libpthread: Implement sem_trywait() and sem_getvalue()

src/untrusted/pthread/nc_semaphore.c

Index: src/untrusted/pthread/nc_semaphore.c
diff --git a/src/untrusted/pthread/nc_semaphore.c b/src/untrusted/pthread/nc_semaphore.c
index 84a31f78f94443929f11709c63cc651070d95cf9..346e928b7c6dd47637fdff3d28c291855aaa5014 100644
--- a/src/untrusted/pthread/nc_semaphore.c
+++ b/src/untrusted/pthread/nc_semaphore.c
@@ -103,6 +103,19 @@ int sem_wait(sem_t *sem) {
   return 0;
 }
 
+int sem_trywait(sem_t *sem) {
+  if (NULL == sem) {

[Roland McGrath, 2013/07/12 21:15:58]

Failing gracefully rather than crashing when passed a bogus sem_t is arguably an
anti-feature.  POSIX makes EINVAL for this case a "may fail" error (vs "shall
fail"), so no conforming application is entitled to rely on it.  Passing a null
pointer here is undefined behavior.  The existing functions diagnosed EINVAL
already and that was historical behavior and so I didn't argue for their
replacement implementations behaving differently.

But these are new functions and there is no possibility of preexisting
applications being broken by a recompile.  So I don't think we should diagnose
EINVAL for null pointers.

Note that glibc doesn't diagnose EINVAL for these cases; it will just crash.

[Mark Seaborn, 2013/07/15 15:51:21]

OK, I've removed the NULL checks.

The NULL check was to be consistent with the existing code, but I've removed
those other NULL checks in a separate change
(https://codereview.chromium.org/18711004/).

+    errno = EINVAL;
+    return -1;
+  }
+
+  if (decrement_if_positive(&sem->count))
+    return 0;
+
+  errno = EAGAIN;
+  return -1;
+}
+
 int sem_post(sem_t *sem) {
   if (NULL == sem) {
     errno = EINVAL;
@@ -132,3 +145,13 @@ int sem_post(sem_t *sem) {
   }
   return 0;
 }
+
+int sem_getvalue(sem_t *sem, int *value) {
+  if (NULL == sem || NULL == value) {

[Roland McGrath, 2013/07/12 21:15:58]

See above about checking |sem| being a null pointer.
There is no excuse whatsoever for checking |value|.
Passing a null pointer is undefined behavior and the best thing to do is crash
(which doesn't require any extra effort to ensure).

[Mark Seaborn, 2013/07/15 15:51:21]

Done.

+    errno = EINVAL;
+    return -1;
+  }
+
+  *value = sem->count;
+  return 0;
+}