This test checks the behaviour of xfs/ext4 filesystems when extsize
hint is set on files with inode size as 0, non-empty
files with allocated and delalloc extents and so on.
This test set also tests one of the scenarios where setting extsize
hints on files with delayed allocation incorrectly succeeds and there is
a recent patch series[1] that got merged[2] which fixes this.
Currently this test only runs in xfs but there is a patch series[3] in review
that adds support for the extsize hints for ext4 as well.
I have tested it on ppc64le (with page size 64k) and x86_64 (with page size 4k).
The block sizes I have tested with are 2k, 4k, 8k, 16k, 32k, 64k with extsize being
twice of the block size or 4 * block size(if the default extsize is 2 * blocksize).
I have also tested with configurations where I added extszinherit mkfs option
on the root. I had to enable CONFIG_TRANSPARENT_HUGEPAGE to enable block size
greater than 4k on x86_64.
Changes since [v2]
- Renamed _test_fsx_xflags_field() to _test_fsxattr_xflag().
- Changed the definition of _test_fsx_xflags_field (now renamed to _test_fsxattr_xflag)
to accept xflag names (like extsize, cowextsize) instead of single characters (like 'e') and
modified the callers of this function accordingly (in xfs/207 and generic/366)
- Some changes in the comments in the setup() function of tests/generic/366.
- Added another helper function _require_scratch_extsize().
- Replaced the usage of _require_scratch() with _require_scratch_extsize() in tests/generic/366.
- Fixed some indentation issues in the test scripts(xfs/207, generic/366) and the commit messages.
Changes since [v1]
- Making the definition of _test_xfs_xflags_field() more compact and renamed
_test_xfs_xflags_field() to _test_fsx_xflags_field() since I moved it to common/rc.
- Removed the explicit import of common/xfs from the test script.
- Renamed the test file from generic/365 to generic/366 since generic/365 was taken.
- Made the following functions in tests/generic/366 accept a second parameter (extsize)
a) check_extsz_xflag_across_remount()
b) check_extsz_and_xflag()
c) read_file_extsize()
d) filter_extsz
This will help filter out any extsize that the test wants. This is required because, this
version no longer assumes any particular fixed value of the default extsize.
- Removed the check for xflags in test_data_allocated() and test_data_delayed() test
- Added an extra line in output that denotes that we are checking that the extent size
hasn't changed if the xfs_io command fails (in test_data_allocated() and test_data_delayed() test).
- Changed the new extsize to 4*blocksize if default extsize size is 2*blocksize to make
sure that the new extsize is not the same as the default extsize so that we can observe it changing.
[1] https://lore.kernel.org/all/20241015094509.678082-1-ojaswin@xxxxxxxxxxxxx/
[2] kernel commit - 2a492ff66673
[3] https://lore.kernel.org/linux-ext4/cover.1726034272.git.ojaswin@xxxxxxxxxxxxx/
[v1] https://lore.kernel.org/all/cover.1729624806.git.nirjhar@xxxxxxxxxxxxx/
[v2] https://lore.kernel.org/all/cover.1731597226.git.nirjhar@xxxxxxxxxxxxx/
Nirjhar Roy (3):
common/rc,xfs/207: Add a common helper function to check xflag bits
common/rc: Add a new _require_scratch_extsize helper function
generic: Addition of new tests for extsize hints
common/rc | 24 ++++++
tests/generic/366 | 175 ++++++++++++++++++++++++++++++++++++++++++
tests/generic/366.out | 26 +++++++
tests/xfs/207 | 15 +---
4 files changed, 229 insertions(+), 11 deletions(-)
create mode 100755 tests/generic/366
create mode 100644 tests/generic/366.out
--
2.43.5
