[pulseaudio-discuss] [PATCH] shm: Don't force the shm files to be read-only

[David Henningsson]

On 04/09/2013 10:32 AM, Tanu Kaskinen wrote:
> Forcing the shm file to be read-only makes shm_unlink() fail on OS X.
> Thanks to Albert Zeyer for finding the bug.
>
> BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=62988
> ---
>   src/pulsecore/shm.c |    3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/src/pulsecore/shm.c b/src/pulsecore/shm.c
> index 9f44df3..8aa34fa 100644
> --- a/src/pulsecore/shm.c
> +++ b/src/pulsecore/shm.c
> @@ -111,6 +111,7 @@ int pa_shm_create_rw(pa_shm *m, size_t size, pa_bool_t shared, mode_t mode) {
>       pa_assert(m);
>       pa_assert(size > 0);
>       pa_assert(size <= MAX_SHM_SIZE);
> +    pa_assert(!(mode & ~0777));
>       pa_assert(mode >= 0600);
>
>       /* Each time we create a new SHM area, let's first drop all stale
> @@ -151,7 +152,7 @@ int pa_shm_create_rw(pa_shm *m, size_t size, pa_bool_t shared, mode_t mode) {
>           pa_random(&m->id, sizeof(m->id));
>           segment_name(fn, sizeof(fn), m->id);
>
> -        if ((fd = shm_open(fn, O_RDWR|O_CREAT|O_EXCL, mode & 0444)) < 0) {
> +        if ((fd = shm_open(fn, O_RDWR|O_CREAT|O_EXCL, mode)) < 0) {
>               pa_log("shm_open() failed: %s", pa_cstrerror(errno));
>               goto fail;
>           }
>

Ack.

I would probably have done "mode & 0744" instead of skipping the mask 
completely, but it's only called with the 0700 argument, so it doesn't 
matter.

-- 
David Henningsson, Canonical Ltd.
https://launchpad.net/~diwic