[PATCH] ovl: skip stale entries in merge dir cache iteration

[PATCH] ovl: skip stale entries in merge dir cache iteration

[Amir Goldstein]

On the first getdents call, ovl_iterate() populates the readdir cache
with a list of entries, but for upper entries with origin lower inode,
p->ino remains zero.

Following getdents calls traverse the readdir cache list and call
ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
in the overlay and return d_ino that is consistent with st_ino.

If the upper file was unlinked between the first getdents call and the
getdents call that lists the file entry, ovl_cache_update_ino() will not
find the entry and fall back to setting d_ino to the upper real st_ino,
which is inconsistent with how this object was presented to users.

Instead of listing a stale entry with inconsistent d_ino, simply skip
the stale entry, which is better for users.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 fs/overlayfs/readdir.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/fs/overlayfs/readdir.c b/fs/overlayfs/readdir.c
index cc1e80257064..10b7780e4bdc 100644
--- a/fs/overlayfs/readdir.c
+++ b/fs/overlayfs/readdir.c
@@ -481,6 +481,8 @@ static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
 	}
 	this = lookup_one_len(p->name, dir, p->len);
 	if (IS_ERR_OR_NULL(this) || !this->d_inode) {
+		/* Mark a stale entry */
+		p->is_whiteout = true;
 		if (IS_ERR(this)) {
 			err = PTR_ERR(this);
 			this = NULL;
@@ -776,6 +778,9 @@ static int ovl_iterate(struct file *file, struct dir_context *ctx)
 				if (err)
 					goto out;
 			}
+		}
+		/* ovl_cache_update_ino() sets is_whiteout on stale entry */
+		if (!p->is_whiteout) {
 			if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 				break;
 		}
-- 
2.25.1

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Amir Goldstein]

On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
>
> On the first getdents call, ovl_iterate() populates the readdir cache
> with a list of entries, but for upper entries with origin lower inode,
> p->ino remains zero.
>
> Following getdents calls traverse the readdir cache list and call
> ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> in the overlay and return d_ino that is consistent with st_ino.
>
> If the upper file was unlinked between the first getdents call and the
> getdents call that lists the file entry, ovl_cache_update_ino() will not
> find the entry and fall back to setting d_ino to the upper real st_ino,
> which is inconsistent with how this object was presented to users.
>
> Instead of listing a stale entry with inconsistent d_ino, simply skip
> the stale entry, which is better for users.
>

Miklos,

I forgot to follow up on this patch.
Upstream xfstest overlay/077 is failing without this patch.
Please add:
Link: https://lore.kernel.org/fstests/CAOQ4uxgR_cLnC_vdU5=seP3fwqVkuZM_-WfD6maFTMbMYq=a9w@mail.gmail.com/

Thanks,
Amir.


> Signed-off-by: Amir Goldstein <amir73il@gmail.com>
> ---
>  fs/overlayfs/readdir.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/fs/overlayfs/readdir.c b/fs/overlayfs/readdir.c
> index cc1e80257064..10b7780e4bdc 100644
> --- a/fs/overlayfs/readdir.c
> +++ b/fs/overlayfs/readdir.c
> @@ -481,6 +481,8 @@ static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
>         }
>         this = lookup_one_len(p->name, dir, p->len);
>         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
> +               /* Mark a stale entry */
> +               p->is_whiteout = true;
>                 if (IS_ERR(this)) {
>                         err = PTR_ERR(this);
>                         this = NULL;
> @@ -776,6 +778,9 @@ static int ovl_iterate(struct file *file, struct dir_context *ctx)
>                                 if (err)
>                                         goto out;
>                         }
> +               }
> +               /* ovl_cache_update_ino() sets is_whiteout on stale entry */
> +               if (!p->is_whiteout) {
>                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
>                                 break;
>                 }
> --
> 2.25.1
>

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Miklos Szeredi]

On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On the first getdents call, ovl_iterate() populates the readdir cache
> > with a list of entries, but for upper entries with origin lower inode,
> > p->ino remains zero.
> >
> > Following getdents calls traverse the readdir cache list and call
> > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > in the overlay and return d_ino that is consistent with st_ino.
> >
> > If the upper file was unlinked between the first getdents call and the
> > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > find the entry and fall back to setting d_ino to the upper real st_ino,
> > which is inconsistent with how this object was presented to users.
> >
> > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > the stale entry, which is better for users.
> >
>
> Miklos,
>
> I forgot to follow up on this patch.
> Upstream xfstest overlay/077 is failing without this patch.

Can't reproduce (on ext4/xfs and "-oxino=on").

Is there some trick?

Thanks,
Miklos

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Amir Goldstein]

On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > >
> > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > with a list of entries, but for upper entries with origin lower inode,
> > > p->ino remains zero.
> > >
> > > Following getdents calls traverse the readdir cache list and call
> > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > in the overlay and return d_ino that is consistent with st_ino.
> > >
> > > If the upper file was unlinked between the first getdents call and the
> > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > which is inconsistent with how this object was presented to users.
> > >
> > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > the stale entry, which is better for users.
> > >
> >
> > Miklos,
> >
> > I forgot to follow up on this patch.
> > Upstream xfstest overlay/077 is failing without this patch.
>
> Can't reproduce (on ext4/xfs and "-oxino=on").
>
> Is there some trick?

Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.

     QA output created by 077
    +entry m100 has inconsistent d_ino (234 != 232)
    +entry f100 has inconsistent d_ino (335 != 16777542)
     Silence is golden

Maybe you need to build src/t_dir_offset2?

Thanks,
Amir.

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Miklos Szeredi]

On Mon, 19 Jul 2021 at 18:43, Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> >
> > On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > >
> > > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > >
> > > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > > with a list of entries, but for upper entries with origin lower inode,
> > > > p->ino remains zero.
> > > >
> > > > Following getdents calls traverse the readdir cache list and call
> > > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > > in the overlay and return d_ino that is consistent with st_ino.
> > > >
> > > > If the upper file was unlinked between the first getdents call and the
> > > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > > which is inconsistent with how this object was presented to users.
> > > >
> > > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > > the stale entry, which is better for users.
> > > >
> > >
> > > Miklos,
> > >
> > > I forgot to follow up on this patch.
> > > Upstream xfstest overlay/077 is failing without this patch.
> >
> > Can't reproduce (on ext4/xfs and "-oxino=on").
> >
> > Is there some trick?
>
> Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.
>
>      QA output created by 077
>     +entry m100 has inconsistent d_ino (234 != 232)
>     +entry f100 has inconsistent d_ino (335 != 16777542)
>      Silence is golden
>
> Maybe you need to build src/t_dir_offset2?

root@kvm:/opt/xfstests-dev# git log -1 --pretty=%h
10f6b231
root@kvm:/opt/xfstests-dev# cat local.config
export TEST_DEV=/dev/vdb1
export TEST_DIR=/test
export SCRATCH_DEV=/dev/vdb2
export SCRATCH_MNT=/scratch
export FSTYP=ext4
export OVERLAY_MOUNT_OPTIONS="-o xino=on"
root@kvm:/opt/xfstests-dev# make src/t_dir_offset2
make: 'src/t_dir_offset2' is up to date.
root@kvm:/opt/xfstests-dev# ./check -overlay overlay/077
FSTYP         -- overlay
PLATFORM      -- Linux/x86_64 kvm 5.14.0-rc2 #276 SMP Tue Jul 20
05:54:44 CEST 2021
MKFS_OPTIONS  -- /scratch
MOUNT_OPTIONS -- -o xino=on /scratch /scratch/ovl-mnt

overlay/077 1s ...  1s
Ran: overlay/077
Passed all 1 tests

root@kvm:/opt/xfstests-dev# cat results/overlay/077.full

Create file in pure upper dir:
getdents at offset 0 returned 192 bytes
created entry p0
entry p0 found as expected

Remove file in pure upper dir:
getdents at offset 0 returned 192 bytes
unlinked entry p100
entry p100 not found as expected

Create file in impure upper dir:
getdents at offset 0 returned 192 bytes
created entry o0
entry o0 found as expected

Remove file in impure upper dir:
getdents at offset 0 returned 192 bytes
unlinked entry o100
entry o100 not found as expected

Create file in merge dir:
getdents at offset 0 returned 192 bytes
created entry m0
entry m0 found as expected

Remove file in merge dir:
getdents at offset 0 returned 192 bytes
unlinked entry m100
entry m100 not found as expected

Create file in former merge dir:
getdents at offset 0 returned 192 bytes
created entry f0
entry f0 found as expected

Remove file in former merge dir:
getdents at offset 0 returned 192 bytes
unlinked entry f100
entry f100 not found as expected

Ideas for further debugging why this test isn't failing for v4.12-rc2?

Thanks,
Miklos

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Miklos Szeredi]

On Tue, 20 Jul 2021 at 06:19, Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Mon, 19 Jul 2021 at 18:43, Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > >
> > > On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > >
> > > > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > >
> > > > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > > > with a list of entries, but for upper entries with origin lower inode,
> > > > > p->ino remains zero.
> > > > >
> > > > > Following getdents calls traverse the readdir cache list and call
> > > > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > > > in the overlay and return d_ino that is consistent with st_ino.
> > > > >
> > > > > If the upper file was unlinked between the first getdents call and the
> > > > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > > > which is inconsistent with how this object was presented to users.
> > > > >
> > > > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > > > the stale entry, which is better for users.
> > > > >
> > > >
> > > > Miklos,
> > > >
> > > > I forgot to follow up on this patch.
> > > > Upstream xfstest overlay/077 is failing without this patch.
> > >
> > > Can't reproduce (on ext4/xfs and "-oxino=on").
> > >
> > > Is there some trick?
> >
> > Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.
> >
> >      QA output created by 077
> >     +entry m100 has inconsistent d_ino (234 != 232)
> >     +entry f100 has inconsistent d_ino (335 != 16777542)
> >      Silence is golden
> >
> > Maybe you need to build src/t_dir_offset2?
>
> root@kvm:/opt/xfstests-dev# git log -1 --pretty=%h
> 10f6b231
> root@kvm:/opt/xfstests-dev# cat local.config
> export TEST_DEV=/dev/vdb1
> export TEST_DIR=/test
> export SCRATCH_DEV=/dev/vdb2
> export SCRATCH_MNT=/scratch
> export FSTYP=ext4
> export OVERLAY_MOUNT_OPTIONS="-o xino=on"
> root@kvm:/opt/xfstests-dev# make src/t_dir_offset2
> make: 'src/t_dir_offset2' is up to date.
> root@kvm:/opt/xfstests-dev# ./check -overlay overlay/077
> FSTYP         -- overlay
> PLATFORM      -- Linux/x86_64 kvm 5.14.0-rc2 #276 SMP Tue Jul 20
> 05:54:44 CEST 2021
> MKFS_OPTIONS  -- /scratch
> MOUNT_OPTIONS -- -o xino=on /scratch /scratch/ovl-mnt
>
> overlay/077 1s ...  1s
> Ran: overlay/077
> Passed all 1 tests
>
> root@kvm:/opt/xfstests-dev# cat results/overlay/077.full
>
> Create file in pure upper dir:
> getdents at offset 0 returned 192 bytes
> created entry p0
> entry p0 found as expected
>
> Remove file in pure upper dir:
> getdents at offset 0 returned 192 bytes
> unlinked entry p100
> entry p100 not found as expected
>
> Create file in impure upper dir:
> getdents at offset 0 returned 192 bytes
> created entry o0
> entry o0 found as expected
>
> Remove file in impure upper dir:
> getdents at offset 0 returned 192 bytes
> unlinked entry o100
> entry o100 not found as expected
>
> Create file in merge dir:
> getdents at offset 0 returned 192 bytes
> created entry m0
> entry m0 found as expected
>
> Remove file in merge dir:
> getdents at offset 0 returned 192 bytes
> unlinked entry m100
> entry m100 not found as expected
>
> Create file in former merge dir:
> getdents at offset 0 returned 192 bytes
> created entry f0
> entry f0 found as expected
>
> Remove file in former merge dir:
> getdents at offset 0 returned 192 bytes
> unlinked entry f100
> entry f100 not found as expected
>
> Ideas for further debugging why this test isn't failing for v4.12-rc2?

v5.14-rc2, that is (got the -rc2 part right the first time, though).

Thanks,
Miklos

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Amir Goldstein]

On Tue, Jul 20, 2021 at 7:22 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Tue, 20 Jul 2021 at 06:19, Miklos Szeredi <miklos@szeredi.hu> wrote:
> >
> > On Mon, 19 Jul 2021 at 18:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > >
> > > On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > >
> > > > On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > >
> > > > > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > >
> > > > > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > > > > with a list of entries, but for upper entries with origin lower inode,
> > > > > > p->ino remains zero.
> > > > > >
> > > > > > Following getdents calls traverse the readdir cache list and call
> > > > > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > > > > in the overlay and return d_ino that is consistent with st_ino.
> > > > > >
> > > > > > If the upper file was unlinked between the first getdents call and the
> > > > > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > > > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > > > > which is inconsistent with how this object was presented to users.
> > > > > >
> > > > > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > > > > the stale entry, which is better for users.
> > > > > >
> > > > >
> > > > > Miklos,
> > > > >
> > > > > I forgot to follow up on this patch.
> > > > > Upstream xfstest overlay/077 is failing without this patch.
> > > >
> > > > Can't reproduce (on ext4/xfs and "-oxino=on").
> > > >
> > > > Is there some trick?
> > >
> > > Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.
> > >
> > >      QA output created by 077
> > >     +entry m100 has inconsistent d_ino (234 != 232)
> > >     +entry f100 has inconsistent d_ino (335 != 16777542)
> > >      Silence is golden
> > >
> > > Maybe you need to build src/t_dir_offset2?
> >
> > root@kvm:/opt/xfstests-dev# git log -1 --pretty=%h
> > 10f6b231
> > root@kvm:/opt/xfstests-dev# cat local.config
> > export TEST_DEV=/dev/vdb1
> > export TEST_DIR=/test
> > export SCRATCH_DEV=/dev/vdb2
> > export SCRATCH_MNT=/scratch
> > export FSTYP=ext4
> > export OVERLAY_MOUNT_OPTIONS="-o xino=on"
> > root@kvm:/opt/xfstests-dev# make src/t_dir_offset2
> > make: 'src/t_dir_offset2' is up to date.
> > root@kvm:/opt/xfstests-dev# ./check -overlay overlay/077
> > FSTYP         -- overlay
> > PLATFORM      -- Linux/x86_64 kvm 5.14.0-rc2 #276 SMP Tue Jul 20
> > 05:54:44 CEST 2021
> > MKFS_OPTIONS  -- /scratch
> > MOUNT_OPTIONS -- -o xino=on /scratch /scratch/ovl-mnt
> >
> > overlay/077 1s ...  1s
> > Ran: overlay/077
> > Passed all 1 tests
> >
> > root@kvm:/opt/xfstests-dev# cat results/overlay/077.full
> >
> > Create file in pure upper dir:
> > getdents at offset 0 returned 192 bytes
> > created entry p0
> > entry p0 found as expected
> >
> > Remove file in pure upper dir:
> > getdents at offset 0 returned 192 bytes
> > unlinked entry p100
> > entry p100 not found as expected
> >
> > Create file in impure upper dir:
> > getdents at offset 0 returned 192 bytes
> > created entry o0
> > entry o0 found as expected
> >
> > Remove file in impure upper dir:
> > getdents at offset 0 returned 192 bytes
> > unlinked entry o100
> > entry o100 not found as expected
> >
> > Create file in merge dir:
> > getdents at offset 0 returned 192 bytes
> > created entry m0
> > entry m0 found as expected
> >
> > Remove file in merge dir:
> > getdents at offset 0 returned 192 bytes
> > unlinked entry m100
> > entry m100 not found as expected
> >
> > Create file in former merge dir:
> > getdents at offset 0 returned 192 bytes
> > created entry f0
> > entry f0 found as expected
> >
> > Remove file in former merge dir:
> > getdents at offset 0 returned 192 bytes
> > unlinked entry f100
> > entry f100 not found as expected
> >
> > Ideas for further debugging why this test isn't failing for v4.12-rc2?
>

It's not you, it's me ;-)

The failure was lost during cleanup of t_dir_offset2 patches
for submission and it is I who was running an older version
of t_dir_offset2. Let me figure this out and get back to you
with a working test.

Thanks,
Amir.

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Amir Goldstein]

[-- Attachment #1: Type: text/plain, Size: 4745 bytes --]

On Tue, Jul 20, 2021 at 10:18 AM Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Tue, Jul 20, 2021 at 7:22 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
> >
> > On Tue, 20 Jul 2021 at 06:19, Miklos Szeredi <miklos@szeredi.hu> wrote:
> > >
> > > On Mon, 19 Jul 2021 at 18:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > >
> > > > On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > > >
> > > > > On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > >
> > > > > > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > >
> > > > > > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > > > > > with a list of entries, but for upper entries with origin lower inode,
> > > > > > > p->ino remains zero.
> > > > > > >
> > > > > > > Following getdents calls traverse the readdir cache list and call
> > > > > > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > > > > > in the overlay and return d_ino that is consistent with st_ino.
> > > > > > >
> > > > > > > If the upper file was unlinked between the first getdents call and the
> > > > > > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > > > > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > > > > > which is inconsistent with how this object was presented to users.
> > > > > > >
> > > > > > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > > > > > the stale entry, which is better for users.
> > > > > > >
> > > > > >
> > > > > > Miklos,
> > > > > >
> > > > > > I forgot to follow up on this patch.
> > > > > > Upstream xfstest overlay/077 is failing without this patch.
> > > > >
> > > > > Can't reproduce (on ext4/xfs and "-oxino=on").
> > > > >
> > > > > Is there some trick?
> > > >
> > > > Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.
> > > >
> > > >      QA output created by 077
> > > >     +entry m100 has inconsistent d_ino (234 != 232)
> > > >     +entry f100 has inconsistent d_ino (335 != 16777542)
> > > >      Silence is golden
> > > >
> > > > Maybe you need to build src/t_dir_offset2?
> > >
> > > root@kvm:/opt/xfstests-dev# git log -1 --pretty=%h
> > > 10f6b231
> > > root@kvm:/opt/xfstests-dev# cat local.config
> > > export TEST_DEV=/dev/vdb1
> > > export TEST_DIR=/test
> > > export SCRATCH_DEV=/dev/vdb2
> > > export SCRATCH_MNT=/scratch
> > > export FSTYP=ext4
> > > export OVERLAY_MOUNT_OPTIONS="-o xino=on"
> > > root@kvm:/opt/xfstests-dev# make src/t_dir_offset2
> > > make: 'src/t_dir_offset2' is up to date.
> > > root@kvm:/opt/xfstests-dev# ./check -overlay overlay/077
> > > FSTYP         -- overlay
> > > PLATFORM      -- Linux/x86_64 kvm 5.14.0-rc2 #276 SMP Tue Jul 20
> > > 05:54:44 CEST 2021
> > > MKFS_OPTIONS  -- /scratch
> > > MOUNT_OPTIONS -- -o xino=on /scratch /scratch/ovl-mnt
> > >
> > > overlay/077 1s ...  1s
> > > Ran: overlay/077
> > > Passed all 1 tests
> > >
> > > root@kvm:/opt/xfstests-dev# cat results/overlay/077.full
> > >
> > > Create file in pure upper dir:
> > > getdents at offset 0 returned 192 bytes
> > > created entry p0
> > > entry p0 found as expected
> > >
> > > Remove file in pure upper dir:
> > > getdents at offset 0 returned 192 bytes
> > > unlinked entry p100
> > > entry p100 not found as expected
> > >
> > > Create file in impure upper dir:
> > > getdents at offset 0 returned 192 bytes
> > > created entry o0
> > > entry o0 found as expected
> > >
> > > Remove file in impure upper dir:
> > > getdents at offset 0 returned 192 bytes
> > > unlinked entry o100
> > > entry o100 not found as expected
> > >
> > > Create file in merge dir:
> > > getdents at offset 0 returned 192 bytes
> > > created entry m0
> > > entry m0 found as expected
> > >
> > > Remove file in merge dir:
> > > getdents at offset 0 returned 192 bytes
> > > unlinked entry m100
> > > entry m100 not found as expected
> > >
> > > Create file in former merge dir:
> > > getdents at offset 0 returned 192 bytes
> > > created entry f0
> > > entry f0 found as expected
> > >
> > > Remove file in former merge dir:
> > > getdents at offset 0 returned 192 bytes
> > > unlinked entry f100
> > > entry f100 not found as expected
> > >
> > > Ideas for further debugging why this test isn't failing for v4.12-rc2?
> >
>
> It's not you, it's me ;-)
>
> The failure was lost during cleanup of t_dir_offset2 patches
> for submission and it is I who was running an older version
> of t_dir_offset2. Let me figure this out and get back to you
> with a working test.
>

How about you try again with a version of the test that actually has
the check and the error print that I reported.... ;-)

Thanks,
Amir.

[-- Attachment #2: fstests-Check-for-inconsistent-d_ino-st_ino.patch --]
[-- Type: text/x-patch, Size: 2156 bytes --]

From 250615da6f01e99c055b7fc3bd9a2c1deac644e0 Mon Sep 17 00:00:00 2001
From: Amir Goldstein <amir73il@gmail.com>
Date: Tue, 20 Jul 2021 17:45:43 +0300
Subject: [PATCH] src/t_dir_offset2: Check for inconsistent d_ino/st_ino

After unlink of a directory entry, that entry may still apear in getdents
results of an already open directory fd, but it should return a d_ino
value that is consistent with the already observed st_ino of that entry.

This is a regression test for kernel commit ....
("ovl: skip stale entries in merge dir cache iteration")

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 src/t_dir_offset2.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/src/t_dir_offset2.c b/src/t_dir_offset2.c
index 75b41c1a..026bc8f3 100644
--- a/src/t_dir_offset2.c
+++ b/src/t_dir_offset2.c
@@ -44,6 +44,7 @@ int main(int argc, char *argv[])
 	char buf[BUF_SIZE];
 	int nread, bufsize = BUF_SIZE;
 	struct linux_dirent64 *d;
+	struct stat st = {};
 	int bpos, total, i;
 	off_t lret;
 	int retval = EXIT_SUCCESS;
@@ -81,9 +82,9 @@ int main(int argc, char *argv[])
 	}
 
 	if (filename) {
-		exists = !faccessat(fd, filename, F_OK, AT_SYMLINK_NOFOLLOW);
+		exists = !fstatat(fd, filename, &st, AT_SYMLINK_NOFOLLOW);
 		if (!exists && errno != ENOENT) {
-			perror("faccessat");
+			perror("fstatat");
 			exit(EXIT_FAILURE);
 		}
 	}
@@ -139,9 +140,6 @@ int main(int argc, char *argv[])
 			continue;
 		}
 
-		if (nread == 0)
-			break;
-
 		for (bpos = 0; bpos < nread; total++) {
 			d = (struct linux_dirent64 *) (buf + bpos);
 
@@ -165,8 +163,16 @@ int main(int argc, char *argv[])
 					printf("entry #%d: %s (d_ino=%lld, d_off=%lld)\n",
 					       i, d->d_name, (long long int)d->d_ino,
 					       (long long int)d->d_off);
-				if (!strcmp(filename, d->d_name))
+				if (!strcmp(filename, d->d_name)) {
 					found = 1;
+					if (st.st_ino && d->d_ino != st.st_ino) {
+						fprintf(stderr, "entry %s has inconsistent d_ino (%lld != %lld)\n",
+								filename,
+								(long long int)d->d_ino,
+								(long long int)st.st_ino);
+					}
+				}
+
 			}
 			bpos += d->d_reclen;
 		}
-- 
2.32.0

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Amir Goldstein]

[-- Attachment #1: Type: text/plain, Size: 5446 bytes --]

On Tue, Jul 20, 2021 at 5:55 PM Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Tue, Jul 20, 2021 at 10:18 AM Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Tue, Jul 20, 2021 at 7:22 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > >
> > > On Tue, 20 Jul 2021 at 06:19, Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > >
> > > > On Mon, 19 Jul 2021 at 18:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > >
> > > > > On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > > > >
> > > > > > On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > >
> > > > > > > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > > >
> > > > > > > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > > > > > > with a list of entries, but for upper entries with origin lower inode,
> > > > > > > > p->ino remains zero.
> > > > > > > >
> > > > > > > > Following getdents calls traverse the readdir cache list and call
> > > > > > > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > > > > > > in the overlay and return d_ino that is consistent with st_ino.
> > > > > > > >
> > > > > > > > If the upper file was unlinked between the first getdents call and the
> > > > > > > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > > > > > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > > > > > > which is inconsistent with how this object was presented to users.
> > > > > > > >
> > > > > > > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > > > > > > the stale entry, which is better for users.
> > > > > > > >
> > > > > > >
> > > > > > > Miklos,
> > > > > > >
> > > > > > > I forgot to follow up on this patch.
> > > > > > > Upstream xfstest overlay/077 is failing without this patch.
> > > > > >
> > > > > > Can't reproduce (on ext4/xfs and "-oxino=on").
> > > > > >
> > > > > > Is there some trick?
> > > > >
> > > > > Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.
> > > > >
> > > > >      QA output created by 077
> > > > >     +entry m100 has inconsistent d_ino (234 != 232)
> > > > >     +entry f100 has inconsistent d_ino (335 != 16777542)
> > > > >      Silence is golden
> > > > >
> > > > > Maybe you need to build src/t_dir_offset2?
> > > >
> > > > root@kvm:/opt/xfstests-dev# git log -1 --pretty=%h
> > > > 10f6b231
> > > > root@kvm:/opt/xfstests-dev# cat local.config
> > > > export TEST_DEV=/dev/vdb1
> > > > export TEST_DIR=/test
> > > > export SCRATCH_DEV=/dev/vdb2
> > > > export SCRATCH_MNT=/scratch
> > > > export FSTYP=ext4
> > > > export OVERLAY_MOUNT_OPTIONS="-o xino=on"
> > > > root@kvm:/opt/xfstests-dev# make src/t_dir_offset2
> > > > make: 'src/t_dir_offset2' is up to date.
> > > > root@kvm:/opt/xfstests-dev# ./check -overlay overlay/077
> > > > FSTYP         -- overlay
> > > > PLATFORM      -- Linux/x86_64 kvm 5.14.0-rc2 #276 SMP Tue Jul 20
> > > > 05:54:44 CEST 2021
> > > > MKFS_OPTIONS  -- /scratch
> > > > MOUNT_OPTIONS -- -o xino=on /scratch /scratch/ovl-mnt
> > > >
> > > > overlay/077 1s ...  1s
> > > > Ran: overlay/077
> > > > Passed all 1 tests
> > > >
> > > > root@kvm:/opt/xfstests-dev# cat results/overlay/077.full
> > > >
> > > > Create file in pure upper dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > created entry p0
> > > > entry p0 found as expected
> > > >
> > > > Remove file in pure upper dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > unlinked entry p100
> > > > entry p100 not found as expected
> > > >
> > > > Create file in impure upper dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > created entry o0
> > > > entry o0 found as expected
> > > >
> > > > Remove file in impure upper dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > unlinked entry o100
> > > > entry o100 not found as expected
> > > >
> > > > Create file in merge dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > created entry m0
> > > > entry m0 found as expected
> > > >
> > > > Remove file in merge dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > unlinked entry m100
> > > > entry m100 not found as expected
> > > >
> > > > Create file in former merge dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > created entry f0
> > > > entry f0 found as expected
> > > >
> > > > Remove file in former merge dir:
> > > > getdents at offset 0 returned 192 bytes
> > > > unlinked entry f100
> > > > entry f100 not found as expected
> > > >
> > > > Ideas for further debugging why this test isn't failing for v4.12-rc2?
> > >
> >
> > It's not you, it's me ;-)
> >
> > The failure was lost during cleanup of t_dir_offset2 patches
> > for submission and it is I who was running an older version
> > of t_dir_offset2. Let me figure this out and get back to you
> > with a working test.
> >
>
> How about you try again with a version of the test that actually has
> the check and the error print that I reported.... ;-)
>

P.S.1:
xino=on is futile for this test as it uses all layers on same fs

P.S.2:
The attached patch to t_dir_offset2 only reproduces the first
inconsistency I reported:
entry m100 has inconsistent d_ino (234 != 232)

The second inconsistency requires another small patch to overlay/077
(attached). I will post both those patches once the kernel fix is in
overlayfs-next.

Thanks,
Amir.

[-- Attachment #2: overlay-077-test-inconsistent-d_ino-in-former-merge-dir.patch --]
[-- Type: text/x-patch, Size: 1577 bytes --]

From e64a7b0c1ce697745c0bdb3ef8f43f0764510ef8 Mon Sep 17 00:00:00 2001
From: Amir Goldstein <amir73il@gmail.com>
Date: Tue, 20 Jul 2021 19:11:03 +0300
Subject: [PATCH] overlay/077: test inconsistent d_ino in former merge dir

For testing of inconsistent d_ino/st_ino we need to unlink an entry
whose st_ino is not that of the upper inode.

In the former merge dir setup we unlink all the files in the lower
dir after copyup, so they all use st_ino of the upper inode.

Let the unlinked file f100 reside in a lower path that is not being
unlinked so it will have the st_ino of the lower inode.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 tests/overlay/077 | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/tests/overlay/077 b/tests/overlay/077
index 58c0f3b5..04cb3881 100755
--- a/tests/overlay/077
+++ b/tests/overlay/077
@@ -61,8 +61,8 @@ mkdir -p $lowerdir/merge $lowerdir/former $upperdir/pure $upperdir/impure
 create_files $lowerdir/merge m
 # Files to be moved into impure upper dir
 create_files $lowerdir o
-# File to be copied up to make former merge dir impure
-touch $lowerdir/former/f100
+# File to be moved into former merge dir to make it impure
+touch $lowerdir/f100
 
 _scratch_mount
 
@@ -72,6 +72,7 @@ create_files $SCRATCH_MNT/former f
 touch $SCRATCH_MNT/merge/m100
 # Move copied up files so readdir will need to lookup origin d_ino
 mv $SCRATCH_MNT/o* $SCRATCH_MNT/impure/
+mv $SCRATCH_MNT/f100 $SCRATCH_MNT/former/
 
 # Remove the lower directory and mount overlay again to create
 # a "former merge dir"
-- 
2.32.0

Re: [PATCH] ovl: skip stale entries in merge dir cache iteration

[Miklos Szeredi]

On Tue, 20 Jul 2021 at 18:17, Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Tue, Jul 20, 2021 at 5:55 PM Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Tue, Jul 20, 2021 at 10:18 AM Amir Goldstein <amir73il@gmail.com> wrote:
> > >
> > > On Tue, Jul 20, 2021 at 7:22 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > >
> > > > On Tue, 20 Jul 2021 at 06:19, Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > > >
> > > > > On Mon, 19 Jul 2021 at 18:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > >
> > > > > > On Mon, Jul 19, 2021 at 6:24 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > > > > > >
> > > > > > > On Fri, 4 Jun 2021 at 12:43, Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > > >
> > > > > > > > On Mon, Apr 26, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
> > > > > > > > >
> > > > > > > > > On the first getdents call, ovl_iterate() populates the readdir cache
> > > > > > > > > with a list of entries, but for upper entries with origin lower inode,
> > > > > > > > > p->ino remains zero.
> > > > > > > > >
> > > > > > > > > Following getdents calls traverse the readdir cache list and call
> > > > > > > > > ovl_cache_update_ino() for entries with zero p->ino to lookup the entry
> > > > > > > > > in the overlay and return d_ino that is consistent with st_ino.
> > > > > > > > >
> > > > > > > > > If the upper file was unlinked between the first getdents call and the
> > > > > > > > > getdents call that lists the file entry, ovl_cache_update_ino() will not
> > > > > > > > > find the entry and fall back to setting d_ino to the upper real st_ino,
> > > > > > > > > which is inconsistent with how this object was presented to users.
> > > > > > > > >
> > > > > > > > > Instead of listing a stale entry with inconsistent d_ino, simply skip
> > > > > > > > > the stale entry, which is better for users.
> > > > > > > > >
> > > > > > > >
> > > > > > > > Miklos,
> > > > > > > >
> > > > > > > > I forgot to follow up on this patch.
> > > > > > > > Upstream xfstest overlay/077 is failing without this patch.
> > > > > > >
> > > > > > > Can't reproduce (on ext4/xfs and "-oxino=on").
> > > > > > >
> > > > > > > Is there some trick?
> > > > > >
> > > > > > Not sure. overlay/077 fails for me on v5.14.0-rc2 on ext4/xfs.
> > > > > >
> > > > > >      QA output created by 077
> > > > > >     +entry m100 has inconsistent d_ino (234 != 232)
> > > > > >     +entry f100 has inconsistent d_ino (335 != 16777542)
> > > > > >      Silence is golden
> > > > > >
> > > > > > Maybe you need to build src/t_dir_offset2?
> > > > >
> > > > > root@kvm:/opt/xfstests-dev# git log -1 --pretty=%h
> > > > > 10f6b231
> > > > > root@kvm:/opt/xfstests-dev# cat local.config
> > > > > export TEST_DEV=/dev/vdb1
> > > > > export TEST_DIR=/test
> > > > > export SCRATCH_DEV=/dev/vdb2
> > > > > export SCRATCH_MNT=/scratch
> > > > > export FSTYP=ext4
> > > > > export OVERLAY_MOUNT_OPTIONS="-o xino=on"
> > > > > root@kvm:/opt/xfstests-dev# make src/t_dir_offset2
> > > > > make: 'src/t_dir_offset2' is up to date.
> > > > > root@kvm:/opt/xfstests-dev# ./check -overlay overlay/077
> > > > > FSTYP         -- overlay
> > > > > PLATFORM      -- Linux/x86_64 kvm 5.14.0-rc2 #276 SMP Tue Jul 20
> > > > > 05:54:44 CEST 2021
> > > > > MKFS_OPTIONS  -- /scratch
> > > > > MOUNT_OPTIONS -- -o xino=on /scratch /scratch/ovl-mnt
> > > > >
> > > > > overlay/077 1s ...  1s
> > > > > Ran: overlay/077
> > > > > Passed all 1 tests
> > > > >
> > > > > root@kvm:/opt/xfstests-dev# cat results/overlay/077.full
> > > > >
> > > > > Create file in pure upper dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > created entry p0
> > > > > entry p0 found as expected
> > > > >
> > > > > Remove file in pure upper dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > unlinked entry p100
> > > > > entry p100 not found as expected
> > > > >
> > > > > Create file in impure upper dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > created entry o0
> > > > > entry o0 found as expected
> > > > >
> > > > > Remove file in impure upper dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > unlinked entry o100
> > > > > entry o100 not found as expected
> > > > >
> > > > > Create file in merge dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > created entry m0
> > > > > entry m0 found as expected
> > > > >
> > > > > Remove file in merge dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > unlinked entry m100
> > > > > entry m100 not found as expected
> > > > >
> > > > > Create file in former merge dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > created entry f0
> > > > > entry f0 found as expected
> > > > >
> > > > > Remove file in former merge dir:
> > > > > getdents at offset 0 returned 192 bytes
> > > > > unlinked entry f100
> > > > > entry f100 not found as expected
> > > > >
> > > > > Ideas for further debugging why this test isn't failing for v4.12-rc2?
> > > >
> > >
> > > It's not you, it's me ;-)
> > >
> > > The failure was lost during cleanup of t_dir_offset2 patches
> > > for submission and it is I who was running an older version
> > > of t_dir_offset2. Let me figure this out and get back to you
> > > with a working test.
> > >
> >
> > How about you try again with a version of the test that actually has
> > the check and the error print that I reported.... ;-)
> >
>
> P.S.1:
> xino=on is futile for this test as it uses all layers on same fs
>
> P.S.2:
> The attached patch to t_dir_offset2 only reproduces the first
> inconsistency I reported:
> entry m100 has inconsistent d_ino (234 != 232)
>
> The second inconsistency requires another small patch to overlay/077
> (attached). I will post both those patches once the kernel fix is in
> overlayfs-next.

Thanks.

Verified and pushed.

Miklos