Hmm, very interesting!
Please try this patch, it must fix the problem and print some numbers to debug.
Ondrej Zary wrote:
On Tuesday 05 June 2012, Konstantin Khlebnikov wrote:
Ondrej Zary wrote:
Hello,
I use NFS for deploying HDD images on new machines. My machine has 2nd
network card just for this, running DHCPD, TFTPD and kernel NFS server.
The target machine is set to boot from LAN and boots SystemRescueCD from
my machine with an autorun script that launches Partimage and deploys the
HDD image (400 to 900 MB compressed).
It worked fine for years, until now. With kernel 3.4, everyting
works only for the first time after boot (and not always). Next time
(next machine), partimage aborts almost immediately as it's probably
unable to decompress the image file. md5sum is different on my machine
vs. on the target (through NFS). Also SystemRescueCD boot aborts with md5
error sometimes. Everything works fine after rebooting back to 3.3.
Bisection found this:
0fc9d1040313047edf6a39fd4d7c7defdca97c62 is the first bad commit
commit 0fc9d1040313047edf6a39fd4d7c7defdca97c62
Author: Konstantin Khlebnikov<khlebnikov@xxxxxxxxxx>
Date: Wed Mar 28 14:42:54 2012 -0700
radix-tree: use iterators in find_get_pages* functions
Reverting this commit in 3.4 fixes the problem.
[all reporters added to CC] let's keep all in one thread
In attachment two patches which might help to debug this regression:
"mm: recheck page index in find_get_pages_contig" adds paranoid check into
find_get_pages_contig(). It can explain everything, but currently I don't
see how this can hapens.
"mm: debug fing_get_pages speculative restart" shows lookup restarting
condition which was removed by bisected commit.
My dmesg (after corruption occured) with these two patches applied:
[ 79.999511] ------------[ cut here ]------------
[ 79.999564] WARNING: at mm/filemap.c:941 find_get_pages_contig+0x177/0x1b0()
[ 79.999611] Hardware name: VT82C694X
[ 79.999617] Modules linked in: nfsd lockd sunrpc des_generic ecb crypto_blkcipher md4 md5 hmac cryptomgr aead cifs crypto_hash crypto_algapi crypto
firewire_ohci firewire_core
[ 79.999653] Pid: 1563, comm: nfsd Not tainted 3.4.0-omega #4
[ 79.999659] Call Trace:
[ 79.999729] [<c011ff88>] ? warn_slowpath_common+0x78/0xb0
[ 79.999744] [<c0175187>] ? find_get_pages_contig+0x177/0x1b0
[ 79.999753] [<c0175187>] ? find_get_pages_contig+0x177/0x1b0
[ 79.999763] [<c011ffd9>] ? warn_slowpath_null+0x19/0x20
[ 79.999772] [<c0175187>] ? find_get_pages_contig+0x177/0x1b0
[ 79.999805] [<c01c544b>] ? __generic_file_splice_read+0xeb/0x510
[ 79.999853] [<c01c4040>] ? page_cache_pipe_buf_release+0x10/0x10
[ 79.999873] [<c04f2589>] ? common_interrupt+0x29/0x30
[ 79.999900] [<f892c710>] ? _fh_update.isra.11.part.12+0x60/0x60 [nfsd]
[ 79.999931] [<c022c9f7>] ? exportfs_decode_fh+0xc7/0x250
[ 79.999981] [<f893133d>] ? exp_get_by_name+0x3d/0x70 [nfsd]
[ 80.000000] [<c0150215>] ? getboottime+0x35/0x40
[ 80.007383] [<c04f0da8>] ? __schedule+0x198/0x470
[ 80.007505] [<f88cbf34>] ? sunrpc_cache_lookup+0x54/0x2d0 [sunrpc]
[ 80.007574] [<c01c58e3>] ? generic_file_splice_read+0x73/0x110
[ 80.007590] [<c01254bf>] ? irq_exit+0x4f/0x90
[ 80.007599] [<c01c5870>] ? __generic_file_splice_read+0x510/0x510
[ 80.007608] [<c01c4330>] ? do_splice_to+0x60/0x90
[ 80.007618] [<c01c459a>] ? splice_direct_to_actor+0xaa/0x1c0
[ 80.007654] [<f892d710>] ? nfsd_buffered_filldir+0x160/0x160 [nfsd]
[ 80.007700] [<f892dc37>] ? nfsd_vfs_read.isra.16+0x117/0x160 [nfsd]
[ 80.007715] [<f892e764>] ? nfsd_read+0x1c4/0x280 [nfsd]
[ 80.007732] [<f89357bf>] ? nfsd3_proc_read+0xcf/0x160 [nfsd]
[ 80.007745] [<f892a7d0>] ? nfsd_dispatch+0xb0/0x190 [nfsd]
[ 80.007779] [<f88c3682>] ? svc_process+0x442/0x7c0 [sunrpc]
[ 80.007825] [<f892a0a3>] ? nfsd+0xa3/0x130 [nfsd]
[ 80.007838] [<f892a000>] ? 0xf8929fff
[ 80.007846] [<f892a000>] ? 0xf8929fff
[ 80.007858] [<c01389bc>] ? kthread+0x6c/0x80
[ 80.007867] [<c0138950>] ? kthread_freezable_should_stop+0x50/0x50
[ 80.007896] [<c04f2596>] ? kernel_thread_helper+0x6/0xd
[ 80.007937] ---[ end trace 0bc8170cf5ac5466 ]---
mm: fix find_get_pages_contig
From: Konstantin Khlebnikov <khlebnikov@xxxxxxxxxx>
Signed-off-by: Konstantin Khlebnikov <khlebnikov@xxxxxxxxxx>
---
mm/filemap.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/mm/filemap.c b/mm/filemap.c
index 79c4b2b..f4343a3 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -928,7 +928,10 @@ repeat:
* otherwise we can get both false positives and false
* negatives, which is just confusing to the caller.
*/
- if (page->mapping == NULL || page->index != iter.index) {
+ if (page->mapping == NULL || page->index != index + ret) {
+ if (iter.index != index + ret)
+ printk("%s %lu %lu %u\n", __func__,
+ iter.index, index, ret);
page_cache_release(page);
break;
}
