Remove unused orig keys from forest read items

These orig copies of the start and end keys serve no purpose.  I think
they were an editing mistake left over from a version where retries
could happen within the read_items call.

Signed-off-by: Zach Brown <zab@versity.com>

ReleaseNotes.md

@@ -1,6 +1,23 @@
 Versity ScoutFS Release Notes
 =============================
 
+---
+v1.30
+\
+*Apr 21, 2026*
+
+Fix a problem reading the accumulated totals of contributing .totl.
+xattrs when log merging is in progress.  The problem would have readers
+of the totals calculate the sums incorrectly.
+
+Fix a problem updating quota rules.  There was a race where updates
+could be corrupted if they happened while a transaction was being
+written.
+
+Fix a problem deleting files with .indx. xattrs.  The internal indexing
+metadata wouldn't be properly deleted so the files would still claim to
+be present and visible in the index, though the file no longer existed.
+
 ---
 v1.29
 \

kmod/src/Makefile

@@ -13,6 +13,7 @@ scoutfs-y +=			\
 	avl.o			\
 	alloc.o			\
 	block.o			\
+	bsearch_index.o		\
 	btree.o			\
 	client.o		\
 	counters.o		\
@@ -36,6 +37,7 @@ scoutfs-y +=			\
 	per_task.o		\
 	quorum.o		\
 	quota.o			\
+	raw.o			\
 	recov.o			\
 	scoutfs_trace.o		\
 	server.o		\

kmod/src/block.h

@@ -1,6 +1,8 @@
 #ifndef _SCOUTFS_BLOCK_H_
 #define _SCOUTFS_BLOCK_H_
 
+struct scoutfs_alloc;
+
 struct scoutfs_block_writer {
 	spinlock_t lock;
 	struct list_head dirty_list;

kmod/src/bsearch_index.c

@@ -0,0 +1,59 @@
+/*
+ * Copyright (C) 2026 Versity Software, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#include <linux/kernel.h>
+#include <linux/bsearch.h>
+
+#include "bsearch_index.h"
+
+struct bsearch_index_key {
+	int (*cmp)(const void *key, const void *elt);
+	/* the key has to be const, so we have to update the index through a pointer */
+	void **index_elt;
+	const void *key;
+	size_t size;
+};
+
+static int cmp_index(const void *key, const void *elt)
+{
+	const struct bsearch_index_key *bik = key;
+	int cmp = bik->cmp(bik->key, elt);
+
+	if (cmp > 0)
+		*(bik->index_elt) = (void *)elt + bik->size;
+	else
+		*(bik->index_elt) = (void *)elt;
+
+	return cmp;
+}
+
+/*
+ * A bsearch() wrapper that returns the index of the element of the
+ * array that the key would be stored in to maintain sort order.  It's
+ * the first element where the existing element is greater than the key.
+ * It returns the size of the array if the key is greater than the last
+ * element in the array.
+ */
+size_t bsearch_index(const void *key, const void *base, size_t num, size_t size,
+		     int (*cmp)(const void *key, const void *elt))
+{
+	void *index_elt = (void *)base;
+	struct bsearch_index_key bik = {
+		.cmp = cmp,
+		.index_elt = &index_elt,
+		.key = key,
+		.size = size,
+	};
+
+	bsearch(&bik, base, num, size, cmp_index);
+	return ((unsigned long)index_elt - (unsigned long)base) / size;
+}

kmod/src/bsearch_index.h

@@ -0,0 +1,7 @@
+#ifndef _SCOUTFS_BSEARCH_INDEX_H_
+#define _SCOUTFS_BSEARCH_INDEX_H_
+
+size_t bsearch_index(const void *key, const void *base, size_t num, size_t size,
+		     int (*cmp)(const void *key, const void *elt));
+
+#endif

kmod/src/btree.c

@@ -1816,6 +1816,11 @@ int scoutfs_btree_dirty(struct super_block *sb,
  * Call the users callback on all the items in the leaf that we find.
  * We also set the caller's keys for the first and last possible keys
  * that could exist in the leaf block.
+ *
+ * The callback can set a new key to continue reading from rather than
+ * iterating over all the items.  It modifies the key and returns
+ * -ESRCH, which performs a new avl search.  If the modified key falls
+ * outside of the range of keys in the block then we return.
  */
 int scoutfs_btree_read_items(struct super_block *sb,
 			     struct scoutfs_btree_root *root,
@@ -1829,6 +1834,7 @@ int scoutfs_btree_read_items(struct super_block *sb,
 	struct scoutfs_avl_node *next_node;
 	struct scoutfs_avl_node *node;
 	struct btree_walk_key_range kr;
+	struct scoutfs_key cb_key;
 	struct scoutfs_block *bl;
 	int ret;
 
@@ -1842,22 +1848,32 @@ int scoutfs_btree_read_items(struct super_block *sb,
 	if (scoutfs_key_compare(&kr.end, end) < 0)
 		*end = kr.end;
 
-	node = scoutfs_avl_search(&bt->item_root, cmp_key_item, start, NULL,
+	cb_key = *start;
+search:
+	node = scoutfs_avl_search(&bt->item_root, cmp_key_item, &cb_key, NULL,
 				  NULL, &next_node, NULL) ?: next_node;
 	while (node) {
 		item = node_item(node);
 		if (scoutfs_key_compare(&item->key, end) > 0)
 			break;
 
-		ret = cb(sb, item_key(item), le64_to_cpu(item->seq), item->flags,
+		cb_key = *item_key(item);
+		ret = cb(sb, &cb_key, le64_to_cpu(item->seq), item->flags,
 			 item_val(bt, item), item_val_len(item), arg);
-		if (ret < 0)
-			break;
+		if (ret < 0) {
+			if (ret == -ESRCH) {
+				if (scoutfs_key_compare(&cb_key, start) >= 0)
+					goto search;
+				ret = 0;
+			}
+			goto out;
+		}
 
 		node = scoutfs_avl_next(&bt->item_root, node);
 	}
 
 	scoutfs_block_put(sb, bl);
+	ret = 0;
 out:
 	return ret;
 }
@@ -2183,6 +2199,8 @@ static int merge_read_item(struct super_block *sb, struct scoutfs_key *key, u64
 		if (ret > 0) {
 			if (ret == SCOUTFS_DELTA_COMBINED) {
 				scoutfs_inc_counter(sb, btree_merge_delta_combined);
+				if (seq > found->seq)
+					found->seq = seq;
 			} else if (ret == SCOUTFS_DELTA_COMBINED_NULL) {
 				scoutfs_inc_counter(sb, btree_merge_delta_null);
 				free_mitem(rng, found);
@@ -2486,6 +2504,14 @@ int scoutfs_btree_merge(struct super_block *sb,
 			mitem = next_mitem(mitem);
 			free_mitem(&rng, tmp);
 		}
+
+		if (mitem && walk_val_len == 0 &&
+		    !(walk_flags & (BTW_INSERT | BTW_DELETE)) &&
+		    scoutfs_trigger(sb, LOG_MERGE_FORCE_PARTIAL)) {
+			ret = -ERANGE;
+			*next_ret = mitem->key;
+			goto out;
+		}
 	}
 
 	ret = 0;

kmod/src/forest.c

@@ -114,6 +114,42 @@ static struct scoutfs_block *read_bloom_ref(struct super_block *sb, struct scout
 	return bl;
 }
 
+/*
+ * Returns >0 if there was a bloom block and all the bits were present.
+ */
+static int all_bloom_bits_present(struct super_block *sb, struct scoutfs_block_ref *ref,
+				  struct forest_bloom_nrs *bloom)
+{
+	struct scoutfs_bloom_block *bb;
+	struct scoutfs_block *bl;
+	int i;
+
+	if (ref->blkno == 0)
+		return 0;
+
+	bl = read_bloom_ref(sb, ref);
+	if (IS_ERR(bl))
+		return PTR_ERR(bl);
+
+	bb = bl->data;
+
+	for (i = 0; i < ARRAY_SIZE(bloom->nrs); i++) {
+		if (!test_bit_le(bloom->nrs[i], bb->bits))
+			break;
+	}
+
+	scoutfs_block_put(sb, bl);
+
+	/* one of the bloom bits wasn't set */
+	if (i != ARRAY_SIZE(bloom->nrs)) {
+		scoutfs_inc_counter(sb, forest_bloom_fail);
+		return 0;
+	}
+
+	scoutfs_inc_counter(sb, forest_bloom_pass);
+	return 1;
+}
+
 /*
  * This is an unlocked iteration across all the btrees to find a hint at
  * the next key that the caller could read.  It's used to find out what
@@ -227,9 +263,13 @@ static int forest_read_items(struct super_block *sb, struct scoutfs_key *key, u6
 }
 
 /*
- * For each forest btree whose bloom block indicates that the lock might
- * have items stored, call the caller's callback for every item in the
- * leaf block in each tree which contains the key.
+ * Call the caller's callback for every item in the leaf blocks in each
+ * forest btree that contain the caller's key.
+ *
+ * If a bloom key is provided then each log tree's bloom block is
+ * checked and only trees with all the bloom key's bloom bits set will
+ * be read from.  When the bloom key is null all trees will be read
+ * from.
  *
  * The btree iter calls clamp the caller's range to the tightest range
  * that covers all the blocks.  Any keys outside of this range can't be
@@ -239,33 +279,26 @@ static int forest_read_items(struct super_block *sb, struct scoutfs_key *key, u6
  * to reset their state and retry with a newer version of the btrees.
  */
 int scoutfs_forest_read_items_roots(struct super_block *sb, struct scoutfs_net_roots *roots,
-				    struct scoutfs_key *key, struct scoutfs_key *bloom_key,
-				    struct scoutfs_key *start, struct scoutfs_key *end,
-				    scoutfs_forest_item_cb cb, void *arg)
+				    u64 merge_input_seq, struct scoutfs_key *key,
+				    struct scoutfs_key *bloom_key, struct scoutfs_key *start,
+				    struct scoutfs_key *end, scoutfs_forest_item_cb cb, void *arg)
 {
 	struct forest_read_items_data rid = {
 		.cb = cb,
 		.cb_arg = arg,
 	};
 	struct scoutfs_log_trees lt;
-	struct scoutfs_bloom_block *bb;
 	struct forest_bloom_nrs bloom;
 	SCOUTFS_BTREE_ITEM_REF(iref);
-	struct scoutfs_block *bl;
 	struct scoutfs_key ltk;
-	struct scoutfs_key orig_start = *start;
-	struct scoutfs_key orig_end = *end;
 	int ret;
-	int i;
 
 	scoutfs_inc_counter(sb, forest_read_items);
-	calc_bloom_nrs(&bloom, bloom_key);
+	if (bloom_key)
+		calc_bloom_nrs(&bloom, bloom_key);
 
 	trace_scoutfs_forest_using_roots(sb, &roots->fs_root, &roots->logs_root);
 
-	*start = orig_start;
-	*end = orig_end;
-
 	/* start with fs root items */
 	rid.fic |= FIC_FS_ROOT;
 	ret = scoutfs_btree_read_items(sb, &roots->fs_root, key, start, end,
@@ -292,40 +325,29 @@ int scoutfs_forest_read_items_roots(struct super_block *sb, struct scoutfs_net_r
 			goto out; /* including stale */
 		}
 
-		if (lt.bloom_ref.blkno == 0)
+		/* we're not expecting -ENOENT from _read_items */
+		if (lt.item_root.ref.blkno == 0)
 			continue;
 
-		bl = read_bloom_ref(sb, &lt.bloom_ref);
-		if (IS_ERR(bl)) {
-			ret = PTR_ERR(bl);
-			goto out;
-		}
-		bb = bl->data;
-
-		for (i = 0; i < ARRAY_SIZE(bloom.nrs); i++) {
-			if (!test_bit_le(bloom.nrs[i], bb->bits))
-				break;
+		if (bloom_key) {
+			ret = all_bloom_bits_present(sb, &lt.bloom_ref, &bloom);
+			if (ret < 0)
+				goto out;
+			if (ret == 0)
+				continue;
 		}
 
-		scoutfs_block_put(sb, bl);
-
-		/* one of the bloom bits wasn't set */
-		if (i != ARRAY_SIZE(bloom.nrs)) {
-			scoutfs_inc_counter(sb, forest_bloom_fail);
-			continue;
-		}
-
-		scoutfs_inc_counter(sb, forest_bloom_pass);
-
-		if ((le64_to_cpu(lt.flags) & SCOUTFS_LOG_TREES_FINALIZED))
-			rid.fic |= FIC_FINALIZED;
+		if ((le64_to_cpu(lt.flags) & SCOUTFS_LOG_TREES_FINALIZED) &&
+		    (merge_input_seq == 0 ||
+		     le64_to_cpu(lt.finalize_seq) < merge_input_seq))
+			rid.fic |= FIC_MERGE_INPUT;
 
 		ret = scoutfs_btree_read_items(sb, &lt.item_root, key, start,
 					       end, forest_read_items, &rid);
 		if (ret < 0)
 			goto out;
 
-		rid.fic &= ~FIC_FINALIZED;
+		rid.fic &= ~FIC_MERGE_INPUT;
 	}
 
 	ret = 0;
@@ -345,7 +367,7 @@ int scoutfs_forest_read_items(struct super_block *sb,
 
 	ret = scoutfs_client_get_roots(sb, &roots);
 	if (ret == 0)
-		ret = scoutfs_forest_read_items_roots(sb, &roots, key, bloom_key, start, end,
+		ret = scoutfs_forest_read_items_roots(sb, &roots, 0, key, bloom_key, start, end,
 						      cb, arg);
 	return ret;
 }

kmod/src/forest.h

@@ -11,7 +11,7 @@ struct scoutfs_lock;
 /* caller gives an item to the callback */
 enum {
 	FIC_FS_ROOT = (1 << 0),
-	FIC_FINALIZED = (1 << 1),
+	FIC_MERGE_INPUT = (1 << 1),
 };
 typedef int (*scoutfs_forest_item_cb)(struct super_block *sb, struct scoutfs_key *key, u64 seq,
 				      u8 flags, void *val, int val_len, int fic, void *arg);
@@ -25,9 +25,9 @@ int scoutfs_forest_read_items(struct super_block *sb,
 			      struct scoutfs_key *end,
 			      scoutfs_forest_item_cb cb, void *arg);
 int scoutfs_forest_read_items_roots(struct super_block *sb, struct scoutfs_net_roots *roots,
-				    struct scoutfs_key *key, struct scoutfs_key *bloom_key,
-				    struct scoutfs_key *start, struct scoutfs_key *end,
-				    scoutfs_forest_item_cb cb, void *arg);
+				    u64 merge_input_seq, struct scoutfs_key *key,
+				    struct scoutfs_key *bloom_key, struct scoutfs_key *start,
+				    struct scoutfs_key *end, scoutfs_forest_item_cb cb, void *arg);
 int scoutfs_forest_set_bloom_bits(struct super_block *sb,
 				  struct scoutfs_lock *lock);
 void scoutfs_forest_set_max_seq(struct super_block *sb, u64 max_seq);

kmod/src/ioctl.c

@@ -49,6 +49,7 @@
 #include "quota.h"
 #include "scoutfs_trace.h"
 #include "util.h"
+#include "raw.h"
 
 /*
  * We make inode index items coherent by locking fixed size regions of
@@ -1739,6 +1740,69 @@ out:
 	return ret;
 }
 
+static long scoutfs_ioc_raw_read_meta_seq(struct file *file, unsigned long arg)
+{
+	struct super_block *sb = file_inode(file)->i_sb;
+	struct scoutfs_ioctl_raw_read_meta_seq __user *urms = (void __user *)arg;
+	struct scoutfs_ioctl_raw_read_meta_seq rms;
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN)) {
+		ret = -EPERM;
+		goto out;
+	}
+
+	if (copy_from_user(&rms, urms, sizeof(rms))) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	if (rms.results_size == 0) {
+		ret = 0;
+		goto out;
+	}
+
+	if (rms.results_size < sizeof(struct scoutfs_ioctl_meta_seq) ||
+	    rms.results_size > INT_MAX) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = scoutfs_raw_read_meta_seq(sb, &rms, &rms.last);
+	if (ret >= 0 && copy_to_user(&urms->last, &rms.last, sizeof(rms.last)))
+		ret = -EFAULT;
+out:
+	return ret;
+}
+
+static long scoutfs_ioc_raw_read_inode_info(struct file *file, unsigned long arg)
+{
+	struct super_block *sb = file_inode(file)->i_sb;
+	struct scoutfs_ioctl_raw_read_inode_info __user *urii = (void __user *)arg;
+	struct scoutfs_ioctl_raw_read_inode_info rii;
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN)) {
+		ret = -EPERM;
+		goto out;
+	}
+
+	if (copy_from_user(&rii, urii, sizeof(rii))) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	if (rii.inos_count == 0 || rii.results_size > INT_MAX ||
+	    !IS_ALIGNED(rii.inos_ptr, __alignof__(__u64))) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = scoutfs_raw_read_inode_info(sb, &rii);
+out:
+	return ret;
+}
+
 long scoutfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
 	switch (cmd) {
@@ -1790,6 +1854,10 @@ long scoutfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 		return scoutfs_ioc_read_xattr_index(file, arg);
 	case SCOUTFS_IOC_PUNCH_OFFLINE:
 		return scoutfs_ioc_punch_offline(file, arg);
+	case SCOUTFS_IOC_RAW_READ_META_SEQ:
+		return scoutfs_ioc_raw_read_meta_seq(file, arg);
+	case SCOUTFS_IOC_RAW_READ_INODE_INFO:
+		return scoutfs_ioc_raw_read_inode_info(file, arg);
 	}
 
 	return -ENOTTY;

kmod/src/ioctl.h

@@ -15,20 +15,6 @@
 
 #define SCOUTFS_IOCTL_MAGIC 0xE8  /* arbitrarily chosen hole in ioctl-number.rst */
 
-/*
- * Packed scoutfs keys rarely cross the ioctl boundary so we have a
- * translation struct.
- */
-struct scoutfs_ioctl_key {
-	__le64	_sk_first;
-	__le64	_sk_second;
-	__le64	_sk_third;
-	__u8	_sk_fourth;
-	__u8	sk_type;
-	__u8	sk_zone;
-	__u8	_pad[5];
-};
-
 struct scoutfs_ioctl_walk_inodes_entry {
 	__u64 major;
 	__u64 ino;
@@ -876,4 +862,169 @@ struct scoutfs_ioctl_punch_offline {
 #define SCOUTFS_IOC_PUNCH_OFFLINE \
 	_IOW(SCOUTFS_IOCTL_MAGIC, 24, struct scoutfs_ioctl_punch_offline)
 
+/*
+ * Read meta_seq items without cluster locking.
+ *
+ * @start is the first meta_seq item value that could be returned.
+ * {0,0} is the minimum.
+ *
+ * @end is the last meta_seq item value that could be returned.
+ * {U64_MAX, U64_MAX} is the maximum.
+ *
+ * @last is only set on success from the call.  It's the last meta_seq
+ * item that could have been returned.  This lets the caller detect that
+ * the full input range wasn't explored.  Another call can be made with
+ * start set to just after this.
+ *
+ * @results_ptr is a pointer to an array of (struct
+ * scoutfs_ioctl_meta_seq) elements that were found in the input range.
+ *
+ * @results_size is the count of elements in the results_ptr array and
+ * the maximum number of results that can be returned.  There must be
+ * room for at least one result.
+ *
+ * Return existing meta_seq items starting from @start until @last.
+ * Partial results can be returned and is indicated by @last being set
+ * to an item before @last.
+ *
+ * The results are sorted first by increasing meta_seq and then by
+ * increasing ino.  All of the results are from one version of file
+ * system metadata.  This means that an inode can not be found multiple
+ * times within the results of one call.
+ *
+ * This call ignores currently dirty transactions and reads persistent
+ * items directly.  A transaction can be written after this call and
+ * cause meta_seq items to appear before or within the results from this
+ * call.
+ *
+ * The number of meta_seq items stored in the results buffer is returned
+ * and @last is updated.  0 items can be returned if none are found
+ * within the input range.
+ *
+ * Unique errors:
+ *
+ *  -EINVAL: The result count was 0 or greater than INT_MAX.
+ *
+ *  -ESTALE: The results could not be read from one stable version of
+ *    file system metadata.  Decrease the number of inodes requested.
+ */
+struct scoutfs_ioctl_meta_seq {
+	__u64 meta_seq;
+	__u64 ino;
+};
+struct scoutfs_ioctl_raw_read_meta_seq {
+	struct scoutfs_ioctl_meta_seq start;
+	struct scoutfs_ioctl_meta_seq end;
+	struct scoutfs_ioctl_meta_seq last;
+	__u64 results_ptr;
+	__u32 results_size;
+	__u32 _pad;
+};
+#define SCOUTFS_IOC_RAW_READ_META_SEQ \
+	_IOR(SCOUTFS_IOCTL_MAGIC, 25, struct scoutfs_ioctl_raw_read_meta_seq)
+
+
+/*
+ * Read inode metadata without cluster locking.
+ *
+ * @inos_ptr is a pointer to an aligned array of 64bit inode numbers.
+ *
+ * @inos_count is the number of elements in the array.  The inode
+ * numbers must not be zero, must strictly increase, and must not
+ * contain any duplicates.
+ *
+ * @names_ptr is a pointer to a byte array of xattr names to return with
+ * each inode.  The names are identical to those used in
+ * {get,set}xattr(2).  The names must be null terminated and no two
+ * names may be equal.
+ *
+ * @names_count is the number of names that will be found in the
+ * names_ptr buffer.
+ *
+ * @results_ptr is a pointer to a buffer that will be filled by the read
+ * inode info results.  The result structs and payloads are not aligned.
+ * Callers will almost certainly need to copy them into aligned
+ * addresses before referencing their contents.
+ *
+ * @results_size is the number of bytes available in the results_ptr
+ * buffer.
+ *
+ * For each inode an _INODE result will always be returned.  Then a
+ * _XATTR result will be returned for each xattr on the inode that
+ * matches one of the given input names.
+ *
+ * Each call will not return partial results. -ERANGE is returned if the
+ * results for the requested inodes do not fit in the results buffer.
+ *
+ * The info for one call is from one consistent version of the file
+ * system metadata.  The call can have to retry if it sees metadata
+ * change during its call.  -ESTALE will be returned if it was not able
+ * to read all the inodes info from one metadata version.  The number of
+ * inodes being read can be decreased to avoid this.
+ *
+ * Inodes with an nlink of 0 are not returned.
+ *
+ * The size in bytes of filled results is returned.  A non-zero return
+ * will always include at least one full
+ * (struct scoutfs_ioctl_raw_read_result) header.
+ *
+ * Unique errors:
+ *
+ *  -EINVAL: The inode count can't be zero. The inos ptr must be aligned
+ *    to __u64 alignment.  The results buffer size can't be larger than
+ *    INT_MAX.  Inode numbers can't be zero, must be sorted, and can't
+ *    have duplicates.  The xattr names must be unique, null terminated,
+ *    and less than 256 bytes long.
+ *
+ *  -ERANGE: The results for the requested inodes do not fit in the
+ *    results buffer.  Increase the buffer size (perhaps allowing for all
+ *    xattrs with large values) or decrease the number of inodes per call.
+ *
+ *  -ESTALE: The results could not be read from one stable version of
+ *    file system metadata.  Decrease the number of inodes requested.
+ *
+ *  -EUCLEAN: Internal xattr metadata is inconsistent.
+ */
+
+struct scoutfs_ioctl_raw_read_inode_info {
+	__u64 inos_ptr;
+	__u32 inos_count;
+	__u32 names_count;
+	__u64 names_ptr;
+	__u64 results_ptr;
+	__u32 results_size;
+	__u8  _pad[4];
+};
+
+/*
+ * @type is one of the enums that determines the type of the following
+ * result payload.
+ *
+ * @size is the number of bytes of result payload immediately following
+ * the result struct.  It does not include the size of the result struct
+ * header.
+ */
+struct scoutfs_ioctl_raw_read_result {
+	__u32 size;
+	__u8  _pad[7];
+	__u8 type;
+};
+
+/*
+ * The _INODE result contains an initial 64bit inode number followed by a
+ * struct scoutfs_inode as defined in format.h.  The size includes the
+ * 8byte initial inode number.  With that subtracted the size of the
+ * inode struct defines its version (and so the fields it supports).
+ */
+#define SCOUTFS_IOC_RAW_READ_RESULT_INODE	1
+/*
+ * The result payload contains the null terminated name and the value.
+ * The value size can be found by subtracting the null terminated name
+ * length from the result size.
+ */
+#define SCOUTFS_IOC_RAW_READ_RESULT_XATTR	2
+
+#define SCOUTFS_IOC_RAW_READ_INODE_INFO \
+	_IOR(SCOUTFS_IOCTL_MAGIC, 25, struct scoutfs_ioctl_raw_read_inode_info)
+
 #endif

kmod/src/lock.c

@@ -1093,19 +1093,24 @@ out_unlock:
 	return ret;
 }
 
+void scoutfs_lock_get_fs_item_range(u64 ino, struct scoutfs_key *start, struct scoutfs_key *end)
+{
+	scoutfs_key_set_zeros(start);
+	start->sk_zone = SCOUTFS_FS_ZONE;
+	start->ski_ino = cpu_to_le64(ino & ~(u64)SCOUTFS_LOCK_INODE_GROUP_MASK);
+
+	scoutfs_key_set_ones(end);
+	end->sk_zone = SCOUTFS_FS_ZONE;
+	end->ski_ino = cpu_to_le64(ino | SCOUTFS_LOCK_INODE_GROUP_MASK);
+}
+
 int scoutfs_lock_ino(struct super_block *sb, enum scoutfs_lock_mode mode, int flags, u64 ino,
 		     struct scoutfs_lock **ret_lock)
 {
 	struct scoutfs_key start;
 	struct scoutfs_key end;
 
-	scoutfs_key_set_zeros(&start);
-	start.sk_zone = SCOUTFS_FS_ZONE;
-	start.ski_ino = cpu_to_le64(ino & ~(u64)SCOUTFS_LOCK_INODE_GROUP_MASK);
-
-	scoutfs_key_set_ones(&end);
-	end.sk_zone = SCOUTFS_FS_ZONE;
-	end.ski_ino = cpu_to_le64(ino | SCOUTFS_LOCK_INODE_GROUP_MASK);
+	scoutfs_lock_get_fs_item_range(ino, &start, &end);
 
 	return lock_key_range(sb, mode, flags, &start, &end, ret_lock);
 }

kmod/src/lock.h

@@ -65,6 +65,7 @@ int scoutfs_lock_invalidate_request(struct super_block *sb, u64 net_id,
 int scoutfs_lock_recover_request(struct super_block *sb, u64 net_id,
 				 struct scoutfs_key *key);
 
+void scoutfs_lock_get_fs_item_range(u64 ino, struct scoutfs_key *start, struct scoutfs_key *end);
 int scoutfs_lock_inode(struct super_block *sb, enum scoutfs_lock_mode mode, int flags,
 		       struct inode *inode, struct scoutfs_lock **ret_lock);
 int scoutfs_lock_ino(struct super_block *sb, enum scoutfs_lock_mode mode, int flags, u64 ino,

kmod/src/quota.c

@@ -34,6 +34,7 @@
 #include "totl.h"
 #include "util.h"
 #include "quota.h"
+#include "trans.h"
 #include "counters.h"
 #include "scoutfs_trace.h"
 
@@ -1086,6 +1087,10 @@ int scoutfs_quota_mod_rule(struct super_block *sb, bool is_add,
 	if (ret < 0)
 		goto out;
 
+	ret = scoutfs_hold_trans(sb, true);
+	if (ret < 0)
+		goto out;
+
 	down_write(&qtinf->rwsem);
 
 	if (is_add) {
@@ -1095,28 +1100,30 @@ int scoutfs_quota_mod_rule(struct super_block *sb, bool is_add,
 		else if (ret == 0)
 			ret = -EEXIST;
 		if (ret < 0)
-			goto unlock;
+			goto release;
 
 		rule_to_rule_val(&rv, &rule);
 		ret = scoutfs_item_create(sb, &key, &rv, sizeof(rv), lock);
 		if (ret < 0)
-			goto unlock;
+			goto release;
 
 	} else {
 		ret = find_rule(sb, &rule, &key, lock) ?:
 		      scoutfs_item_delete(sb, &key, lock);
 		if (ret < 0)
-			goto unlock;
+			goto release;
 	}
 
 	scoutfs_quota_invalidate(sb);
 	ret = 0;
 
-unlock:
+release:
 	up_write(&qtinf->rwsem);
-	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_WRITE);
+	scoutfs_release_trans(sb);
 
 out:
+	scoutfs_unlock(sb, lock, SCOUTFS_LOCK_WRITE);
+
 	if (is_add)
 		trace_scoutfs_quota_add_rule(sb, &rule, ret);
 	else

kmod/src/raw.c

@@ -0,0 +1,744 @@
+/*
+ * Copyright (C) 2026 Versity Software, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#include <linux/kernel.h>
+#include <linux/fs.h>
+#include <linux/list_sort.h>
+#include <linux/sort.h>
+
+#include "format.h"
+#include "key.h"
+#include "block.h"
+#include "inode.h"
+#include "forest.h"
+#include "client.h"
+#include "ioctl.h"
+#include "lock.h"
+#include "xattr.h"
+#include "attr_x.h"
+#include "bsearch_index.h"
+#include "raw.h"
+
+struct fs_item {
+	struct list_head head;
+	struct scoutfs_key key;
+	u64 seq;
+	int val_len;
+	bool deletion;
+	/* val is aligned so we can deref structs in vals */
+	u8 val[0] __aligned(ARCH_KMALLOC_MINALIGN);
+};
+
+static int save_fs_item(struct list_head *list, struct scoutfs_key *key, u64 seq, u8 flags,
+			void *val, int val_len)
+{
+	struct fs_item *fsi;
+
+	/* max btree val len is hundreds of bytes */
+	fsi = kmalloc(offsetof(struct fs_item, val[val_len]), GFP_NOFS);
+	if (!fsi)
+		return -ENOMEM;
+
+	fsi->key = *key;
+	fsi->seq = seq;
+	fsi->val_len = val_len;
+	fsi->deletion = !!(flags & SCOUTFS_ITEM_FLAG_DELETION);
+	if (val_len > 0)
+		memcpy(fsi->val, val, val_len);
+	list_add_tail(&fsi->head, list);
+
+	return 0;
+}
+
+static void free_fs_item(struct fs_item *fsi)
+{
+	if (!list_empty(&fsi->head))
+		list_del_init(&fsi->head);
+	kfree(fsi);
+}
+
+static void free_fs_items(struct list_head *list)
+{
+	struct fs_item *fsi;
+	struct fs_item *tmp;
+
+	list_for_each_entry_safe(fsi, tmp, list, head)
+		free_fs_item(fsi);
+}
+
+static struct fs_item *next_fs_item(struct list_head *list, struct fs_item *fsi)
+{
+	list_for_each_entry_continue(fsi, list, head)
+		return fsi;
+	return NULL;
+}
+
+static int cmp_fs_items(void *priv, KC_LIST_CMP_CONST struct list_head *A,
+			KC_LIST_CMP_CONST struct list_head *B)
+{
+	KC_LIST_CMP_CONST struct fs_item *a =
+		container_of(A, KC_LIST_CMP_CONST struct fs_item, head);
+	KC_LIST_CMP_CONST struct fs_item *b =
+		container_of(B, KC_LIST_CMP_CONST struct fs_item, head);
+
+	return scoutfs_key_compare(&a->key, &b->key) ?: -scoutfs_cmp(a->seq, b->seq);
+}
+
+static void sort_and_remove(struct list_head *list, struct scoutfs_key *end)
+{
+	struct fs_item *prev;
+	struct fs_item *fsi;
+	struct fs_item *tmp;
+
+	list_sort(NULL, list, cmp_fs_items);
+
+	/* start by removing any items read before end was decreased by later blocks */
+	list_for_each_entry_safe_reverse(fsi, tmp, list, head) {
+		if (scoutfs_key_compare(&fsi->key, end) > 0)
+			free_fs_item(fsi);
+		else
+			break;
+	}
+
+	prev = NULL;
+	list_for_each_entry_safe(fsi, tmp, list, head) {
+		/* remove this item if it's an older version of previous item */
+		if (prev && scoutfs_key_compare(&prev->key, &fsi->key) == 0) {
+			free_fs_item(fsi);
+			continue;
+		}
+
+		/* remove previous deletion item once it has removed all older versions */
+		if (prev && prev->deletion)
+			free_fs_item(prev);
+
+		/* next item might match this, record to compare */
+		prev = fsi;
+	}
+
+	/* remove the last item if it's a deletion */
+	list_for_each_entry_reverse(fsi, list, head) {
+		if (fsi->deletion)
+			free_fs_item(fsi);
+		break;
+	}
+}
+
+static int save_all_items(struct super_block *sb, struct scoutfs_key *key, u64 seq, u8 flags,
+			  void *val, int val_len, int fic, void *arg)
+{
+	struct list_head *list = arg;
+
+	return save_fs_item(list, key, seq, flags, val, val_len);
+}
+
+/* -------------- */
+
+static void ms_from_key(struct scoutfs_ioctl_meta_seq *ms, struct scoutfs_key *key)
+{
+	ms->meta_seq = le64_to_cpu(key->skii_major);
+	ms->ino = le64_to_cpu(key->skii_ino);
+}
+
+/*
+ * Increment the key's ino->meta_seq so that we don't land between items.
+ */
+static void inc_meta_seq(struct scoutfs_key *key)
+{
+	le64_add_cpu(&key->skii_ino, 1);
+	if (key->skii_ino == 0)
+		le64_add_cpu(&key->skii_major, 1);
+}
+
+int scoutfs_raw_read_meta_seq(struct super_block *sb,
+			      struct scoutfs_ioctl_raw_read_meta_seq *rms,
+			      struct scoutfs_ioctl_meta_seq *last_ret)
+{
+	struct scoutfs_ioctl_meta_seq __user *ums;
+	struct scoutfs_ioctl_meta_seq ms;
+	struct scoutfs_net_roots roots;
+	DECLARE_SAVED_REFS(saved);
+	struct scoutfs_key start;
+	struct scoutfs_key last;
+	struct scoutfs_key key;
+	struct scoutfs_key end;
+	struct fs_item *fsi;
+	struct fs_item *tmp;
+	LIST_HEAD(list);
+	int retries;
+	int copied;
+	int count;
+	int ret;
+
+	ums = (void __user *)rms->results_ptr;
+	count = rms->results_size / sizeof(struct scoutfs_ioctl_meta_seq);
+	retries = 10;
+	copied = 0;
+
+	scoutfs_inode_init_index_key(&last, SCOUTFS_INODE_INDEX_META_SEQ_TYPE,
+				     rms->end.meta_seq, 0, rms->end.ino);
+
+retry:
+	ret = scoutfs_client_get_roots(sb, &roots);
+	if (ret)
+		goto out;
+
+	scoutfs_inode_init_index_key(&key, SCOUTFS_INODE_INDEX_META_SEQ_TYPE,
+				     rms->start.meta_seq, 0, rms->start.ino);
+
+	for (;;) {
+		start = key;
+		end = last;
+		ret = scoutfs_forest_read_items_roots(sb, &roots, 0, &key, NULL, &start, &end,
+						      save_all_items, &list);
+		if (ret < 0)
+			goto out;
+
+		sort_and_remove(&list, &end);
+
+		list_for_each_entry_safe(fsi, tmp, &list, head) {
+
+			if (copied == count) {
+				/* results are full, set end to before item can't return */
+				end = fsi->key;
+				le64_add_cpu(&end.skii_ino, -1ULL);
+				ret = 0;
+				goto out;
+			}
+
+			ms_from_key(&ms, &fsi->key);
+			if (copy_to_user(&ums[copied], &ms, sizeof(ms))) {
+				ret = -EFAULT;
+				goto out;
+			}
+
+			free_fs_item(fsi);
+			copied++;
+		}
+
+		if (scoutfs_key_compare(&end, &last) >= 0) {
+			end = last;
+			break;
+		}
+
+		key = end;
+		inc_meta_seq(&key);
+	}
+
+	ret = 0;
+out:
+	free_fs_items(&list);
+
+	ret = scoutfs_block_check_stale(sb, ret, &saved, &roots.fs_root.ref, &roots.logs_root.ref);
+	if (ret == -ESTALE && copied == 0 && retries-- > 0)
+		goto retry;
+
+	ms_from_key(last_ret, &end);
+
+	return ret ?: copied;
+}
+
+/* -------------- */
+
+struct inode_info_context {
+	size_t nr_inos;
+	u64 *inos;
+
+	size_t nr_names;
+	struct xattr_name {
+		u64 hash;
+		char *name;
+		u8 name_len; /* no null */
+	} *names;
+
+	struct list_head fs_items;
+};
+
+static int cmp_u64(const void *A, const void *B)
+{
+	const u64 *a = A;
+	const u64 *b = B;
+
+	return scoutfs_cmp(*a, *b);
+}
+
+static int cmp_name_hash(const void *A, const void *B)
+{
+	const struct xattr_name *a = A;
+	const struct xattr_name *b = B;
+
+	return scoutfs_cmp(a->hash, b->hash);
+}
+
+static int cmp_name_string(const void *A, const void *B)
+{
+	const struct xattr_name *a = A;
+	const struct xattr_name *b = B;
+
+	return scoutfs_cmp(a->name_len, b->name_len) ?: memcmp(a->name, b->name, a->name_len);
+}
+
+static int setup_context(struct inode_info_context *ctx,
+			 struct scoutfs_ioctl_raw_read_inode_info *rii)
+{
+	__u64 __user *uinos = (void __user *)rii->inos_ptr;
+	char __user *uname;
+	long len_null;
+	long len;
+	int ret;
+	u32 i;
+
+	ctx->nr_inos = rii->inos_count;
+	ctx->nr_names = rii->names_count;
+	INIT_LIST_HEAD(&ctx->fs_items);
+
+	ctx->inos = kvmalloc_array(ctx->nr_inos, sizeof(ctx->inos[0]), GFP_KERNEL);
+	ctx->names = kvcalloc(ctx->nr_names, sizeof(ctx->names[0]), GFP_KERNEL);
+	if (!ctx->inos || !ctx->names) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	if (copy_from_user(ctx->inos, uinos, ctx->nr_inos * sizeof(ctx->inos[0]))) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	/* inos must not be 0 and must increase and contain no duplicates */
+	if (ctx->inos[0] == 0) {
+		ret = -EINVAL;
+		goto out;
+	}
+	for (i = 1; i < ctx->nr_inos; i++) {
+		if (ctx->inos[i] <= ctx->inos[i - 1]) {
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	uname = (void __user *)rii->names_ptr;
+	for (i = 0; i < ctx->nr_names; i++) {
+		len_null = SCOUTFS_XATTR_MAX_NAME_LEN + 1;
+		ret = strnlen_user(uname, len_null);
+		if (ret <= 1 || ret > len_null) {
+			if (ret >= 0)
+				ret = -EINVAL;
+			goto out;
+		}
+		len_null = ret;
+		len = len_null - 1;
+
+		ctx->names[i].name_len = len;
+		ctx->names[i].name = kmalloc(len_null, GFP_KERNEL);
+		if (!ctx->names[i].name) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ret = strncpy_from_user(ctx->names[i].name, uname, len_null);
+		if (ret != len) {
+			if (ret >= 0)
+				ret = -EINVAL;
+			goto out;
+		}
+
+		ctx->names[i].hash = scoutfs_xattr_name_hash(ctx->names[i].name, len);
+		uname += len_null;
+	}
+
+	/* make sure all the names differ */
+	sort(ctx->names, ctx->nr_names, sizeof(ctx->names[0]), cmp_name_string, NULL);
+	for (i = 1; i < ctx->nr_names; i++) {
+		if (cmp_name_string(&ctx->names[i - 1], &ctx->names[i]) == 0) {
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	/* then leave them sorted by hash */
+	sort(ctx->names, ctx->nr_names, sizeof(ctx->names[0]), cmp_name_hash, NULL);
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static void free_context(struct inode_info_context *ctx)
+{
+	int i;
+
+	kvfree(ctx->inos);
+
+	if (ctx->names) {
+		for (i = 0; i < ctx->nr_names; i++) {
+			if (!ctx->names[i].name)
+				break;
+			kfree(ctx->names[i].name);
+		}
+		kvfree(ctx->names);
+	}
+}
+
+/*
+ * Iterate over fs items and save any that we're interested in.  We want
+ * inode struct items and any xattr items whose hashes collide with the
+ * xattr names we're searching for.
+ *
+ * Our forest calls can be advancing through the key space as we see
+ * slices that intersect with blocks in trees.  And each forest caller
+ * can be resetting the key position to the start of each forest block
+ * it reads in an intersection.
+ *
+ * From this callback's perspective, the key can be jumping all over the
+ * place.  We don't have any iterative position state.  For each key we
+ * decide if we want to save it and then set the key to the next key we
+ * want after the current key.  We'll combine all the saved keys later.
+ */
+static int save_info_items(struct super_block *sb, struct scoutfs_key *key, u64 seq,
+			   u8 flags, void *val, int val_len, int fic, void *arg)
+{
+	u64 ino = le64_to_cpu(key->_sk_first);
+	struct inode_info_context *ctx = arg;
+	struct xattr_name name;
+	size_t name_ind;
+	size_t ino_ind;
+	bool hash_match;
+	bool ino_match;
+	int ret;
+
+	ino_ind = bsearch_index(&ino, ctx->inos, ctx->nr_inos, sizeof(ctx->inos[0]), cmp_u64);
+	ino_match = ino_ind < ctx->nr_inos && ctx->inos[ino_ind] == ino;
+
+	/* jump to to next ino, could be for this key if we're before the ino struct */
+	if (!ino_match || key->sk_type < SCOUTFS_INODE_TYPE)
+		goto next_inode;
+
+	/* find our search position in xattrs */
+	if (key->sk_type < SCOUTFS_XATTR_TYPE) {
+		name_ind = 0;
+		hash_match = false;
+
+	} else if (key->sk_type == SCOUTFS_XATTR_TYPE) {
+		name = (struct xattr_name) { .hash = le64_to_cpu(key->skx_name_hash) };
+		name_ind = bsearch_index(&name, ctx->names, ctx->nr_names, sizeof(ctx->names[0]),
+					 cmp_name_hash);
+		hash_match = name_ind < ctx->nr_names && ctx->names[name_ind].hash == name.hash;
+	} else {
+		name_ind = ctx->nr_names;
+		hash_match = false;
+	}
+
+	/* save inode items for our search and all xattr items that match search hashes */
+	if (key->sk_type == SCOUTFS_INODE_TYPE || hash_match) {
+		ret = save_fs_item(&ctx->fs_items, key,  seq, flags, val, val_len);
+		if (ret < 0)
+			goto out;
+	}
+
+	/* let the caller continue iterating through matching xattr items */
+	if (hash_match) {
+		ret = 0;
+		goto out;
+	}
+
+	/* jump to the next xattr */
+	if (name_ind < ctx->nr_names) {
+		scoutfs_xattr_init_key(key, ino, ctx->names[name_ind].hash, 0);
+		ret = -ESRCH;
+		goto out;
+	}
+
+	/* no more xattrs, must be done with this ino */
+	ino_ind++;
+
+next_inode:
+	/* now jump to next inode struct key, or we're done */
+	if (ino_ind < ctx->nr_inos)
+		scoutfs_inode_init_key(key, ctx->inos[ino_ind]);
+	else
+		scoutfs_key_set_ones(key);
+
+	ret = -ESRCH;
+out:
+	return ret;
+}
+
+static int copy_to_user_off(void __user *dst, size_t *dst_off, size_t dst_size,
+			    void *src, size_t copy_size)
+{
+	if (copy_size == 0)
+		return 0;
+	if (*dst_off + copy_size > dst_size)
+		return -ERANGE;
+	if (copy_to_user(dst + *dst_off, src, copy_size))
+		return -EFAULT;
+
+	*dst_off += copy_size;
+	return 0;
+}
+
+static int copy_result_to_user(void __user *ures, size_t *off, size_t size, u8 type,
+			       void *a_data, size_t a_len, void *b_data, size_t b_len,
+			       size_t extra_size)
+{
+	struct scoutfs_ioctl_raw_read_result res;
+	const size_t szof_res = sizeof(struct scoutfs_ioctl_raw_read_result);
+
+	memzero_explicit(&res, szof_res);
+	res = (struct scoutfs_ioctl_raw_read_result) {
+		.size = a_len + b_len + extra_size,
+		.type = type,
+	};
+
+	return copy_to_user_off(ures, off, size, &res, szof_res) ?:
+	       (a_len ? copy_to_user_off(ures, off, size, a_data, a_len) : 0) ?:
+	       (b_len ? copy_to_user_off(ures, off, size, b_data, b_len) : 0);
+}
+
+static int copy_item_results_to_user(struct super_block *sb, struct inode_info_context *ctx,
+				     void __user *ures, size_t *off, size_t size,
+				     struct fs_item *fsi)
+{
+	struct scoutfs_inode *cinode;
+	struct scoutfs_xattr *xat;
+	static char null = '\0';
+	size_t len;
+	u64 ino;
+	int ret = 0;
+
+	if (fsi->key.sk_type == SCOUTFS_INODE_TYPE) {
+		cinode = (void *)fsi->val;
+		ino = le64_to_cpu(fsi->key.ski_ino);
+
+		ret = copy_result_to_user(ures, off, size, SCOUTFS_IOC_RAW_READ_RESULT_INODE,
+					  &ino, sizeof(ino), cinode, sizeof(struct scoutfs_inode),
+					  0);
+
+	} else if (fsi->key.sk_type == SCOUTFS_XATTR_TYPE) {
+		if (fsi->key.skx_part == 0) {
+			xat = (void *)fsi->val;
+			ret = copy_result_to_user(ures, off, size,
+						  SCOUTFS_IOC_RAW_READ_RESULT_XATTR, xat->name,
+						  xat->name_len, &null, sizeof(null),
+						  le16_to_cpu(xat->val_len));
+			if (ret == 0 && xat->val_len != 0) {
+				/* then append the start of the value */
+				len = fsi->val_len -
+				      offsetof(struct scoutfs_xattr, name[xat->name_len]);
+				ret = copy_to_user_off(ures, off, size, xat->name + xat->name_len,
+						       len);
+			}
+		} else {
+			/* continue appending partial values */
+			ret = copy_to_user_off(ures, off, size, fsi->val, fsi->val_len);
+		}
+	}
+
+	return ret;
+}
+
+static bool ignore_zero_nlink(struct inode_info_context *ctx, struct fs_item *fsi)
+{
+	struct scoutfs_inode *cinode = (void *)fsi->val;
+
+	return cinode->nlink == 0;
+}
+
+static bool ignore_xattr_name(struct inode_info_context *ctx, struct fs_item *fsi)
+{
+	struct scoutfs_xattr *xat = (void *)fsi->val;
+	struct xattr_name name = {
+		.hash = le64_to_cpu(fsi->key.skx_name_hash),
+		.name = xat->name,
+		.name_len = xat->name_len,
+	};
+	size_t i;
+
+	for (i = bsearch_index(&name, ctx->names, ctx->nr_names, sizeof(ctx->names[0]),
+			       cmp_name_hash);
+	     i < ctx->nr_names && name.hash == ctx->names[i].hash; i++) {
+		if (cmp_name_string(&name, &ctx->names[i]) == 0)
+			return false;
+	}
+
+	return true;
+}
+
+static int copy_results_to_user(struct super_block *sb, struct inode_info_context *ctx,
+				struct scoutfs_ioctl_raw_read_inode_info *rii)
+{
+	void __user *ures = (void __user *)rii->results_ptr;
+	struct scoutfs_xattr *xat;
+	struct fs_item *next;
+	struct fs_item *fsi;
+	struct fs_item *tmp;
+	size_t xattr_end;
+	size_t off;
+	__le64 in_ino;
+	__le64 in_id;
+	int ret;
+
+	in_ino = 0;
+	xattr_end = 0;
+	in_id = 0;
+	off = 0;
+
+	list_for_each_entry_safe(fsi, tmp, &ctx->fs_items, head) {
+		/*
+		 * ignore:
+		 *  - inodes with an nlink of 0
+		 *  - all items for an ino after the inode struct that we're ignoring 
+		 *  - first xattr parts with a name we don't need
+		 *  - additional xattr parts when we ignored the first
+		 */
+		if ((fsi->key.sk_type == SCOUTFS_INODE_TYPE && ignore_zero_nlink(ctx, fsi)) ||
+		    (fsi->key.sk_type > SCOUTFS_INODE_TYPE && fsi->key._sk_first != in_ino) ||
+		    (fsi->key.sk_type == SCOUTFS_XATTR_TYPE &&
+		     ((fsi->key.skx_part == 0 && ignore_xattr_name(ctx, fsi)) ||
+		      (fsi->key.skx_part > 0 && fsi->key.skx_id != in_id)))) {
+			free_fs_item(fsi);
+			in_ino = 0;
+			in_id = 0;
+			continue;
+		}
+
+		/* advance ino/xattr stream context state machine */
+		if (fsi->key.sk_type == SCOUTFS_INODE_TYPE) {
+			in_ino = fsi->key.ski_ino;
+			in_id = 0;
+		} else if (fsi->key.sk_type == SCOUTFS_XATTR_TYPE && fsi->key.skx_part == 0) {
+			in_id = fsi->key.skx_id;
+			/* save the required offset after the complete xattr */
+			xat = (void *)fsi->val;
+			xattr_end = off + sizeof(struct scoutfs_ioctl_raw_read_result) +
+				    xat->name_len + 1 + le16_to_cpu(xat->val_len);
+		}
+
+		/* copy results, usually with header, but additional xattr parts copied raw */
+		ret = copy_item_results_to_user(sb, ctx, ures, &off, rii->results_size, fsi);
+		if (ret < 0)
+			goto out;
+
+		/* make sure we saw all xattr parts and copied the correct size */
+		if (xattr_end > 0 &&
+		    !((next = next_fs_item(&ctx->fs_items, fsi)) &&
+		      next->key.sk_type == SCOUTFS_XATTR_TYPE && next->key.skx_ino == in_ino &&
+		      next->key.skx_id == in_id)) {
+			if (off != xattr_end) {
+				ret = -EUCLEAN;
+				goto out;
+			}
+			xattr_end = 0;
+		}
+	}
+
+	ret = 0;
+out:
+	return ret ?: off;
+}
+
+/*
+ * If the key is for an inode we're not interested in, or if its past
+ * the xattr items, then advance to the next inode.  This is used
+ * between forest read items calls to avoid leaf blocks.  The callback
+ * takes care of iterating through the items for an inode across
+ * multiple leaves.
+ */
+static void advance_key_ino(struct scoutfs_key *key, struct inode_info_context *ctx)
+{
+	u64 ino = le64_to_cpu(key->_sk_first);
+	size_t ino_ind;
+
+	ino_ind = bsearch_index(&ino, ctx->inos, ctx->nr_inos, sizeof(ctx->inos[0]), cmp_u64);
+	if (ino_ind < ctx->nr_inos && ctx->inos[ino_ind] == ino) {
+		if (key->sk_type <= SCOUTFS_XATTR_TYPE)
+			return;
+		else
+			ino_ind++;
+	}
+
+	if (ino_ind < ctx->nr_inos)
+		scoutfs_inode_init_key(key, ctx->inos[ino_ind]);
+	else
+		scoutfs_key_set_ones(key);
+}
+
+int scoutfs_raw_read_inode_info(struct super_block *sb,
+				struct scoutfs_ioctl_raw_read_inode_info *rii)
+{
+	struct inode_info_context ctx = {0, };
+	struct scoutfs_net_roots roots;
+	DECLARE_SAVED_REFS(saved);
+	struct scoutfs_key lock_start;
+	struct scoutfs_key lock_end;
+	struct scoutfs_key start;
+	struct scoutfs_key last;
+	struct scoutfs_key key;
+	struct scoutfs_key end;
+	LIST_HEAD(list);
+	int retries = 10;
+	int ret;
+
+	ret = setup_context(&ctx, rii);
+	if (ret < 0)
+		goto out;
+
+	if (ctx.nr_names > 0)
+		scoutfs_xattr_init_key(&last, ctx.inos[ctx.nr_inos -1],
+				       ctx.names[ctx.nr_names - 1].hash, U64_MAX);
+	else
+		scoutfs_inode_init_key(&last, ctx.inos[ctx.nr_inos - 1]);
+
+retry:
+	ret = scoutfs_client_get_roots(sb, &roots);
+	if (ret)
+		goto out;
+
+	scoutfs_inode_init_key(&key, ctx.inos[0]);
+
+	while (scoutfs_key_compare(&key, &last) <= 0) {
+		scoutfs_lock_get_fs_item_range(le64_to_cpu(key._sk_first), &lock_start, &lock_end);
+
+		start = key;
+		end = last;
+		if (scoutfs_key_compare(&lock_end, &end) < 0)
+			end = lock_end;
+
+		ret = scoutfs_forest_read_items_roots(sb, &roots, 0, &key, &lock_start,
+						      &start, &end, save_info_items, &ctx);
+		if (ret < 0)
+			goto out;
+
+		/* save each sorted batch, might have partial results for an inode */
+		sort_and_remove(&ctx.fs_items, &end);
+		list_splice_tail_init(&ctx.fs_items, &list);
+
+		key = end;
+		if (!scoutfs_key_is_ones(&key)) {
+			scoutfs_key_inc(&key);
+			advance_key_ino(&key, &ctx);
+		}
+	}
+
+	list_splice_tail_init(&list, &ctx.fs_items);
+	ret = copy_results_to_user(sb, &ctx, rii);
+out:
+	free_fs_items(&list);
+	free_fs_items(&ctx.fs_items);
+
+	ret = scoutfs_block_check_stale(sb, ret, &saved, &roots.fs_root.ref, &roots.logs_root.ref);
+	if (ret == -ESTALE && retries-- > 0)
+		goto retry;
+
+	free_context(&ctx);
+	return ret;
+}

kmod/src/raw.h

@@ -0,0 +1,10 @@
+#ifndef _SCOUTFS_RAW_H_
+#define _SCOUTFS_RAW_H_
+
+int scoutfs_raw_read_meta_seq(struct super_block *sb,
+			      struct scoutfs_ioctl_raw_read_meta_seq *rms,
+			      struct scoutfs_ioctl_meta_seq *last_ret);
+int scoutfs_raw_read_inode_info(struct super_block *sb,
+				struct scoutfs_ioctl_raw_read_inode_info *rii);
+
+#endif

kmod/src/totl.c

@@ -30,6 +30,11 @@ void scoutfs_totl_merge_init(struct scoutfs_totl_merging *merg)
 	memset(merg, 0, sizeof(struct scoutfs_totl_merging));
 }
 
+/*
+ * bin the incoming merge inputs so that we can resolve delta items
+ * properly. Finalized logs that are merge inputs are kept separately
+ * from those that are not.
+ */
 void scoutfs_totl_merge_contribute(struct scoutfs_totl_merging *merg,
 				   u64 seq, u8 flags, void *val, int val_len, int fic)
 {
@@ -39,10 +44,10 @@ void scoutfs_totl_merge_contribute(struct scoutfs_totl_merging *merg,
 		merg->fs_seq = seq;
 		merg->fs_total = le64_to_cpu(tval->total);
 		merg->fs_count = le64_to_cpu(tval->count);
-	} else if (fic & FIC_FINALIZED) {
-		merg->fin_seq = seq;
-		merg->fin_total += le64_to_cpu(tval->total);
-		merg->fin_count += le64_to_cpu(tval->count);
+	} else if (fic & FIC_MERGE_INPUT) {
+		merg->inp_seq = seq;
+		merg->inp_total += le64_to_cpu(tval->total);
+		merg->inp_count += le64_to_cpu(tval->count);
 	} else {
 		merg->log_seq = seq;
 		merg->log_total += le64_to_cpu(tval->total);
@@ -53,15 +58,18 @@ void scoutfs_totl_merge_contribute(struct scoutfs_totl_merging *merg,
 /*
  * .totl. item merging has to be careful because the log btree merging
  * code can write partial results to the fs_root.  This means that a
- * reader can see both cases where new finalized logs should be applied
- * to the old fs items and where old finalized logs have already been
- * applied to the partially merged fs items.  Currently active logged
- * items are always applied on top of all cases.
+ * reader can see both cases where merge input deltas should be applied
+ * to the old fs items and where they have already been applied to the
+ * partially merged fs items.
+ *
+ * Only finalized log trees that are inputs to the current merge cycle
+ * are tracked in the inp_ bucket.  Finalized trees that aren't merge
+ * inputs and active log trees are always applied unconditionally since
+ * they cannot be in fs_root.
  *
  * These cases are differentiated with a combination of sequence numbers
- * in items, the count of contributing xattrs, and a flag
- * differentiating finalized and active logged items.  This lets us
- * recognize all cases, including when finalized logs were merged and
+ * in items and the count of contributing xattrs.  This lets us
+ * recognize all cases, including when merge inputs were merged and
  * deleted the fs item.
  */
 void scoutfs_totl_merge_resolve(struct scoutfs_totl_merging *merg, __u64 *total, __u64 *count)
@@ -75,14 +83,14 @@ void scoutfs_totl_merge_resolve(struct scoutfs_totl_merging *merg, __u64 *total,
 		*count = merg->fs_count;
 	}
 
-	/* apply finalized logs if they're newer or creating */
-	if (((merg->fs_seq != 0) && (merg->fin_seq > merg->fs_seq)) ||
-	    ((merg->fs_seq == 0) && (merg->fin_count > 0))) {
-		*total += merg->fin_total;
-		*count += merg->fin_count;
+	/* apply merge input deltas if they're newer or creating */
+	if (((merg->fs_seq != 0) && (merg->inp_seq > merg->fs_seq)) ||
+	    ((merg->fs_seq == 0) && (merg->inp_count > 0))) {
+		*total += merg->inp_total;
+		*count += merg->inp_count;
 	}
 
-	/* always apply active logs which must be newer than fs and finalized */
+	/* always apply non-input finalized and active logs */
 	if (merg->log_seq > 0) {
 		*total += merg->log_total;
 		*count += merg->log_count;

kmod/src/totl.h

@@ -7,9 +7,9 @@ struct scoutfs_totl_merging {
 	u64 fs_seq;
 	u64 fs_total;
 	u64 fs_count;
-	u64 fin_seq;
-	u64 fin_total;
-	s64 fin_count;
+	u64 inp_seq;
+	u64 inp_total;
+	s64 inp_count;
 	u64 log_seq;
 	u64 log_total;
 	s64 log_count;

kmod/src/triggers.c

@@ -46,6 +46,7 @@ static char *names[] = {
 	[SCOUTFS_TRIGGER_SRCH_MERGE_STOP_SAFE] = "srch_merge_stop_safe",
 	[SCOUTFS_TRIGGER_STATFS_LOCK_PURGE] = "statfs_lock_purge",
 	[SCOUTFS_TRIGGER_RECLAIM_SKIP_FINALIZE] = "reclaim_skip_finalize",
+	[SCOUTFS_TRIGGER_LOG_MERGE_FORCE_PARTIAL] = "log_merge_force_partial",
 };
 
 bool scoutfs_trigger_test_and_clear(struct super_block *sb, unsigned int t)

kmod/src/triggers.h

@@ -9,6 +9,7 @@ enum scoutfs_trigger {
 	SCOUTFS_TRIGGER_SRCH_MERGE_STOP_SAFE,
 	SCOUTFS_TRIGGER_STATFS_LOCK_PURGE,
 	SCOUTFS_TRIGGER_RECLAIM_SKIP_FINALIZE,
+	SCOUTFS_TRIGGER_LOG_MERGE_FORCE_PARTIAL,
 	SCOUTFS_TRIGGER_NR,
 };
 

kmod/src/wkic.c

@@ -30,7 +30,6 @@
 #include "counters.h"
 #include "scoutfs_trace.h"
 #include "wkic.h"
-#include "msg.h"
 
 /*
  * This weaker item cache differs from the core item cache in item.c:
@@ -96,6 +95,7 @@ struct wkic_info {
 	/* block reading slow path */
 	struct mutex roots_mutex;
 	struct scoutfs_net_roots roots;
+	u64 merge_input_seq;
 	u64 roots_read_seq;
 	ktime_t roots_expire;
 
@@ -747,16 +747,6 @@ static void fill_page_items(struct super_block *sb, struct wkic_page *wpage, str
 				rb_erase(&witem->node, root);
 				kfree(witem);
 				continue;
-			} else if (tval->count == 0) {
-				/*
-				 * BUG: there are no contributing items but count != 0,
-				 * which shouldn't happen - we've gone off kilt.
-				 */
-				scoutfs_err(sb, "non-zero value for zero count totl "SK_FMT", dropping item",
-					    SK_ARG(&witem->key));
-				rb_erase(&witem->node, root);
-				kfree(witem);
-				continue;
 			}
 		}
 
@@ -816,29 +806,79 @@ static void free_page_list(struct super_block *sb, struct list_head *list)
  * read_seq number so that we can compare the age of the items in cached
  * pages.  Only one request to refresh the roots is in progress at a
  * time.  This is the slow path that's only used when the cache isn't
- * populated and the roots aren't cached.  The root request is fast
- * enough, especially compared to the resulting item reading IO, that we
- * don't mind hiding it behind a trivial mutex.
+ * populated and the roots aren't cached.
+ *
+ * We read roots directly from the on-disk superblock rather than
+ * requesting them from the server so that we can also read the
+ * log_merge btree from the same superblock.  The merge status item
+ * seq tells us which finalized log trees are inputs to the current
+ * merge, which is needed to correctly resolve totl delta items.
  */
-static int get_roots(struct super_block *sb, struct wkic_info *winf,
-		     struct scoutfs_net_roots *roots_ret, u64 *read_seq, bool force_new)
+static int refresh_roots(struct super_block *sb, struct wkic_info *winf)
+{
+	struct scoutfs_super_block *super;
+	struct scoutfs_log_merge_status *stat;
+	SCOUTFS_BTREE_ITEM_REF(iref);
+	struct scoutfs_key key;
+	int ret;
+
+	super = kmalloc(sizeof(*super), GFP_NOFS);
+	if (!super)
+		return -ENOMEM;
+
+	ret = scoutfs_read_super(sb, super);
+	if (ret < 0)
+		goto out;
+
+	winf->roots = (struct scoutfs_net_roots){
+		.fs_root = super->fs_root,
+		.logs_root = super->logs_root,
+		.srch_root = super->srch_root,
+	};
+
+	winf->merge_input_seq = 0;
+	if (super->log_merge.ref.blkno) {
+		scoutfs_key_set_zeros(&key);
+		key.sk_zone = SCOUTFS_LOG_MERGE_STATUS_ZONE;
+		ret = scoutfs_btree_lookup(sb, &super->log_merge, &key, &iref);
+		if (ret == 0) {
+			if (iref.val_len == sizeof(*stat)) {
+				stat = iref.val;
+				winf->merge_input_seq = le64_to_cpu(stat->seq);
+			} else {
+				ret = -EUCLEAN;
+			}
+			scoutfs_btree_put_iref(&iref);
+		} else if (ret == -ENOENT) {
+			ret = 0;
+		}
+		if (ret < 0)
+			goto out;
+	}
+
+	winf->roots_read_seq++;
+	winf->roots_expire = ktime_add_ms(ktime_get_raw(), WKIC_CACHE_LIFETIME_MS);
+out:
+	kfree(super);
+	return ret;
+}
+
+static int get_roots(struct super_block *sb, struct wkic_info *winf,
+		     struct scoutfs_net_roots *roots_ret, u64 *merge_input_seq,
+		     u64 *read_seq, bool force_new)
 {
-	struct scoutfs_net_roots roots;
 	int ret;
 
 	mutex_lock(&winf->roots_mutex);
 
 	if (force_new || ktime_before(winf->roots_expire, ktime_get_raw())) {
-		ret = scoutfs_client_get_roots(sb, &roots);
+		ret = refresh_roots(sb, winf);
 		if (ret < 0)
 			goto out;
-
-		winf->roots = roots;
-		winf->roots_read_seq++;
-		winf->roots_expire = ktime_add_ms(ktime_get_raw(), WKIC_CACHE_LIFETIME_MS);
 	}
 
 	*roots_ret = winf->roots;
+	*merge_input_seq = winf->merge_input_seq;
 	*read_seq = winf->roots_read_seq;
 	ret = 0;
 out:
@@ -881,24 +921,30 @@ static int insert_read_pages(struct super_block *sb, struct wkic_info *winf,
 	struct scoutfs_key end;
 	struct wkic_page *wpage;
 	LIST_HEAD(pages);
-	u64 read_seq;
+	u64 merge_input_seq;
+	u64 read_seq = 0;
 	int ret;
 
 	ret = 0;
 retry_stale:
-	ret = get_roots(sb, winf, &roots, &read_seq, ret == -ESTALE);
+	ret = get_roots(sb, winf, &roots, &merge_input_seq, &read_seq, ret == -ESTALE);
 	if (ret < 0)
-		goto out;
+		goto check_stale;
 
 	start = *range_start;
 	end = *range_end;
-	ret = scoutfs_forest_read_items_roots(sb, &roots, key, range_start, &start, &end,
-					      read_items_cb, &root);
+	ret = scoutfs_forest_read_items_roots(sb, &roots, merge_input_seq, key, range_start,
+					      &start, &end, read_items_cb, &root);
 	trace_scoutfs_wkic_read_items(sb, key, &start, &end);
+check_stale:
 	ret = scoutfs_block_check_stale(sb, ret, &saved, &roots.fs_root.ref, &roots.logs_root.ref);
 	if (ret < 0) {
-		if (ret == -ESTALE)
+		if (ret == -ESTALE) {
+			/* not safe to retry due to delta items, must restart clean */
+			free_item_tree(&root);
+			root = RB_ROOT;
 			goto retry_stale;
+		}
 		goto out;
 	}
 

kmod/src/xattr.c

@@ -47,7 +47,7 @@
  *  - add acl support and call generic xattr->handlers for SYSTEM
  */
 
-static u32 xattr_name_hash(const char *name, unsigned int name_len)
+u32 scoutfs_xattr_name_hash(const char *name, unsigned int name_len)
 {
 	return crc32c(U32_MAX, name, name_len);
 }
@@ -65,8 +65,7 @@ static unsigned int xattr_nr_parts(struct scoutfs_xattr *xat)
 				      le16_to_cpu(xat->val_len));
 }
 
-static void init_xattr_key(struct scoutfs_key *key, u64 ino, u32 name_hash,
-			   u64 id)
+void scoutfs_xattr_init_key(struct scoutfs_key *key, u64 ino, u32 name_hash, u64 id)
 {
 	*key = (struct scoutfs_key) {
 		.sk_zone = SCOUTFS_FS_ZONE,
@@ -187,10 +186,10 @@ static int get_next_xattr(struct inode *inode, struct scoutfs_key *key,
 		return -EINVAL;
 
 	if (name_len)
-		name_hash = xattr_name_hash(name, name_len);
+		name_hash = scoutfs_xattr_name_hash(name, name_len);
 
-	init_xattr_key(key, scoutfs_ino(inode), name_hash, id);
-	init_xattr_key(&last, scoutfs_ino(inode), U32_MAX, U64_MAX);
+	scoutfs_xattr_init_key(key, scoutfs_ino(inode), name_hash, id);
+	scoutfs_xattr_init_key(&last, scoutfs_ino(inode), U32_MAX, U64_MAX);
 
 	for (;;) {
 		ret = scoutfs_item_next(sb, key, &last, xat, xat_bytes, lock);
@@ -335,8 +334,8 @@ static int create_xattr_items(struct inode *inode, u64 id, struct scoutfs_xattr
 	int len;
 	int i;
 
-	init_xattr_key(&key, scoutfs_ino(inode),
-		       xattr_name_hash(xat->name, xat->name_len), id);
+	scoutfs_xattr_init_key(&key, scoutfs_ino(inode),
+		       scoutfs_xattr_name_hash(xat->name, xat->name_len), id);
 
 	for (i = 0; i < new_parts; i++) {
 		key.skx_part = i;
@@ -365,7 +364,7 @@ static int delete_xattr_items(struct inode *inode, u32 name_hash, u64 id,
 	int ret = 0;
 	int i;
 
-	init_xattr_key(&key, scoutfs_ino(inode), name_hash, id);
+	scoutfs_xattr_init_key(&key, scoutfs_ino(inode), name_hash, id);
 
 	/* dirty additional existing old items */
 	for (i = 1; i < nr_parts; i++) {
@@ -407,8 +406,8 @@ static int change_xattr_items(struct inode *inode, u64 id,
 	int i;
 	int ret;
 
-	init_xattr_key(&key, scoutfs_ino(inode),
-		       xattr_name_hash(xat->name, xat->name_len), id);
+	scoutfs_xattr_init_key(&key, scoutfs_ino(inode),
+			       scoutfs_xattr_name_hash(xat->name, xat->name_len), id);
 
 	/* dirty existing old items */
 	for (i = 0; i < old_parts; i++) {
@@ -1224,8 +1223,8 @@ int scoutfs_xattr_drop(struct super_block *sb, u64 ino,
 		goto out;
 	}
 
-	init_xattr_key(&key, ino, 0, 0);
-	init_xattr_key(&last, ino, U32_MAX, U64_MAX);
+	scoutfs_xattr_init_key(&key, ino, 0, 0);
+	scoutfs_xattr_init_key(&last, ino, U32_MAX, U64_MAX);
 
 	for (;;) {
 		ret = scoutfs_item_next(sb, &key, &last, (void *)xat, bytes,
@@ -1265,6 +1264,7 @@ int scoutfs_xattr_drop(struct super_block *sb, u64 ino,
 			ret = parse_indx_key(&tag_key, xat->name, xat->name_len, ino);
 			if (ret < 0)
 				goto out;
+			scoutfs_xattr_set_indx_key_xid(&tag_key, le64_to_cpu(key.skx_id));
 		}
 
 		if ((tgs.totl || tgs.indx) && locked_zone != tag_key.sk_zone) {

kmod/src/xattr.h

@@ -10,6 +10,9 @@ struct scoutfs_xattr_prefix_tags {
 
 extern const struct xattr_handler *scoutfs_xattr_handlers[];
 
+u32 scoutfs_xattr_name_hash(const char *name, unsigned int name_len);
+void scoutfs_xattr_init_key(struct scoutfs_key *key, u64 ino, u32 name_hash, u64 id);
+
 int scoutfs_xattr_get_locked(struct inode *inode, const char *name, void *buffer, size_t size,
 			     struct scoutfs_lock *lck);
 int scoutfs_xattr_set_locked(struct inode *inode, const char *name, size_t name_len,

tests/.gitignore

@@ -12,3 +12,4 @@ src/o_tmpfile_umask
 src/o_tmpfile_linkat
 src/mmap_stress
 src/mmap_validate
+src/watch_raw_inode_change

tests/Makefile

@@ -15,7 +15,8 @@ BIN := src/createmany			\
 	src/o_tmpfile_umask		\
 	src/o_tmpfile_linkat		\
 	src/mmap_stress			\
-	src/mmap_validate
+	src/mmap_validate		\
+	src/watch_raw_inode_change
 
 DEPS := $(wildcard src/*.d)
 

tests/funcs/filter.sh

@@ -20,9 +20,6 @@ t_filter_fs()
 # [ 2687.691366] BUG: KASAN: stack-out-of-bounds in get_reg+0x1bc/0x230
 # ...
 # [ 2687.706220] ==================================================================
-# [ 2687.707284] Disabling lock debugging due to kernel taint
-#
-# That final lock debugging message may not be included.
 #
 ignore_harmless_unwind_kasan_stack_oob()
 {
@@ -46,10 +43,6 @@ awk '
 		saved=""
         }
         ( in_soob == 2 && $0 ~ /==================================================================/ ) {
-                in_soob = 3
-                soob_nr = NR
-        }
-        ( in_soob == 3 && NR > soob_nr && $0 !~ /Disabling lock debugging/ ) {
                 in_soob = 0
         }
         ( !in_soob ) { print $0 }
@@ -61,6 +54,58 @@ awk '
 '
 }
 
+#
+# in el97+, XFS can generate a spurious lockdep circular dependency
+# warning about reclaim. Fixed upstream in e.g. v5.7-rc4-129-g6dcde60efd94
+#
+ignore_harmless_xfs_lockdep_warning()
+{
+awk '
+	BEGIN {
+		in_block = 0
+		block_nr = 0
+		buf = ""
+	}
+	( !in_block && $0 ~ /======================================================/ ) {
+		in_block = 1
+		block_nr = NR
+		buf = $0 "\n"
+		next
+	}
+	( in_block == 1 && NR == (block_nr + 1) ) {
+		if (match($0, /WARNING: possible circular locking dependency detected/) != 0) {
+			in_block = 2
+			buf = buf $0 "\n"
+		} else {
+			in_block = 0
+			printf "%s", buf
+			print $0
+			buf = ""
+		}
+		next
+	}
+	( in_block == 2 ) {
+		buf = buf $0 "\n"
+		if ($0 ~ /<\/TASK>/) {
+			if (buf ~ /xfs_(nondir_|dir_)?ilock_class/ && buf ~ /fs_reclaim/) {
+				# known xfs lockdep false positive, discard
+			} else {
+				printf "%s", buf
+			}
+			in_block = 0
+			buf = ""
+		}
+		next
+	}
+	{ print $0 }
+	END {
+		if (buf) {
+			printf "%s", buf
+		}
+	}
+'
+}
+
 #
 # Filter out expected messages.  Putting messages here implies that
 # tests aren't relying on messages to discover failures.. they're
@@ -176,6 +221,10 @@ t_filter_dmesg()
 	# creating block devices may trigger this
 	re="$re|block device autoloading is deprecated and will be removed."
 
+	# lockdep or kasan warnings can cause this
+	re="$re|Disabling lock debugging due to kernel taint"
+
 	egrep -v "($re)" | \
-		ignore_harmless_unwind_kasan_stack_oob
+		ignore_harmless_unwind_kasan_stack_oob | \
+		ignore_harmless_xfs_lockdep_warning
 }

tests/golden/basic-xattr-indx

@@ -0,0 +1,54 @@
+== testing invalid read-xattr-index arguments
+bad index position entry argument 'bad', it must be in the form "a.b.ino" where each value can be prefixed by '0' for octal or '0x' for hex
+scoutfs: read-xattr-index failed: Invalid argument (22)
+bad index position entry argument '1.2', it must be in the form "a.b.ino" where each value can be prefixed by '0' for octal or '0x' for hex
+scoutfs: read-xattr-index failed: Invalid argument (22)
+initial major index position '256' must be between 0 and 255, inclusive.
+scoutfs: read-xattr-index failed: Invalid argument (22)
+first index position 1.2.3 must be less than last index position 0.0.0
+scoutfs: read-xattr-index failed: Invalid argument (22)
+first index position 1.2.0 must be less than last index position 1.1.2
+scoutfs: read-xattr-index failed: Invalid argument (22)
+first index position 2.2.2 must be less than last index position 2.2.1
+scoutfs: read-xattr-index failed: Invalid argument (22)
+== testing invalid names
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/invalid: Numerical result out of range
+== testing boundary values
+0.0 found
+255.max found
+== indx xattr must have no value
+setfattr: /mnt/test/test/basic-xattr-indx/noval: Invalid argument
+setfattr: /mnt/test/test/basic-xattr-indx/noval: Invalid argument
+== set indx xattr and verify index entry
+found
+== setting same indx xattr again is a no-op
+found
+== removing non-existent indx xattr succeeds
+setfattr: /mnt/test/test/basic-xattr-indx/file: No such attribute
+still found
+== explicit xattr removal cleans up index entry
+== file deletion cleans up index entry
+found before delete
+== multiple indx xattrs on one file cleaned up by deletion
+entries before delete: 2
+entries after delete: 0
+== partial removal leaves other entries
+300 found
+== multiple files at same index position
+files at same position: 2
+surviving file found
+== cross-mount visibility
+found on mount 1
+== duplicate position deduplication
+entries for same position: 1

tests/golden/totl-merge-read

@@ -0,0 +1,3 @@
+== setup
+expected 4681
+== cleanup

tests/run-tests.sh

@@ -694,8 +694,8 @@ for t in $tests; do
 		if [ "$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" > \
-				"$T_TMPDIR/dmesg.new"
+				"$T_TMPDIR/dmesg.before" "$T_TMPDIR/dmesg.after" | \
+				grep -v '^$' > "$T_TMPDIR/dmesg.new"
 
 			if [ -s "$T_TMPDIR/dmesg.new" ]; then
 				message="unexpected messages in dmesg"

tests/sequence

@@ -26,7 +26,9 @@ srch-basic-functionality.sh
 simple-xattr-unit.sh
 retention-basic.sh
 totl-xattr-tag.sh
+basic-xattr-indx.sh
 quota.sh
+totl-merge-read.sh
 lock-refleak.sh
 lock-shrink-consistency.sh
 lock-shrink-read-race.sh

tests/src/watch_raw_inode_change.c

@@ -0,0 +1,664 @@
+/*
+ * Copyright (C) 2026 Versity Software, Inc.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <unistd.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <time.h>
+#include <linux/types.h>
+#include <assert.h>
+#include <stdbool.h>
+
+#include "../../utils/src/util.h"
+#include "ioctl.h"
+#include "format.h"
+
+/*
+ * This is a quick example of using the raw reading ioctls to get info
+ * on inodes as they change.  We maintain an array of meta_seq items for
+ * inodes that we've seen.  If we read the current meta_seq items and
+ * see differences then we get inode info and update our array with what
+ * we find.
+ *
+ * This only maintains one array and sorts it back and forth as we walk
+ * the meta_seq items and then search by inode number.  This will
+ * eventually use far too much cpu as the number of inodes increases.
+ */
+
+#define MSF		"%llu.%llu"
+#define MSA(ms)		(ms)->meta_seq, (ms)->ino
+#define NERRF		"nerr %d (\"%s\")"
+#define NERRA(nerr)	nerr, strerror(-nerr)
+
+#define prerror(fmt, args...) \
+	fprintf(stderr, "error: "fmt"\n", ##args)
+
+#define prdebug(fmt, args...) \
+do { \
+	if (opts.debug) \
+		printf(fmt"\n", ##args); \
+} while (0)
+
+static struct opts {
+	bool debug;
+	char *path;
+	char *names;
+	size_t names_size;
+	size_t names_count;
+} opts;
+
+struct stats {
+	__u64 start;
+	__u64 last;
+
+	struct per_call {
+		__u64 begin;
+		__u64 calls;
+		__u64 time;
+		__u64 inos;
+	} rms, rii;
+
+	__u64 inodes;
+	__u64 add;
+	__u64 remove;
+	__u64 update;
+
+	unsigned lines;
+} stats;
+
+struct meta_seq_array {
+	size_t nr;
+	size_t alloc;
+	struct scoutfs_ioctl_meta_seq *ms;
+};
+
+#define INO_BATCH	1000
+/* *2 for gratuitous allowance for struct expansion */
+#define RESULTS_SIZE	(INO_BATCH * 2 * (sizeof(struct scoutfs_ioctl_raw_read_result) + \
+		                          sizeof(__u64) + \
+		                          180 /* ~= sizeof(struct scoutfs_inode) */ + \
+		                          sizeof(struct scoutfs_ioctl_inode_attr_x)))
+
+#define NSEC_PER_SEC 1000000000
+
+static __u64 get_ns(void)
+{
+	struct timespec tp;
+	int ret;
+
+	ret = clock_gettime(CLOCK_MONOTONIC, &tp);
+	if (ret != 0) {
+		ret = -errno;
+		prerror("clock_gettime() error: "NERRF, NERRA(ret));
+		exit(2);
+	}
+
+	return ((__u64)tp.tv_sec * NSEC_PER_SEC) + (__u64)tp.tv_nsec;
+}
+static void begin_call(struct per_call *pc)
+{
+	pc->begin = get_ns();
+}
+
+static void end_call(struct per_call *pc)
+{
+	pc->calls++;
+	pc->time += get_ns() - pc->begin;
+}
+
+static int expand_array(struct meta_seq_array *arr, size_t additional)
+{
+#define ALLOC_BATCH	(1024 * 1024 / (sizeof(struct scoutfs_ioctl_meta_seq)))
+	struct scoutfs_ioctl_meta_seq *ms;
+	size_t expand;
+
+	if (arr->nr + additional <= arr->alloc)
+		return 0;
+
+	expand = arr->alloc + ALLOC_BATCH;
+	ms = reallocarray(arr->ms, expand, sizeof(arr->ms[0]));
+	if (!ms) {
+		prerror("allocating ms array with %zu elements failed", expand);
+		return -ENOMEM;
+	}
+
+	arr->alloc = expand;
+	arr->ms = ms;
+
+	return 0;
+}
+
+static void inc_ms(struct scoutfs_ioctl_meta_seq *ms)
+{
+	if (++ms->ino == 0)
+		ms->meta_seq++;
+}
+
+static void set_ms(struct scoutfs_ioctl_meta_seq *ms, __u64 meta_seq, __u64 ino)
+{
+	ms->meta_seq = meta_seq;
+	ms->ino = ino;
+}
+
+static int compar_ms_ino(const void *A, const void *B)
+{
+	const struct scoutfs_ioctl_meta_seq *a = A;
+	const struct scoutfs_ioctl_meta_seq *b = B;
+
+	return a->ino < b->ino ? -1 : a->ino > b->ino ? 1 : 0;
+}
+
+static int compar_ms_meta_seq(const void *A, const void *B)
+{
+	const struct scoutfs_ioctl_meta_seq *a = A;
+	const struct scoutfs_ioctl_meta_seq *b = B;
+
+	return a->meta_seq < b->meta_seq ? -1 : a->meta_seq > b->meta_seq ? 1 :
+	       compar_ms_ino(A, B);
+}
+
+static int compar_u64(const void *A, const void *B)
+{
+	const __u64 *a = A;
+	const __u64 *b = B;
+
+	return *a < *b ? -1 : *a > *b ? 1 : 0;
+}
+
+struct bsearch_ind_key {
+	int (*compar)(const void *a, const void *b);
+	void *key;
+	size_t size;
+	void **index;
+};
+
+static int bsearch_ind_compar(const void *a, const void *b)
+{
+	const struct bsearch_ind_key *bik = (const void *)((unsigned long)a ^ 1);
+	int cmp;
+
+	/* this key hack only works if compar is always called where a is key and b is &base[..] */
+	assert((unsigned long)a & 1);
+	assert(!((unsigned long)b & 1));
+
+	cmp = bik->compar(bik->key, b);
+	if (cmp > 0)
+		*(bik->index) = (void *)b + bik->size;
+	else
+		*(bik->index) = (void *)b;
+
+	return cmp;
+}
+
+static size_t bsearch_ind(const void *key, const void *base, size_t nmemb, size_t size,
+			  int (*compar)(const void *a, const void *b))
+{
+	void *index = (void *)base;
+	struct bsearch_ind_key bik = {
+		.compar = compar,
+		.key = (void *)key,
+		.size = size,
+		.index = &index,
+	};
+
+	bsearch((void *)(((unsigned long)&bik) | 1), base, nmemb, size, bsearch_ind_compar);
+
+	return (index - base) / size;
+}
+
+/*
+ * Generate a sorted list of inode numbers for the meta_seq items that
+ * differ between the results from raw_read_meta_seq and the items we
+ * have saved in our array. 
+ */
+static int differing_inos(__u64 *inos, struct meta_seq_array *arr,
+			  struct scoutfs_ioctl_meta_seq *start,
+			  struct scoutfs_ioctl_meta_seq *last,
+			  struct scoutfs_ioctl_meta_seq *ms, size_t nr)
+{
+	size_t arr_last;
+	size_t a;
+	size_t m;
+	int nr_inos;
+	int cmp;
+	int i;
+	int n;
+
+	/* find where we're going to stop in arr */
+	arr_last = bsearch_ind(last, arr->ms, arr->nr, sizeof(arr->ms[0]), compar_ms_meta_seq);
+	if (arr_last < arr->nr && compar_ms_meta_seq(&arr->ms[arr_last], last) == 0)
+		arr_last++;
+
+	a = bsearch_ind(start, arr->ms, arr->nr, sizeof(arr->ms[0]), compar_ms_meta_seq);
+
+	for (m = 0, nr_inos = 0; (a < arr_last || m < nr) && nr_inos < INO_BATCH; ) {
+
+		prdebug("diffing: m %zu nr %zu | a %zu arr_last %zu | nr_inos %d",
+			m, nr, a, arr_last, nr_inos);
+		if (a < arr_last)
+			prdebug("  arr->ms[%zu] = "MSF, a, MSA(&arr->ms[a]));
+		if (m < nr)
+			prdebug("  ms[%zu] = "MSF, m, MSA(&ms[m]));
+
+		/* setup comparison to copy lesser or only */
+		if (a < arr_last && m < nr)
+			cmp = compar_ms_meta_seq(&arr->ms[a], &ms[m]);
+		else if (a < arr_last)
+			cmp = -1;
+		else
+			cmp = 1;
+
+		prdebug("  cmp %d", cmp);
+
+		if (cmp == 0) {
+			/* ignore both when they match */
+			a++;
+			m++;
+		} else if (cmp < 0) {
+			inos[nr_inos++] = arr->ms[a++].ino;
+		} else { /* cmp > 0 */
+			inos[nr_inos++] = ms[m++].ino;
+		}
+	}
+
+	/* if we didn't consume all the read meta_seq then we might need to clamp last */
+	if (m < nr && compar_ms_meta_seq(&ms[m], last) <= 0) {
+		*last = ms[m];
+		last->ino--; /* must be non-zero, can't wrap */
+	}
+
+	/* sort and remove duplicate inode numbers */
+	if (nr_inos > 0) {
+		qsort(inos, nr_inos, sizeof(inos[0]), compar_u64);
+		for (i = 1, n = 1; i < nr_inos; i++) {
+			if (inos[i] != inos[n - 1])
+				inos[n++] = inos[i];
+		}
+		nr_inos = n;
+	}
+
+	return nr_inos;
+}
+
+/*
+ * We're not really validating the result stream.  We assume that the offset currently
+ * points at an inode.  We fill the caller's ms with its info then iterate through
+ * all its results until the next ino.
+ */
+static ssize_t read_inode_results(void *buf, size_t off, size_t size,
+				  struct scoutfs_ioctl_meta_seq *found)
+{
+	struct scoutfs_ioctl_raw_read_result res;
+	size_t len;
+	__le64 ms;
+
+	found->ino = 0;
+
+	while (off < size) {
+		memcpy(&res, buf + off, sizeof(res));
+		prdebug("res %u %u", res.type, res.size);
+
+		if (res.type == SCOUTFS_IOC_RAW_READ_RESULT_INODE && found->ino != 0)
+			break;
+
+		off += sizeof(res);
+
+		switch(res.type) {
+			case SCOUTFS_IOC_RAW_READ_RESULT_INODE:
+				memcpy(&found->ino, buf + off, sizeof(__u64));
+				memcpy(&ms, buf + off + sizeof(__u64) +
+				       offsetof(struct scoutfs_inode, meta_seq), sizeof(__le64));
+				found->meta_seq = le64_to_cpu(ms);
+				prdebug("res ino %llu ms %llu", found->ino, found->meta_seq);
+				break;
+
+			case SCOUTFS_IOC_RAW_READ_RESULT_XATTR:
+				len = strlen((char *)buf + off) + 1;
+				prdebug("res xattr '%s' len %d: '%.*s'",
+					(char *)buf + off, 
+					(int)(res.size - len),
+					(int)(res.size - len),
+					(char *)buf + off + len);
+				break;
+		};
+		off += res.size;
+	}
+
+	return off;
+}
+
+/*
+ * inos[] contains the inode numbers that we're interested in.  Get
+ * their info and update our array with what we find.
+ */
+static int read_inode_info(int fd, void *buf, struct meta_seq_array *arr, __u64 *inos, int nr_inos)
+{
+	struct scoutfs_ioctl_raw_read_inode_info rii;
+	struct scoutfs_ioctl_meta_seq found;
+	struct scoutfs_ioctl_meta_seq ms;
+	ssize_t off;
+	size_t size;
+	size_t ind;
+	size_t added;
+	int i;
+	int ret;
+
+	rii = (struct scoutfs_ioctl_raw_read_inode_info) {
+		.inos_ptr = (unsigned long)inos,
+		.inos_count = nr_inos,
+		.names_ptr = (unsigned long)opts.names,
+		.names_count = opts.names_count,
+		.results_ptr = (unsigned long)buf,
+		.results_size = RESULTS_SIZE,
+	};
+
+	begin_call(&stats.rii);
+	ret = ioctl(fd, SCOUTFS_IOC_RAW_READ_INODE_INFO, &rii);
+	if (ret < 0) {
+		ret = -errno;
+		prerror("READ_INODE_INFO ioctl failed: "NERRF, NERRA(ret));
+		goto out;
+	}
+	end_call(&stats.rii);
+
+	prdebug("gii ret %d", ret);
+
+	off = 0;
+	size = ret;
+	set_ms(&found, 0, 0);
+	added = 0;
+	i = 0;
+
+	/* sort by ino so we can search by ino for updates */
+	qsort(arr->ms, arr->nr, sizeof(arr->ms[0]), compar_ms_ino);
+
+	while (i < nr_inos) {
+		/* find next ino */
+		if (!found.ino && off < size) {
+			off = read_inode_results(buf, off, size, &found);
+			if (off < 0) {
+				ret = off;
+				goto out;
+			}
+			stats.rii.inos++;
+		}
+
+		if (i < nr_inos && (!found.ino || inos[i] < found.ino)) {
+			/* delete any record of inodes we didn't find */
+			set_ms(&ms, UINT64_MAX, inos[i]);
+			i++;
+
+		} else if (found.ino) {
+			/* update/add arr to match the found ino */
+			ms = found;
+			if (i < nr_inos && inos[i] == found.ino)
+				i++;
+			set_ms(&found, 0, 0);
+		}
+
+		/* find existing record */
+		ind = bsearch_ind(&ms, arr->ms, arr->nr, sizeof(arr->ms[0]), compar_ms_ino);
+		if (ind < arr->nr && arr->ms[ind].ino == ms.ino) {
+			/* update existing ino, can be marking for deletion */
+			prdebug("updating arr [%zu] ino %llu ms %llu -> %llu",
+					ind, ms.ino, arr->ms[ind].meta_seq, ms.meta_seq);
+			arr->ms[ind].meta_seq = ms.meta_seq;
+			if (ms.meta_seq == UINT64_MAX)
+				stats.remove++;
+			else
+				stats.update++;
+
+		} else if (ms.meta_seq != UINT64_MAX) {
+			/* append new found, maintaining existing sorting */
+			arr->ms[arr->nr + added] = ms;
+			prdebug("adding arr [%zu] ino %llu ms %llu",
+					arr->nr + added, ms.ino, ms.meta_seq);
+			added++;
+			stats.add++;
+		}
+	}
+
+	/* sort by seq again for next meta seq read */
+	arr->nr += added;
+	qsort(arr->ms, arr->nr, sizeof(arr->ms[0]), compar_ms_meta_seq);
+
+	/* and trim off any deletions */
+	while (arr->nr > 0 && arr->ms[arr->nr - 1].meta_seq == UINT64_MAX)
+		arr->nr--;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static double secs(u64 a_ns, u64 b_ns)
+{
+	return (double)(a_ns - b_ns) / NSEC_PER_SEC;
+}
+
+static double nr_per_sec(u64 nr, __u64 nsec)
+{
+	if (nsec == 0)
+		return 0;
+
+	return (double)nr / secs(nsec, 0);
+}
+
+static void print_stats(void)
+{
+	u64 now = get_ns();
+
+	if (secs(now, stats.last) < 1.0)
+		return;
+
+	if ((stats.lines++ % 16) == 0) {
+		printf("%6s | %-29s | %-23s | %-23s\n",
+			"", "inodes", "meta_seq", "inode_info");
+		printf("%6s | %8s %6s %6s %6s | %7s %7s %7s | %7s %7s %7s\n",
+			"now",
+			"total", "add", "remove", "update",
+			"calls", "inos", "inos/s",
+			"calls", "inos", "inos/s");
+	}
+
+	printf("%6.3lf | %8llu %6llu %6llu %6llu | %7llu %7llu %7.0lf | %7llu %7llu %7.0lf\n",
+		secs(now, stats.start),
+		stats.inodes, stats.add, stats.remove, stats.update,
+		stats.rms.calls, stats.rms.inos, nr_per_sec(stats.rms.inos, stats.rms.time),
+		stats.rii.calls, stats.rii.inos, nr_per_sec(stats.rms.inos, stats.rii.time));
+
+	stats.last = now;
+
+	{
+		struct stats save = stats;
+		stats = (struct stats) {
+			.start = save.start,
+			.last = save.last,
+			.lines = save.lines,
+		};
+	}
+}
+
+static void add_xattr(char *name)
+{
+	size_t len_null;
+	char *names;
+	int ret;
+
+	len_null = strlen(name) + 1;
+	names = realloc(opts.names, opts.names_size + len_null);
+	if (!names) {
+		ret = -errno;
+		prerror("allocation of xattr names buffer failed: "NERRF, NERRA(ret));
+		exit(3);
+	}
+
+	memcpy(names + opts.names_size, name, len_null);
+
+	opts.names = names;
+	opts.names_size += len_null;
+	opts.names_count++;
+}
+
+static bool parse_opts(int argc, char **argv)
+{
+	bool usage = false;
+	int c;
+
+	opts = (struct opts) {
+		.debug = false,
+	};
+
+        while ((c = getopt(argc, argv, "dp:x:")) != -1) {
+                switch(c) {
+                case 'd':
+                        opts.debug = true;
+                        break;
+                case 'p':
+                        opts.path = strdup(optarg);
+                        break;
+                case 'x':
+			add_xattr(optarg);
+                        break;
+                case '?':
+                        printf("Unknown option '%c'\n", optopt);
+			usage = true;
+                }
+	}
+
+	if (!usage) {
+		usage = true;
+		if (!opts.path)
+			printf("need -p path option\n");
+		else
+			usage = false;
+	}
+
+	if (usage) {
+		printf("\nusage:\n"
+		       " -d      | enable verbose debugging output\n"
+		       " -p PATH | path to file system to watch\n"
+		       " -x NAME | try to read named xattr with inodes, can be many\n"
+		      );
+		return false;
+	}
+
+	return true;
+}
+
+int main(int argc, char **argv)
+{
+	struct scoutfs_ioctl_raw_read_meta_seq rms = {0,};
+	struct scoutfs_ioctl_meta_seq *ms;
+	struct meta_seq_array arr = {0,};
+	__u64 *inos = NULL;
+	void *buf = NULL;
+	int fd = -1;
+	int nr_inos;
+	int nr;
+	int i;
+	int ret;
+
+	if (!parse_opts(argc, argv))
+		exit(1);
+
+	inos = calloc(INO_BATCH, sizeof(inos[0]));
+	buf = malloc(RESULTS_SIZE);
+	if (!inos || !buf) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	rms.results_ptr = (unsigned long)buf;
+	rms.results_size = min(RESULTS_SIZE, INO_BATCH * sizeof(struct scoutfs_ioctl_meta_seq));
+
+	fd = open(opts.path, O_RDONLY);
+	if (fd == -1) {
+		perror("error");
+		exit(1);
+	}
+
+	stats.start = get_ns();
+
+	for (;;) {
+		set_ms(&rms.start, 0, 0);
+		set_ms(&rms.end, UINT64_MAX, UINT64_MAX);
+
+		do {
+			begin_call(&stats.rms);
+			ret = ioctl(fd, SCOUTFS_IOC_RAW_READ_META_SEQ, &rms);
+			if (ret < 0) {
+				ret = -errno;
+				prerror("READ_META_SEQ ioctl failed, "
+					"start "MSF" end "MSF", "NERRF,
+					MSA(&rms.start), MSA(&rms.end), NERRA(ret));
+				goto out;
+			}
+			end_call(&stats.rms);
+			stats.rms.inos += ret;
+
+			prdebug("RMS last "MSF" ret %d:", MSA(&rms.last), ret);
+
+			nr = ret;
+			ms = buf;
+
+			if (opts.debug && nr > 0) {
+				for (i = 0; i < nr; i++)
+					prdebug(" [%u] "MSF"", i, MSA(&ms[i]));
+			}
+
+			nr_inos = differing_inos(inos, &arr, &rms.start, &rms.last, ms, nr);
+
+			if (nr_inos > 0) {
+				prdebug("diff inos %d:", nr_inos);
+				for (i = 0; i < nr_inos; i++)
+					prdebug(" [%u] %llu", i, inos[i]);
+
+				ret = expand_array(&arr, nr_inos) ?:
+				      read_inode_info(fd, buf, &arr, inos, nr_inos);
+				if (ret < 0)
+					goto out;
+			}
+
+			stats.inodes = arr.nr;
+			print_stats();
+
+			rms.start = rms.last;
+			inc_ms(&rms.start);
+
+		} while (rms.last.meta_seq != UINT64_MAX || rms.last.ino != UINT64_MAX);
+
+
+		sleep(1);
+	}
+
+	ret = 0;
+out:
+	if (fd >= 0)
+		close(fd);
+
+	free(inos);
+	free(buf);
+	free(arr.ms);
+	free(opts.names);
+
+	return ret;
+}

tests/tests/basic-xattr-indx.sh

@@ -0,0 +1,143 @@
+#
+# Test basic .indx. xattr tag functionality and index entry lifecycle
+#
+
+t_require_commands touch rm setfattr scoutfs stat
+t_require_mounts 2
+
+# query index from a specific mount, default mount 0
+read_xattr_index()
+{
+	local nr="${1:-0}"
+	local mnt="$(eval echo \$T_M$nr)"
+	shift
+
+	sync
+	echo 1 > $(t_debugfs_path $nr)/drop_weak_item_cache
+	scoutfs read-xattr-index -p "$mnt" "$@"
+}
+
+MAJOR=5
+MINOR=100
+
+echo "== testing invalid read-xattr-index arguments"
+scoutfs read-xattr-index -p "$T_M0" bad 2>&1
+scoutfs read-xattr-index -p "$T_M0" 1.2 2>&1
+scoutfs read-xattr-index -p "$T_M0" 1.2.3 256.0.0 2>&1
+scoutfs read-xattr-index -p "$T_M0" 1.2.3 0.0.0 2>&1
+scoutfs read-xattr-index -p "$T_M0" 1.2.0 1.1.2 2>&1
+scoutfs read-xattr-index -p "$T_M0" 2.2.2 2.2.1 2>&1
+
+echo "== testing invalid names"
+touch "$T_D0/invalid"
+setfattr -n scoutfs.hide.indx.test.$MAJOR "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.. "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test..$MINOR "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.$MAJOR. "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.256.$MINOR "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.abc.$MINOR "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.$MAJOR.abc "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.-1.$MINOR "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.$MAJOR.-1 "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.18446744073709551616.$MINOR "$T_D0/invalid" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.$(printf 'x%.0s' $(seq 1 240)).$MAJOR.$MINOR "$T_D0/invalid" 2>&1 | t_filter_fs
+rm -f "$T_D0/invalid"
+
+echo "== testing boundary values"
+touch "$T_D0/boundary"
+INO=$(stat -c "%i" "$T_D0/boundary")
+setfattr -n scoutfs.hide.indx.test.0.0 "$T_D0/boundary"
+read_xattr_index 0 0.0.0 0.0.-1 | awk '($3 == "'$INO'") {print "0.0 found"}'
+setfattr -x scoutfs.hide.indx.test.0.0 "$T_D0/boundary"
+setfattr -n scoutfs.hide.indx.test.255.18446744073709551615 "$T_D0/boundary"
+read_xattr_index 0 255.0.0 255.-1.-1 | awk '($3 == "'$INO'") {print "255.max found"}'
+setfattr -x scoutfs.hide.indx.test.255.18446744073709551615 "$T_D0/boundary"
+rm -f "$T_D0/boundary"
+
+echo "== indx xattr must have no value"
+touch "$T_D0/noval"
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR -v "" "$T_D0/noval" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR -v 0 "$T_D0/noval" 2>&1 | t_filter_fs
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR -v 1 "$T_D0/noval" 2>&1 | t_filter_fs
+rm -f "$T_D0/noval"
+
+echo "== set indx xattr and verify index entry"
+touch "$T_D0/file"
+INO=$(stat -c "%i" "$T_D0/file")
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/file"
+read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found"}'
+
+echo "== setting same indx xattr again is a no-op"
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/file"
+read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found"}'
+
+echo "== removing non-existent indx xattr succeeds"
+setfattr -x scoutfs.hide.indx.nonexistent.$MAJOR.999 "$T_D0/file" 2>&1 | t_filter_fs
+read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "still found"}'
+
+echo "== explicit xattr removal cleans up index entry"
+setfattr -x scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/file"
+read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found orphan"}'
+rm -f "$T_D0/file"
+
+echo "== file deletion cleans up index entry"
+touch "$T_D0/file2"
+INO=$(stat -c "%i" "$T_D0/file2")
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/file2"
+read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found before delete"}'
+rm -f "$T_D0/file2"
+read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found orphan after delete"}'
+
+echo "== multiple indx xattrs on one file cleaned up by deletion"
+touch "$T_D0/file3"
+INO=$(stat -c "%i" "$T_D0/file3")
+setfattr -n scoutfs.hide.indx.a.$MAJOR.200 "$T_D0/file3"
+setfattr -n scoutfs.hide.indx.b.$MAJOR.300 "$T_D0/file3"
+BEFORE=$(read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'")' | wc -l)
+echo "entries before delete: $BEFORE"
+rm -f "$T_D0/file3"
+AFTER=$(read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'")' | wc -l)
+echo "entries after delete: $AFTER"
+
+echo "== partial removal leaves other entries"
+touch "$T_D0/partial"
+INO=$(stat -c "%i" "$T_D0/partial")
+setfattr -n scoutfs.hide.indx.a.$MAJOR.200 "$T_D0/partial"
+setfattr -n scoutfs.hide.indx.b.$MAJOR.300 "$T_D0/partial"
+setfattr -x scoutfs.hide.indx.a.$MAJOR.200 "$T_D0/partial"
+read_xattr_index 0 $MAJOR.200.0 $MAJOR.200.-1 | awk '($3 == "'$INO'") {print "200 found"}'
+read_xattr_index 0 $MAJOR.300.0 $MAJOR.300.-1 | awk '($3 == "'$INO'") {print "300 found"}'
+rm -f "$T_D0/partial"
+
+echo "== multiple files at same index position"
+touch "$T_D0/multi_a" "$T_D0/multi_b"
+INO_A=$(stat -c "%i" "$T_D0/multi_a")
+INO_B=$(stat -c "%i" "$T_D0/multi_b")
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/multi_a"
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/multi_b"
+COUNT=$(read_xattr_index 0 $MAJOR.$MINOR.0 $MAJOR.$MINOR.-1 | wc -l)
+echo "files at same position: $COUNT"
+rm -f "$T_D0/multi_a"
+read_xattr_index 0 $MAJOR.$MINOR.0 $MAJOR.$MINOR.-1 | awk '($3 == "'$INO_A'") {print "deleted file still found"}'
+read_xattr_index 0 $MAJOR.$MINOR.0 $MAJOR.$MINOR.-1 | awk '($3 == "'$INO_B'") {print "surviving file found"}'
+rm -f "$T_D0/multi_b"
+
+echo "== cross-mount visibility"
+touch "$T_D0/file4"
+INO=$(stat -c "%i" "$T_D0/file4")
+setfattr -n scoutfs.hide.indx.test.$MAJOR.$MINOR "$T_D0/file4"
+read_xattr_index 1 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found on mount 1"}'
+rm -f "$T_D0/file4"
+read_xattr_index 1 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'") {print "found orphan on mount 1"}'
+
+echo "== duplicate position deduplication"
+touch "$T_D0/file5"
+INO=$(stat -c "%i" "$T_D0/file5")
+setfattr -n scoutfs.hide.indx.aa.$MAJOR.$MINOR "$T_D0/file5"
+setfattr -n scoutfs.hide.indx.bb.$MAJOR.$MINOR "$T_D0/file5"
+COUNT=$(read_xattr_index 0 $MAJOR.0.0 $MAJOR.-1.-1 | awk '($3 == "'$INO'")' | wc -l)
+echo "entries for same position: $COUNT"
+rm -f "$T_D0/file5"
+
+t_pass

tests/tests/totl-merge-read.sh

@@ -0,0 +1,50 @@
+#
+# Test that merge_read_item() correctly updates the sequence number when
+# combining delta items from multiple finalized log trees.  Each mount
+# sets a totl value in its own 3-bit lane (powers of 8) so that any
+# double-counting overflows the lane and is caught by: or(v, exp) != exp.
+#
+
+t_require_commands setfattr scoutfs
+t_require_mounts 5
+
+echo "== setup"
+for nr in $(t_fs_nrs); do
+	d=$(eval echo \$T_D$nr)
+	for i in $(seq 1 2500); do : > "$d/f$nr$i"; done
+done
+sync
+t_force_log_merge
+
+vals=(1 8 64 512 4096)
+expected=4681
+n=0
+for nr in $(t_fs_nrs); do
+	d=$(eval echo \$T_D$nr)
+	v=${vals[$((n++))]}
+	for i in $(seq 1 2500); do
+		setfattr -n "scoutfs.totl.t.$i.0.0" -v $v "$d/f$nr$i"
+	done
+done
+
+t_trigger_arm_silent log_merge_force_partial $(t_server_nr)
+
+bad="$T_TMPDIR/bad"
+for nr in $(t_fs_nrs); do
+	( while true; do
+		echo 1 > "$(t_debugfs_path $nr)/drop_weak_item_cache"
+		scoutfs read-xattr-totals -p "$(eval echo \$T_M$nr)" | \
+			awk -F'[ =,]+' -v e=$expected 'or($2+0,e) != e'
+	done ) >> "$bad" &
+done
+
+echo "expected $expected"
+t_force_log_merge
+t_silent_kill $(jobs -p)
+test -s "$bad" && echo "double-counted:" && cat "$bad"
+
+echo "== cleanup"
+for nr in $(t_fs_nrs); do
+	find "$(eval echo \$T_D$nr)" -name "f$nr*" -delete
+done
+t_pass