tests/cp_files: fallback for C libs w/o SEEK_DATA

[jlsalvador]

Not all C libraries support SEEK_DATA (e.g., uClibc). When SEEK_DATA is not defined, fallback to SEEK_END.

Motivation and Context

Currently, uClibc does not compile OpenZFS because the test cp_files uses SEEK_DATA as the whence parameter for the lseek method.

Description

When SEEK_DATA is not defined, use SEEK_END as the fallback value.

How Has This Been Tested?

• Compiling Buildroot with a preliminary patch version using GLIBC, UCLIBC, and MUSL as toolchains.
• Running the ZFS Test Suite with this change applied, although the affected test was skipped due to a kernel configuration for Debian 12.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Library ABI change (libzfs, libzfs_core, libnvpair, libuutil and libzfsbootenv)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the OpenZFS code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• I have run the ZFS Test Suite with this change applied.
• All commit messages are properly formatted and contain Signed-off-by.

[jlsalvador]

Hello all,

Any alternative instead of SEEK_END?
Thoughts?

[rincebrain]

I think the right thing to do, if SEEK_DATA/SEEK_HOLE are not available, is to make all the tests that rely on that constant get marked SKIP, rather than substituting one with different semantics to test.

I'd probably do this with a configure time check that checked for support for those and then defined something for the test suite and that test program to change their behavior around, but that's off the top of my head, I don't know offhand how we do this for other platform-specific checks.

[behlendorf]

Right, instead of changing the behavior of this test we should disable it as @rincebrain suggested.

[robn]

Another option would be to define those two if they're missing:

#ifndef SEEK_DATA
#define SEEK_DATA 3
#endif
#ifndef SEEK_HOLE
#define SEEK_HOLE 4
#endif

That particular value gets passed directly into SYS__llseek in the kernel, so it should always work on any Linux kernel.

Though as I type is, I feel a little uncomfortable about it. Disabling makes more sense. @jlsalvador if you'd like a hand with the detection bits, let me know - I do know my way around that stuff.

[jlsalvador]

Hello again,

I force-pushed a commit with another solution as @rincebrain suggested.

I have made a few assumptions. Could someone please check if the patch follows the ZFS guidelines?

Perhaps someone could improve it by skipping the test case at runtime (printing [SKIP]) instead of ignoring it with a new AutoMake conditional (WANT_UNISTD_SEEK_DATA).

I tested it by building Buildroot using GLIBC, UCLIBC, and MUSL toolchains.
Currently I am running the ZFS Test Suite on a Debian 12 (./configure --disable-pyzfs), it requires a few hours to finish.

---

@robn, thank you for offering to help. Could you please confirm if the patch is correct?

[behlendorf]

Perhaps someone could improve it by skipping the test case at runtime (printing [SKIP]) instead of ignoring it with a new

The way we've handled this before is to only build the cp_files and seekflood binaries when SEEK_DATA / SEEK_HOLE are available. Which is what you've already done. Then at the top of each relevant test add a check for the binary. If it doesn't exist call log_unsupported with a meaningful error message which will cause the script to exist. You'll also need to update zts-report.py.in to understand these tests may be skipped. The Linux specific tmpfile tests provide a good example of this.