pclose() hangs and never return in multi-threaded program

zxtole · 08-29-2011, 02:20 AM

hi guys,

i'm using popen() and redirection to generate a ".gz" file for ftp download like this:

Code:

FILE* testfd;
testfd = popen("/bin/gzip - -fq > /home/ftp/testfile.gz","w");
...
fprintf(testfd, "blablabla");
...
pclose(testfd);

it just worked fine for a few days, then i found something wield and the program would not response anymore:

Code:

#:/ps -ef
UID        PID  PPID  C STIME TTY          TIME CMD
....
root      1065  1064  0 Aug26 ?        00:00:00 myserver
....
root      1077  1065  0 Aug26 ?        00:00:00 sh -c /bin/gzip - -fq > /home/ftp/testfile.gz
root      1078  1077  0 Aug26 ?        00:00:00 /bin/gzip - -fq
....

the file /home/ftp/testfile.gz looks ok but its size is 0.

i try to look up into it and attach to PID 1065 with gdb, the backtrace comes out as below:

Code:

(gdb) bt
#0  0x40109588 in _IO_proc_close () from /lib/libc.so.6
#1  0x40110ec8 in _IO_file_close_it () from /lib/libc.so.6
#2  0x40107860 in fclose () from /lib/libc.so.6
#3  0x000f9a64 in generate_testfile ()
#4  0x00101ddc in handleMsgFromClient ()
#5  0x00102c94 in work_pthreadMain ()
#6  0x40056a74 in pthread_start_thread () from /lib/libpthread.so.0
#7  0x401776c8 in clone () from /lib/libc.so.6
#8  0x401776c8 in clone () from /lib/libc.so.6
Previous frame identical to this frame (corrupt stack?)
(gdb)

btw, "myserver" is a multi-threaded program using pthread library.

i have no idea why fclose() hangs here, and the child process would not exit as before.

anyone have any idea?

chrism01 · 08-29-2011, 06:25 PM

I'd recommend asking the Mods to move this to the Programming forum, using the Report button.

zxtole · 08-29-2011, 11:47 PM

Quote:

Originally Posted by chrism01

I'd recommend asking the Mods to move this to the Programming forum, using the Report button.

thx, chrism01. You mean "Report LQ Bug"? That's the only "report" button i could find on this page....

ta0kira · 08-30-2011, 08:04 AM

When it hangs, please ps the gzip process (using its pid) and post it. Specifically, the "STAT" column is of interest.
Kevin Barry

zxtole · 08-30-2011, 07:54 PM

Quote:

Originally Posted by ta0kira

When it hangs, please ps the gzip process (using its pid) and post it. Specifically, the "STAT" column is of interest.
Kevin Barry

The "STAT" column is "S" for these two "gzip" processes both, i've looked into it already. They were not zombies, definitely.

It seems that the "gzip" command was still waiting for an "EOF" from standard input, but "pclose()" hung during "fclose()" call and never returned.

Is that true?

ta0kira · 08-30-2011, 08:54 PM

Quote:

Originally Posted by zxtole

The "STAT" column is "S" for these two "gzip" processes both, i've looked into it already. They were not zombies, definitely.

It seems that the "gzip" command was still waiting for an "EOF" from standard input, but "pclose()" hung during "fclose()" call and never returned.

Is that true?

Yes, this is the interpretation I would have given.

I would strace the server program and pipe the output through egrep 'pipe|fork|close|wait' (remember strace writes to stderr). With a little work you should be able to make sure the server-side descriptor created by popen (with pipe) is closed (with pclose) before it waits for the shell process to exit. You might also make sure the other end of the pipe is closed by the server after forking.

If you can, run the server with strace just enough for the hang to happen. If it isn't too lengthy, post the strace output filtered through the egrep command above.
Kevin Barry

zxtole · 08-30-2011, 09:56 PM

ahh, thank you, ta0kira. That's a good advice, i'll have a try asap.