bamliquidator crashes after failed absurdly large memory allocation about pipeline HOT 3 CLOSED

I confirmed that this crash for smaller bins doesn't occur for other bams. There seems to be something special about this bam.

from pipeline.

I tracked this down to some internals of samtools while reading the region chr1:30000-60000. In function bam_read1 in bam.c, bam_read/bgzf_read is called to populate a block_len variable. The value is calculated to be about 1.4 billion, which is not a valid value. When the call to kroundup32 rounds it up to 2.8 billion it overflows (since it is a 32 bit signed integer) and it becomes -2147483648, then it is passed as an unsigned 64 bit size argument to realloc, which is interpreted as the absurdly large number 18446744071562067968. This of course fails, and returns a nullptr but samtools doesn't check for the error nullptr case. The block_len calculation of 1.4 billion in bgzf_read is the result of reading 4 bytes at a time from an apparently corrupt samtools structure.

If you run samtools view for this region a similar crash occurs:

jdm@tod:~/Gunk/tmp_large_alloc$ samtools view test.bam chr1:30000-60000
Segmentation fault (core dumped)

We are using samtools version 0.1.19-96b5f2294a from March 2013.

If we run this same command on a more recent samtools version 1.2-56-g5e6d7bd we get a more helpful error message:

samtools view ~/gunk/del_me/test.bam chr1:30000-60000
[main_samview] retrieval of region "chr1:30000-60000" failed due to truncated file or corrupt BAM index file

We might want to consider upgrading to the latest stable version of samtools, which would likely better handle this bam file. But I'm going to go ahead and close this defect since the input is bad and there is nothing straightforward to do to better handle this bad input.

from pipeline.

Thanks John!

-Charles

Charles Y. Lin, Ph.D.
Dana-Farber Cancer Institute
Department of Medical Oncology
[email protected]:[email protected]
http://bradnerlab.com

On Wed, May 27, 2015 at 1:51 AM, John DiMatteo [email protected]
wrote:

Closed #46 #46.

—
Reply to this email directly or view it on GitHub
#46 (comment).

from pipeline.