padsp's access wrapper doesn't check for NULL, breaks "padsp firefox"

[alon]

Hi,

Short summary: trying "padsp firefox" failed, gdb discovered that the failure was in access. I'll spare you the stacktrace - the crux of it is that pathname was ==NULL, and padsp failed in the call to strcmp.

Solution: this is the unified diff patch I used to fix it. After applying it I successfully used "padsp firefox" to play some music (using the mplayer embeddable plugin for firefox).

--- pulseaudio-0.9.5/src/utils/padsp.c 2006-08-19 00:29:26.000000000 +0300 +++ pulseaudio-0.9.5-alon/src/utils/padsp.c 2006-08-29 01:36:30.010620500 +0300 @@ -2122,7 +2122,12 @@

int access(const char *pathname, int mode) {

debug(DEBUG_LEVEL_VERBOSE, FILE": access(%s)\n", pathname);

- + if (pathname == NULL) { + // according to the man page from ubuntu the closest error condition: + // EFAULT pathname points outside your accessible address space + errno = EFAULT; + return -1; + }

if (strcmp(pathname, "/dev/dsp") != 0 &&

strcmp(pathname, "/dev/adsp") != 0 && strcmp(pathname, "/dev/sndstat") != 0 &&

[alon]

the patch

[alon]

Added the diff as an attachment since trac killed the indentation when the diff was put in the comment.

[alon]

I just checked with the latest svn sources, seems it isn't fixed there yet too, so I think this applies not only to 0.9.5 but to the latest as well.

[lennart]

(In [1352]) work around bug in firefox which apparently misuses access() as NULL pointer test. Original patch by "alon". (Closes #27)

[coling]

See #227 (200 bugs in the future ;)) for a possibly more generic fix for this issue.