+				augment_rotate(parent, sibling);
+				sibling = tmp1;
+			}
+			tmp1 = sibling->rb_left;
+			if (!tmp1 || rb_is_black(tmp1)) {
+				tmp2 = sibling->rb_right;
+				if (!tmp2 || rb_is_black(tmp2)) {
+					/* Case 2 - sibling color flip */
+					rb_set_parent_color(sibling, parent,
+							    RB_RED);
+					if (rb_is_red(parent))
+						rb_set_black(parent);
+					else {
+						node = parent;
+						parent = rb_parent(node);
+						if (parent)
+							continue;
+					}
+					break;
+				}
+				/* Case 3 - left rotate at sibling */
+				tmp1 = tmp2->rb_left;
+				WRITE_ONCE(sibling->rb_right, tmp1);
+				WRITE_ONCE(tmp2->rb_left, sibling);
+				WRITE_ONCE(parent->rb_left, tmp2);
+				if (tmp1)
+					rb_set_parent_color(tmp1, sibling,
+							    RB_BLACK);
+				augment_rotate(sibling, tmp2);
+				tmp1 = sibling;
+				sibling = tmp2;
+			}
+			/* Case 4 - right rotate at parent + color flips */
+			tmp2 = sibling->rb_right;
+			WRITE_ONCE(parent->rb_left, tmp2);
+			WRITE_ONCE(sibling->rb_right, parent);
+			rb_set_parent_color(tmp1, sibling, RB_BLACK);
+			if (tmp2)
+				rb_set_parent(tmp2, parent);
+			__rb_rotate_set_parents(parent, sibling, root,
+						RB_BLACK);
+			augment_rotate(parent, sibling);
+			break;
+		}
+	}
+}
+
+/* Non-inline version for rb_erase_augmented() use */
+void __rb_erase_color(struct rb_node *parent, struct rb_root *root,
+	void (*augment_rotate)(struct rb_node *old, struct rb_node *new))
+{
+	____rb_erase_color(parent, root, augment_rotate);
+}
+EXPORT_SYMBOL(__rb_erase_color);
+
+/*
+ * Non-augmented rbtree manipulation functions.
+ *
+ * We use dummy augmented callbacks here, and have the compiler optimize them
+ * out of the rb_insert_color() and rb_erase() function definitions.
+ */
+
+static inline void dummy_propagate(struct rb_node *node, struct rb_node *stop) {}
+static inline void dummy_copy(struct rb_node *old, struct rb_node *new) {}
+static inline void dummy_rotate(struct rb_node *old, struct rb_node *new) {}
+
+static const struct rb_augment_callbacks dummy_callbacks = {
+	.propagate = dummy_propagate,
+	.copy = dummy_copy,
+	.rotate = dummy_rotate
+};
+
+void rb_insert_color(struct rb_node *node, struct rb_root *root)
+{
+	__rb_insert(node, root, dummy_rotate);
+}
+EXPORT_SYMBOL(rb_insert_color);
+
+void rb_erase(struct rb_node *node, struct rb_root *root)
+{
+	struct rb_node *rebalance;
+	rebalance = __rb_erase_augmented(node, root, &dummy_callbacks);
+	if (rebalance)
+		____rb_erase_color(rebalance, root, dummy_rotate);
+}
+EXPORT_SYMBOL(rb_erase);
+
+/*
+ * Augmented rbtree manipulation functions.
+ *
+ * This instantiates the same __always_inline functions as in the non-augmented
+ * case, but this time with user-defined callbacks.
+ */
+
+void __rb_insert_augmented(struct rb_node *node, struct rb_root *root,
+	void (*augment_rotate)(struct rb_node *old, struct rb_node *new))
+{
+	__rb_insert(node, root, augment_rotate);
+}
+EXPORT_SYMBOL(__rb_insert_augmented);
+
+/*
+ * This function returns the first node (in sort order) of the tree.
+ */
+struct rb_node *rb_first(const struct rb_root *root)
+{
+	struct rb_node	*n;
+
+	n = root->rb_node;
+	if (!n)
+		return NULL;
+	while (n->rb_left)
+		n = n->rb_left;
+	return n;
+}
+EXPORT_SYMBOL(rb_first);
+
+struct rb_node *rb_last(const struct rb_root *root)
+{
+	struct rb_node	*n;
+
+	n = root->rb_node;
+	if (!n)
+		return NULL;
+	while (n->rb_right)
+		n = n->rb_right;
+	return n;
+}
+EXPORT_SYMBOL(rb_last);
+
+struct rb_node *rb_next(const struct rb_node *node)
+{
+	struct rb_node *parent;
+
+	if (RB_EMPTY_NODE(node))
+		return NULL;
+
+	/*
+	 * If we have a right-hand child, go down and then left as far
+	 * as we can.
+	 */
+	if (node->rb_right) {
+		node = node->rb_right;
+		while (node->rb_left)
+			node = node->rb_left;
+		return (struct rb_node *)node;
+	}
+
+	/*
+	 * No right-hand children. Everything down and left is smaller than us,
+	 * so any 'next' node must be in the general direction of our parent.
+	 * Go up the tree; any time the ancestor is a right-hand child of its
+	 * parent, keep going up. First time it's a left-hand child of its
+	 * parent, said parent is our 'next' node.
+	 */
+	while ((parent = rb_parent(node)) && node == parent->rb_right)
+		node = parent;
+
+	return parent;
+}
+EXPORT_SYMBOL(rb_next);
+
+struct rb_node *rb_prev(const struct rb_node *node)
+{
+	struct rb_node *parent;
+
+	if (RB_EMPTY_NODE(node))
+		return NULL;
+
+	/*
+	 * If we have a left-hand child, go down and then right as far
+	 * as we can.
+	 */
+	if (node->rb_left) {
+		node = node->rb_left;
+		while (node->rb_right)
+			node = node->rb_right;
+		return (struct rb_node *)node;
+	}
+
+	/*
+	 * No left-hand children. Go up till we find an ancestor which
+	 * is a right-hand child of its parent.
+	 */
+	while ((parent = rb_parent(node)) && node == parent->rb_left)
+		node = parent;
+
+	return parent;
+}
+EXPORT_SYMBOL(rb_prev);
+
+void rb_replace_node(struct rb_node *victim, struct rb_node *new,
+		     struct rb_root *root)
+{
+	struct rb_node *parent = rb_parent(victim);
+
+	/* Copy the pointers/colour from the victim to the replacement */
+	*new = *victim;
+
+	/* Set the surrounding nodes to point to the replacement */
+	if (victim->rb_left)
+		rb_set_parent(victim->rb_left, new);
+	if (victim->rb_right)
+		rb_set_parent(victim->rb_right, new);
+	__rb_change_child(victim, new, parent, root);
+}
+EXPORT_SYMBOL(rb_replace_node);
+
+void rb_replace_node_rcu(struct rb_node *victim, struct rb_node *new,
+			 struct rb_root *root)
+{
+	struct rb_node *parent = rb_parent(victim);
+
+	/* Copy the pointers/colour from the victim to the replacement */
+	*new = *victim;
+
+	/* Set the surrounding nodes to point to the replacement */
+	if (victim->rb_left)
+		rb_set_parent(victim->rb_left, new);
+	if (victim->rb_right)
+		rb_set_parent(victim->rb_right, new);
+
+	/* Set the parent's pointer to the new node last after an RCU barrier
+	 * so that the pointers onwards are seen to be set correctly when doing
+	 * an RCU walk over the tree.
+	 */
+	__rb_change_child_rcu(victim, new, parent, root);
+}
+EXPORT_SYMBOL(rb_replace_node_rcu);
+
+static struct rb_node *rb_left_deepest_node(const struct rb_node *node)
+{
+	for (;;) {
+		if (node->rb_left)
+			node = node->rb_left;
+		else if (node->rb_right)
+			node = node->rb_right;
+		else
+			return (struct rb_node *)node;
+	}
+}
+
+struct rb_node *rb_next_postorder(const struct rb_node *node)
+{
+	const struct rb_node *parent;
+	if (!node)
+		return NULL;
+	parent = rb_parent(node);
+
+	/* If we're sitting on node, we've already seen our children */
+	if (parent && node == parent->rb_left && parent->rb_right) {
+		/* If we are the parent's left node, go to the parent's right
+		 * node then all the way down to the left */
+		return rb_left_deepest_node(parent->rb_right);
+	} else
+		/* Otherwise we are the parent's right node, and the parent
+		 * should be next */
+		return (struct rb_node *)parent;
+}
+EXPORT_SYMBOL(rb_next_postorder);
+
+struct rb_node *rb_first_postorder(const struct rb_root *root)
+{
+	if (!root->rb_node)
+		return NULL;
+
+	return rb_left_deepest_node(root->rb_node);
+}
+EXPORT_SYMBOL(rb_first_postorder);

utils/src/rbtree.h

@@ -0,0 +1,328 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+  Red Black Trees
+  (C) 1999  Andrea Arcangeli <andrea@suse.de>
+  
+
+  linux/include/linux/rbtree.h
+
+  To use rbtrees you'll have to implement your own insert and search cores.
+  This will avoid us to use callbacks and to drop drammatically performances.
+  I know it's not the cleaner way,  but in C (not in C++) to get
+  performances and genericity...
+
+  See Documentation/core-api/rbtree.rst for documentation and samples.
+*/
+
+#ifndef	_LINUX_RBTREE_H
+#define	_LINUX_RBTREE_H
+
+#define rb_parent(r)   ((struct rb_node *)((r)->__rb_parent_color & ~3))
+
+#define	rb_entry(ptr, type, member) container_of(ptr, type, member)
+
+#define RB_EMPTY_ROOT(root)  (READ_ONCE((root)->rb_node) == NULL)
+
+/* 'empty' nodes are nodes that are known not to be inserted in an rbtree */
+#define RB_EMPTY_NODE(node)  \
+	((node)->__rb_parent_color == (unsigned long)(node))
+#define RB_CLEAR_NODE(node)  \
+	((node)->__rb_parent_color = (unsigned long)(node))
+
+
+extern void rb_insert_color(struct rb_node *, struct rb_root *);
+extern void rb_erase(struct rb_node *, struct rb_root *);
+
+
+/* Find logical next and previous nodes in a tree */
+extern struct rb_node *rb_next(const struct rb_node *);
+extern struct rb_node *rb_prev(const struct rb_node *);
+extern struct rb_node *rb_first(const struct rb_root *);
+extern struct rb_node *rb_last(const struct rb_root *);
+
+/* Postorder iteration - always visit the parent after its children */
+extern struct rb_node *rb_first_postorder(const struct rb_root *);
+extern struct rb_node *rb_next_postorder(const struct rb_node *);
+
+/* Fast replacement of a single node without remove/rebalance/add/rebalance */
+extern void rb_replace_node(struct rb_node *victim, struct rb_node *new,
+			    struct rb_root *root);
+extern void rb_replace_node_rcu(struct rb_node *victim, struct rb_node *new,
+				struct rb_root *root);
+
+static inline void rb_link_node(struct rb_node *node, struct rb_node *parent,
+				struct rb_node **rb_link)
+{
+	node->__rb_parent_color = (unsigned long)parent;
+	node->rb_left = node->rb_right = NULL;
+
+	*rb_link = node;
+}
+
+static inline void rb_link_node_rcu(struct rb_node *node, struct rb_node *parent,
+				    struct rb_node **rb_link)
+{
+	node->__rb_parent_color = (unsigned long)parent;
+	node->rb_left = node->rb_right = NULL;
+
+	rcu_assign_pointer(*rb_link, node);
+}
+
+#define rb_entry_safe(ptr, type, member) \
+	({ typeof(ptr) ____ptr = (ptr); \
+	   ____ptr ? rb_entry(____ptr, type, member) : NULL; \
+	})
+
+/**
+ * rbtree_postorder_for_each_entry_safe - iterate in post-order over rb_root of
+ * given type allowing the backing memory of @pos to be invalidated
+ *
+ * @pos:	the 'type *' to use as a loop cursor.
+ * @n:		another 'type *' to use as temporary storage
+ * @root:	'rb_root *' of the rbtree.
+ * @field:	the name of the rb_node field within 'type'.
+ *
+ * rbtree_postorder_for_each_entry_safe() provides a similar guarantee as
+ * list_for_each_entry_safe() and allows the iteration to continue independent
+ * of changes to @pos by the body of the loop.
+ *
+ * Note, however, that it cannot handle other modifications that re-order the
+ * rbtree it is iterating over. This includes calling rb_erase() on @pos, as
+ * rb_erase() may rebalance the tree, causing us to miss some nodes.
+ */
+#define rbtree_postorder_for_each_entry_safe(pos, n, root, field) \
+	for (pos = rb_entry_safe(rb_first_postorder(root), typeof(*pos), field); \
+	     pos && ({ n = rb_entry_safe(rb_next_postorder(&pos->field), \
+			typeof(*pos), field); 1; }); \
+	     pos = n)
+
+/* Same as rb_first(), but O(1) */
+#define rb_first_cached(root) (root)->rb_leftmost
+
+static inline void rb_insert_color_cached(struct rb_node *node,
+					  struct rb_root_cached *root,
+					  bool leftmost)
+{
+	if (leftmost)
+		root->rb_leftmost = node;
+	rb_insert_color(node, &root->rb_root);
+}
+
+
+static inline struct rb_node *
+rb_erase_cached(struct rb_node *node, struct rb_root_cached *root)
+{
+	struct rb_node *leftmost = NULL;
+
+	if (root->rb_leftmost == node)
+		leftmost = root->rb_leftmost = rb_next(node);
+
+	rb_erase(node, &root->rb_root);
+
+	return leftmost;
+}
+
+static inline void rb_replace_node_cached(struct rb_node *victim,
+					  struct rb_node *new,
+					  struct rb_root_cached *root)
+{
+	if (root->rb_leftmost == victim)
+		root->rb_leftmost = new;
+	rb_replace_node(victim, new, &root->rb_root);
+}
+
+/*
+ * The below helper functions use 2 operators with 3 different
+ * calling conventions. The operators are related like:
+ *
+ *	comp(a->key,b) < 0  := less(a,b)
+ *	comp(a->key,b) > 0  := less(b,a)
+ *	comp(a->key,b) == 0 := !less(a,b) && !less(b,a)
+ *
+ * If these operators define a partial order on the elements we make no
+ * guarantee on which of the elements matching the key is found. See
+ * rb_find().
+ *
+ * The reason for this is to allow the find() interface without requiring an
+ * on-stack dummy object, which might not be feasible due to object size.
+ */
+
+/**
+ * rb_add_cached() - insert @node into the leftmost cached tree @tree
+ * @node: node to insert
+ * @tree: leftmost cached tree to insert @node into
+ * @less: operator defining the (partial) node order
+ *
+ * Returns @node when it is the new leftmost, or NULL.
+ */
+static __always_inline struct rb_node *
+rb_add_cached(struct rb_node *node, struct rb_root_cached *tree,
+	      bool (*less)(struct rb_node *, const struct rb_node *))
+{
+	struct rb_node **link = &tree->rb_root.rb_node;
+	struct rb_node *parent = NULL;
+	bool leftmost = true;
+
+	while (*link) {
+		parent = *link;
+		if (less(node, parent)) {
+			link = &parent->rb_left;
+		} else {
+			link = &parent->rb_right;
+			leftmost = false;
+		}
+	}
+
+	rb_link_node(node, parent, link);
+	rb_insert_color_cached(node, tree, leftmost);
+
+	return leftmost ? node : NULL;
+}
+
+/**
+ * rb_add() - insert @node into @tree
+ * @node: node to insert
+ * @tree: tree to insert @node into
+ * @less: operator defining the (partial) node order
+ */
+static __always_inline void
+rb_add(struct rb_node *node, struct rb_root *tree,
+       bool (*less)(struct rb_node *, const struct rb_node *))
+{
+	struct rb_node **link = &tree->rb_node;
+	struct rb_node *parent = NULL;
+
+	while (*link) {
+		parent = *link;
+		if (less(node, parent))
+			link = &parent->rb_left;
+		else
+			link = &parent->rb_right;
+	}
+
+	rb_link_node(node, parent, link);
+	rb_insert_color(node, tree);
+}
+
+/**
+ * rb_find_add() - find equivalent @node in @tree, or add @node
+ * @node: node to look-for / insert
+ * @tree: tree to search / modify
+ * @cmp: operator defining the node order
+ *
+ * Returns the rb_node matching @node, or NULL when no match is found and @node
+ * is inserted.
+ */
+static __always_inline struct rb_node *
+rb_find_add(struct rb_node *node, struct rb_root *tree,
+	    int (*cmp)(struct rb_node *, const struct rb_node *))
+{
+	struct rb_node **link = &tree->rb_node;
+	struct rb_node *parent = NULL;
+	int c;
+
+	while (*link) {
+		parent = *link;
+		c = cmp(node, parent);
+
+		if (c < 0)
+			link = &parent->rb_left;
+		else if (c > 0)
+			link = &parent->rb_right;
+		else
+			return parent;
+	}
+
+	rb_link_node(node, parent, link);
+	rb_insert_color(node, tree);
+	return NULL;
+}
+
+/**
+ * rb_find() - find @key in tree @tree
+ * @key: key to match
+ * @tree: tree to search
+ * @cmp: operator defining the node order
+ *
+ * Returns the rb_node matching @key or NULL.
+ */
+static __always_inline struct rb_node *
+rb_find(const void *key, const struct rb_root *tree,
+	int (*cmp)(const void *key, const struct rb_node *))
+{
+	struct rb_node *node = tree->rb_node;
+
+	while (node) {
+		int c = cmp(key, node);
+
+		if (c < 0)
+			node = node->rb_left;
+		else if (c > 0)
+			node = node->rb_right;
+		else
+			return node;
+	}
+
+	return NULL;
+}
+
+/**
+ * rb_find_first() - find the first @key in @tree
+ * @key: key to match
+ * @tree: tree to search
+ * @cmp: operator defining node order
+ *
+ * Returns the leftmost node matching @key, or NULL.
+ */
+static __always_inline struct rb_node *
+rb_find_first(const void *key, const struct rb_root *tree,
+	      int (*cmp)(const void *key, const struct rb_node *))
+{
+	struct rb_node *node = tree->rb_node;
+	struct rb_node *match = NULL;
+
+	while (node) {
+		int c = cmp(key, node);
+
+		if (c <= 0) {
+			if (!c)
+				match = node;
+			node = node->rb_left;
+		} else if (c > 0) {
+			node = node->rb_right;
+		}
+	}
+
+	return match;
+}
+
+/**
+ * rb_next_match() - find the next @key in @tree
+ * @key: key to match
+ * @tree: tree to search
+ * @cmp: operator defining node order
+ *
+ * Returns the next node matching @key, or NULL.
+ */
+static __always_inline struct rb_node *
+rb_next_match(const void *key, struct rb_node *node,
+	      int (*cmp)(const void *key, const struct rb_node *))
+{
+	node = rb_next(node);
+	if (node && cmp(key, node))
+		node = NULL;
+	return node;
+}
+
+/**
+ * rb_for_each() - iterates a subtree matching @key
+ * @node: iterator
+ * @key: key to match
+ * @tree: tree to search
+ * @cmp: operator defining node order
+ */
+#define rb_for_each(node, key, tree, cmp) \
+	for ((node) = rb_find_first((key), (tree), (cmp)); \
+	     (node); (node) = rb_next_match((key), (node), (cmp)))
+
+#endif	/* _LINUX_RBTREE_H */

utils/src/rbtree_augmented.h

@@ -0,0 +1,313 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+  Red Black Trees
+  (C) 1999  Andrea Arcangeli <andrea@suse.de>
+  (C) 2002  David Woodhouse <dwmw2@infradead.org>
+  (C) 2012  Michel Lespinasse <walken@google.com>
+
+
+  linux/include/linux/rbtree_augmented.h
+*/
+
+#ifndef _LINUX_RBTREE_AUGMENTED_H
+#define _LINUX_RBTREE_AUGMENTED_H
+
+/*
+ * Please note - only struct rb_augment_callbacks and the prototypes for
+ * rb_insert_augmented() and rb_erase_augmented() are intended to be public.
+ * The rest are implementation details you are not expected to depend on.
+ *
+ * See Documentation/core-api/rbtree.rst for documentation and samples.
+ */
+
+struct rb_augment_callbacks {
+	void (*propagate)(struct rb_node *node, struct rb_node *stop);
+	void (*copy)(struct rb_node *old, struct rb_node *new);
+	void (*rotate)(struct rb_node *old, struct rb_node *new);
+};
+
+extern void __rb_insert_augmented(struct rb_node *node, struct rb_root *root,
+	void (*augment_rotate)(struct rb_node *old, struct rb_node *new));
+
+/*
+ * Fixup the rbtree and update the augmented information when rebalancing.
+ *
+ * On insertion, the user must update the augmented information on the path
+ * leading to the inserted node, then call rb_link_node() as usual and
+ * rb_insert_augmented() instead of the usual rb_insert_color() call.
+ * If rb_insert_augmented() rebalances the rbtree, it will callback into
+ * a user provided function to update the augmented information on the
+ * affected subtrees.
+ */
+static inline void
+rb_insert_augmented(struct rb_node *node, struct rb_root *root,
+		    const struct rb_augment_callbacks *augment)
+{
+	__rb_insert_augmented(node, root, augment->rotate);
+}
+
+static inline void
+rb_insert_augmented_cached(struct rb_node *node,
+			   struct rb_root_cached *root, bool newleft,
+			   const struct rb_augment_callbacks *augment)
+{
+	if (newleft)
+		root->rb_leftmost = node;
+	rb_insert_augmented(node, &root->rb_root, augment);
+}
+
+/*
+ * Template for declaring augmented rbtree callbacks (generic case)
+ *
+ * RBSTATIC:    'static' or empty
+ * RBNAME:      name of the rb_augment_callbacks structure
+ * RBSTRUCT:    struct type of the tree nodes
+ * RBFIELD:     name of struct rb_node field within RBSTRUCT
+ * RBAUGMENTED: name of field within RBSTRUCT holding data for subtree
+ * RBCOMPUTE:   name of function that recomputes the RBAUGMENTED data
+ */
+
+#define RB_DECLARE_CALLBACKS(RBSTATIC, RBNAME,				\
+			     RBSTRUCT, RBFIELD, RBAUGMENTED, RBCOMPUTE)	\
+static inline void							\
+RBNAME ## _propagate(struct rb_node *rb, struct rb_node *stop)		\
+{									\
+	while (rb != stop) {						\
+		RBSTRUCT *node = rb_entry(rb, RBSTRUCT, RBFIELD);	\
+		if (RBCOMPUTE(node, true))				\
+			break;						\
+		rb = rb_parent(&node->RBFIELD);				\
+	}								\
+}									\
+static inline void							\
+RBNAME ## _copy(struct rb_node *rb_old, struct rb_node *rb_new)		\
+{									\
+	RBSTRUCT *old = rb_entry(rb_old, RBSTRUCT, RBFIELD);		\
+	RBSTRUCT *new = rb_entry(rb_new, RBSTRUCT, RBFIELD);		\
+	new->RBAUGMENTED = old->RBAUGMENTED;				\
+}									\
+static void								\
+RBNAME ## _rotate(struct rb_node *rb_old, struct rb_node *rb_new)	\
+{									\
+	RBSTRUCT *old = rb_entry(rb_old, RBSTRUCT, RBFIELD);		\
+	RBSTRUCT *new = rb_entry(rb_new, RBSTRUCT, RBFIELD);		\
+	new->RBAUGMENTED = old->RBAUGMENTED;				\
+	RBCOMPUTE(old, false);						\
+}									\
+RBSTATIC const struct rb_augment_callbacks RBNAME = {			\
+	.propagate = RBNAME ## _propagate,				\
+	.copy = RBNAME ## _copy,					\
+	.rotate = RBNAME ## _rotate					\
+};
+
+/*
+ * Template for declaring augmented rbtree callbacks,
+ * computing RBAUGMENTED scalar as max(RBCOMPUTE(node)) for all subtree nodes.
+ *
+ * RBSTATIC:    'static' or empty
+ * RBNAME:      name of the rb_augment_callbacks structure
+ * RBSTRUCT:    struct type of the tree nodes
+ * RBFIELD:     name of struct rb_node field within RBSTRUCT
+ * RBTYPE:      type of the RBAUGMENTED field
+ * RBAUGMENTED: name of RBTYPE field within RBSTRUCT holding data for subtree
+ * RBCOMPUTE:   name of function that returns the per-node RBTYPE scalar
+ */
+
+#define RB_DECLARE_CALLBACKS_MAX(RBSTATIC, RBNAME, RBSTRUCT, RBFIELD,	      \
+				 RBTYPE, RBAUGMENTED, RBCOMPUTE)	      \
+static inline bool RBNAME ## _compute_max(RBSTRUCT *node, bool exit)	      \
+{									      \
+	RBSTRUCT *child;						      \
+	RBTYPE max = RBCOMPUTE(node);					      \
+	if (node->RBFIELD.rb_left) {					      \
+		child = rb_entry(node->RBFIELD.rb_left, RBSTRUCT, RBFIELD);   \
+		if (child->RBAUGMENTED > max)				      \
+			max = child->RBAUGMENTED;			      \
+	}								      \
+	if (node->RBFIELD.rb_right) {					      \
+		child = rb_entry(node->RBFIELD.rb_right, RBSTRUCT, RBFIELD);  \
+		if (child->RBAUGMENTED > max)				      \
+			max = child->RBAUGMENTED;			      \
+	}								      \
+	if (exit && node->RBAUGMENTED == max)				      \
+		return true;						      \
+	node->RBAUGMENTED = max;					      \
+	return false;							      \
+}									      \
+RB_DECLARE_CALLBACKS(RBSTATIC, RBNAME,					      \
+		     RBSTRUCT, RBFIELD, RBAUGMENTED, RBNAME ## _compute_max)
+
+
+#define	RB_RED		0
+#define	RB_BLACK	1
+
+#define __rb_parent(pc)    ((struct rb_node *)(pc & ~3))
+
+#define __rb_color(pc)     ((pc) & 1)
+#define __rb_is_black(pc)  __rb_color(pc)
+#define __rb_is_red(pc)    (!__rb_color(pc))
+#define rb_color(rb)       __rb_color((rb)->__rb_parent_color)
+#define rb_is_red(rb)      __rb_is_red((rb)->__rb_parent_color)
+#define rb_is_black(rb)    __rb_is_black((rb)->__rb_parent_color)
+
+static inline void rb_set_parent(struct rb_node *rb, struct rb_node *p)
+{
+	rb->__rb_parent_color = rb_color(rb) | (unsigned long)p;
+}
+
+static inline void rb_set_parent_color(struct rb_node *rb,
+				       struct rb_node *p, int color)
+{
+	rb->__rb_parent_color = (unsigned long)p | color;
+}
+
+static inline void
+__rb_change_child(struct rb_node *old, struct rb_node *new,
+		  struct rb_node *parent, struct rb_root *root)
+{
+	if (parent) {
+		if (parent->rb_left == old)
+			WRITE_ONCE(parent->rb_left, new);
+		else
+			WRITE_ONCE(parent->rb_right, new);
+	} else
+		WRITE_ONCE(root->rb_node, new);
+}
+
+static inline void
+__rb_change_child_rcu(struct rb_node *old, struct rb_node *new,
+		      struct rb_node *parent, struct rb_root *root)
+{
+	if (parent) {
+		if (parent->rb_left == old)
+			rcu_assign_pointer(parent->rb_left, new);
+		else
+			rcu_assign_pointer(parent->rb_right, new);
+	} else
+		rcu_assign_pointer(root->rb_node, new);
+}
+
+extern void __rb_erase_color(struct rb_node *parent, struct rb_root *root,
+	void (*augment_rotate)(struct rb_node *old, struct rb_node *new));
+
+static __always_inline struct rb_node *
+__rb_erase_augmented(struct rb_node *node, struct rb_root *root,
+		     const struct rb_augment_callbacks *augment)
+{
+	struct rb_node *child = node->rb_right;
+	struct rb_node *tmp = node->rb_left;
+	struct rb_node *parent, *rebalance;
+	unsigned long pc;
+
+	if (!tmp) {
+		/*
+		 * Case 1: node to erase has no more than 1 child (easy!)
+		 *
+		 * Note that if there is one child it must be red due to 5)
+		 * and node must be black due to 4). We adjust colors locally
+		 * so as to bypass __rb_erase_color() later on.
+		 */
+		pc = node->__rb_parent_color;
+		parent = __rb_parent(pc);
+		__rb_change_child(node, child, parent, root);
+		if (child) {
+			child->__rb_parent_color = pc;
+			rebalance = NULL;
+		} else
+			rebalance = __rb_is_black(pc) ? parent : NULL;
+		tmp = parent;
+	} else if (!child) {
+		/* Still case 1, but this time the child is node->rb_left */
+		tmp->__rb_parent_color = pc = node->__rb_parent_color;
+		parent = __rb_parent(pc);
+		__rb_change_child(node, tmp, parent, root);
+		rebalance = NULL;
+		tmp = parent;
+	} else {
+		struct rb_node *successor = child, *child2;
+
+		tmp = child->rb_left;
+		if (!tmp) {
+			/*
+			 * Case 2: node's successor is its right child
+			 *
+			 *    (n)          (s)
+			 *    / \          / \
+			 *  (x) (s)  ->  (x) (c)
+			 *        \
+			 *        (c)
+			 */
+			parent = successor;
+			child2 = successor->rb_right;
+
+			augment->copy(node, successor);
+		} else {
+			/*
+			 * Case 3: node's successor is leftmost under
+			 * node's right child subtree
+			 *
+			 *    (n)          (s)
+			 *    / \          / \
+			 *  (x) (y)  ->  (x) (y)
+			 *      /            /
+			 *    (p)          (p)
+			 *    /            /
+			 *  (s)          (c)
+			 *    \
+			 *    (c)
+			 */
+			do {
+				parent = successor;
+				successor = tmp;
+				tmp = tmp->rb_left;
+			} while (tmp);
+			child2 = successor->rb_right;
+			WRITE_ONCE(parent->rb_left, child2);
+			WRITE_ONCE(successor->rb_right, child);
+			rb_set_parent(child, successor);
+
+			augment->copy(node, successor);
+			augment->propagate(parent, successor);
+		}
+
+		tmp = node->rb_left;
+		WRITE_ONCE(successor->rb_left, tmp);
+		rb_set_parent(tmp, successor);
+
+		pc = node->__rb_parent_color;
+		tmp = __rb_parent(pc);
+		__rb_change_child(node, successor, tmp, root);
+
+		if (child2) {
+			rb_set_parent_color(child2, parent, RB_BLACK);
+			rebalance = NULL;
+		} else {
+			rebalance = rb_is_black(successor) ? parent : NULL;
+		}
+		successor->__rb_parent_color = pc;
+		tmp = successor;
+	}
+
+	augment->propagate(tmp, NULL);
+	return rebalance;
+}
+
+static __always_inline void
+rb_erase_augmented(struct rb_node *node, struct rb_root *root,
+		   const struct rb_augment_callbacks *augment)
+{
+	struct rb_node *rebalance = __rb_erase_augmented(node, root, augment);
+	if (rebalance)
+		__rb_erase_color(rebalance, root, augment->rotate);
+}
+
+static __always_inline void
+rb_erase_augmented_cached(struct rb_node *node, struct rb_root_cached *root,
+			  const struct rb_augment_callbacks *augment)
+{
+	if (root->rb_leftmost == node)
+		root->rb_leftmost = rb_next(node);
+	rb_erase_augmented(node, &root->rb_root, augment);
+}
+
+#endif	/* _LINUX_RBTREE_AUGMENTED_H */

utils/src/rbtree_types.h

@@ -0,0 +1,34 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef _LINUX_RBTREE_TYPES_H
+#define _LINUX_RBTREE_TYPES_H
+
+struct rb_node {
+	unsigned long  __rb_parent_color;
+	struct rb_node *rb_right;
+	struct rb_node *rb_left;
+} __attribute__((aligned(sizeof(long))));
+/* The alignment might seem pointless, but allegedly CRIS needs it */
+
+struct rb_root {
+	struct rb_node *rb_node;
+};
+
+/*
+ * Leftmost-cached rbtrees.
+ *
+ * We do not cache the rightmost node based on footprint
+ * size vs number of potential users that could benefit
+ * from O(1) rb_last(). Just not worth it, users that want
+ * this feature can always implement the logic explicitly.
+ * Furthermore, users that want to cache both pointers may
+ * find it a bit asymmetric, but that's ok.
+ */
+struct rb_root_cached {
+	struct rb_root rb_root;
+	struct rb_node *rb_leftmost;
+};
+
+#define RB_ROOT (struct rb_root) { NULL, }
+#define RB_ROOT_CACHED (struct rb_root_cached) { {NULL, }, NULL }
+
+#endif

utils/src/srch.c

@@ -44,3 +44,37 @@ int srch_decode_entry(void *buf, struct scoutfs_srch_entry *sre,
 
 	return tot;
 }
+
+static int encode_u64(__le64 *buf, u64 val)
+{
+	int bytes;
+
+	val = (val << 1) ^ ((s64)val >> 63); /* shift sign extend */
+	bytes = (fls64(val) + 7) >> 3;
+
+	put_unaligned_le64(val, buf);
+	return bytes;
+}
+
+int srch_encode_entry(void *buf, struct scoutfs_srch_entry *sre, struct scoutfs_srch_entry *prev)
+{
+	u64 diffs[] = {
+		le64_to_cpu(sre->hash) - le64_to_cpu(prev->hash),
+		le64_to_cpu(sre->ino) - le64_to_cpu(prev->ino),
+		le64_to_cpu(sre->id) - le64_to_cpu(prev->id),
+	};
+	u16 lengths = 0;
+	int bytes;
+	int tot = 2;
+	int i;
+
+	for (i = 0; i < array_size(diffs); i++) {
+		bytes = encode_u64(buf + tot, diffs[i]);
+		lengths |= bytes << (i << 2);
+		tot += bytes;
+	}
+
+	put_unaligned_le16(lengths, buf);
+
+	return tot;
+}

utils/src/srch.h

@@ -3,5 +3,6 @@
 
 int srch_decode_entry(void *buf, struct scoutfs_srch_entry *sre,
 		      struct scoutfs_srch_entry *prev);
+int srch_encode_entry(void *buf, struct scoutfs_srch_entry *sre, struct scoutfs_srch_entry *prev);
 
 #endif

utils/src/util.h

@@ -70,6 +70,8 @@ do {				\
 #define container_of(ptr, type, memb) \
 	((type *)((void *)(ptr) - offsetof(type, memb)))
 
+#define NSEC_PER_SEC 1000000000
+
 #define BITS_PER_LONG (sizeof(long) * 8)
 #define U8_MAX ((u8)~0ULL)
 #define U16_MAX ((u16)~0ULL)
@@ -82,6 +84,7 @@ do {				\
 							\
 	(_x == 0 ? 0 : 64 - __builtin_clzll(_x));	\
 })
+#define fls64(x) flsll(x)
 
 #define ilog2(x)					\
 ({							\
@@ -99,6 +102,16 @@ emit_get_unaligned_le(16)
 emit_get_unaligned_le(32)
 emit_get_unaligned_le(64)
 
+#define emit_put_unaligned_le(nr)				\
+static inline void put_unaligned_le##nr(u##nr val, void *buf)	\
+{								\
+	__le##nr x = cpu_to_le##nr(val);			\
+	memcpy(buf, &x, sizeof(x));				\
+}
+emit_put_unaligned_le(16)
+emit_put_unaligned_le(32)
+emit_put_unaligned_le(64)
+
 /*
  * return -1,0,+1 based on the memcmp comparison of the minimum of their
  * two lengths.  If their min shared bytes are equal but the lengths