v4l2 device won't open after opening/closing it 16 times

[tmatth]

Summary of the bug:

In the attached case, I open and close a v4l2 device (/dev/video0) 17 times. On the 17th time, it fails to open with the following errors:

libv4l2: error attempting to open more then 16 video devices
[video4linux2,v4l2 @ 0x1f6a5c0] ioctl(VIDIOC_QUERYCAP): Bad file descriptor
Could not open input "/dev/video0"

How to reproduce:
Build and run the attached test case with a v4l2 device connected.

I'm using the ffmpeg git master branch:

ffmpeg version N-42636-ga8d8e86 Copyright (c) 2000-2012 the FFmpeg developers
  built on Jul 19 2012 17:32:59 with gcc 4.7.0 (GCC) 20120507 (Red Hat 4.7.0-5)
  configuration: --prefix=/opt/ --enable-bzlib --enable-gnutls --enable-libv4l2 --enable-libx264 --enable-x11grab --enable-postproc --enable-pthreads --disable-static --enable-shared --enable-gpl --disable-debug --disable-stripping --enable-pic

[tmatth]

compile with "cc v4l2_testcase.c -o v4l2_testcase pkg-config --cflags --libs libavdevice libavformat"

[tmatth]

I've also realized that this bug is only present when the --enable-libv4l2 option is given to configure.

[Cigaes]

Same thing on one of my boxes. libavdevice compiled without libv4l2 works.
Try removing the paragraph under #ifdef CONFIG_LIBV4L2 in device_open() in libavdevice/v4l2.c and see if it fixes things.

[tmatth]

Yup that did it, I'm not sure how the code should be adjusted though, should we only be calling v4l2_fd_open if fd is < 0? If so, I've attached a patch.

[tmatth]

Earlier patch was no good, I've replaced it with this patch which just fixes the error handling. The code currently in master is off, fd would never be < 0 since the function would have already returned.

The original problem still stands. From the documentation it seems that v4l2_fd_open takes an already open file descriptor and tries to make it open for further use by libv4l2, but I can't see why it's not being closed properly.

[tmatth]

fix error handling for v4l2

[Cigaes]

Your patch would clearly be right by itself, but the whole paragraph is just redundant:
 http://git.linuxtv.org/v4l-utils.git/blob/HEAD:/lib/libv4l2/libv4l2.c#l578
v4l2_open already calls v4l2_fd_open, calling a second time only uses an additional slot in libv4l2 structures that will never be deallocated.
I sent a patch to the devel mailing list.

[tmatth]

Replying to Cigaes:

Your patch would clearly be right by itself, but the whole paragraph is just redundant:
 http://git.linuxtv.org/v4l-utils.git/blob/HEAD:/lib/libv4l2/libv4l2.c#l578
v4l2_open already calls v4l2_fd_open, calling a second time only uses an additional slot in libv4l2 structures that will never be deallocated.
I sent a patch to the devel mailing list.

Well spotted, it wasn't until I went through with gdb that I noticed v4l2_fd_open was getting called twice. Thanks for sending the patch.