]> git.baikalelectronics.ru Git - kernel.git/commit
binfmt_misc: work around gcc-4.9 warning
authorArnd Bergmann <arnd@arndb.de>
Mon, 13 Oct 2014 22:52:08 +0000 (15:52 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 Oct 2014 00:18:16 +0000 (02:18 +0200)
commit84a6791916059de83b3451efaa0ac64ee3a10602
tree1974c13d945203c4042e06133f7b7229e61d6598
parentb184ed8d751fe88b4852a935d83a55380c7eb152
binfmt_misc: work around gcc-4.9 warning

gcc-4.9 on ARM gives us a mysterious warning about the binfmt_misc
parse_command function:

  fs/binfmt_misc.c: In function 'parse_command.part.3':
  fs/binfmt_misc.c:405:7: warning: array subscript is above array bounds [-Warray-bounds]

I've managed to trace this back to the ARM implementation of memset,
which is called from copy_from_user in case of a fault and which does

 #define memset(p,v,n)                                                  \
        ({                                                              \
                void *__p = (p); size_t __n = n;                        \
                if ((__n) != 0) {                                       \
                        if (__builtin_constant_p((v)) && (v) == 0)      \
                                __memzero((__p),(__n));                 \
                        else                                            \
                                memset((__p),(v),(__n));                \
                }                                                       \
                (__p);                                                  \
        })

Apparently gcc gets confused by the check for "size != 0" and believes
that the size might be zero when it gets to the line that does "if
(s[count-1] == '\n')", so it would access data outside of the array.

gcc is clearly wrong here, since this condition was already checked
earlier in the function and the 'size' value can not change in the
meantime.

Fortunately, we can work around it and get rid of the warning by
rearranging the function to check for zero size after doing the
copy_from_user.  It is still safe to pass a zero size into
copy_from_user, so it does not cause any side effects.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/binfmt_misc.c