[PATCH] ext4: call fallocate may cause process to hang when using fast commit

Xin Yin posted 1 patch 4 years, 6 months ago
fs/ext4/extents.c | 4 ++++
1 file changed, 4 insertions(+)
[PATCH] ext4: call fallocate may cause process to hang when using fast commit
Posted by Xin Yin 4 years, 6 months ago
If open a file with O_SYNC, and call fallocate with mode=0.when using
fast commit, will cause the process to hang.

During the fast_commit procedure, it will wait for inode update done.
call ext4_fc_stop_update() before ext4_fc_commit() to mark inode
complete update.

Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
---
 fs/ext4/extents.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 4108896d471b..92db33887b6c 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -4707,8 +4707,12 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 		goto out;
 
 	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
+		ext4_fc_stop_update(inode);
 		ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
 					EXT4_I(inode)->i_sync_tid);
+		inode_unlock(inode);
+		trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
+		return ret;
 	}
 out:
 	inode_unlock(inode);
-- 
2.20.1

Re: [PATCH] ext4: call fallocate may cause process to hang when using fast commit
Posted by harshad shirwadkar 4 years, 6 months ago
Thanks for the patch Xin, it looks good to me. I think there are a few
other places where we are not stopping the transaction before calling
commit. I'm trying to find them out.

Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>

On Thu, Dec 16, 2021 at 12:04 AM Xin Yin <yinxin.x@bytedance.com> wrote:
>
> If open a file with O_SYNC, and call fallocate with mode=0.when using
> fast commit, will cause the process to hang.
>
> During the fast_commit procedure, it will wait for inode update done.
> call ext4_fc_stop_update() before ext4_fc_commit() to mark inode
> complete update.
>
> Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
> ---
>  fs/ext4/extents.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> index 4108896d471b..92db33887b6c 100644
> --- a/fs/ext4/extents.c
> +++ b/fs/ext4/extents.c
> @@ -4707,8 +4707,12 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
>                 goto out;
>
>         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
> +               ext4_fc_stop_update(inode);
>                 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
>                                         EXT4_I(inode)->i_sync_tid);
> +               inode_unlock(inode);
> +               trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
> +               return ret;
>         }
>  out:
>         inode_unlock(inode);
> --
> 2.20.1
>
Re: [PATCH] ext4: call fallocate may cause process to hang when using fast commit
Posted by harshad shirwadkar 4 years, 6 months ago
Once the patch series "ext4 fast commit API cleanup" gets merged in
(https://patchwork.ozlabs.org/project/linux-ext4/list/?series=277672),
this patch won't be required anymore. So, I'll undo my review in favor
of merging the API cleanup patch series first.

- Harshad

On Fri, Dec 17, 2021 at 12:17 PM harshad shirwadkar
<harshadshirwadkar@gmail.com> wrote:
>
> Thanks for the patch Xin, it looks good to me. I think there are a few
> other places where we are not stopping the transaction before calling
> commit. I'm trying to find them out.
>
> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
>
> On Thu, Dec 16, 2021 at 12:04 AM Xin Yin <yinxin.x@bytedance.com> wrote:
> >
> > If open a file with O_SYNC, and call fallocate with mode=0.when using
> > fast commit, will cause the process to hang.
> >
> > During the fast_commit procedure, it will wait for inode update done.
> > call ext4_fc_stop_update() before ext4_fc_commit() to mark inode
> > complete update.
> >
> > Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
> > ---
> >  fs/ext4/extents.c | 4 ++++
> >  1 file changed, 4 insertions(+)
> >
> > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> > index 4108896d471b..92db33887b6c 100644
> > --- a/fs/ext4/extents.c
> > +++ b/fs/ext4/extents.c
> > @@ -4707,8 +4707,12 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
> >                 goto out;
> >
> >         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
> > +               ext4_fc_stop_update(inode);
> >                 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
> >                                         EXT4_I(inode)->i_sync_tid);
> > +               inode_unlock(inode);
> > +               trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
> > +               return ret;
> >         }
> >  out:
> >         inode_unlock(inode);
> > --
> > 2.20.1
> >
Re: [External] Re: [PATCH] ext4: call fallocate may cause process to hang when using fast commit
Posted by Xin Yin 4 years, 6 months ago
After applying Harshad's patch series "ext4 fast commit API cleanup" ,
this issue has been fixed.
So , please ignore this patch.

Thank,
Xin Yin

>On Thu, Dec 23, 2021 at 10:44 AM harshad shirwadkar <harshadshirwadkar@gmail.com> wrote:
>
> Once the patch series "ext4 fast commit API cleanup" gets merged in
> (https://patchwork.ozlabs.org/project/linux-ext4/list/?series=277672),
> this patch won't be required anymore. So, I'll undo my review in favor
> of merging the API cleanup patch series first.
>
> - Harshad
>
> On Fri, Dec 17, 2021 at 12:17 PM harshad shirwadkar
> <harshadshirwadkar@gmail.com> wrote:
> >
> > Thanks for the patch Xin, it looks good to me. I think there are a few
> > other places where we are not stopping the transaction before calling
> > commit. I'm trying to find them out.
> >
> > Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
> >
> > On Thu, Dec 16, 2021 at 12:04 AM Xin Yin <yinxin.x@bytedance.com> wrote:
> > >
> > > If open a file with O_SYNC, and call fallocate with mode=0.when using
> > > fast commit, will cause the process to hang.
> > >
> > > During the fast_commit procedure, it will wait for inode update done.
> > > call ext4_fc_stop_update() before ext4_fc_commit() to mark inode
> > > complete update.
> > >
> > > Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
> > > ---
> > >  fs/ext4/extents.c | 4 ++++
> > >  1 file changed, 4 insertions(+)
> > >
> > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> > > index 4108896d471b..92db33887b6c 100644
> > > --- a/fs/ext4/extents.c
> > > +++ b/fs/ext4/extents.c
> > > @@ -4707,8 +4707,12 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
> > >                 goto out;
> > >
> > >         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
> > > +               ext4_fc_stop_update(inode);
> > >                 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
> > >                                         EXT4_I(inode)->i_sync_tid);
> > > +               inode_unlock(inode);
> > > +               trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
> > > +               return ret;
> > >         }
> > >  out:
> > >         inode_unlock(inode);
> > > --
> > > 2.20.1
> > >