576467 - return value from clock_getres should be checked

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[timeless]

Most bugs I've filed based on Coverity rely on information Coverity gains from Mozilla's codebase. This is an exception. Here, Coverity was run against a larger system and has determined that everyone else (dbus, glibc, gstreamer, perl, and pulseaudio) checks it.

Unchecked call
A function was called without checking the return value. Prevent inferred that the return value should be checked based on the checked calls shown below.

Called function "clock_getres" whose return value should be checked (checked 6 out of 7 times)

xulrunner/xpcom/ds/TimeStamp_posix.cpp
112  	    // measurable resolution is either incredibly low, ~1ns, or very
113  	    // high.  fall back on clock_getres()
114  	    struct timespec ts;
Event check_return: Called function "clock_getres" whose return value should be checked (checked 6 out of 7 times)
Event unchecked_value: Return value of "clock_getres(1, &ts)" is not checked
115  	    clock_getres(CLOCK_MONOTONIC, &ts);
116  	
117  	    minres = TimespecToNs(ts);
118  	  }

dbus/dbus-sysdeps-pthread.c
350  	{
351  	#ifdef HAVE_MONOTONIC_CLOCK
352  	  struct timespec dummy;
353  	  if (clock_getres (CLOCK_MONOTONIC, &dummy) == 0)
354  	    have_monotonic_clock = TRUE;
355  	#endif
356  	}

glibc/nscd/connections.c
1835 	# if _POSIX_CLOCK_SELECTION == 0
1836 	    if (sysconf (_SC_CLOCK_SELECTION) > 0)
1837 	# endif
1838 	      if (clock_getres (CLOCK_MONOTONIC, &dummy) == 0
1839 		  && pthread_condattr_setclock (&condattr, CLOCK_MONOTONIC) == 0)
1840 		timeout_clock = CLOCK_MONOTONIC;
1841 	#endif

gstreamer/gst/gstsystemclock.c
534  	
535  	    ptype = clock_type_to_posix_id (sysclock->priv->clock_type);
536  	
537  	    if (G_UNLIKELY (clock_getres (ptype, &ts)))
538  	      return GST_CLOCK_TIME_NONE;
539  	
540  	    return GST_TIMESPEC_TO_TIME (ts);

perl/ext/Time-HiRes/HiRes.xs
1144 	#ifdef TIME_HIRES_CLOCK_GETRES_SYSCALL
1145 		status = syscall(SYS_clock_getres, clock_id, &ts);
1146 	#else
1147 		status = clock_getres(clock_id, &ts);
1148 	#endif
1149 		RETVAL = status == 0 ? ts.tv_sec + (NV) ts.tv_nsec / (NV) 1e9 : -1;
1150 	

pulseaudio/src/pulsecore/core-rtclock.c
81   	    struct timespec ts;
82   	
83   	#ifdef CLOCK_MONOTONIC
84   	    if (clock_getres(CLOCK_MONOTONIC, &ts) >= 0)
85   	        return ts.tv_sec == 0 && ts.tv_nsec <= PA_HRTIMER_THRESHOLD_USEC*1000;
86   	#endif
87   	
88          pa_assert_se(clock_getres(CLOCK_REALTIME, &ts) == 0);
89          return ts.tv_sec == 0 && ts.tv_nsec <= PA_HRTIMER_THRESHOLD_USEC*1000;

Comment 1

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Check the return value of clock_getres()

FWIW clock_getres() on Linux is infallible if the specified clock exists, which is already checked by the time we call clock_getres().  But we don't gain anything by not checking the return value, and I'm all for making coverity happy.

Comment 2

[timeless]

yeah, in a previous run through coverity i checked this code and decided it wasn't interesting. but obviously that's a losing proposition since each coverity run user wastes time doing that.

thanks for agreeing to save scan readers :)

Comment 3

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

http://hg.mozilla.org/mozilla-central/rev/0fe45cf61aeb