On Mon, Nov 07, 2022 at 11:35:16AM -0500, Theodore Ts'o wrote: > On Mon, Nov 07, 2022 at 10:02:36AM +0800, Zorro Lang wrote: > > I think it doesn't make sense to use $FSX_AVOID in `fsx --replay-ops` cases. > > Due to generally the operations which a cases would like to replay are exact > > steps to reproduce to a known bug. If we skip some operations (e.g. -F), it > > doesn't make sense for this reproducer. > > > > The recommended way for this kind of cases is making sure current fs/system > > support the operations will be run by fsx, especially those features are not > > common on different fs/system.... > > > > So it uses below _require_* helpers to make sure these operations are supported, > > before testing: > > > > _require_xfs_io_command "falloc" > > _require_xfs_io_command "falloc" "-k" > > _require_xfs_io_command "fzero" > > _require_xfs_io_command "fcollapse" > > > > That's my point, hope I didn't misunderstand what you said :) > > No, you didn't understand me. :-) > > For context, I have an out of tree patch (see attached), which I had > tried upstreaming a while back, but it got rejected, so I've continued > to keep it in my personal tree. The basic idea is sometimes you might > want to suppress a test even *though* _require_xfs_io_command seems to > indicate that operation was supported. > > This might either be because the test didn't know about ext4 > bigalloc's cluster alignment requirements, or because a particular > operation might just be *buggy* and being able to run tests as if a > particular command wasn't supported was useful. > > It was rejected because the claim was that you could just exclude by > group instead (e.g., "punch", "collapse") but I didn't trust that the > group list would be kept up to date, so I never really agreed with > that line of reasoning. These days, given that group declaration are > kept in the test script, it's much less likely to happen, but I've > kept the patch in my tree because it's occasionally useful. > > At this point, it's admittedly pretty rarely needed since ext4's > collapse and insert range commands are pretty solid modulo tests not > understanding cluster alignment, but still, it's not much effort for > me to keep carrying the patch and I don't expect it will ever get > upstreamed. If it's collapse/insert range you're specifically worried about, perhaps its time to implement _get_file_block_size for ext4 so that _test_congruent_file_oplen can exclude those tests that will get the alignment wrong? --D > > - Ted > > commit c9d25475a94d5e53d7f18d247a17088999522862 > Author: Theodore Ts'o <tytso@xxxxxxx> > Date: Sat Oct 17 14:39:26 2015 -0400 > > common: introduce XFS_IO_AVOID env var > > Like FSSTRESS_AVOID and FSX_AVOID, XFS_IO_AVOID can be used to avoid > using various advanced file system features such as "fpunch" > "fcollapse", "finsert", or "zero". Tests that require an xfs_io > command which is included in the space-separated list found in the > XFS_IO_AVOID environment variable will be skipped using _notrun. > > Signed-off-by: Theodore Ts'o <tytso@xxxxxxx> > > diff --git a/README b/README > index 4c4f22f85..42baff07b 100644 > --- a/README > +++ b/README > @@ -245,6 +245,10 @@ Misc: > this option is supported for all filesystems currently only -overlay is > expected to run without issues. For other filesystems additional patches > and fixes to the test suite might be needed. > + - setenv XFS_IO_AVOID, which may contain a list of space separated > + xfs_io commands which will be avoided in case you want to exclude > + tests that require the use of certain file system operations such > + as "fpunch", "fcollapse", "finsert", or "zero". > > ______________________ > USING THE FSQA SUITE > diff --git a/common/rc b/common/rc > index eb67e0cdc..d1c07a4d0 100644 > --- a/common/rc > +++ b/common/rc > @@ -2485,6 +2485,11 @@ _require_xfs_io_command() > local opts="" > local attr_info="" > > + if echo "$XFS_IO_AVOID" | grep -wq -- "$command" > + then > + _notrun "Avoiding xfs_io $command" > + fi > + > local testfile=$TEST_DIR/$$.xfs_io > local testio > case $command in
