On Thu, Sep 21, 2023 at 11:00:37AM +0300, Amir Goldstein wrote: > On Thu, Sep 21, 2023 at 9:26 AM Zorro Lang <zlang@xxxxxxxxxx> wrote: > > > > On Wed, Sep 20, 2023 at 06:34:21PM +0300, Amir Goldstein wrote: > > > On Wed, Sep 20, 2023 at 6:14 PM Zorro Lang <zlang@xxxxxxxxxx> wrote: > > > > > > > > On Wed, Sep 20, 2023 at 04:03:55PM +0300, Amir Goldstein wrote: > > > > > A test for a regression from v5.15 reported by Ruiwen Zhao: > > > > > https://lore.kernel.org/linux-unionfs/CAKd=y5Hpg7J2gxrFT02F94o=FM9QvGp=kcH1Grctx8HzFYvpiA@xxxxxxxxxxxxxx/ > > > > Could you give one more sentence to tell what kind of regression > > does this case test for? Not only a link address. > > > > > > > > > > > > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx> > > > > > --- > > > > > > > > > > Zorro, > > > > > > > > > > This is a test for a regression in kernel v5.15. > > > > > The fix was merged for 6.6-rc2 and has been picked for > > > > > the upcoming LTS releases 5.15, 6.1, 6.5. > > > > > > > > > > The reproducer only manifests the bug in fs that inherit noatime flag, > > > > > namely ext4, btrfs, ... but not xfs. > > FYI, I made a mistake in the statement above. > xfs does support inherit of noatime flag, but > it does not inherit noatime for *symlinks*. > > I added the _require_chattr_inherit helper that you suggested > in v2, but it only checks for inherit of noatime flag (the 2nd _notrun). > I did not add a helper for _require_chattr_inherit_symlink > because it was too specific and so I left the 3rd _notrun > open coded in the test in v2. OK, if xfs thinks it's an expected behavior which won't be changed :) Thanks, Zorro > > > > > > > > > > > The test does _notrun on xfs for that reason. > > > > > > > > > > Thanks, > > > > > Amir. > > > > > > > > > > tests/overlay/082 | 68 +++++++++++++++++++++++++++++++++++++++++++ > > > > > tests/overlay/082.out | 2 ++ > > > > > 2 files changed, 70 insertions(+) > > > > > create mode 100755 tests/overlay/082 > > > > > create mode 100644 tests/overlay/082.out > > > > > > > > > > diff --git a/tests/overlay/082 b/tests/overlay/082 > > > > > new file mode 100755 > > > > > index 00000000..abea3c2b > > > > > --- /dev/null > > > > > +++ b/tests/overlay/082 > > > > > @@ -0,0 +1,68 @@ > > > > > +#! /bin/bash > > > > > +# SPDX-License-Identifier: GPL-2.0 > > > > > +# Copyright (C) 2023 CTERA Networks. All Rights Reserved. > > > > > +# > > > > > +# FS QA Test 082 > > > > > +# > > > > > +# kernel commit 72db82115d2b ("ovl: copy up sync/noatime fileattr flags") > > > > > +# from v5.15 introduced a regression. > > > > Hi Amir, > > > > Thanks for this new regression test. More detailed (picky:) review points > > as below ... > > > > So this commit is the one which introduced a regression. But we generally > > say what kind of regression does this case test for, in this comment. > > > > > > > +# > > > > > +. ./common/preamble > > > > > +_begin_fstest auto quick > > > > According the source code of this case, please think about more detailed group > > names, likes: "symlink", "copyup" and "atime". > > > > > > > + > > > > > +# Import common functions. > > > > > +. ./common/filter > > > > I think this case doesn't use any filter helpers, right? > > > > > > > + > > > > > +# real QA test starts here > > > > > +_supported_fs overlay > > > > > +_fixed_by_kernel_commit ab048302026d \ > > > > > + "ovl: fix failed copyup of fileattr on a symlink" > > > > > + > > > > > +_require_scratch > > > > > +_require_chattr A > > > > > + > > > > > +# remove all files from previous runs > > > > > +_scratch_mkfs > > > > > + > > > > > +# prepare lower test dir with NOATIME flag > > > > > +lowerdir=$OVL_BASE_SCRATCH_MNT/$OVL_LOWER > > > > > +mkdir -p $lowerdir/testdir > > > > > +$CHATTR_PROG +A $lowerdir/testdir >> $seqres.full 2>&1 || > > > > > + _notrun "base fs $OVL_BASE_FSTYP does not support No_Atime flag" > > > > > + > > > > > +# The NOATIME is inheritted to children symlink in ext4/fs2fs > > > > > +# (and on tmpfs on recent kernels). > > > > > +# The overlayfs test will not fail unless base fs is > > > > > +# one of those filesystems. > > > > > +# > > > > > +# The problem with this inheritence is that the NOATIME flag is inheritted > > > > > +# to a symlink and the flag does take effect, but there is no way to query > > > > > +# the flag (lsattr) or change it (chattr) on a symlink, so overlayfs will > > > > > +# fail when trying to copy up NOATIME flag from lower to upper symlink. > > > > > +# > > > > > +touch $lowerdir/testdir/foo > > > > > +ln -sf foo $lowerdir/testdir/lnk > > > > > + > > > > > +$LSATTR_PROG -l $lowerdir/testdir/foo >> $seqres.full 2>&1 > > > > > +$LSATTR_PROG -l $lowerdir/testdir/foo | grep -q No_Atime || \ > > > > > + _notrun "base fs $OVL_BASE_FSTYP does not inherit No_Atime flag" > > > > > + > > > > > +before=$(stat -c %x $lowerdir/testdir/lnk) > > > > > +echo "symlink atime before readlink: $before" >> $seqres.full 2>&1 > > > > I remember some filesystems' timestamp for atime (e.g. exfat) might have more > > seconds granularity. So it would be better to `sleep 2s` at here. > > > > Correct me if someone fs need more or less :) > > > > That would be a futile waste of 2 seconds IMO, because > Those niche fs probably do not support chattr and because > This is an overlayfs regression test and overlayfs is highly > unlikely to be running in production on those niche fs and it > probably does not support many of them. > > > > > > > +cat $lowerdir/testdir/lnk > > > > > +after=$(stat -c %x $lowerdir/testdir/lnk) > > > > > +echo "symlink atime after readlink: $after" >> $seqres.full 2>&1 > > > > > + > > > > > +[ "$before" == "$after" ] || \ > > > > > + _notrun "base fs $OVL_BASE_FSTYP does not inherit No_Atime flag on symlink" > > > > > + > > > > > +# mounting overlay > > > > > +_scratch_mount > > > > > + > > > > > +# moving symlink will try to copy up lower symlink flags > > > > > +mv $SCRATCH_MNT/testdir/lnk $SCRATCH_MNT/ > > > > > > > > Lots of above codes are checking if the underlying fs supports No_Atime (and inherit), > > > > and _notrun if not support. How about do these checking steps in a require_* > > > > function locally or in common/, likes _require_noatime_inheritance(). And we also > > > > can let _require_chattr accept one more argument to specify a test directory. > > > > > > > > > > ok. > > > > > > > The "mv ..." command looks like the final testing step. If there's not that bug, > > > > nothing happen, but I'm wondering what should happen if there's a bug? > > > > > > mv fails with error ENXIO, see linked bug report in commit message. > > > > Thanks, I think we can add "fails with error ENXIO at here, if the bug is reproduced" in > > the comment of that "mv ..." command. > > > > Sure. no problem. > I will post v2 soon with _require_chattr_inherit and > above comments fixed. > > Thanks for the review! > Amir. >