Bug 28947
| Summary: | dup2 closing pthread file descriptor | ||
|---|---|---|---|
| Product: | Base System | Reporter: | sdalu <sdalu> |
| Component: | misc | Assignee: | Maxim Konovalov <maxim> |
| Status: | Closed FIXED | ||
| Severity: | Affects Only Me | ||
| Priority: | Normal | ||
| Version: | 4.2-RELEASE | ||
| Hardware: | Any | ||
| OS: | Any | ||
Responsible Changed From-To: freebsd-bugs->jasone I'll look at this. Could you please try the patch below:
Index: uthread_dup2.c
===================================================================
RCS file: /home/ncvs/src/lib/libc_r/uthread/uthread_dup2.c,v
retrieving revision 1.11
diff -u -r1.11 uthread_dup2.c
--- uthread_dup2.c 10 Apr 2001 04:19:20 -0000 1.11
+++ uthread_dup2.c 13 Feb 2002 18:12:40 -0000
@@ -45,7 +45,8 @@
int newfd_opened;
/* Check if the file descriptor is out of range: */
- if (newfd < 0 || newfd >= _thread_dtablesize) {
+ if (newfd < 0 || newfd >= _thread_dtablesize ||
+ newfd == _thread_kern_pipe[0] || newfd == _thread_kern_pipe[1]) {
/* Return a bad file descriptor error: */
errno = EBADF;
ret = -1;
--
Maxim Konovalov, MAcomnet, Internet-Intranet Dept., system engineer
phone: +7 (095) 796-9079, mailto:maxim@macomnet.ru
On Wed, Feb 13, 2002 at 09:22:34PM +0300, Maxim Konovalov wrote: > > Could you please try the patch below: Tested on FreeBSD 4.4-RELEASE i386. Works fine. Would it be reasonable to say that file descriptors internally used by the thread library should not be visible to programmer, and move them to other places when there is a clash with legitimate "syscall" requests like dup2? Not sure the effort is worth it anyway. > > Index: uthread_dup2.c > =================================================================== > RCS file: /home/ncvs/src/lib/libc_r/uthread/uthread_dup2.c,v > retrieving revision 1.11 > diff -u -r1.11 uthread_dup2.c > --- uthread_dup2.c 10 Apr 2001 04:19:20 -0000 1.11 > +++ uthread_dup2.c 13 Feb 2002 18:12:40 -0000 > @@ -45,7 +45,8 @@ > int newfd_opened; > > /* Check if the file descriptor is out of range: */ > - if (newfd < 0 || newfd >= _thread_dtablesize) { > + if (newfd < 0 || newfd >= _thread_dtablesize || > + newfd == _thread_kern_pipe[0] || newfd == _thread_kern_pipe[1]) { > /* Return a bad file descriptor error: */ > errno = EBADF; > ret = -1; > > -- > Maxim Konovalov, MAcomnet, Internet-Intranet Dept., system engineer > phone: +7 (095) 796-9079, mailto:maxim@macomnet.ru -- Stephane D'Alu Responsible Changed From-To: jasone->maxim I have a patch for -current. State Changed From-To: open->analyzed The fix is committed to -current. State Changed From-To: analyzed->closed The fix committed to -current and -stable. |
when using pthread 2 file descriptors are used for pipes (see lsof output), these file descriptors can't be closed by close(), but dup2() manage to duplicate other fd at these positions (fcntl/F_DUPFD cannot) the real problem occurs when doing an exec, when presumably the pthread is doing some cleanup and closes its files descriptor, unfortunatelly these descriptors are not its own anymore but the copy done but dup2. would have expected dup2() to fail (the problem is perhaps more with the pthread librairy than the kernel) How-To-Repeat: -- a ------------------------ #include <pthread.h> int main() { if (close(3) < 0) perror("close"); if (dup2(0, 3) < 0) perror("dup2"); } ----------------------------- gcc -o a a.c -pthread a 71332 sdalu 0u VCHR 5,14 0t4766 8311 /dev/ttype a 71332 sdalu 1u VCHR 5,14 0t4766 8311 /dev/ttype a 71332 sdalu 2u VCHR 5,14 0t4766 8311 /dev/ttype a 71332 sdalu 3u PIPE 0xd734f260 16384 ->0xd734f300 a 71332 sdalu 4u PIPE 0xd734f300 16384 ->0xd734f260 ./a close: Bad file descriptor