This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Linux: Use mmap instead of malloc in dirent/tst-getdents64
* Adhemerval Zanella:
> On 28/06/2019 05:49, Florian Weimer wrote:
>> malloc dirties the entire allocated memory region due to M_PERTURB
>> in the test harness.
>>
>> 2019-06-28 Florian Weimer <fweimer@redhat.com>
>>
>> * sysdeps/unix/sysv/linux/tst-getdents64.c (large_buffer_checks):
>> Use mmap instead of malloc. malloc with M_PERTURB writes to the
>> entire allocated memory range.
>
> LGTM.
>
> Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Thanks!
>> diff --git a/sysdeps/unix/sysv/linux/tst-getdents64.c b/sysdeps/unix/sysv/linux/tst-getdents64.c
>> index 24e77e04d8..8a28e6c67c 100644
>> --- a/sysdeps/unix/sysv/linux/tst-getdents64.c
>> +++ b/sysdeps/unix/sysv/linux/tst-getdents64.c
>> @@ -27,6 +27,7 @@
>> #include <support/check.h>
>> #include <support/support.h>
>> #include <support/xunistd.h>
>> +#include <sys/mman.h>
>> #include <unistd.h>
>>
>> /* Called by large_buffer_checks below. */
>> @@ -53,8 +54,13 @@ large_buffer_checks (int fd)
>> size_t large_buffer_size;
>> if (!__builtin_add_overflow (UINT_MAX, 2, &large_buffer_size))
>> {
>> - char *large_buffer = malloc (large_buffer_size);
>> - if (large_buffer == NULL)
>> + int flags = MAP_ANONYMOUS | MAP_PRIVATE;
>> +#ifdef MAP_NORESERVE
>> + flags |= MAP_NORESERVE;
>> +#endif
>> + void *large_buffer = mmap (NULL, large_buffer_size,
>> + PROT_READ | PROT_WRITE, flags, -1, 0);
>> + if (large_buffer == MAP_FAILED)
>
> Should we really skip the test instead of using xmmap? The only case I think of
> that it might fail is if kernel can't allocate bookeeping memory for the page
> table itself, but I really think it unlikely (sanitizer usually allocates a
> lot of VMA as well).
MAP_NORESERVE is a misnomer; it does not do what it says. It disables
the overcommit heuristics for vm.overcommit_memory=0, but not for
vm.overcommit_memory=2. mmap can also fail due to resource limits on
address space, not residential memory. So I think the code above is the
best we can do.
Thanks,
Florian