| Summary: | xfs/300 fails with xfsprogs-4.5.0-6.el7; not built against libattr | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Red Hat Enterprise Linux 7 | Reporter: | Eric Sandeen <esandeen> | ||||
| Component: | xfsprogs | Assignee: | Eric Sandeen <esandeen> | ||||
| Status: | CLOSED ERRATA | QA Contact: | Zorro Lang <zlang> | ||||
| Severity: | unspecified | Docs Contact: | |||||
| Priority: | unspecified | ||||||
| Version: | 7.3 | CC: | eguan, fweimer, jbastian, jcm, tlavigne, xzhou, yselkowi, zlang | ||||
| Target Milestone: | rc | Keywords: | Patch, Regression | ||||
| Target Release: | --- | ||||||
| Hardware: | All | ||||||
| OS: | Unspecified | ||||||
| Whiteboard: | |||||||
| Fixed In Version: | xfsprogs-4.5.0-7.el7 | Doc Type: | If docs needed, set a value | ||||
| Doc Text: | Story Points: | --- | |||||
| Clone Of: | Environment: | ||||||
| Last Closed: | 2016-11-04 06:24:48 UTC | Type: | Bug | ||||
| Regression: | --- | Mount Type: | --- | ||||
| Documentation: | --- | CRM: | |||||
| Verified Versions: | Category: | --- | |||||
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |||||
| Cloudforms Team: | --- | Target Upstream Version: | |||||
| Attachments: |
|
||||||
|
Description
Eric Sandeen
2016-09-04 02:01:30 UTC
Thanks for find this, Eric. We can't find this problem due to another known issue. When auto test with beaker, xfs/300 get below output: --- QA output created by 300 --- When I run xfs/300 manually, I got below output: --- QA output created by 300 SCRATCH_MNT/300.test XFS_IOC_SWAPEXT failed: SCRATCH_MNT/300.test: Invalid argument --- Both of them are unexpected. Thanks, Zorro I may need to dig into this more; I did a scratch build of xfsprogs: https://brewweb.engineering.redhat.com/brew/taskinfo?taskID=11707699 which I think used the new gcc, and that build still fails. This fails on all arches because we somehow lost a dependency on libattr, and fsr now #ifdefs away calls if configure didn't find the attr headers.
commit c14c7b795c0bdb92d925bab44e8239867b2883c6
Author: Jan Tulak <jtulak>
Date: Wed Oct 14 10:58:03 2015 +1100
build: Add autoconf check for fsetxattr call
This is a regression, and needs to be fixed for all arches.
Link to the commit: http://oss.sgi.com/cgi-bin/gitweb.cgi?p=xfs/cmds/xfsprogs.git;a=commitdiff;h=c14c7b795c0bdb92d925bab44e8239867b2883c6;hp=7f7fd8ac1f5cc0d3af7c2f050ac070a321581de5 Supposed rationale: > OS X has fsetxattr() in another header and with different arguments. > For now, check for the Linux variant and if not available, skip > the code using the call. So the configure test looks for attr/xattr.h (from libattr-devel) to determine if it's on Linux, but then uses sys/xattr.h in the sources. Honestly, this looks bogus to me; why not just use #ifndef __APPLE__ around that code? Looks like we need BuildRequires: libattr-devel for attr/xattr.h in the configure test. Nothing appears to link with libattr, so glibc's implementation should still be used. (In reply to Yaakov Selkowitz from comment #5) Yep, not a big fan of that change - I hadn't looked at the two different headers, ugh. > Looks like we need BuildRequires: libattr-devel for attr/xattr.h in the > configure test. Nothing appears to link with libattr, so glibc's > implementation should still be used. Yes, that's the simplest/fastest fix for now I think. I'll look harder at the commit and why we're looking at 2 different headers, thanks! -Eric This does probably fix it as well, w/o the new dependency:
diff --git a/m4/package_libcdev.m4 b/m4/package_libcdev.m4
index 7a847e9..55946ab 100644
--- a/m4/package_libcdev.m4
+++ b/m4/package_libcdev.m4
@@ -239,7 +239,7 @@ AC_DEFUN([AC_HAVE_FSETXATTR],
have_fsetxattr=yes,
[],
[#include <sys/types.h>
- #include <attr/xattr.h>]
+ #include <sys/xattr.h>]
)
AC_SUBST(have_fsetxattr)
])
(In reply to Eric Sandeen from comment #6) > (In reply to Yaakov Selkowitz from comment #5) > > Yep, not a big fan of that change - I hadn't looked at the two different > headers, ugh. > > Yes, that's the simplest/fastest fix for now I think. Scratch build in progress. > I'll look harder at the commit and why we're looking at 2 different headers, > thanks! IIUC, both Darwin[1] and Linux have <sys/xattr.h> but libattr is not available on both and it provides <attr/xattr.h>, so someone thought this was a "smart" way of detecting Linux vs. Darwin fsetxattr. IMHO the entire commit is incorrect; it should really be reverted, and the #ifdef HAVE_FSETXATTR block conditional on #ifndef __APPLE__ instead. Alternatively, a more complicated configure test can be written to determine how many arguments fsetxattr accepts, but I'm not sure if any other systems have the extra argument. (In reply to Eric Sandeen from comment #7) > This does probably fix it as well, w/o the new dependency: > > [#include <sys/types.h> > - #include <attr/xattr.h>] > + #include <sys/xattr.h>] Darwin has <sys/xattr.h> too, hence that wouldn't help. Created attachment 1198407 [details] Patch for rhel-7.3 Successful scratch build: https://brewweb.engineering.redhat.com/brew/taskinfo?taskID=11709316 Most notably, in build.log: checking whether fsetxattr is declared... yes and libattr.so.1 does NOT show up as a dependency. Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://rhn.redhat.com/errata/RHBA-2016-2431.html |