Skip to contentSkip to conversation

aplay: fix lurking capture file overwrite bug

PATCH Applied David Fries 3 participants 4 messages Apr 14, 2016 – Apr 18, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 aplay: fix lurking capture file overwrite bug

If -d was given to arecord while commit 8aa13eec80eac312e4b99423909387660fb99b8f (now reverted) was in effect, the last read would be shorter than the chunk size, but pcm_read would read and return the chunk size, the samples were discarded, and capture() continued in a loop because count never reached 0. arecord opens a new file each loop iteration, if arecord is dynamically naming files, --use-strftime option or beyond the wave 2GB limit, this will generate a series of header only wave files. If the file is unique the originally recorded data is lost and it will continue overwriting the same file with a header only wave file.

While the current pcm_read can't fail (it can exit), it is better to just fix this lurking bug in case it is "fixed" again.
---
In my case I wanted to record for 16 hours, I did get the audio, but I also found I had 3.1 million 44 byte files, which took the system 3 hours to unlink, I'm glad I at least discovered this problem the next day.. Debian has the effectively broken pcm_read, after figuring out what was going wrong, and that the current arecord doesn't exhibit the problem I wanted to fix it anyway. There are remaining problems, pcm_read will read chunk_size, but if the requested size is smaller samples are lost, I'm not trying to fix that here. This will at least gracefully close a wave file (update the counts) if read or write returns a size different from requested.

 aplay/aplay.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
aplay/aplay.c +6 -3
index 7acaa83..2da7dda 100644
@@ -3067,11 +3067,14 @@ static void capture(char *orig_name)
 			size_t c = (rest <= (off64_t)chunk_bytes) ?
 				(size_t)rest : chunk_bytes;
 			size_t f = c * 8 / bits_per_frame;
-			if (pcm_read(audiobuf, f) != f)
+			if (pcm_read(audiobuf, f) != f) {
+				in_aborting = 1;
 				break;
+			}
 			if (write(fd, audiobuf, c) != c) {
 				perror(name);
-				prg_exit(EXIT_FAILURE);
+				in_aborting = 1;
+				break;
 			}
 			count -= c;
 			rest -= c;
@@ -3091,7 +3094,7 @@ static void capture(char *orig_name)
 		}
 
 		if (in_aborting)
-			break;
+			prg_exit(EXIT_FAILURE);
 
 		/* repeat the loop when format is raw without timelimit or
 		 * requested counts of data are recorded
Signature · 1 line
2.1.4
Applied
#1 lore raw
Quoted 12 lines · David Fries

If -d was given to arecord while commit 8aa13eec80eac312e4b99423909387660fb99b8f (now reverted) was in effect, the last read would be shorter than the chunk size, but pcm_read would read and return the chunk size, the samples were discarded, and capture() continued in a loop because count never reached 0. arecord opens a new file each loop iteration, if arecord is dynamically naming files, --use-strftime option or beyond the wave 2GB limit, this will generate a series of header only wave files. If the file is unique the originally recorded data is lost and it will continue overwriting the same file with a header only wave file.

Quoted 13 lines

While the current pcm_read can't fail (it can exit), it is better to just fix this lurking bug in case it is "fixed" again.
---
In my case I wanted to record for 16 hours, I did get the audio, but I also found I had 3.1 million 44 byte files, which took the system 3 hours to unlink, I'm glad I at least discovered this problem the next day.. Debian has the effectively broken pcm_read, after figuring out what was going wrong, and that the current arecord doesn't exhibit the problem I wanted to fix it anyway. There are remaining problems, pcm_read will read chunk_size, but if the requested size is smaller samples are lost, I'm not trying to fix that here. This will at least gracefully close a wave file (update the counts) if read or write returns a size different from requested.

Applied, thanks.

Takashi

Quoted 42 lines
 aplay/aplay.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
aplay/aplay.c +6 -3
index 7acaa83..2da7dda 100644
@@ -3067,11 +3067,14 @@ static void capture(char *orig_name)
 			size_t c = (rest <= (off64_t)chunk_bytes) ?
 				(size_t)rest : chunk_bytes;
 			size_t f = c * 8 / bits_per_frame;
-			if (pcm_read(audiobuf, f) != f)
+			if (pcm_read(audiobuf, f) != f) {
+				in_aborting = 1;
 				break;
+			}
 			if (write(fd, audiobuf, c) != c) {
 				perror(name);
-				prg_exit(EXIT_FAILURE);
+				in_aborting = 1;
+				break;
 			}
 			count -= c;
 			rest -= c;
@@ -3091,7 +3094,7 @@ static void capture(char *orig_name)
 		}
 
 		if (in_aborting)
-			break;
+			prg_exit(EXIT_FAILURE);
 
 		/* repeat the loop when format is raw without timelimit or
 		 * requested counts of data are recorded
Signature · 6 lines
2.1.4

_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
#2 lore raw
In reply to David Fries
Quoted 8 lines · David Fries

If -d was given to arecord while commit 8aa13eec80eac312e4b99423909387660fb99b8f (now reverted) was in effect, the last read would be shorter than the chunk size, but pcm_read would read and return the chunk size, the samples were discarded, and capture() continued in a loop because count never reached 0. arecord opens a new file each loop iteration, if arecord is dynamically naming files, --use-strftime option or beyond the wave 2GB limit, this will generate a series of header only wave files.

<snip>

Quoted 1 line

Debian has the effectively broken pcm_read,

It is fortunate for me that you mentioned those details as I have just been bitten by exactly that, using arecord whilst tracking down a Portaudio issue in Debian Jessie! I was gearing up to report the bug, assuming it was in the capture system.

It seems a pity that Debian Jessie stable doesn't have a backport of the fix/revert.

Now back to tracing the original issue...

Regards

Alan

#3 lore raw
Quoted 14 lines · Alan Horstmann
Quoted 8 lines · David Fries

If -d was given to arecord while commit 8aa13eec80eac312e4b99423909387660fb99b8f (now reverted) was in effect, the last read would be shorter than the chunk size, but pcm_read would read and return the chunk size, the samples were discarded, and capture() continued in a loop because count never reached 0. arecord opens a new file each loop iteration, if arecord is dynamically naming files, --use-strftime option or beyond the wave 2GB limit, this will generate a series of header only wave files.

<snip>

Quoted 1 line

Debian has the effectively broken pcm_read,

Quoted 7 lines

It is fortunate for me that you mentioned those details as I have just been bitten by exactly that, using arecord whilst tracking down a Portaudio issue in Debian Jessie! I was gearing up to report the bug, assuming it was in the capture system.

It seems a pity that Debian Jessie stable doesn't have a backport of the fix/revert.

Glad to hear I helped.

I filed a Debian bug report with the two patches that would avoid it as Bug#821048, but it isn't going to be fixed. Denial of service is usually seen as a security issue, and this bug could run the system out of inodes, I guess it isn't seen as a high enough priority. I was able to download and compile 1.1.0-2 on Jessie which doesn't have this problem.

----------------
Please update to alsa-utils 1.1.0-2 first. In stable there will be only security patches accepted. Bug closed herewith. Backports are not maintained from us.

Elimar
----------------

Quoted 9 lines

Now back to tracing the original issue...

Regards

Alan
_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

Signature · 1 line
David Fries <[email protected]>