1. Patchew
  2. linux
  3. View series

[PATCH] f2fs: avoid setting SBI_NEED_FSCK on transient resize failure with -EAGAIN

Daeho Jeong posted 1 patch 1 month, 1 week ago 
fs/f2fs/gc.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
[PATCH] f2fs: avoid setting SBI_NEED_FSCK on transient resize failure with -EAGAIN
Posted by Daeho Jeong 1 month, 1 week ago 
From: Daeho Jeong <daehojeong@google.com>

When f2fs_resize_fs() fails due to transient lock contention or retryable
GC failure in free_segment_range() returning -EAGAIN, no filesystem
metadata has been modified on-disk yet. The filesystem remains completely
consistent and clean.

However, the current error recovery path unconditionally sets the
SBI_NEED_FSCK flag on any error, forcing an unnecessary and time-consuming
fsck.f2fs repair on the subsequent mount/reboot.

Fix this by guarding set_sbi_flag(sbi, SBI_NEED_FSCK) with
`if (err != -EAGAIN)`, avoiding false-positive filesystem corruption
flags on transient resize retries.

Signed-off-by: Daeho Jeong <daehojeong@google.com>
Signed-off-by: Sunmin Jeong <s_min.jeong@samsung.com>
---
 fs/f2fs/gc.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index 192b16ac02f8..787133ee2eb2 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -2446,8 +2446,10 @@ int f2fs_resize_fs(struct file *filp, __u64 block_count)
 recover_out:
 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
 	if (err) {
-		set_sbi_flag(sbi, SBI_NEED_FSCK);
-		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
+		if (err != -EAGAIN) {
+			set_sbi_flag(sbi, SBI_NEED_FSCK);
+			f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
+		}
 
 		spin_lock(&sbi->stat_lock);
 		sbi->user_block_count += shrunk_blocks;
-- 
2.55.0.691.gc56d675ccc-goog
Re: [f2fs-dev] [PATCH] f2fs: avoid setting SBI_NEED_FSCK on transient resize failure with -EAGAIN
Posted by Chao Yu 1 month, 1 week ago 
On 8/19/26 01:05, Daeho Jeong wrote:
> From: Daeho Jeong <daehojeong@google.com>
> 
> When f2fs_resize_fs() fails due to transient lock contention or retryable
> GC failure in free_segment_range() returning -EAGAIN, no filesystem
> metadata has been modified on-disk yet. The filesystem remains completely
> consistent and clean.

Any way to make sure EAGAIN is from free_segment_range(), in case we return
EAGAIN from 1) any other places that we may miss to check now or 2) we changed
the code to return EAGAIN in future.

Thanks,

> 
> However, the current error recovery path unconditionally sets the
> SBI_NEED_FSCK flag on any error, forcing an unnecessary and time-consuming
> fsck.f2fs repair on the subsequent mount/reboot.
> 
> Fix this by guarding set_sbi_flag(sbi, SBI_NEED_FSCK) with
> `if (err != -EAGAIN)`, avoiding false-positive filesystem corruption
> flags on transient resize retries.
> 
> Signed-off-by: Daeho Jeong <daehojeong@google.com>
> Signed-off-by: Sunmin Jeong <s_min.jeong@samsung.com>
> ---
>  fs/f2fs/gc.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> index 192b16ac02f8..787133ee2eb2 100644
> --- a/fs/f2fs/gc.c
> +++ b/fs/f2fs/gc.c
> @@ -2446,8 +2446,10 @@ int f2fs_resize_fs(struct file *filp, __u64 block_count)
>  recover_out:
>  	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
>  	if (err) {
> -		set_sbi_flag(sbi, SBI_NEED_FSCK);
> -		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
> +		if (err != -EAGAIN) {
> +			set_sbi_flag(sbi, SBI_NEED_FSCK);
> +			f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
> +		}
>  
>  		spin_lock(&sbi->stat_lock);
>  		sbi->user_block_count += shrunk_blocks;
Re: [f2fs-dev] [PATCH] f2fs: avoid setting SBI_NEED_FSCK on transient resize failure with -EAGAIN
Posted by Daeho Jeong 1 month, 1 week ago 
On Tue, Aug 18, 2026 at 7:34 PM Chao Yu <chao@kernel.org> wrote:
>
> On 8/19/26 01:05, Daeho Jeong wrote:
> > From: Daeho Jeong <daehojeong@google.com>
> >
> > When f2fs_resize_fs() fails due to transient lock contention or retryable
> > GC failure in free_segment_range() returning -EAGAIN, no filesystem
> > metadata has been modified on-disk yet. The filesystem remains completely
> > consistent and clean.
>
> Any way to make sure EAGAIN is from free_segment_range(), in case we return
> EAGAIN from 1) any other places that we may miss to check now or 2) we changed
> the code to return EAGAIN in future.

Makes sense.

In fact, if free_segment_range() fails with ANY error (e.g., -EAGAIN,
-ENOMEM, -ERESTARTSYS, etc.), no on-disk superblock or filesystem metadata
has been modified yet, and free_segment_range() safely restores all in-memory
counters (MAIN_SECS, free_sections) before returning. Therefore, setting
SBI_NEED_FSCK is not needed for all errors occurring in free_segment_range().

To make this explicit and defensive against any future changes, we can
separate the error recovery path with dedicated labels as follows:

  err = free_segment_range(sbi, secs, false);
  if (err)
- goto recover_out;
+ goto recover_user_blocks;

  update_sb_metadata(sbi, -secs);

...

 recover_out:
+ set_sbi_flag(sbi, SBI_NEED_FSCK);
+ f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
+recover_user_blocks:
  clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
  if (err) {
- set_sbi_flag(sbi, SBI_NEED_FSCK);
- f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
-
  spin_lock(&sbi->stat_lock);
  sbi->user_block_count += shrunk_blocks;
  spin_unlock(&sbi->stat_lock);

Thanks,

>
> Thanks,
>
> >
> > However, the current error recovery path unconditionally sets the
> > SBI_NEED_FSCK flag on any error, forcing an unnecessary and time-consuming
> > fsck.f2fs repair on the subsequent mount/reboot.
> >
> > Fix this by guarding set_sbi_flag(sbi, SBI_NEED_FSCK) with
> > `if (err != -EAGAIN)`, avoiding false-positive filesystem corruption
> > flags on transient resize retries.
> >
> > Signed-off-by: Daeho Jeong <daehojeong@google.com>
> > Signed-off-by: Sunmin Jeong <s_min.jeong@samsung.com>
> > ---
> >  fs/f2fs/gc.c | 6 ++++--
> >  1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> > index 192b16ac02f8..787133ee2eb2 100644
> > --- a/fs/f2fs/gc.c
> > +++ b/fs/f2fs/gc.c
> > @@ -2446,8 +2446,10 @@ int f2fs_resize_fs(struct file *filp, __u64 block_count)
> >  recover_out:
> >       clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
> >       if (err) {
> > -             set_sbi_flag(sbi, SBI_NEED_FSCK);
> > -             f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
> > +             if (err != -EAGAIN) {
> > +                     set_sbi_flag(sbi, SBI_NEED_FSCK);
> > +                     f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
> > +             }
> >
> >               spin_lock(&sbi->stat_lock);
> >               sbi->user_block_count += shrunk_blocks;
>
Re: [f2fs-dev] [PATCH] f2fs: avoid setting SBI_NEED_FSCK on transient resize failure with -EAGAIN
Posted by Chao Yu 1 month, 1 week ago 
On 8/20/26 02:21, Daeho Jeong wrote:
> On Tue, Aug 18, 2026 at 7:34 PM Chao Yu <chao@kernel.org> wrote:
>>
>> On 8/19/26 01:05, Daeho Jeong wrote:
>>> From: Daeho Jeong <daehojeong@google.com>
>>>
>>> When f2fs_resize_fs() fails due to transient lock contention or retryable
>>> GC failure in free_segment_range() returning -EAGAIN, no filesystem
>>> metadata has been modified on-disk yet. The filesystem remains completely
>>> consistent and clean.
>>
>> Any way to make sure EAGAIN is from free_segment_range(), in case we return
>> EAGAIN from 1) any other places that we may miss to check now or 2) we changed
>> the code to return EAGAIN in future.
> 
> Makes sense.
> 
> In fact, if free_segment_range() fails with ANY error (e.g., -EAGAIN,
> -ENOMEM, -ERESTARTSYS, etc.), no on-disk superblock or filesystem metadata
> has been modified yet, and free_segment_range() safely restores all in-memory
> counters (MAIN_SECS, free_sections) before returning. Therefore, setting
> SBI_NEED_FSCK is not needed for all errors occurring in free_segment_range().

Yeah, better.

> 
> To make this explicit and defensive against any future changes, we can
> separate the error recovery path with dedicated labels as follows:
> 
>   err = free_segment_range(sbi, secs, false);
>   if (err)
> - goto recover_out;
> + goto recover_user_blocks;
> 
>   update_sb_metadata(sbi, -secs);
> 
> ...
> 
>  recover_out:
> + set_sbi_flag(sbi, SBI_NEED_FSCK);
> + f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
> +recover_user_blocks:

How about this?

recover_out:
	if (err) {
		f2fs_bug_on(sbi, err == -EAGAIN);
		set_sbi_flag(sbi, SBI_NEED_FSCK);
		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
	}

recover_user_blocks:

Thanks,

>   clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
>   if (err) {
> - set_sbi_flag(sbi, SBI_NEED_FSCK);
> - f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
> -
>   spin_lock(&sbi->stat_lock);
>   sbi->user_block_count += shrunk_blocks;
>   spin_unlock(&sbi->stat_lock);
> 
> Thanks,
> 
>>
>> Thanks,
>>
>>>
>>> However, the current error recovery path unconditionally sets the
>>> SBI_NEED_FSCK flag on any error, forcing an unnecessary and time-consuming
>>> fsck.f2fs repair on the subsequent mount/reboot.
>>>
>>> Fix this by guarding set_sbi_flag(sbi, SBI_NEED_FSCK) with
>>> `if (err != -EAGAIN)`, avoiding false-positive filesystem corruption
>>> flags on transient resize retries.
>>>
>>> Signed-off-by: Daeho Jeong <daehojeong@google.com>
>>> Signed-off-by: Sunmin Jeong <s_min.jeong@samsung.com>
>>> ---
>>>  fs/f2fs/gc.c | 6 ++++--
>>>  1 file changed, 4 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>> index 192b16ac02f8..787133ee2eb2 100644
>>> --- a/fs/f2fs/gc.c
>>> +++ b/fs/f2fs/gc.c
>>> @@ -2446,8 +2446,10 @@ int f2fs_resize_fs(struct file *filp, __u64 block_count)
>>>  recover_out:
>>>       clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
>>>       if (err) {
>>> -             set_sbi_flag(sbi, SBI_NEED_FSCK);
>>> -             f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
>>> +             if (err != -EAGAIN) {
>>> +                     set_sbi_flag(sbi, SBI_NEED_FSCK);
>>> +                     f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
>>> +             }
>>>
>>>               spin_lock(&sbi->stat_lock);
>>>               sbi->user_block_count += shrunk_blocks;
>>

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.