Subject: | corrupted mmaped-files in linux |
Date: | Tue, 16 Dec 2014 17:24:05 +0100 |
To: | openafs-bugs@openafs.org |
From: | Christof Hanke <christof.hanke@rzg.mpg.de> |
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hi all,
some users told me that sometimes files which are created and filled with mmap are corrupted.
(blocks are filled with zeros)
The exact conditions for this are unclear.
However, I could verify the problem on
openSUSE 13.2, SLES 11 SP3 and CentOS 7
using the attached program (sorry for the bad commandline-parsing).
So you create a eg. 257Mbyte large testfile with "a.out 257 testfile"
and check it with "a.out testfile 257 check"
I could created corrupted files with the
starting the afsd with "-blocks 262144"
memcache or diskcache is irrelevant.
You should see problems for file-sizes of 513M, 1029M 2049M...
But that may depend on the size of your main-memory.
My observations so far:
The problem is that one or more blocks are filled with zeros.
The size of this zero-filled blocks is a multiple of 512.
The position of these blocks is random (?)
This error only happened to me on physical machines
running the OSs mentioned above in a KVM-instance does not show the error.
Hope I provided all that's necessary to fix this.
Many thanks,
Christof
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAEBAgAGBQJUkFytAAoJEHOAkeR8FKk+qssH/0Kmw8Z7LXLgRHWuA6bTLA2M
B0+ZIDz3Mq42mivN07LSu5oFQxi839hjw4AjNUvvW2O8iUiczzeDcrp/NjZbQYXy
jHYKF9fns/pKkuCYSpYRFOdYTj/5qbne02Sfo7hB9E5V/vmHvbyyI9wqM+z7xjW0
ATCfoCNBrtPtE2VYUyRyHjzZIbiAUEPs7W8y5j4XzcNJIqSadkjy6lWVPGiQ0W/E
9FDBisu99RNl0UpHX9OM47aBgZFz0mMT7n9OirLPjqXwGycBkjTN0/iFuKXQXJF5
wL1OgZsivlKlHniExO2gAPUF2/5nsFyGu0dM6ok8RNLBdA/h7FK+YrObfJ/zZkA=
=pU6t
-----END PGP SIGNATURE-----
Hash: SHA1
Hi all,
some users told me that sometimes files which are created and filled with mmap are corrupted.
(blocks are filled with zeros)
The exact conditions for this are unclear.
However, I could verify the problem on
openSUSE 13.2, SLES 11 SP3 and CentOS 7
using the attached program (sorry for the bad commandline-parsing).
So you create a eg. 257Mbyte large testfile with "a.out 257 testfile"
and check it with "a.out testfile 257 check"
I could created corrupted files with the
starting the afsd with "-blocks 262144"
memcache or diskcache is irrelevant.
You should see problems for file-sizes of 513M, 1029M 2049M...
But that may depend on the size of your main-memory.
My observations so far:
The problem is that one or more blocks are filled with zeros.
The size of this zero-filled blocks is a multiple of 512.
The position of these blocks is random (?)
This error only happened to me on physical machines
running the OSs mentioned above in a KVM-instance does not show the error.
Hope I provided all that's necessary to fix this.
Many thanks,
Christof
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAEBAgAGBQJUkFytAAoJEHOAkeR8FKk+qssH/0Kmw8Z7LXLgRHWuA6bTLA2M
B0+ZIDz3Mq42mivN07LSu5oFQxi839hjw4AjNUvvW2O8iUiczzeDcrp/NjZbQYXy
jHYKF9fns/pKkuCYSpYRFOdYTj/5qbne02Sfo7hB9E5V/vmHvbyyI9wqM+z7xjW0
ATCfoCNBrtPtE2VYUyRyHjzZIbiAUEPs7W8y5j4XzcNJIqSadkjy6lWVPGiQ0W/E
9FDBisu99RNl0UpHX9OM47aBgZFz0mMT7n9OirLPjqXwGycBkjTN0/iFuKXQXJF5
wL1OgZsivlKlHniExO2gAPUF2/5nsFyGu0dM6ok8RNLBdA/h7FK+YrObfJ/zZkA=
=pU6t
-----END PGP SIGNATURE-----
Message body is not shown because sender requested not to inline it.