Skip to contentSkip to conversation

swap: Add exfat to filesystems requiring tst_fill_file

PATCH In review Jan Polensky 6 participants 22 messages Jun 8 – Aug 26

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 swap: Add exfat to filesystems requiring tst_fill_file

exfat, like XFS, requires actual data writes (not just fallocate) to fully allocate swap files. Without this, swapon(2) fails with EINVAL due to unallocated extents.

This matches the kernel's requirement that swap files must be fully allocated before activation.

---
 libs/swap/tse_swap.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
libs/swap/tse_swap.c +3 -2
index ae3c193132e6..e321cb1dfe63 100644
@@ -177,8 +177,9 @@ int make_swapfile(const char *file, const int lineno,
 	if (prealloc_contiguous_file(swapfile, blk_size, blocks) != 0)
 		tst_brk_(file, lineno, TBROK, "Failed to create swapfile");
 
-	/* Fill the file if needed (specific to old xfs filesystems) */
-	if (tst_fs_type(swapfile) == TST_XFS_MAGIC) {
+	/* Fill the file if needed (specific to xfs and exfat filesystems) */
+	long fs_type = tst_fs_type(swapfile);
+	if (fs_type == TST_XFS_MAGIC || fs_type == TST_EXFAT_MAGIC) {
 		if (tst_fill_file(swapfile, 0, blk_size, blocks) != 0)
 			tst_brk_(file, lineno, TBROK, "Failed to fill swapfile");
 	}
Signature · 5 lines
2.54.0


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

Hi Jan,

Quoted 1 line · Jan Polensky

swap: Add exfat to filesystems requiring tst_fill_file

Verdict: Reviewed

---
Note:

The agent can sometimes produce false positives although often its findings are genuine. If you find issues with the review, please comment this email or ignore the suggestions.

Regards,
LTP AI Reviewer

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw
In reply to Jan Polensky

Hi!

Quoted 18 lines

exfat, like XFS, requires actual data writes (not just fallocate) to fully allocate swap files. Without this, swapon(2) fails with EINVAL due to unallocated extents.

This matches the kernel's requirement that swap files must be fully allocated before activation.

---
 libs/swap/tse_swap.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
libs/swap/tse_swap.c +0 -0
index ae3c193132e6..e321cb1dfe63 100644
@@ -177,8 +177,9 @@ int make_swapfile(const char *file, const int lineno,
 	if (prealloc_contiguous_file(swapfile, blk_size, blocks) != 0)

libs/swap/tse_swap.c int make_swapfile(const char *file, const int lineno, · line 185

Quoted 10 lines
libs/swap/tse_swap.c +3 -2
 		tst_brk_(file, lineno, TBROK, "Failed to create swapfile");
 
-	/* Fill the file if needed (specific to old xfs filesystems) */
-	if (tst_fs_type(swapfile) == TST_XFS_MAGIC) {
+	/* Fill the file if needed (specific to xfs and exfat filesystems) */
+	long fs_type = tst_fs_type(swapfile);
+	if (fs_type == TST_XFS_MAGIC || fs_type == TST_EXFAT_MAGIC) {
 		if (tst_fill_file(swapfile, 0, blk_size, blocks) != 0)
 			tst_brk_(file, lineno, TBROK, "Failed to fill swapfile");
 	}

The swapon01 works for me for exfat without this change. Where does this fail (arch, kernel, OS, is exfat in kernel or FUSE, etc.)?

Signature · 5 lines
Cyril Hrubis
[email protected]

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#3 lore raw
Quoted 21 lines · Cyril Hrubis

Hi!

Quoted 20 lines

exfat, like XFS, requires actual data writes (not just fallocate) to fully allocate swap files. Without this, swapon(2) fails with EINVAL due to unallocated extents.

This matches the kernel's requirement that swap files must be fully allocated before activation.

---
 libs/swap/tse_swap.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
libs/swap/tse_swap.c +0 -0
index ae3c193132e6..e321cb1dfe63 100644
@@ -177,8 +177,9 @@ int make_swapfile(const char *file, const int lineno,
 	if (prealloc_contiguous_file(swapfile, blk_size, blocks) != 0)
 		tst_brk_(file, lineno, TBROK, "Failed to create swapfile");
 
Quoted 9 lines

libs/swap/tse_swap.c int make_swapfile(const char *file, const int lineno, · line 185

Quoted 8 lines
libs/swap/tse_swap.c +3 -2
-	/* Fill the file if needed (specific to old xfs filesystems) */
-	if (tst_fs_type(swapfile) == TST_XFS_MAGIC) {
+	/* Fill the file if needed (specific to xfs and exfat filesystems) */
+	long fs_type = tst_fs_type(swapfile);
+	if (fs_type == TST_XFS_MAGIC || fs_type == TST_EXFAT_MAGIC) {
 		if (tst_fill_file(swapfile, 0, blk_size, blocks) != 0)
 			tst_brk_(file, lineno, TBROK, "Failed to fill swapfile");
 	}

Hi Cyril,

Quoted 2 lines

The swapon01 works for me for exfat without this change. Where does this fail (arch, kernel, OS, is exfat in kernel or FUSE, etc.)?

The failure occurred on:
- Architecture: s390x
- Kernel: 7.1.0-20260608.rc7.git374.a87737435cfa.300.fc44.s390x+next (linux-next and since ~ 17 days)
- OS: Fedora Linux 44
- exfat: Kernel module (not FUSE)

The issue manifests when swapon(2) is called on an exfat swap file that was created using only fallocate() without actual data writes. The kernel rejects the swap file with EINVAL because it detects unallocated extents.

This is similar to the existing XFS behavior where tst_fill_file() is already required. Both filesystems need actual data writes (not just metadata allocation via fallocate) to ensure swap files are fully allocated before swapon(2) can activate them.

The patch ensures consistent behavior across filesystems that have this requirement.

Best regards,
Jan

Quoted 4 lines

--
Cyril Hrubis
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#4 lore raw

Hi Jan,

maybe I reproduced it, but it would be nice to have the test output as well. Also Kernel 7.1 is still in RC stage, so this patch must wait kernel release before being considered valid.

The next time please add more information to the commit, so we know exactly what kernel version is having issues, as well as the arch and distro.

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#5 lore raw

Hi!

Quoted 3 lines

maybe I reproduced it, but it would be nice to have the test output as well. Also Kernel 7.1 is still in RC stage, so this patch must wait kernel release before being considered valid.

If it's reproducible only in 7.1 isn't it a kernel regression?

Signature · 5 lines
Cyril Hrubis
[email protected]

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#6 lore raw

Hi Cyril,

Quoted 5 lines
Quoted 3 lines

maybe I reproduced it, but it would be nice to have the test output as well. Also Kernel 7.1 is still in RC stage, so this patch must wait kernel release before being considered valid.

If it's reproducible only in 7.1 isn't it a kernel regression?

Had time to go back looking at the results. Error was caused by an another issue on my side. I also can't reprouce the issue on kernel 7.1.0 after the release. I ran swapon01, but it runs fine.

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#7 lore raw

An update. I had to go through the code to understand if we are having issues from our side. Indeed, there's a commit which is responsible for this exfat behaviour and it's causing a regression-like issue:

bf1797960c20 - exfat: add fallocate FALLOC_FL_ALLOCATE_RANGE support

tst_prealloc_size_fd() is called by make_swapfile() which is called by our swapon/swapoff testing suites. Inside it we use fallocate() that is now supported by exfat and returns 0 after the kernel patch.

This naturally falls down into the next:

	/* Fill the file if needed (specific to old xfs filesystems) */
	if (tst_fs_type(swapfile) == TST_XFS_MAGIC) {
		if (tst_fill_file(swapfile, 0, blk_size, blocks) != 0)
			tst_brk_(file, lineno, TBROK, "Failed to fill swapfile");
	}

but there's no data written and we get EINVAL at the end of the tests because swapon() is using a file created with fallocate() instead.

The Jan assumption is correct and the patch as well, but it's strange we can't reproduce it. The EINVAL is the expected behaviour in this case.

I'm just wondering how to implement it better.

First of all, we should mention the commit that introduced fallocate() support into exfat inside the patch commit message.

Secondly, since the code was already working before v7.1, maybe we should fill the file in exfat only for kernel >= 7.1. But I don't want to touch the core library.

@Cyril WDYT?

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#8 lore raw
Quoted 8 lines

I'm just wondering how to implement it better.

First of all, we should mention the commit that introduced fallocate() support into exfat inside the patch commit message.

Secondly, since the code was already working before v7.1, maybe we should fill the file in exfat only for kernel >= 7.1. But I don't want to touch the core library.

Second update: .swap_activate is not implemented in the linux kernel for exfat and the commit causes EINVAL only for swap allocation. That is caused by bmap() due to the generic swap activation.

So instead of working around the issue, we probably should have a feature in the kernel along with fallocate() support in exfat.

Probably a kernel bug/missing feature.

--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#9 lore raw
Quoted 16 lines · Andrea Cervesato
Quoted 8 lines

I'm just wondering how to implement it better.

First of all, we should mention the commit that introduced fallocate() support into exfat inside the patch commit message.

Secondly, since the code was already working before v7.1, maybe we should fill the file in exfat only for kernel >= 7.1. But I don't want to touch the core library.

Second update: .swap_activate is not implemented in the linux kernel for exfat and the commit causes EINVAL only for swap allocation. That is caused by bmap() due to the generic swap activation.

So instead of working around the issue, we probably should have a feature in the kernel along with fallocate() support in exfat.

Quoted 6 lines

Probably a kernel bug/missing feature.

--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Hi Andrea,

just to add some context: the regression itself was on the kernel side. The relevant upstream patch is
https://lore.kernel.org/all/[email protected]/.
My understanding is that the kernel-side change did not resolve the issue for the LTP test suite, and that there was also a separate LTP-relevant aspect for exfat swapfiles, where actual data writes are required before swapon() succeeds.

Best regards,
Jan

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#10 lore raw

Hi Jan,

Quoted 9 lines

Hi Andrea,

just to add some context: the regression itself was on the kernel side. The relevant upstream patch is
https://lore.kernel.org/all/[email protected]/.
My understanding is that the kernel-side change did not resolve the issue for the LTP test suite, and that there was also a separate LTP-relevant aspect for exfat swapfiles, where actual data writes are required before swapon() succeeds.

Sorry, I forgot to properly reply to this email.

Today I sent a patch to fix the kernel, you can take a look at it here:
https://lore.kernel.org/all/[email protected]/

This patch should fix the issue, let me know if you are still seeing a failure in the LTP tests after applying it.

Thanks,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#11 lore raw
Quoted 17 lines · Andrea Cervesato

Hi Jan,

Quoted 9 lines

Hi Andrea,

just to add some context: the regression itself was on the kernel side. The relevant upstream patch is
https://lore.kernel.org/all/[email protected]/.
My understanding is that the kernel-side change did not resolve the issue for the LTP test suite, and that there was also a separate LTP-relevant aspect for exfat swapfiles, where actual data writes are required before swapon() succeeds.

Sorry, I forgot to properly reply to this email.

Today I sent a patch to fix the kernel, you can take a look at it here:
https://lore.kernel.org/all/[email protected]/

Quoted 8 lines

This patch should fix the issue, let me know if you are still seeing a failure in the LTP tests after applying it.

Thanks,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Hi Andrea,

thanks, this fixes the swapping issue for me on s390.

There is just one remaining problem: when exfat is built as a module (CONFIG_EXFAT_FS=m), the build fails because generic_swapfile_activate() is not exported.

Please add EXPORT_SYMBOL_GPL(generic_swapfile_activate) in mm/page_io.c.

Best regards,
Jan

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#12 lore raw
Quoted 1 line

thanks, this fixes the swapping issue for me on s390.

Great :-)

Quoted 6 lines

There is just one remaining problem: when exfat is built as a module (CONFIG_EXFAT_FS=m), the build fails because generic_swapfile_activate() is not exported.

Please add EXPORT_SYMBOL_GPL(generic_swapfile_activate) in mm/page_io.c.

Ah that's right, I will wait for kernel devs reply and then point it out in the conversation. Thanks for the suggestion.

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#13 lore raw
In reply to Jan Polensky

Hi Jan,

I seen you implemented in v7.2-rc1:

commit 03a43677ca91e3997355a13b1da6e47329b025e1

Author: Jan Polensky <[email protected]>
Date:   Tue Jun 9 18:21:49 2026 +0900

exfat: add swap_activate support

Commit 07d67f3e9083 ("exfat: add iomap buffered I/O support") converted exfat buffered I/O to iomap, but did not add a .swap_activate handler to the address_space_operations.

swapon(2) on an exfat swapfile then fails with EINVAL, which causes LTP swap tests to fail.

Add exfat_iomap_swap_activate() and hook it into exfat_aops so exfat uses iomap_swapfile_activate() for swapfile activation.

Fixes 614f71ca1bdf ("exfat: add iomap buffered I/O support")

what should we do with your patch in LTP?

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#14 lore raw
Quoted 25 lines · Andrea Cervesato

Hi Jan,

I seen you implemented in v7.2-rc1:

commit 03a43677ca91e3997355a13b1da6e47329b025e1

Author: Jan Polensky <[email protected]>
Date:   Tue Jun 9 18:21:49 2026 +0900

exfat: add swap_activate support

Commit 07d67f3e9083 ("exfat: add iomap buffered I/O support") converted exfat buffered I/O to iomap, but did not add a .swap_activate handler to the address_space_operations.

swapon(2) on an exfat swapfile then fails with EINVAL, which causes LTP swap tests to fail.

Add exfat_iomap_swap_activate() and hook it into exfat_aops so exfat uses iomap_swapfile_activate() for swapfile activation.

Fixes 614f71ca1bdf ("exfat: add iomap buffered I/O support")
Quoted 7 lines

what should we do with your patch in LTP?

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Hi Andrea,

Thanks for following up on this.

On my side (s390x, Fedora 44, kernel v7.2-rc1), I still need both patches for the LTP swap tests to pass:
1. The kernel fix (03a43677ca91 "exfat: add swap_activate support")
2. This LTP patch (adding exfat to tst_fill_file)

Without the LTP patch, the tests still fail even with the kernel fix applied.

Could you validate on your setup whether the kernel fix alone is sufficient, or if you also need this LTP patch? It would be good to understand if this is architecture-specific or a general requirement.

If you can confirm the tests pass with only the kernel fix, we can close this LTP patch. Otherwise, one should merge it.

Best regards,
Jan

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
Reviewed-by
#15 lore raw
Quoted 46 lines · Andrea Cervesato
Quoted 25 lines

Hi Jan,

I seen you implemented in v7.2-rc1:

commit 03a43677ca91e3997355a13b1da6e47329b025e1

Author: Jan Polensky <[email protected]>
Date:   Tue Jun 9 18:21:49 2026 +0900

exfat: add swap_activate support

Commit 07d67f3e9083 ("exfat: add iomap buffered I/O support") converted exfat buffered I/O to iomap, but did not add a .swap_activate handler to the address_space_operations.

swapon(2) on an exfat swapfile then fails with EINVAL, which causes LTP swap tests to fail.

Add exfat_iomap_swap_activate() and hook it into exfat_aops so exfat uses iomap_swapfile_activate() for swapfile activation.

Fixes 614f71ca1bdf ("exfat: add iomap buffered I/O support")
Quoted 7 lines

what should we do with your patch in LTP?

Regards,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Hi Andrea,

Thanks for following up on this.

On my side (s390x, Fedora 44, kernel v7.2-rc1), I still need both patches for the LTP swap tests to pass:
1. The kernel fix (03a43677ca91 "exfat: add swap_activate support")
2. This LTP patch (adding exfat to tst_fill_file)

Without the LTP patch, the tests still fail even with the kernel fix applied.

Quoted 6 lines

Could you validate on your setup whether the kernel fix alone is sufficient, or if you also need this LTP patch? It would be good to understand if this is architecture-specific or a general requirement.

If you can confirm the tests pass with only the kernel fix, we can close this LTP patch. Otherwise, one should merge it.

And yeah, we need it in kernel 7.2 on any distro :).

Should we somehow mention relevant kernel commit at least in the commit message?

e.g.: Change needed since 03a43677ca91 ("exfat: add swap_activate support") from v7.2-rc1.

Kind regards,
Petr

Quoted 2 lines

Best regards,
Jan

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#16 lore raw

Hi Petr,

Quoted 8 lines

And yeah, we need it in kernel 7.2 on any distro :).

Should we somehow mention relevant kernel commit at least in the commit message?

e.g.: Change needed since 03a43677ca91 ("exfat: add swap_activate support") from v7.2-rc1.

yes it makes sense to do it before merge.

Thanks,
--
Andrea Cervesato
SUSE QE Automation Engineer Linux
[email protected]

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#17 lore raw
In reply to Petr Vorel

Hi!

Quoted 8 lines

And yeah, we need it in kernel 7.2 on any distro :).

Should we somehow mention relevant kernel commit at least in the commit message?

e.g.: Change needed since 03a43677ca91 ("exfat: add swap_activate support") from v7.2-rc1.

That does not make much sense, if swap_activate was added to kernel, pre-filling the fallocated file shouldn't be needed anymore.

Signature · 5 lines
Cyril Hrubis
[email protected]

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#18 lore raw

Hi all,

[ Cc kernel folks and LKML ]

Quoted 4 lines

Hi!

Quoted 3 lines

And yeah, we need it in kernel 7.2 on any distro :).

... and on any arch, e.g. x86_64 (not just s390x as Jan previously reported [1])

Quoted 7 lines
Quoted 4 lines

Should we somehow mention relevant kernel commit at least in the commit message?

e.g.: Change needed since 03a43677ca91 ("exfat: add swap_activate support") from v7.2-rc1.

That does not make much sense, if swap_activate was added to kernel, pre-filling the fallocated file shouldn't be needed anymore.

Good catch, you're right. I'm also confused, before v7.2 swap worked on exfat, now even with 82a81a7352bc ("exfat: add iomap buffered I/O support") [2] and your fix 03a43677ca91 ("exfat: add swap_activate support") [3] is not working. Obviously the fix was not good enough.

Kind regards,
Petr

[1] https://lore.kernel.org/ltp/[email protected]/
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=82a81a7352bcf5f2756ac33d47ee0582737e9a85
[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=03a43677ca91e3997355a13b1da6e47329b025e1

#19 lore raw

Hi!

Quoted 7 lines
Quoted 2 lines

That does not make much sense, if swap_activate was added to kernel, pre-filling the fallocated file shouldn't be needed anymore.

Good catch, you're right. I'm also confused, before v7.2 swap worked on exfat, now even with 82a81a7352bc ("exfat: add iomap buffered I/O support") [2] and your fix 03a43677ca91 ("exfat: add swap_activate support") [3] is not working. Obviously the fix was not good enough.

There were two bugs I suppose, one is that fallocate() support for exfat broke the swapon on fallocated file. The second bug was introduced later when exfat was rewritten to use iomap. The iomap case was fixed in the patch above, however it looks like the original bug caused by fallocate() implementation for exfat is still there.

Signature · 2 lines
Cyril Hrubis
[email protected]
maintainer
#20 lore raw
Quoted 2 lines · Cyril Hrubis

Hi!

Hi Cyril, Petr,

Quoted 8 lines
Quoted 7 lines
Quoted 2 lines

That does not make much sense, if swap_activate was added to kernel, pre-filling the fallocated file shouldn't be needed anymore.

Good catch, you're right. I'm also confused, before v7.2 swap worked on exfat, now even with 82a81a7352bc ("exfat: add iomap buffered I/O support") [2] and your fix 03a43677ca91 ("exfat: add swap_activate support") [3] is not working. Obviously the fix was not good enough.

Quoted 5 lines

There were two bugs I suppose, one is that fallocate() support for exfat broke the swapon on fallocated file. The second bug was introduced later when exfat was rewritten to use iomap. The iomap case was fixed in the patch above, however it looks like the original bug caused by fallocate() implementation for exfat is still there.

Can you check if an attached patch improves this issue? Thanks.

From 5679a624816454c8e7683842eb0ab3f1adf0dcca Mon Sep 17 00:00:00 2001
From: Namjae Jeon <[email protected]>
Date: Wed, 26 Aug 2026 09:14:14 +0900
Subject: [PATCH] exfat: map allocated extents for swap activation

exFAT reports allocated ranges beyond valid_size as IOMAP_HOLE when IOMAP_REPORT is set. iomap_swapfile_activate() also uses IOMAP_REPORT while collecting physical extents for a swap file, so it treats the preallocated tail as unallocated and rejects the file with -EINVAL.

Use swap-specific iomap operations that suppress IOMAP_REPORT before mapping the file. This reports physically allocated ranges as IOMAP_UNWRITTEN during swap activation without changing the byte-accurate SEEK_HOLE and SEEK_DATA behavior of the regular iomap operations.

Fixes 03a43677ca91 ("exfat: add swap_activate support")
---
 fs/exfat/iomap.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)
fs/exfat/iomap.c +18 -1
index 0c805bf6676a..bf8baf78a9b8 100644
@@ -155,6 +155,22 @@ const struct iomap_ops exfat_iomap_ops = {
 	.iomap_begin = exfat_iomap_begin,
 };
 
+static int exfat_swap_iomap_begin(struct inode *inode, loff_t offset,
+		loff_t length, unsigned int flags, struct iomap *iomap,
+		struct iomap *srcmap)
+{
+	/*
+	 * Swap activation needs the physical mappings of preallocated
+	 * ranges. Do not report the VDL tail as a hole.
+	 */
+	return __exfat_iomap_begin(inode, offset, length,
+			flags & ~IOMAP_REPORT, iomap, false);
+}
+
+static const struct iomap_ops exfat_swap_iomap_ops = {
+	.iomap_begin = exfat_swap_iomap_begin,
+};
+
 /*
  * exfat_write_iomap_end - Update the state after write
  *
@@ -274,5 +290,6 @@ const struct iomap_read_ops exfat_iomap_bio_read_ops = {
 int exfat_iomap_swap_activate(struct swap_info_struct *sis,
 			       struct file *file, sector_t *span)
 {
-	return iomap_swapfile_activate(sis, file, span, &exfat_iomap_ops);
+	return iomap_swapfile_activate(sis, file, span,
+				       &exfat_swap_iomap_ops);
 }
Signature · 1 line
2.34.1
Tested-by
#21 lore raw

Hi Namjae,

Quoted 12 lines · Cyril Hrubis
Quoted 2 lines

Hi!

Hi Cyril, Petr,

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

That does not make much sense, if swap_activate was added to kernel, pre-filling the fallocated file shouldn't be needed anymore.

Good catch, you're right. I'm also confused, before v7.2 swap worked on exfat, now even with 82a81a7352bc ("exfat: add iomap buffered I/O support") [2] and your fix 03a43677ca91 ("exfat: add swap_activate support") [3] is not working. Obviously the fix was not good enough.

Quoted 7 lines
Quoted 5 lines

There were two bugs I suppose, one is that fallocate() support for exfat broke the swapon on fallocated file. The second bug was introduced later when exfat was rewritten to use iomap. The iomap case was fixed in the patch above, however it looks like the original bug caused by fallocate() implementation for exfat is still there.

Can you check if an attached patch improves this issue? Thanks.

LGTM.

Tested on rapido-linux with:
CONFIG_EXFAT_FS=y
CONFIG_EXFAT_DEFAULT_IOCHARSET="utf8"

Hopefully Jan will test it as well, built as module.

Kind regards,
Petr

Quoted 57 lines

From 5679a624816454c8e7683842eb0ab3f1adf0dcca Mon Sep 17 00:00:00 2001
From: Namjae Jeon <[email protected]>
Date: Wed, 26 Aug 2026 09:14:14 +0900
Subject: [PATCH] exfat: map allocated extents for swap activation

exFAT reports allocated ranges beyond valid_size as IOMAP_HOLE when IOMAP_REPORT is set. iomap_swapfile_activate() also uses IOMAP_REPORT while collecting physical extents for a swap file, so it treats the preallocated tail as unallocated and rejects the file with -EINVAL.

Use swap-specific iomap operations that suppress IOMAP_REPORT before mapping the file. This reports physically allocated ranges as IOMAP_UNWRITTEN during swap activation without changing the byte-accurate SEEK_HOLE and SEEK_DATA behavior of the regular iomap operations.

Fixes 03a43677ca91 ("exfat: add swap_activate support")
---
 fs/exfat/iomap.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)
fs/exfat/iomap.c +18 -1
index 0c805bf6676a..bf8baf78a9b8 100644
@@ -155,6 +155,22 @@ const struct iomap_ops exfat_iomap_ops = {
 	.iomap_begin = exfat_iomap_begin,
 };
 
+static int exfat_swap_iomap_begin(struct inode *inode, loff_t offset,
+		loff_t length, unsigned int flags, struct iomap *iomap,
+		struct iomap *srcmap)
+{
+	/*
+	 * Swap activation needs the physical mappings of preallocated
+	 * ranges. Do not report the VDL tail as a hole.
+	 */
+	return __exfat_iomap_begin(inode, offset, length,
+			flags & ~IOMAP_REPORT, iomap, false);
+}
+
+static const struct iomap_ops exfat_swap_iomap_ops = {
+	.iomap_begin = exfat_swap_iomap_begin,
+};
+
 /*
  * exfat_write_iomap_end - Update the state after write
  *
@@ -274,5 +290,6 @@ const struct iomap_read_ops exfat_iomap_bio_read_ops = {
 int exfat_iomap_swap_activate(struct swap_info_struct *sis,
 			       struct file *file, sector_t *span)
 {
-	return iomap_swapfile_activate(sis, file, span, &exfat_iomap_ops);
+	return iomap_swapfile_activate(sis, file, span,
+				       &exfat_swap_iomap_ops);
 }