compat/linux: return EBADF for O_PATH mmap() and xattr operations

[ilovegrape]

This fixes LTP open13, which expects O_PATH mmap() and fgetxattr()
to fail with EBADF, but FreeBSD returned EACCES for mmap() and
EOPNOTSUPP for fgetxattr().

Apply the same EBADF handling to fsetxattr(), fremovexattr(), and
flistxattr() so the xattr paths stay consistent.

[ricardobranco777]

I suggest to open a bug report in https://bugs.freebsd.org with what is failing and then add a tag with PR: 1234567 in the commit message so it gets more visibility. :)

[ilovegrape]

I suggest to open a bug report in https://bugs.freebsd.org with what is failing and then add a tag with PR: 1234567 in the commit message so it gets more visibility. :)

I don't have an account, and I need to request one...
Maybe the same is required for #2181 ?

@ricardobranco777 How do you think if I create a patch in Phabricator?

[ricardobranco777]

I suggest to open a bug report in https://bugs.freebsd.org with what is failing and then add a tag with PR: 1234567 in the commit message so it gets more visibility. :)

I don't have an account, and I need to request one... Maybe the same is required for #2181 ?

You can use any e-mail to create an account in Bugzilla.

@ricardobranco777 How do you think if I create a patch in Phabricator?

I dunno. Haven't used it yet.

[ilovegrape]

You can use any e-mail to create an account in Bugzilla.

I requested the account once again according to the process just now

[ilovegrape]

Hi @ricardobranco777

Bug fired with commit msg updated, any further comments?

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=295537

[ricardobranco777]

Hi @ricardobranco777

Bug fired with commit msg updated, any further comments?

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=295537

Thank you. Including a minimal reproducer also helps. :)

[ilovegrape]

Updated with the comments below on the PR case.

Using the curren FreeBSD 16.0 code, there are 4 failures of the LTP test case open13

$ /opt/ltp/testcases/bin/open13
tst_tmpdir.c:317: TINFO: Using /tmp/LTP_opeJ6gQkD as tmpdir (tmpfs filesystem)
tst_test.c:2047: TINFO: LTP version: 20260130
tst_test.c:2050: TINFO: Tested kernel: 5.15.0 FreeBSD 16.0-CURRENT #2 main-n286042-28d85db46b48: Sun May 24 21 x86_64
tst_kconfig.c:71: TINFO: Couldn't locate kernel config!
tst_test.c:1877: TINFO: Overall timeout per run is 0h 00m 30s
open13.c:77: TPASS: read() on original FD : EBADF (9)
open13.c:79: TPASS: read() on duplicated FD : EBADF (9)
open13.c:77: TPASS: write() on original FD : EBADF (9)
open13.c:79: TPASS: write() on duplicated FD : EBADF (9)
open13.c:77: TPASS: fchmod() on original FD : EBADF (9)
open13.c:79: TPASS: fchmod() on duplicated FD : EBADF (9)
open13.c:77: TPASS: fchown() on original FD : EBADF (9)
open13.c:79: TPASS: fchown() on duplicated FD : EBADF (9)
open13.c:77: TPASS: ioctl() on original FD : EBADF (9)
open13.c:79: TPASS: ioctl() on duplicated FD : EBADF (9)
open13.c:77: TFAIL: mmap() on original FD expected EBADF: EACCES (13)
open13.c:79: TFAIL: mmap() on duplicated FD expected EBADF: EACCES (13)
open13.c:77: TFAIL: fgetxattr() on original FD expected EBADF: EOPNOTSUPP (95)
open13.c:79: TFAIL: fgetxattr() on duplicated FD expected EBADF: EOPNOTSUPP (95)

Summary:
passed 10
failed 4
broken 0
skipped 0
warnings 0

And all passed after the fix

$ /opt/ltp/testcases/bin/open13
tst_tmpdir.c:317: TINFO: Using /tmp/LTP_opeiIAOHE as tmpdir (tmpfs filesystem)
tst_test.c:2047: TINFO: LTP version: 20260130
tst_test.c:2050: TINFO: Tested kernel: 5.15.0 FreeBSD 16.0-CURRENT #10 open13-fix-src-22b97fbf255b-dirty: Thu x86_64
tst_kconfig.c:71: TINFO: Couldn't locate kernel config!
tst_test.c:1877: TINFO: Overall timeout per run is 0h 00m 30s
open13.c:77: TPASS: read() on original FD : EBADF (9)
open13.c:79: TPASS: read() on duplicated FD : EBADF (9)
open13.c:77: TPASS: write() on original FD : EBADF (9)
open13.c:79: TPASS: write() on duplicated FD : EBADF (9)
open13.c:77: TPASS: fchmod() on original FD : EBADF (9)
open13.c:79: TPASS: fchmod() on duplicated FD : EBADF (9)
open13.c:77: TPASS: fchown() on original FD : EBADF (9)
open13.c:79: TPASS: fchown() on duplicated FD : EBADF (9)
open13.c:77: TPASS: ioctl() on original FD : EBADF (9)
open13.c:79: TPASS: ioctl() on duplicated FD : EBADF (9)
open13.c:77: TPASS: mmap() on original FD : EBADF (9)
open13.c:79: TPASS: mmap() on duplicated FD : EBADF (9)
open13.c:77: TPASS: fgetxattr() on original FD : EBADF (9)
open13.c:79: TPASS: fgetxattr() on duplicated FD : EBADF (9)

Summary:
passed 14
failed 0
broken 0
skipped 0
warnings 0

[shkhln]

The description makes it seem like it would be enough to convert EOPNOTSUPP to EBADF, but the patch itself is a bit more elaborate. Is there any reason for that?

[ilovegrape]

The description makes it seem like it would be enough to convert EOPNOTSUPP to EBADF, but the patch itself is a bit more elaborate. Is there any reason for that?

The patch is only intended to return EBADF for O_PATH file descriptors, not to rewrite all EOPNOTSUPP errors.

[kostikbel]

There is a TOTTOU race.

[ilovegrape]

Good point, I need to think a little deeply into it...