[Bug malloc/22611] New: malloc might succeed and set errno to ENOMEM in case of heap allocation failure

["aurelien at aurel32 dot net" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=22611

            Bug ID: 22611
           Summary: malloc might succeed and set errno to ENOMEM in case
                    of heap allocation failure
           Product: glibc
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: malloc
          Assignee: unassigned at sourceware dot org
          Reporter: aurelien at aurel32 dot net
  Target Milestone: ---

tst-realloc tests, among others things that malloc and friends to not set errno
to ENOMEM in case of successful memory allocation. This test always fail on
sparc64 with "Error: errno is set but should not be".

On sparc64 it happens that, for a reason I don't really understand, when
running processes by explicitly calling ld.so, all the heap allocation fail, ie
the call to brk return an unchanged address. sysmalloc first call sbrk through
the call to MORECORE (size). This sets errno to ENOMEM. Then mmap is used as a
fallback and succeed, but doesn't reset errno to its initial value.

While the way to trigger the issue within the testsuite is sparc64 specific,
the issue itself is architecture independent and can easily be reproduced for
example on x86-64 by playing with ulimit:

$ cat > test.c << EOF
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>

int main()
{
    errno = 0;
    void *data = malloc(32*1024*1024);
    int saved_errno = errno;
    printf("errno = %i\n", saved_errno);
}
EOF
$ gcc test.c
$ ./a.out
errno = 0
$ (ulimit -d 1024 ; ./a.out)
errno = 12

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