I didn't see anything obvious that "intended" to set the thread specific errno value based on system calls made on that thread's behalf (i.e., I would expect this in the syscall wrappers, perhaps..?), but I'm fairly new to this library so I'm not sure whether this is considered a bug, misconfiguration or simple lack of functionality.
For completeness, I compared the behavior with an _old_ glibc-2.2 system which looked like it had the same general design. The application worked in this case, but I noticed that the thread_self() call (while requesting the errno value in the thread, following the failed syscall) ended up returning a pointer to the __pthread_initial_thread structure rather than the specific thread structure I assume it created when it cloned (which is what happens in uClibc's thread_self()).
I also tested a quick change in uClibc by setting __pthread_manager_thread's p_errnop to '&_errno' as well as new threads created in pthread_handle_create(). This change allowed expected behavior, but it's probably not the appropriate fix. Any information on this is greatly appreciated, and let me know if I've been unclear or missed any important points. Thanks.
Brian
-------
#include <stdio.h>
#include <errno.h>
#include <pthread.h>
void *test(void *data);
void error_test();
int main()
{
pthread_t tid;
if (pthread_create(&tid, NULL, test, NULL) < 0)
printf("error creating thread\n");
pthread_join(tid, NULL);
error_test();
return 0;
}
void *test(void *data)
{
error_test();
}
void error_test()
{
if (send(23, NULL, 0) < 0) {
printf("send(): %d (%p): %s\n", errno, &errno, strerror(errno));
}
}
----------------------------------------------
main3.c [^] (983 bytes) 08-23-07 17:18 pthread-errno.patch [^] (18,038 bytes) 10-19-07 12:09 
