[PATCH] pthread_sigmask: check 'how' only when 'set' is not NULL

[PATCH] pthread_sigmask: check 'how' only when 'set' is not NULL

[Yousong Zhou]

According to POSIX document

    If set is a null pointer, the value of the argument how is not
    significant and the thread's signal mask shall be unchanged; thus
    the call can be used to enquire about currently blocked signals.

This is also how the current Linux kernel syscall is doing.  So the
following function call from binutils-gdb should not fail

    sigprocmask (0,  NULL, &original_signal_mask);
---
 src/thread/pthread_sigmask.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/thread/pthread_sigmask.c b/src/thread/pthread_sigmask.c
index 88c333f..f188782 100644
--- a/src/thread/pthread_sigmask.c
+++ b/src/thread/pthread_sigmask.c
@@ -5,7 +5,7 @@
 int pthread_sigmask(int how, const sigset_t *restrict set, sigset_t *restrict old)
 {
 	int ret;
-	if ((unsigned)how - SIG_BLOCK > 2U) return EINVAL;
+	if (set && (unsigned)how - SIG_BLOCK > 2U) return EINVAL;
 	ret = -__syscall(SYS_rt_sigprocmask, how, set, old, _NSIG/8);
 	if (!ret && old) {
 		if (sizeof old->__bits[0] == 8) {
-- 
2.6.4

Re: [PATCH] pthread_sigmask: check 'how' only when 'set' is not NULL

[Rich Felker]

On Wed, Mar 22, 2017 at 08:19:43PM +0800, Yousong Zhou wrote:
> According to POSIX document
> 
>     If set is a null pointer, the value of the argument how is not
>     significant and the thread's signal mask shall be unchanged; thus
>     the call can be used to enquire about currently blocked signals.
> 
> This is also how the current Linux kernel syscall is doing.  So the
> following function call from binutils-gdb should not fail
> 
>     sigprocmask (0,  NULL, &original_signal_mask);
> ---
>  src/thread/pthread_sigmask.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/thread/pthread_sigmask.c b/src/thread/pthread_sigmask.c
> index 88c333f..f188782 100644
> --- a/src/thread/pthread_sigmask.c
> +++ b/src/thread/pthread_sigmask.c
> @@ -5,7 +5,7 @@
>  int pthread_sigmask(int how, const sigset_t *restrict set, sigset_t *restrict old)
>  {
>  	int ret;
> -	if ((unsigned)how - SIG_BLOCK > 2U) return EINVAL;
> +	if (set && (unsigned)how - SIG_BLOCK > 2U) return EINVAL;
>  	ret = -__syscall(SYS_rt_sigprocmask, how, set, old, _NSIG/8);
>  	if (!ret && old) {
>  		if (sizeof old->__bits[0] == 8) {
> -- 
> 2.6.4

I don't think this change is conforming. There is a requirement to
produce an error ("shall fail") independent of whether the set
argument is a null pointer:

    The pthread_sigmask() and sigprocmask() functions shall fail if:

    [EINVAL]
        The value of the how argument is not equal to one of the defined
        values.

http://pubs.opengroup.org/onlinepubs/9699919799/functions/pthread_sigmask.html

If gdb is calling it with an invalid argument (rather, one that's
conditionally invalid depending on the platform's definitions of the
how macros), a patch should be sent to gdb to fix it.

Rich

Re: [PATCH] pthread_sigmask: check 'how' only when 'set' is not NULL

[Yousong Zhou]

On 22 March 2017 at 20:49, Rich Felker <dalias@libc.org> wrote:
> On Wed, Mar 22, 2017 at 08:19:43PM +0800, Yousong Zhou wrote:
>> According to POSIX document
>>
>>     If set is a null pointer, the value of the argument how is not
>>     significant and the thread's signal mask shall be unchanged; thus
>>     the call can be used to enquire about currently blocked signals.
>>
>> This is also how the current Linux kernel syscall is doing.  So the
>> following function call from binutils-gdb should not fail
>>
>>     sigprocmask (0,  NULL, &original_signal_mask);
>> ---
>>  src/thread/pthread_sigmask.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/src/thread/pthread_sigmask.c b/src/thread/pthread_sigmask.c
>> index 88c333f..f188782 100644
>> --- a/src/thread/pthread_sigmask.c
>> +++ b/src/thread/pthread_sigmask.c
>> @@ -5,7 +5,7 @@
>>  int pthread_sigmask(int how, const sigset_t *restrict set, sigset_t *restrict old)
>>  {
>>       int ret;
>> -     if ((unsigned)how - SIG_BLOCK > 2U) return EINVAL;
>> +     if (set && (unsigned)how - SIG_BLOCK > 2U) return EINVAL;
>>       ret = -__syscall(SYS_rt_sigprocmask, how, set, old, _NSIG/8);
>>       if (!ret && old) {
>>               if (sizeof old->__bits[0] == 8) {
>> --
>> 2.6.4
>
> I don't think this change is conforming. There is a requirement to
> produce an error ("shall fail") independent of whether the set
> argument is a null pointer:
>
>     The pthread_sigmask() and sigprocmask() functions shall fail if:
>
>     [EINVAL]
>         The value of the how argument is not equal to one of the defined
>         values.
>
> http://pubs.opengroup.org/onlinepubs/9699919799/functions/pthread_sigmask.html
>
> If gdb is calling it with an invalid argument (rather, one that's
> conditionally invalid depending on the platform's definitions of the
> how macros), a patch should be sent to gdb to fix it.
>
> Rich

I think you are right in the interpretation of POSIX document on this
topic.  I missed the "shall fail" part in ERRORS section.

But it came to me that darwin was thought to be a certificated
POSIX-compliant environment, so I just ran tested that pthread_sigmask
call on my osx 10.11.6 and it worked without error.

  https://en.wikipedia.org/wiki/Darwin_(operating_system)

I will patch gdb anyway, to be correct.


Regards,
               yousong