[PATCH v2] xfs: stop exchanging reflink flags when exchanging mappings

[PATCH v2] xfs: stop exchanging reflink flags when exchanging mappings

[Norbert Szetei]

When an exchange covers both files from offset 0 to i_disk_size,
xmi_can_exchange_reflink_flags() assumes every mapping moved and hands
the reflink flag over to the other file.

That assumption is wrong.  A file can still own shared blocks past
i_disk_size, e.g. after EXCHANGE_RANGE_TO_EOF sets a smaller size
without unmapping anything.  The flag is cleared anyway, and the next
write to such a block happens in place instead of CoW, corrupting the
other file that shares it.

Commit a23eca88448e ("xfs: fix exchange-range reflink flag clearing
issue with INO1_WRITTEN") fixed one instance of this.  Rather than add
another special case, stop moving the flags.
xfs_exchmaps_ensure_reflink() already sets the flag on both inodes
before the exchange, so both just keep it.

Nothing sets CLEAR_INO{1,2}_REFLINK anymore.  Intents logged by older
kernels may still carry them; recovery already masks them off, so they
stay in the log format mask and are ignored.

XFS_IOC_SWAPEXT has the same problem: it moves the reflink flag from
one inode to the other along with the extents.  Just removing that is
not enough here: the exchange path flags both inodes up front in
xfs_exchmaps_ensure_reflink(), but swapext has no such step, so the
inode receiving the shared extents would be left without the flag.
Set it on both inodes instead, and make sure both have a CoW fork,
since an inode with the reflink flag is expected to have one.

The flag now stays until FALLOC_FL_UNSHARE_RANGE, xfs_scrub, or a
truncate to zero removes it.

Fixes: 966ceafc7a43 ("xfs: create deferred log items for file mapping exchanges")
Fixes: f0bc4d134b46 ("xfs: swap inode reflink flags when swapping inode extents")
Cc: stable@vger.kernel.org # v6.10
Assisted-by: LLM
Signed-off-by: Norbert Szetei <norbert@doyensec.com>
---
v2: also handle XFS_IOC_SWAPEXT
v1: https://lore.kernel.org/linux-xfs/1E196589-DEBE-40AC-AFEA-D420DAAB067F@doyensec.com/
Discussion: https://lore.kernel.org/linux-xfs/20261006050914.GU2705364@frogsfrogsfrogs/

 fs/xfs/libxfs/xfs_exchmaps.c   | 80 ++--------------------------------
 fs/xfs/libxfs/xfs_log_format.h |  4 +-
 fs/xfs/xfs_bmap_util.c         | 15 +++----
 3 files changed, 10 insertions(+), 89 deletions(-)

diff --git a/fs/xfs/libxfs/xfs_exchmaps.c b/fs/xfs/libxfs/xfs_exchmaps.c
index 6a66b6075e0a..7af8ee35c89d 100644
--- a/fs/xfs/libxfs/xfs_exchmaps.c
+++ b/fs/xfs/libxfs/xfs_exchmaps.c
@@ -49,7 +49,7 @@ struct xfs_exchmaps_adjacent {
 	.right2 = { .br_startblock = HOLESTARTBLOCK }, \
 }
 
-/* Information to reset reflink flag / CoW fork state after an exchange. */
+/* Information to reset CoW fork state after an exchange. */
 
 /*
  * If the reflink flag is set on either inode, make sure it has an incore CoW
@@ -127,9 +127,7 @@ xmi_has_more_exchange_work(const struct xfs_exchmaps_intent *xmi)
 static inline bool
 xmi_has_postop_work(const struct xfs_exchmaps_intent *xmi)
 {
-	return xmi->xmi_flags & (XFS_EXCHMAPS_CLEAR_INO1_REFLINK |
-				 XFS_EXCHMAPS_CLEAR_INO2_REFLINK |
-				 __XFS_EXCHMAPS_INO2_SHORTFORM);
+	return xmi->xmi_flags & __XFS_EXCHMAPS_INO2_SHORTFORM;
 }
 
 /* Check all mappings to make sure we can actually exchange them. */
@@ -525,18 +523,6 @@ xfs_exchmaps_link_to_sf(
 	return error;
 }
 
-/* Clear the reflink flag after an exchange. */
-static inline void
-xfs_exchmaps_clear_reflink(
-	struct xfs_trans	*tp,
-	struct xfs_inode	*ip)
-{
-	trace_xfs_reflink_unset_inode_flag(ip);
-
-	ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
-	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
-}
-
 /* Finish whatever work might come after an exchange operation. */
 static int
 xfs_exchmaps_do_postop_work(
@@ -557,16 +543,6 @@ xfs_exchmaps_do_postop_work(
 			return error;
 	}
 
-	if (xmi->xmi_flags & XFS_EXCHMAPS_CLEAR_INO1_REFLINK) {
-		xfs_exchmaps_clear_reflink(tp, xmi->xmi_ip1);
-		xmi->xmi_flags &= ~XFS_EXCHMAPS_CLEAR_INO1_REFLINK;
-	}
-
-	if (xmi->xmi_flags & XFS_EXCHMAPS_CLEAR_INO2_REFLINK) {
-		xfs_exchmaps_clear_reflink(tp, xmi->xmi_ip2);
-		xmi->xmi_flags &= ~XFS_EXCHMAPS_CLEAR_INO2_REFLINK;
-	}
-
 	return 0;
 }
 
@@ -948,46 +924,13 @@ xfs_exchmaps_intent_destroy_cache(void)
 }
 
 /*
- * Decide if we will exchange the reflink flags between the two files after the
- * exchange.  The only time we want to do this is if we're exchanging all
- * mappings under EOF and the inode reflink flags have different states.
+ * Allocate and initialize a new incore intent item from a request.
  */
-static inline bool
-xmi_can_exchange_reflink_flags(
-	const struct xfs_exchmaps_req	*req,
-	unsigned int			reflink_state)
-{
-	struct xfs_mount		*mp = req->ip1->i_mount;
-
-	/*
-	 * The INO1_WRITTEN optimization can skip exchanging hole and
-	 * unwritten mappings, which means we cannot guarantee that all
-	 * shared extents actually moved to the other file.  Clearing the
-	 * reflink flag of an inode that still holds shared extents breaks
-	 * the CoW write path, so refuse to exchange the flags in that case.
-	 */
-	if (req->flags & XFS_EXCHMAPS_INO1_WRITTEN)
-		return false;
-
-	if (hweight32(reflink_state) != 1)
-		return false;
-	if (req->startoff1 != 0 || req->startoff2 != 0)
-		return false;
-	if (req->blockcount != XFS_B_TO_FSB(mp, req->ip1->i_disk_size))
-		return false;
-	if (req->blockcount != XFS_B_TO_FSB(mp, req->ip2->i_disk_size))
-		return false;
-	return true;
-}
-
-
-/* Allocate and initialize a new incore intent item from a request. */
 struct xfs_exchmaps_intent *
 xfs_exchmaps_init_intent(
 	const struct xfs_exchmaps_req	*req)
 {
 	struct xfs_exchmaps_intent	*xmi;
-	unsigned int			rs = 0;
 
 	xmi = kmem_cache_zalloc(xfs_exchmaps_intent_cache,
 			GFP_NOFS | __GFP_NOFAIL);
@@ -1011,23 +954,6 @@ xfs_exchmaps_init_intent(
 		xmi->xmi_isize2 = req->ip1->i_disk_size;
 	}
 
-	/* Record the state of each inode's reflink flag before the op. */
-	if (xfs_is_reflink_inode(req->ip1))
-		rs |= 1;
-	if (xfs_is_reflink_inode(req->ip2))
-		rs |= 2;
-
-	/*
-	 * Figure out if we're clearing the reflink flags (which effectively
-	 * exchanges them) after the operation.
-	 */
-	if (xmi_can_exchange_reflink_flags(req, rs)) {
-		if (rs & 1)
-			xmi->xmi_flags |= XFS_EXCHMAPS_CLEAR_INO1_REFLINK;
-		if (rs & 2)
-			xmi->xmi_flags |= XFS_EXCHMAPS_CLEAR_INO2_REFLINK;
-	}
-
 	if (S_ISDIR(VFS_I(xmi->xmi_ip2)->i_mode) ||
 	    S_ISLNK(VFS_I(xmi->xmi_ip2)->i_mode))
 		xmi->xmi_flags |= __XFS_EXCHMAPS_INO2_SHORTFORM;
diff --git a/fs/xfs/libxfs/xfs_log_format.h b/fs/xfs/libxfs/xfs_log_format.h
index a4e1b3eb425c..6e620380d5dc 100644
--- a/fs/xfs/libxfs/xfs_log_format.h
+++ b/fs/xfs/libxfs/xfs_log_format.h
@@ -927,10 +927,8 @@ struct xfs_xmi_log_format {
  */
 #define XFS_EXCHMAPS_INO1_WRITTEN	(1ULL << 2)
 
-/* Clear the reflink flag from inode1 after the operation. */
+/* No longer set, but older intents may carry them; keep them valid. */
 #define XFS_EXCHMAPS_CLEAR_INO1_REFLINK	(1ULL << 3)
-
-/* Clear the reflink flag from inode2 after the operation. */
 #define XFS_EXCHMAPS_CLEAR_INO2_REFLINK	(1ULL << 4)
 
 #define XFS_EXCHMAPS_LOGGED_FLAGS	(XFS_EXCHMAPS_ATTR_FORK | \
diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
index 268d159339d0..865aec3ae5ab 100644
--- a/fs/xfs/xfs_bmap_util.c
+++ b/fs/xfs/xfs_bmap_util.c
@@ -1544,7 +1544,6 @@ xfs_swap_extents(
 	struct xfs_bstat	*sbp = &sxp->sx_stat;
 	int			src_log_flags, target_log_flags;
 	int			error = 0;
-	uint64_t		f;
 	int			resblks = 0;
 	unsigned int		flags = 0;
 	struct timespec64	ctime, mtime;
@@ -1700,14 +1699,12 @@ xfs_swap_extents(
 	if (error)
 		goto out_trans_cancel;
 
-	/* Do we have to swap reflink flags? */
-	if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
-	    (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
-		f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
-		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
-		ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
-		tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
-		tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
+	/* Either inode may now own shared blocks; flag both, with CoW forks. */
+	if ((ip->i_diflags2 | tip->i_diflags2) & XFS_DIFLAG2_REFLINK) {
+		ip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
+		tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
+		xfs_ifork_init_cow(ip);
+		xfs_ifork_init_cow(tip);
 	}
 
 	/* Swap the cow forks. */
-- 
2.55.0

Re: [PATCH v2] xfs: stop exchanging reflink flags when exchanging mappings

[Norbert Szetei]

Here is a reproducer for the XFS_IOC_SWAPEXT case, since there wasn't
one on the list yet, in case it's useful for fstests.

	$ truncate -s 512M img && mkfs.xfs -q -m rmapbt=1,reflink=1 img
	$ mkdir -p /tmp/m && sudo mount -o loop img /tmp/m && sudo chmod 1777 /tmp/m
	$ sudo xfs_io -f -c "pwrite -q -S 0x50 0 4096" /tmp/m/peer
	$ sudo chmod 644 /tmp/m/peer
	$ cd /tmp/m && ./xfs_swapext_poc

On an unpatched kernel it reports peer as rewritten. With v2 applied,
peer is left intact. It needs rmapbt=1.

// SPDX-License-Identifier: GPL-2.0
#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/vfs.h>
#include <linux/fs.h>
#include <linux/fiemap.h>

#ifndef FICLONERANGE
#define FICLONERANGE _IOW(0x94, 13, struct file_clone_range)
#endif

/* fs/xfs/libxfs/xfs_fs.h */
struct xfs_exchange_range {
	__s32		file1_fd;
	__u32		pad;
	__u64		file1_offset;
	__u64		file2_offset;
	__u64		length;
	__u64		flags;
};
#define XFS_IOC_EXCHANGE_RANGE		_IOW('X', 129, struct xfs_exchange_range)
#define XFS_EXCHANGE_RANGE_TO_EOF	(1ULL << 0)

typedef struct xfs_bstime {
	long		tv_sec;
	__s32		tv_nsec;
} xfs_bstime_t;

struct xfs_bstat {
	__u64		bs_ino;
	__u16		bs_mode;
	__u16		bs_nlink;
	__u32		bs_uid;
	__u32		bs_gid;
	__u32		bs_rdev;
	__s32		bs_blksize;
	__s64		bs_size;
	xfs_bstime_t	bs_atime;
	xfs_bstime_t	bs_mtime;
	xfs_bstime_t	bs_ctime;
	__s64		bs_blocks;
	__u32		bs_xflags;
	__s32		bs_extsize;
	__s32		bs_extents;
	__u32		bs_gen;
	__u16		bs_projid_lo;
	__u16		bs_forkoff;
	__u16		bs_projid_hi;
	__u16		bs_sick;
	__u16		bs_checked;
	unsigned char	bs_pad[2];
	__u32		bs_cowextsize;
	__u32		bs_dmevmask;
	__u16		bs_dmstate;
	__u16		bs_aextents;
};

struct xfs_swapext {
	__s64			sx_version;
#define XFS_SX_VERSION		0
	__s64			sx_fdtarget;
	__s64			sx_fdtmp;
	__s64			sx_offset;
	__s64			sx_length;
	char			sx_pad[16];
	struct xfs_bstat	sx_stat;
};
#define XFS_IOC_SWAPEXT		_IOWR('X', 109, struct xfs_swapext)

static unsigned long blksz;
static int peer, f1;

static unsigned long long phys(int fd, unsigned long long off)
{
	struct { struct fiemap f; struct fiemap_extent e[1]; } q;

	memset(&q, 0, sizeof q);
	q.f.fm_start = off;
	q.f.fm_length = blksz;
	q.f.fm_extent_count = 1;
	if (ioctl(fd, FS_IOC_FIEMAP, &q.f) || q.f.fm_mapped_extents < 1)
		return 0;
	return (q.e[0].fe_physical + (off - q.e[0].fe_logical)) / blksz;
}

static void show(const char *tag)
{
	struct stat st;
	int i;

	fstat(f1, &st);
	printf("%-26s peer [%llu]   file1 size %5lld  [", tag, phys(peer, 0),
	       (long long)st.st_size);
	for (i = 0; i < 3; i++)
		printf("%llu%s", phys(f1, (unsigned long long)i * blksz),
		       i == 2 ? "]\n" : "] [");
}

static int mkfile(const char *name, char fill, int blocks)
{
	char *buf;
	int fd;

	unlink(name);
	fd = open(name, O_RDWR | O_CREAT | O_EXCL, 0600);
	if (fd < 0)
		return fprintf(stderr, "open %s: %m\n", name), -1;
	buf = malloc(blocks * blksz);
	memset(buf, fill, blocks * blksz);
	if (pwrite(fd, buf, blocks * blksz, 0) != (ssize_t)(blocks * blksz))
		return fprintf(stderr, "pwrite %s: %m\n", name), -1;
	free(buf);
	fsync(fd);
	return fd;
}

int main(void)
{
	struct xfs_exchange_range xr;
	struct file_clone_range cr;
	struct xfs_swapext sx;
	char before[65536], after[65536], *buf;
	struct statfs sfs;
	struct stat st;
	int d1, d2;

	if (statfs(".", &sfs))
		return fprintf(stderr, "statfs: %m\n"), 2;
	blksz = sfs.f_bsize;

	peer = open("peer", O_RDONLY);
	if (peer < 0)
		return fprintf(stderr, "open peer: %m\n"), 2;
	fstat(peer, &st);
	printf("peer uid %u mode %04o size %lld, block size %lu\n\n",
	       st.st_uid, st.st_mode & 07777, (long long)st.st_size, blksz);
	if (pread(peer, before, blksz, 0) != (ssize_t)blksz)
		return fprintf(stderr, "read peer: %m\n"), 2;
	syncfs(peer);	/* so FIEMAP reports peer's real block */

	f1 = mkfile("file1", 'A', 3);
	d1 = mkfile("donor1", 'B', 1);
	d2 = mkfile("donor2", 'C', 1);
	if (f1 < 0 || d1 < 0 || d2 < 0)
		return 2;
	show("start");

	/* 1. share peer's block into the middle of file1 */
	cr = (struct file_clone_range){ .src_fd = peer, .src_offset = 0,
					.src_length = blksz,
					.dest_offset = blksz };
	if (ioctl(f1, FICLONERANGE, &cr))
		return fprintf(stderr, "FICLONERANGE: %m\n"), 2;
	show("1 FICLONERANGE");

	/*
	 * 2. exchange file1's last block against the whole of donor1 with
	 * TO_EOF.  The sizes are exchanged too, so file1 claims one block
	 * while still owning three.  file2_offset is block 2, so this exchange
	 * cannot clear a reflink flag.
	 */
	xr = (struct xfs_exchange_range){ .file1_fd = d1, .file1_offset = 0,
					  .file2_offset = 2 * blksz,
					  .length = 0,
					  .flags = XFS_EXCHANGE_RANGE_TO_EOF };
	if (ioctl(f1, XFS_IOC_EXCHANGE_RANGE, &xr))
		return fprintf(stderr, "EXCHANGE_RANGE TO_EOF: %m\n"), 2;
	show("2 EXCHANGE_RANGE TO_EOF");

	/*
	 * 3. XFS_IOC_SWAPEXT with file1 as the target.  sx_offset is 0 and
	 * sx_length equals both inodes' i_disk_size, so the "Verify all data
	 * are being swapped" test passes.  xfs_swap_extent_rmap() then bounds
	 * its remap loop with XFS_B_TO_FSB(mp, i_size_read(VFS_I(ip))), which
	 * is one block, so file1's two mappings above its size are left in
	 * place -- and the reflink flags are swapped anyway, clearing file1's.
	 */
	fsync(f1);
	fsync(d2);
	if (fstat(f1, &st))
		return fprintf(stderr, "fstat file1: %m\n"), 2;
	memset(&sx, 0, sizeof sx);
	sx.sx_version		= XFS_SX_VERSION;
	sx.sx_fdtarget		= f1;
	sx.sx_fdtmp		= d2;
	sx.sx_offset		= 0;
	sx.sx_length		= st.st_size;
	sx.sx_stat.bs_mtime.tv_sec  = st.st_mtim.tv_sec;
	sx.sx_stat.bs_mtime.tv_nsec = st.st_mtim.tv_nsec;
	sx.sx_stat.bs_ctime.tv_sec  = st.st_ctim.tv_sec;
	sx.sx_stat.bs_ctime.tv_nsec = st.st_ctim.tv_nsec;
	if (ioctl(f1, XFS_IOC_SWAPEXT, &sx))
		return fprintf(stderr, "SWAPEXT: %m\n"), 2;
	show("3 SWAPEXT");

	/* 4. write to the block file1 still shares with peer */
	buf = malloc(blksz);
	memset(buf, 'X', blksz);
	if (pwrite(f1, buf, blksz, blksz) != (ssize_t)blksz)
		return fprintf(stderr, "pwrite: %m\n"), 2;
	fsync(f1);
	show("4 pwrite(file1, 4096)");

	posix_fadvise(peer, 0, 0, POSIX_FADV_DONTNEED);
	if (pread(peer, after, blksz, 0) != (ssize_t)blksz)
		return fprintf(stderr, "re-read peer: %m\n"), 2;
	printf("\npeer first byte %c -> %c   %s\n", before[0], after[0],
	       memcmp(before, after, blksz) ? "PEER REWRITTEN" : "peer intact");
	return memcmp(before, after, blksz) ? 0 : 1;
}

Re: [PATCH v2] xfs: stop exchanging reflink flags when exchanging mappings

[Darrick J. Wong]

On Wed, Oct 07, 2026 at 04:43:22PM +0200, Norbert Szetei wrote:
> When an exchange covers both files from offset 0 to i_disk_size,
> xmi_can_exchange_reflink_flags() assumes every mapping moved and hands
> the reflink flag over to the other file.
> 
> That assumption is wrong.  A file can still own shared blocks past
> i_disk_size, e.g. after EXCHANGE_RANGE_TO_EOF sets a smaller size
> without unmapping anything.  The flag is cleared anyway, and the next
> write to such a block happens in place instead of CoW, corrupting the
> other file that shares it.
> 
> Commit a23eca88448e ("xfs: fix exchange-range reflink flag clearing
> issue with INO1_WRITTEN") fixed one instance of this.  Rather than add
> another special case, stop moving the flags.
> xfs_exchmaps_ensure_reflink() already sets the flag on both inodes
> before the exchange, so both just keep it.
> 
> Nothing sets CLEAR_INO{1,2}_REFLINK anymore.  Intents logged by older
> kernels may still carry them; recovery already masks them off, so they
> stay in the log format mask and are ignored.
> 
> XFS_IOC_SWAPEXT has the same problem: it moves the reflink flag from
> one inode to the other along with the extents.  Just removing that is
> not enough here: the exchange path flags both inodes up front in
> xfs_exchmaps_ensure_reflink(), but swapext has no such step, so the
> inode receiving the shared extents would be left without the flag.
> Set it on both inodes instead, and make sure both have a CoW fork,
> since an inode with the reflink flag is expected to have one.
> 
> The flag now stays until FALLOC_FL_UNSHARE_RANGE, xfs_scrub, or a
> truncate to zero removes it.
> 
> Fixes: 966ceafc7a43 ("xfs: create deferred log items for file mapping exchanges")
> Fixes: f0bc4d134b46 ("xfs: swap inode reflink flags when swapping inode extents")
> Cc: stable@vger.kernel.org # v6.10
> Assisted-by: LLM
> Signed-off-by: Norbert Szetei <norbert@doyensec.com>

Yeah, let's shut off this whole class of attacks...
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
> v2: also handle XFS_IOC_SWAPEXT
> v1: https://lore.kernel.org/linux-xfs/1E196589-DEBE-40AC-AFEA-D420DAAB067F@doyensec.com/
> Discussion: https://lore.kernel.org/linux-xfs/20261006050914.GU2705364@frogsfrogsfrogs/
> 
>  fs/xfs/libxfs/xfs_exchmaps.c   | 80 ++--------------------------------
>  fs/xfs/libxfs/xfs_log_format.h |  4 +-
>  fs/xfs/xfs_bmap_util.c         | 15 +++----
>  3 files changed, 10 insertions(+), 89 deletions(-)
> 
> diff --git a/fs/xfs/libxfs/xfs_exchmaps.c b/fs/xfs/libxfs/xfs_exchmaps.c
> index 6a66b6075e0a..7af8ee35c89d 100644
> --- a/fs/xfs/libxfs/xfs_exchmaps.c
> +++ b/fs/xfs/libxfs/xfs_exchmaps.c
> @@ -49,7 +49,7 @@ struct xfs_exchmaps_adjacent {
>  	.right2 = { .br_startblock = HOLESTARTBLOCK }, \
>  }
>  
> -/* Information to reset reflink flag / CoW fork state after an exchange. */
> +/* Information to reset CoW fork state after an exchange. */
>  
>  /*
>   * If the reflink flag is set on either inode, make sure it has an incore CoW
> @@ -127,9 +127,7 @@ xmi_has_more_exchange_work(const struct xfs_exchmaps_intent *xmi)
>  static inline bool
>  xmi_has_postop_work(const struct xfs_exchmaps_intent *xmi)
>  {
> -	return xmi->xmi_flags & (XFS_EXCHMAPS_CLEAR_INO1_REFLINK |
> -				 XFS_EXCHMAPS_CLEAR_INO2_REFLINK |
> -				 __XFS_EXCHMAPS_INO2_SHORTFORM);
> +	return xmi->xmi_flags & __XFS_EXCHMAPS_INO2_SHORTFORM;
>  }
>  
>  /* Check all mappings to make sure we can actually exchange them. */
> @@ -525,18 +523,6 @@ xfs_exchmaps_link_to_sf(
>  	return error;
>  }
>  
> -/* Clear the reflink flag after an exchange. */
> -static inline void
> -xfs_exchmaps_clear_reflink(
> -	struct xfs_trans	*tp,
> -	struct xfs_inode	*ip)
> -{
> -	trace_xfs_reflink_unset_inode_flag(ip);
> -
> -	ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
> -	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
> -}
> -
>  /* Finish whatever work might come after an exchange operation. */
>  static int
>  xfs_exchmaps_do_postop_work(
> @@ -557,16 +543,6 @@ xfs_exchmaps_do_postop_work(
>  			return error;
>  	}
>  
> -	if (xmi->xmi_flags & XFS_EXCHMAPS_CLEAR_INO1_REFLINK) {
> -		xfs_exchmaps_clear_reflink(tp, xmi->xmi_ip1);
> -		xmi->xmi_flags &= ~XFS_EXCHMAPS_CLEAR_INO1_REFLINK;
> -	}
> -
> -	if (xmi->xmi_flags & XFS_EXCHMAPS_CLEAR_INO2_REFLINK) {
> -		xfs_exchmaps_clear_reflink(tp, xmi->xmi_ip2);
> -		xmi->xmi_flags &= ~XFS_EXCHMAPS_CLEAR_INO2_REFLINK;
> -	}
> -
>  	return 0;
>  }
>  
> @@ -948,46 +924,13 @@ xfs_exchmaps_intent_destroy_cache(void)
>  }
>  
>  /*
> - * Decide if we will exchange the reflink flags between the two files after the
> - * exchange.  The only time we want to do this is if we're exchanging all
> - * mappings under EOF and the inode reflink flags have different states.
> + * Allocate and initialize a new incore intent item from a request.
>   */
> -static inline bool
> -xmi_can_exchange_reflink_flags(
> -	const struct xfs_exchmaps_req	*req,
> -	unsigned int			reflink_state)
> -{
> -	struct xfs_mount		*mp = req->ip1->i_mount;
> -
> -	/*
> -	 * The INO1_WRITTEN optimization can skip exchanging hole and
> -	 * unwritten mappings, which means we cannot guarantee that all
> -	 * shared extents actually moved to the other file.  Clearing the
> -	 * reflink flag of an inode that still holds shared extents breaks
> -	 * the CoW write path, so refuse to exchange the flags in that case.
> -	 */
> -	if (req->flags & XFS_EXCHMAPS_INO1_WRITTEN)
> -		return false;
> -
> -	if (hweight32(reflink_state) != 1)
> -		return false;
> -	if (req->startoff1 != 0 || req->startoff2 != 0)
> -		return false;
> -	if (req->blockcount != XFS_B_TO_FSB(mp, req->ip1->i_disk_size))
> -		return false;
> -	if (req->blockcount != XFS_B_TO_FSB(mp, req->ip2->i_disk_size))
> -		return false;
> -	return true;
> -}
> -
> -
> -/* Allocate and initialize a new incore intent item from a request. */
>  struct xfs_exchmaps_intent *
>  xfs_exchmaps_init_intent(
>  	const struct xfs_exchmaps_req	*req)
>  {
>  	struct xfs_exchmaps_intent	*xmi;
> -	unsigned int			rs = 0;
>  
>  	xmi = kmem_cache_zalloc(xfs_exchmaps_intent_cache,
>  			GFP_NOFS | __GFP_NOFAIL);
> @@ -1011,23 +954,6 @@ xfs_exchmaps_init_intent(
>  		xmi->xmi_isize2 = req->ip1->i_disk_size;
>  	}
>  
> -	/* Record the state of each inode's reflink flag before the op. */
> -	if (xfs_is_reflink_inode(req->ip1))
> -		rs |= 1;
> -	if (xfs_is_reflink_inode(req->ip2))
> -		rs |= 2;
> -
> -	/*
> -	 * Figure out if we're clearing the reflink flags (which effectively
> -	 * exchanges them) after the operation.
> -	 */
> -	if (xmi_can_exchange_reflink_flags(req, rs)) {
> -		if (rs & 1)
> -			xmi->xmi_flags |= XFS_EXCHMAPS_CLEAR_INO1_REFLINK;
> -		if (rs & 2)
> -			xmi->xmi_flags |= XFS_EXCHMAPS_CLEAR_INO2_REFLINK;
> -	}
> -
>  	if (S_ISDIR(VFS_I(xmi->xmi_ip2)->i_mode) ||
>  	    S_ISLNK(VFS_I(xmi->xmi_ip2)->i_mode))
>  		xmi->xmi_flags |= __XFS_EXCHMAPS_INO2_SHORTFORM;
> diff --git a/fs/xfs/libxfs/xfs_log_format.h b/fs/xfs/libxfs/xfs_log_format.h
> index a4e1b3eb425c..6e620380d5dc 100644
> --- a/fs/xfs/libxfs/xfs_log_format.h
> +++ b/fs/xfs/libxfs/xfs_log_format.h
> @@ -927,10 +927,8 @@ struct xfs_xmi_log_format {
>   */
>  #define XFS_EXCHMAPS_INO1_WRITTEN	(1ULL << 2)
>  
> -/* Clear the reflink flag from inode1 after the operation. */
> +/* No longer set, but older intents may carry them; keep them valid. */
>  #define XFS_EXCHMAPS_CLEAR_INO1_REFLINK	(1ULL << 3)
> -
> -/* Clear the reflink flag from inode2 after the operation. */
>  #define XFS_EXCHMAPS_CLEAR_INO2_REFLINK	(1ULL << 4)
>  
>  #define XFS_EXCHMAPS_LOGGED_FLAGS	(XFS_EXCHMAPS_ATTR_FORK | \
> diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
> index 268d159339d0..865aec3ae5ab 100644
> --- a/fs/xfs/xfs_bmap_util.c
> +++ b/fs/xfs/xfs_bmap_util.c
> @@ -1544,7 +1544,6 @@ xfs_swap_extents(
>  	struct xfs_bstat	*sbp = &sxp->sx_stat;
>  	int			src_log_flags, target_log_flags;
>  	int			error = 0;
> -	uint64_t		f;
>  	int			resblks = 0;
>  	unsigned int		flags = 0;
>  	struct timespec64	ctime, mtime;
> @@ -1700,14 +1699,12 @@ xfs_swap_extents(
>  	if (error)
>  		goto out_trans_cancel;
>  
> -	/* Do we have to swap reflink flags? */
> -	if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
> -	    (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
> -		f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
> -		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
> -		ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
> -		tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
> -		tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
> +	/* Either inode may now own shared blocks; flag both, with CoW forks. */
> +	if ((ip->i_diflags2 | tip->i_diflags2) & XFS_DIFLAG2_REFLINK) {
> +		ip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
> +		tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
> +		xfs_ifork_init_cow(ip);
> +		xfs_ifork_init_cow(tip);
>  	}
>  
>  	/* Swap the cow forks. */
> -- 
> 2.55.0
> 
> 

Re: [PATCH v2] xfs: stop exchanging reflink flags when exchanging mappings

[Darrick J. Wong]

On Wed, Oct 07, 2026 at 04:46:39PM +0200, Norbert Szetei wrote:
> Here is a reproducer for the XFS_IOC_SWAPEXT case, since there wasn't
> one on the list yet, in case it's useful for fstests.
> 
> 	$ truncate -s 512M img && mkfs.xfs -q -m rmapbt=1,reflink=1 img
> 	$ mkdir -p /tmp/m && sudo mount -o loop img /tmp/m && sudo chmod 1777 /tmp/m
> 	$ sudo xfs_io -f -c "pwrite -q -S 0x50 0 4096" /tmp/m/peer
> 	$ sudo chmod 644 /tmp/m/peer
> 	$ cd /tmp/m && ./xfs_swapext_poc
> 
> On an unpatched kernel it reports peer as rewritten. With v2 applied,
> peer is left intact. It needs rmapbt=1.

Thanks for the third repro, I'll kludge that into a fstests too.

--D

> // SPDX-License-Identifier: GPL-2.0
> #define _GNU_SOURCE
> #include <stdio.h>
> #include <stdlib.h>
> #include <string.h>
> #include <unistd.h>
> #include <fcntl.h>
> #include <sys/ioctl.h>
> #include <sys/stat.h>
> #include <sys/vfs.h>
> #include <linux/fs.h>
> #include <linux/fiemap.h>
> 
> #ifndef FICLONERANGE
> #define FICLONERANGE _IOW(0x94, 13, struct file_clone_range)
> #endif
> 
> /* fs/xfs/libxfs/xfs_fs.h */
> struct xfs_exchange_range {
> 	__s32		file1_fd;
> 	__u32		pad;
> 	__u64		file1_offset;
> 	__u64		file2_offset;
> 	__u64		length;
> 	__u64		flags;
> };
> #define XFS_IOC_EXCHANGE_RANGE		_IOW('X', 129, struct xfs_exchange_range)
> #define XFS_EXCHANGE_RANGE_TO_EOF	(1ULL << 0)
> 
> typedef struct xfs_bstime {
> 	long		tv_sec;
> 	__s32		tv_nsec;
> } xfs_bstime_t;
> 
> struct xfs_bstat {
> 	__u64		bs_ino;
> 	__u16		bs_mode;
> 	__u16		bs_nlink;
> 	__u32		bs_uid;
> 	__u32		bs_gid;
> 	__u32		bs_rdev;
> 	__s32		bs_blksize;
> 	__s64		bs_size;
> 	xfs_bstime_t	bs_atime;
> 	xfs_bstime_t	bs_mtime;
> 	xfs_bstime_t	bs_ctime;
> 	__s64		bs_blocks;
> 	__u32		bs_xflags;
> 	__s32		bs_extsize;
> 	__s32		bs_extents;
> 	__u32		bs_gen;
> 	__u16		bs_projid_lo;
> 	__u16		bs_forkoff;
> 	__u16		bs_projid_hi;
> 	__u16		bs_sick;
> 	__u16		bs_checked;
> 	unsigned char	bs_pad[2];
> 	__u32		bs_cowextsize;
> 	__u32		bs_dmevmask;
> 	__u16		bs_dmstate;
> 	__u16		bs_aextents;
> };
> 
> struct xfs_swapext {
> 	__s64			sx_version;
> #define XFS_SX_VERSION		0
> 	__s64			sx_fdtarget;
> 	__s64			sx_fdtmp;
> 	__s64			sx_offset;
> 	__s64			sx_length;
> 	char			sx_pad[16];
> 	struct xfs_bstat	sx_stat;
> };
> #define XFS_IOC_SWAPEXT		_IOWR('X', 109, struct xfs_swapext)
> 
> static unsigned long blksz;
> static int peer, f1;
> 
> static unsigned long long phys(int fd, unsigned long long off)
> {
> 	struct { struct fiemap f; struct fiemap_extent e[1]; } q;
> 
> 	memset(&q, 0, sizeof q);
> 	q.f.fm_start = off;
> 	q.f.fm_length = blksz;
> 	q.f.fm_extent_count = 1;
> 	if (ioctl(fd, FS_IOC_FIEMAP, &q.f) || q.f.fm_mapped_extents < 1)
> 		return 0;
> 	return (q.e[0].fe_physical + (off - q.e[0].fe_logical)) / blksz;
> }
> 
> static void show(const char *tag)
> {
> 	struct stat st;
> 	int i;
> 
> 	fstat(f1, &st);
> 	printf("%-26s peer [%llu]   file1 size %5lld  [", tag, phys(peer, 0),
> 	       (long long)st.st_size);
> 	for (i = 0; i < 3; i++)
> 		printf("%llu%s", phys(f1, (unsigned long long)i * blksz),
> 		       i == 2 ? "]\n" : "] [");
> }
> 
> static int mkfile(const char *name, char fill, int blocks)
> {
> 	char *buf;
> 	int fd;
> 
> 	unlink(name);
> 	fd = open(name, O_RDWR | O_CREAT | O_EXCL, 0600);
> 	if (fd < 0)
> 		return fprintf(stderr, "open %s: %m\n", name), -1;
> 	buf = malloc(blocks * blksz);
> 	memset(buf, fill, blocks * blksz);
> 	if (pwrite(fd, buf, blocks * blksz, 0) != (ssize_t)(blocks * blksz))
> 		return fprintf(stderr, "pwrite %s: %m\n", name), -1;
> 	free(buf);
> 	fsync(fd);
> 	return fd;
> }
> 
> int main(void)
> {
> 	struct xfs_exchange_range xr;
> 	struct file_clone_range cr;
> 	struct xfs_swapext sx;
> 	char before[65536], after[65536], *buf;
> 	struct statfs sfs;
> 	struct stat st;
> 	int d1, d2;
> 
> 	if (statfs(".", &sfs))
> 		return fprintf(stderr, "statfs: %m\n"), 2;
> 	blksz = sfs.f_bsize;
> 
> 	peer = open("peer", O_RDONLY);
> 	if (peer < 0)
> 		return fprintf(stderr, "open peer: %m\n"), 2;
> 	fstat(peer, &st);
> 	printf("peer uid %u mode %04o size %lld, block size %lu\n\n",
> 	       st.st_uid, st.st_mode & 07777, (long long)st.st_size, blksz);
> 	if (pread(peer, before, blksz, 0) != (ssize_t)blksz)
> 		return fprintf(stderr, "read peer: %m\n"), 2;
> 	syncfs(peer);	/* so FIEMAP reports peer's real block */
> 
> 	f1 = mkfile("file1", 'A', 3);
> 	d1 = mkfile("donor1", 'B', 1);
> 	d2 = mkfile("donor2", 'C', 1);
> 	if (f1 < 0 || d1 < 0 || d2 < 0)
> 		return 2;
> 	show("start");
> 
> 	/* 1. share peer's block into the middle of file1 */
> 	cr = (struct file_clone_range){ .src_fd = peer, .src_offset = 0,
> 					.src_length = blksz,
> 					.dest_offset = blksz };
> 	if (ioctl(f1, FICLONERANGE, &cr))
> 		return fprintf(stderr, "FICLONERANGE: %m\n"), 2;
> 	show("1 FICLONERANGE");
> 
> 	/*
> 	 * 2. exchange file1's last block against the whole of donor1 with
> 	 * TO_EOF.  The sizes are exchanged too, so file1 claims one block
> 	 * while still owning three.  file2_offset is block 2, so this exchange
> 	 * cannot clear a reflink flag.
> 	 */
> 	xr = (struct xfs_exchange_range){ .file1_fd = d1, .file1_offset = 0,
> 					  .file2_offset = 2 * blksz,
> 					  .length = 0,
> 					  .flags = XFS_EXCHANGE_RANGE_TO_EOF };
> 	if (ioctl(f1, XFS_IOC_EXCHANGE_RANGE, &xr))
> 		return fprintf(stderr, "EXCHANGE_RANGE TO_EOF: %m\n"), 2;
> 	show("2 EXCHANGE_RANGE TO_EOF");
> 
> 	/*
> 	 * 3. XFS_IOC_SWAPEXT with file1 as the target.  sx_offset is 0 and
> 	 * sx_length equals both inodes' i_disk_size, so the "Verify all data
> 	 * are being swapped" test passes.  xfs_swap_extent_rmap() then bounds
> 	 * its remap loop with XFS_B_TO_FSB(mp, i_size_read(VFS_I(ip))), which
> 	 * is one block, so file1's two mappings above its size are left in
> 	 * place -- and the reflink flags are swapped anyway, clearing file1's.
> 	 */
> 	fsync(f1);
> 	fsync(d2);
> 	if (fstat(f1, &st))
> 		return fprintf(stderr, "fstat file1: %m\n"), 2;
> 	memset(&sx, 0, sizeof sx);
> 	sx.sx_version		= XFS_SX_VERSION;
> 	sx.sx_fdtarget		= f1;
> 	sx.sx_fdtmp		= d2;
> 	sx.sx_offset		= 0;
> 	sx.sx_length		= st.st_size;
> 	sx.sx_stat.bs_mtime.tv_sec  = st.st_mtim.tv_sec;
> 	sx.sx_stat.bs_mtime.tv_nsec = st.st_mtim.tv_nsec;
> 	sx.sx_stat.bs_ctime.tv_sec  = st.st_ctim.tv_sec;
> 	sx.sx_stat.bs_ctime.tv_nsec = st.st_ctim.tv_nsec;
> 	if (ioctl(f1, XFS_IOC_SWAPEXT, &sx))
> 		return fprintf(stderr, "SWAPEXT: %m\n"), 2;
> 	show("3 SWAPEXT");
> 
> 	/* 4. write to the block file1 still shares with peer */
> 	buf = malloc(blksz);
> 	memset(buf, 'X', blksz);
> 	if (pwrite(f1, buf, blksz, blksz) != (ssize_t)blksz)
> 		return fprintf(stderr, "pwrite: %m\n"), 2;
> 	fsync(f1);
> 	show("4 pwrite(file1, 4096)");
> 
> 	posix_fadvise(peer, 0, 0, POSIX_FADV_DONTNEED);
> 	if (pread(peer, after, blksz, 0) != (ssize_t)blksz)
> 		return fprintf(stderr, "re-read peer: %m\n"), 2;
> 	printf("\npeer first byte %c -> %c   %s\n", before[0], after[0],
> 	       memcmp(before, after, blksz) ? "PEER REWRITTEN" : "peer intact");
> 	return memcmp(before, after, blksz) ? 0 : 1;
> }
> 
> 

Re: [PATCH v2] xfs: stop exchanging reflink flags when exchanging mappings

[Dave Chinner]

On Wed, Oct 07, 2026 at 04:43:22PM +0200, Norbert Szetei wrote:
> When an exchange covers both files from offset 0 to i_disk_size,
> xmi_can_exchange_reflink_flags() assumes every mapping moved and hands
> the reflink flag over to the other file.
> 
> That assumption is wrong.  A file can still own shared blocks past
> i_disk_size, e.g. after EXCHANGE_RANGE_TO_EOF sets a smaller size
> without unmapping anything.  The flag is cleared anyway, and the next
> write to such a block happens in place instead of CoW, corrupting the
> other file that shares it.
> 
> Commit a23eca88448e ("xfs: fix exchange-range reflink flag clearing
> issue with INO1_WRITTEN") fixed one instance of this.  Rather than add
> another special case, stop moving the flags.
> xfs_exchmaps_ensure_reflink() already sets the flag on both inodes
> before the exchange, so both just keep it.
> 
> Nothing sets CLEAR_INO{1,2}_REFLINK anymore.  Intents logged by older
> kernels may still carry them; recovery already masks them off, so they
> stay in the log format mask and are ignored.
> 
> XFS_IOC_SWAPEXT has the same problem: it moves the reflink flag from
> one inode to the other along with the extents.  Just removing that is
> not enough here: the exchange path flags both inodes up front in
> xfs_exchmaps_ensure_reflink(), but swapext has no such step, so the
> inode receiving the shared extents would be left without the flag.
> Set it on both inodes instead, and make sure both have a CoW fork,
> since an inode with the reflink flag is expected to have one.
> 
> The flag now stays until FALLOC_FL_UNSHARE_RANGE, xfs_scrub, or a
> truncate to zero removes it.
> 
> Fixes: 966ceafc7a43 ("xfs: create deferred log items for file mapping exchanges")
> Fixes: f0bc4d134b46 ("xfs: swap inode reflink flags when swapping inode extents")
> Cc: stable@vger.kernel.org # v6.10
> Assisted-by: LLM
> Signed-off-by: Norbert Szetei <norbert@doyensec.com>
> ---
> v2: also handle XFS_IOC_SWAPEXT
> v1: https://lore.kernel.org/linux-xfs/1E196589-DEBE-40AC-AFEA-D420DAAB067F@doyensec.com/
> Discussion: https://lore.kernel.org/linux-xfs/20261006050914.GU2705364@frogsfrogsfrogs/

Agreed, this seems like the best course of action given the recurring issues
in this area.

Reviewed-by: Dave Chinner <dchinner@redhat.com>

-- 
Dave Chinner
dgc@kernel.org