[musl] __pthread_mutex_unlock uninitialized value

[musl] __pthread_mutex_unlock uninitialized value

[Leonid Shamis]

[-- Attachment #1: Type: text/plain, Size: 298 bytes --]

https://github.com/bminor/musl/blob/54ca677983d47529bab8752315ac1a2b49888870/src/thread/pthread_mutex_unlock.c#L34

In the case where a mutex:
is one of PTHREAD_MUTEX_ERRORCHECK or PTHREAD_MUTEX_RECURSIVE
and PTHREAD_PRIO_INHERIT

an uninitialized value of 'old' is used to check whether to futex.

[-- Attachment #2: Type: text/html, Size: 553 bytes --]

Re: [musl] __pthread_mutex_unlock uninitialized value

[Rich Felker]

On Fri, Mar 27, 2020 at 10:52:58AM -0700, Leonid Shamis wrote:
> https://github.com/bminor/musl/blob/54ca677983d47529bab8752315ac1a2b49888870/src/thread/pthread_mutex_unlock.c#L34

BTW official git is here:

https://git.musl-libc.org/cgit/musl/tree/src/thread/pthread_mutex_unlock.c?id=v1.2.0

> In the case where a mutex:
> is one of PTHREAD_MUTEX_ERRORCHECK or PTHREAD_MUTEX_RECURSIVE
> and PTHREAD_PRIO_INHERIT
> 
> an uninitialized value of 'old' is used to check whether to futex.

Can you elaborate on this? In line 15, old is assigned; this applies
to all mutex types except plain boring normal (without PI and without
robust). The condition in line 33 can only be true if type is nonzero
(not plain boring normal mutex) so I don't see any way it can be used
uninitialized in line 34. Is your report based on your own reading or
a static analysis tool?

Rich

Re: [musl] __pthread_mutex_unlock uninitialized value

[Leonid Shamis]

[-- Attachment #1: Type: text/plain, Size: 1485 bytes --]

Apologies. First post. I'll use the official repo from now on.

This was originally brought to my attention via static analysis:

warning: ‘old’ may be used uninitialized in this function
[-Wmaybe-uninitialized]
   __asm__ __volatile__("lock ; cmpxchg %3, %1" : "=a"(t), "=m"(*p) :
"a"(t), "r"(s) : "memory");
   ^~~~~~~

And in my reading, I thought the (type != PTHREAD_MUTEX_NORMAL) only
checked the bottom three bits.

Please disregard this email chain :)

On Fri, Mar 27, 2020 at 11:16 AM Rich Felker <dalias@libc.org> wrote:

> On Fri, Mar 27, 2020 at 10:52:58AM -0700, Leonid Shamis wrote:
> >
> https://github.com/bminor/musl/blob/54ca677983d47529bab8752315ac1a2b49888870/src/thread/pthread_mutex_unlock.c#L34
>
> BTW official git is here:
>
>
> https://git.musl-libc.org/cgit/musl/tree/src/thread/pthread_mutex_unlock.c?id=v1.2.0
>
> > In the case where a mutex:
> > is one of PTHREAD_MUTEX_ERRORCHECK or PTHREAD_MUTEX_RECURSIVE
> > and PTHREAD_PRIO_INHERIT
> >
> > an uninitialized value of 'old' is used to check whether to futex.
>
> Can you elaborate on this? In line 15, old is assigned; this applies
> to all mutex types except plain boring normal (without PI and without
> robust). The condition in line 33 can only be true if type is nonzero
> (not plain boring normal mutex) so I don't see any way it can be used
> uninitialized in line 34. Is your report based on your own reading or
> a static analysis tool?
>
> Rich
>

[-- Attachment #2: Type: text/html, Size: 2282 bytes --]