diff mbox series

Fix stringop-overflow warning in bug-regex19.c.

Message ID 20210512063345.2269779-1-stli@linux.ibm.com
State Superseded
Headers show
Series Fix stringop-overflow warning in bug-regex19.c. | expand

Checks

Context Check Description
dj/TryBot-apply_patch success Patch applied to master at the time it was sent

Commit Message

Stefan Liebler May 12, 2021, 6:33 a.m. UTC
Starting with commit
26492c0a14966c32c43cd6ca1d0dca5e62c6cfef
"Annotate additional APIs with GCC attribute access.",
gcc emits this warning on s390x:
In function ‘do_one_test’,
    inlined from ‘do_mb_tests’ at bug-regex19.c:385:11:
bug-regex19.c:271:9: error: ‘re_search’ specified size 18446744073709551615 exceeds maximum object size 9223372036854775807 [-Werror=stringop-overflow=]
  271 |   res = re_search (&regbuf, test->string, strlen (test->string),
      |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  272 |      test->start, strlen (test->string) - test->start, NULL);
      |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from ../include/regex.h:2,
                 from bug-regex19.c:22:
bug-regex19.c: In function ‘do_mb_tests’:
../posix/regex.h:554:17: note: in a call to function ‘re_search’ declared with attribute ‘read_only (2, 3)’
  554 | extern regoff_t re_search (struct re_pattern_buffer *__buffer,
      |                 ^~~~~~~~~
...

The function do_one_test is inlined into do_mb_tests on s390x (at least with
gcc 10).  If do_one_test is marked with __attribute__ ((noinline)), there are
no warnings on s390x. If do_one_test is marked with
__attribute__ ((always_inline)), there are the same warnings on x86_64.

test->string points to a variable length array on stack of do_mb_tests
and the content is generated based on the passed test struct.
---
 posix/bug-regex19.c | 1 +
 1 file changed, 1 insertion(+)

Comments

Martin Sebor May 12, 2021, 3:56 p.m. UTC | #1
On 5/12/21 12:33 AM, Stefan Liebler via Libc-alpha wrote:
> Starting with commit
> 26492c0a14966c32c43cd6ca1d0dca5e62c6cfef
> "Annotate additional APIs with GCC attribute access.",
> gcc emits this warning on s390x:
> In function ‘do_one_test’,
>      inlined from ‘do_mb_tests’ at bug-regex19.c:385:11:
> bug-regex19.c:271:9: error: ‘re_search’ specified size 18446744073709551615 exceeds maximum object size 9223372036854775807 [-Werror=stringop-overflow=]
>    271 |   res = re_search (&regbuf, test->string, strlen (test->string),
>        |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>    272 |      test->start, strlen (test->string) - test->start, NULL);
>        |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> In file included from ../include/regex.h:2,
>                   from bug-regex19.c:22:
> bug-regex19.c: In function ‘do_mb_tests’:
> ../posix/regex.h:554:17: note: in a call to function ‘re_search’ declared with attribute ‘read_only (2, 3)’
>    554 | extern regoff_t re_search (struct re_pattern_buffer *__buffer,
>        |                 ^~~~~~~~~
> ...
> 
> The function do_one_test is inlined into do_mb_tests on s390x (at least with
> gcc 10).  If do_one_test is marked with __attribute__ ((noinline)), there are
> no warnings on s390x. If do_one_test is marked with
> __attribute__ ((always_inline)), there are the same warnings on x86_64.
> 
> test->string points to a variable length array on stack of do_mb_tests
> and the content is generated based on the passed test struct.

This is a false positive caused by the same bug as the one in
nss/makedb.c.  It's fixed in GCC 11 but the whole change is too
intrusive to backport to 10.  I'll see if I can extract just
the fix itself and backport it.

Disabling inlining for the test function as a workaround seems
reasonable to me (a comment should probably be added mentioning
why it's being done).  An alternative is to suppress the warning
using #pragma GCC diagnostic (as was done in nss/makedb.c).

Martin


> ---
>   posix/bug-regex19.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/posix/bug-regex19.c b/posix/bug-regex19.c
> index 9bbffb17e3..fcae533762 100644
> --- a/posix/bug-regex19.c
> +++ b/posix/bug-regex19.c
> @@ -251,6 +251,7 @@ static struct test_s
>   };
>   
>   int
> +__attribute__ ((noinline))
>   do_one_test (const struct test_s *test, const char *fail)
>   {
>     int res;
>
Martin Sebor May 13, 2021, 5:04 p.m. UTC | #2
On 5/12/21 9:56 AM, Martin Sebor wrote:
> On 5/12/21 12:33 AM, Stefan Liebler via Libc-alpha wrote:
>> Starting with commit
>> 26492c0a14966c32c43cd6ca1d0dca5e62c6cfef
>> "Annotate additional APIs with GCC attribute access.",
>> gcc emits this warning on s390x:
>> In function ‘do_one_test’,
>>      inlined from ‘do_mb_tests’ at bug-regex19.c:385:11:
>> bug-regex19.c:271:9: error: ‘re_search’ specified size 
>> 18446744073709551615 exceeds maximum object size 9223372036854775807 
>> [-Werror=stringop-overflow=]
>>    271 |   res = re_search (&regbuf, test->string, strlen (test->string),
>>        |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>    272 |      test->start, strlen (test->string) - test->start, NULL);
>>        |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> In file included from ../include/regex.h:2,
>>                   from bug-regex19.c:22:
>> bug-regex19.c: In function ‘do_mb_tests’:
>> ../posix/regex.h:554:17: note: in a call to function ‘re_search’ 
>> declared with attribute ‘read_only (2, 3)’
>>    554 | extern regoff_t re_search (struct re_pattern_buffer *__buffer,
>>        |                 ^~~~~~~~~
>> ...
>>
>> The function do_one_test is inlined into do_mb_tests on s390x (at 
>> least with
>> gcc 10).  If do_one_test is marked with __attribute__ ((noinline)), 
>> there are
>> no warnings on s390x. If do_one_test is marked with
>> __attribute__ ((always_inline)), there are the same warnings on x86_64.
>>
>> test->string points to a variable length array on stack of do_mb_tests
>> and the content is generated based on the passed test struct.
> 
> This is a false positive caused by the same bug as the one in
> nss/makedb.c.  It's fixed in GCC 11 but the whole change is too
> intrusive to backport to 10.  I'll see if I can extract just
> the fix itself and backport it.

I did that in r10-9819.  The fix should be in GCC 10.4 if/when it's
released.

Martin

> 
> Disabling inlining for the test function as a workaround seems
> reasonable to me (a comment should probably be added mentioning
> why it's being done).  An alternative is to suppress the warning
> using #pragma GCC diagnostic (as was done in nss/makedb.c).
> 
> Martin
> 
> 
>> ---
>>   posix/bug-regex19.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/posix/bug-regex19.c b/posix/bug-regex19.c
>> index 9bbffb17e3..fcae533762 100644
>> --- a/posix/bug-regex19.c
>> +++ b/posix/bug-regex19.c
>> @@ -251,6 +251,7 @@ static struct test_s
>>   };
>>   int
>> +__attribute__ ((noinline))
>>   do_one_test (const struct test_s *test, const char *fail)
>>   {
>>     int res;
>>
>
Stefan Liebler May 17, 2021, 2:23 p.m. UTC | #3
On 12/05/2021 17:56, Martin Sebor wrote:
> 
> This is a false positive caused by the same bug as the one in
> nss/makedb.c.  It's fixed in GCC 11 but the whole change is too
> intrusive to backport to 10.  I'll see if I can extract just
> the fix itself and backport it.
> 
> Disabling inlining for the test function as a workaround seems
> reasonable to me (a comment should probably be added mentioning
> why it's being done).  An alternative is to suppress the warning
> using #pragma GCC diagnostic (as was done in nss/makedb.c).
> 
> Martin

Thanks for the hint. Then I prefer to use the same approach as for
nss/makedb.c.
Please have a look at V2:
https://sourceware.org/pipermail/libc-alpha/2021-May/126413.html

Thanks,
Stefan
Martin Sebor May 17, 2021, 3:22 p.m. UTC | #4
On 5/17/21 8:23 AM, Stefan Liebler wrote:
> On 12/05/2021 17:56, Martin Sebor wrote:
>>
>> This is a false positive caused by the same bug as the one in
>> nss/makedb.c.  It's fixed in GCC 11 but the whole change is too
>> intrusive to backport to 10.  I'll see if I can extract just
>> the fix itself and backport it.
>>
>> Disabling inlining for the test function as a workaround seems
>> reasonable to me (a comment should probably be added mentioning
>> why it's being done).  An alternative is to suppress the warning
>> using #pragma GCC diagnostic (as was done in nss/makedb.c).
>>
>> Martin
> 
> Thanks for the hint. Then I prefer to use the same approach as for
> nss/makedb.c.
> Please have a look at V2:
> https://sourceware.org/pipermail/libc-alpha/2021-May/126413.html

That looks good to me.

Martin

> 
> Thanks,
> Stefan
>
diff mbox series

Patch

diff --git a/posix/bug-regex19.c b/posix/bug-regex19.c
index 9bbffb17e3..fcae533762 100644
--- a/posix/bug-regex19.c
+++ b/posix/bug-regex19.c
@@ -251,6 +251,7 @@  static struct test_s
 };
 
 int
+__attribute__ ((noinline))
 do_one_test (const struct test_s *test, const char *fail)
 {
   int res;