[Bug libc/21019] [mips] n32 lseek incorrect on overflow

["cvs-commit at gcc dot gnu.org" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=21019

--- Comment #1 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "GNU C Library master sources".

The branch, master has been updated
       via  df19fdcfec7143073b50f9f01af712528bed6d26 (commit)
      from  4179178bf0cfafe72c00647b79e69e47bd5c6a89 (commit)

Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.

- Log -----------------------------------------------------------------
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=df19fdcfec7143073b50f9f01af712528bed6d26

commit df19fdcfec7143073b50f9f01af712528bed6d26
Author: Joseph Myers <joseph@codesourcery.com>
Date:   Mon Jan 2 23:09:25 2017 +0000

    Fix MIPS n32 lseek, lseek64 (bug 21019).

    The lseek consolidation broke lseek64 for MIPS n32, so resulting in
    io/test-lfs failing with an incorrect return from ftello64.  This
    configuration uses the lseek syscall with a 64-bit return value; as
    the C syscall macros return long, they cannot be used in this case and
    so an assembly implementation is needed; accordingly, this patch adds
    lseek64 back to syscalls.list for this configuration.

    lseek was also broken, truncating the result without checking for
    overflow.  lseek however was already broken before the consolidation;
    it aliased lseek64 so would return an out-of-range value, resulting in
    architecturally undefined behavior in the caller if it tried to use a
    non-sign-extended value with a 32-bit instruction.  This patch adds a
    custom lseek implementation in C for n32, which calls __lseek64 to get
    the 64-bit value then checks for overflow.

    Because the prior lseek breakage did not show in test results, and the
    lseek64 breakage showed only indirectly through tests of ftello64,
    test coverage was clearly inadequate.  This patch extends
    io/test-lfs.c to test the lseek64 return value (at a point where it
    has already seeked over 2GB into a file), and then to test the lseek
    return value (with the latter's expectations depending on whether
    off_t is smaller than off64_t).

    Tested for mips64 n32.  Also tested test-lfs for x86_64 and x86, where
    as expected it passes.

        [BZ #21019]
        * sysdeps/unix/sysv/linux/mips/mips64/n32/syscalls.list (lseek64):
        New syscall entry.
        * sysdeps/unix/sysv/linux/mips/mips64/n32/lseek.c: New file.
        * io/test-lfs.c (do_test): Test offset returned from lseek64 and
        lseek.

-----------------------------------------------------------------------

Summary of changes:
 ChangeLog                                          |    7 +++++
 io/test-lfs.c                                      |   21 +++++++++++++++
 .../epoll_create.c => mips/mips64/n32/lseek.c}     |   28 +++++++++----------
 .../unix/sysv/linux/mips/mips64/n32/syscalls.list  |    4 +++
 4 files changed, 45 insertions(+), 15 deletions(-)
 copy sysdeps/unix/sysv/linux/{generic/epoll_create.c =>
mips/mips64/n32/lseek.c} (68%)

-- 
You are receiving this mail because:
You are on the CC list for the bug.