[LU-15798] RHEL9 / gcc-11 adds -Wstringop-overread false warnings

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Assignee: Shaun Tancheff
• Reporter: Shaun Tancheff

Description

gcc-11 stringop-overread reports a false positive on
changelog record name.

Use _Pragma to exclude the diagnostic for this case.

Issue Links

• is related to: LU-17400 lfs changelog 08RENME records changelog_rec_sname string overrun
• is related to: LU-15420 Linux kernel 5.15 support

Comments

[James A Simmons]

All patches landed.

[Xinliang Liu]

This issue can be fixed by patch: https://review.whamcloud.com/c/fs/lustre-release/+/53624

[Gerrit Updater]

"xinliang <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53685
Subject: LU-15798 build: fix gcc-11 -Wstringop-overread warnings
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: d14dc3ce31e06bacdb79e2c2457b65669c554358

[Xinliang Liu]

Find that the below commit breaks the compiling on rhel9.2.

efc5c8d4de60 LU-16335 build: remove _GNU_SOURCE dependency in lustre_user.h

~/work/lustre/lustre-release$ git log -p efc5c8d4de60 --  lustre//include/uapi/linux/lustre/lustre_user.h
Author: Lai Siyao <[e-mail]>
Date:   Thu Dec 1 03:17:00 2022 -0500


    LU-16335 build: remove _GNU_SOURCE dependency in lustre_user.h
    
    The lustre_user.h header uses the non-standard strchrnul() function
    in userspace.  This will always leads to LC_IOC_REMOVE_ENTRY configure
    check to fail, and in the end "lfs rm_entry" always returns -ENOTSUP.
    
    Implement an alternative approach to avoid external dependencies on
    the lustre_user.h header.  Also, LC_IOC_REMOVE_ENTRY is itself
    unnecessary, the code can check for LL_IOC_REMOVE_ENTRY directly.
    
    Replace the NFS-specific -ENOTSUP error return code with -EOPNOTSUPP.
    
    Fix the compile test_400[ab] checks to not use "-std=c99" to verify
    that the uapi headers are usable without this dependency.
    
...
diff --git a/lustre/include/uapi/linux/lustre/lustre_user.h b/lustre/include/uapi/linux/lustre/lustre_user.h
index 1b1e60acf805..36e1e70b9224 100644
--- a/lustre/include/uapi/linux/lustre/lustre_user.h
+++ b/lustre/include/uapi/linux/lustre/lustre_user.h
@@ -2033,7 +2033,11 @@ static inline char *changelog_rec_name(const struct changelog_rec *rec)
 
 static inline char *changelog_rec_sname(const struct changelog_rec *rec)
 {
-       return strchrnul(changelog_rec_name(rec), '\0') + 1;
+       char *str = changelog_rec_name(rec);
+
+       while (*str != '\0')
+               str++;
+       return str + 1;
 }
  

[Xinliang Liu]

Thanks, stancheff for the info, it seems an Aarch64 specific issue.

[Shaun Tancheff]

Yes, make and make rpms both work.

With gcc-12 I do see a few issues with configure and with compiling:

In function ‘strnlen’,
    inlined from ‘lov_iocontrol’ at /home/shaun/lustre-release/lustre/lov/lov_obd.c:1079:4:
include/linux/fortify-string.h:23:34: error: array subscript 112 is outside array bounds of ‘void[112]’ [-Werror=array-bounds]
   23 |             __builtin_constant_p(*__p)) {                       \
      |                                  ^~~~
include/linux/fortify-string.h:141:24: note: in expansion of macro ‘__compiletime_strlen’
  141 |         size_t p_len = __compiletime_strlen(p);
      |                        ^~~~~~~~~~~~~~~~~~~~
In file included from include/linux/fs.h:44,
                 from /home/shaun/lustre-release/libcfs/include/libcfs/linux/linux-misc.h:35,
                 from /home/shaun/lustre-release/libcfs/include/libcfs/libcfs.h:43:
In function ‘kmalloc’,
    inlined from ‘lov_iocontrol’ at /home/shaun/lustre-release/lustre/lov/lov_obd.c:1072:3:
include/linux/slab.h:611:24: note: at offset 112 into object of size 112 allocated by ‘kmem_cache_alloc_trace’
  611 |                 return kmem_cache_alloc_trace(
      |                        ^~~~~~~~~~~~~~~~~~~~~~~
  612 |                                 kmalloc_caches[kmalloc_type(flags)][index],
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                                 flags, size);
      |                                 ~~~~~~~~~~~~
cc1: all warnings being treated as errors

[Xinliang Liu]

@stancheff have tried make rpms? Just make no compile error.

[Xinliang Liu]

Yes, it is aarch64.

[Shaun Tancheff]

Strange that I do not see this on x86_64 ... it this aarch64?

[Xinliang Liu]

This issue still exists on rhel9.2:

/tmp/rpmbuild-lustre-jenkins-SCTCWlBO/BUILD/lustre-2.15.58_59_g2ddb1d3_dirty/lustre/include/uapi/linux/lustre/lustre_user.h:2095:16: error: 'strlen' reading 1 or more bytes from a region of size 0 [-Werror=stringop-overread]
 2095 |         return strlen(changelog_rec_sname(rec));
      |                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~