Jake Maciejewski wrote:
On Wed, 2007-07-11 at 23:48 +0400, Edward Shishkin wrote:
Jake Maciejewski wrote:
I've hit the same panic looping kernel builds (while true ; do make
mrproper ; make allmodconfig ; make -j4 ; done) on 2.6.21.1 with the
Namesys patch and reiser4 debug enabled. I've seen it on my amd64
desktop and x86 laptop.
Another one I've seen is:
reiser4 panicked cowardly: reiser4[fixdep(16043)]: sibling_list_remove (fs/reiser4/tree_walk.c:814)[zam-32245]
In both cases the fsck didn't find anything, as you observed.
On Wed, 2007-07-11 at 06:46 +0200, Ingo Bormuth wrote:
Hmm, whenever I try to build busybox (1.4.2) I get nikita-191 panics:
[...]
cc console_tools/clear.o
reiser4 panicked cowardly: reiser4[cc1(13066)]: save_file_hint (fs/reiser4/plugin/file.c:705) [nikity-1991]:
kernel panic - not syncing: reiser4[cc1(13066)]: save_file_hint (fs/reiser4/plugin/file.c:705) [nikity-1991]:
Somebody missed set_file_hint(), which synchronizes the coords.
err, sorry, its name is reiser4_set_hint
Unfortunately I can not reproduce it. Would you please (if possible)
catch the stack with the attached patch?
[<ffffffff88186b5e>] :reiser4:save_file_hint+0xee/0x3c0
[<ffffffff88189c60>] :reiser4:read_unix_file+0x940/0xa10
[<ffffffff80276bbb>] vfs_read+0xdb/0x180
[<ffffffff80277083>] sys_read+0x53/0x90
[<ffffffff8020993e>] system_call+0x7e/0x83
Thanks!
Indeed, the coords are not synchronized when reading tails. However,
it is not a fatal bug: we are victims of brain damaged and unreadable
hint interface.
The possible fix is attached. Would you please test it?
Also don't forget to apply this patch:
http://lkml.org/lkml/diff/2007/7/11/396/1
as it also can be related to the problem.
Edward.
As for reproducing it, I think I should mention that:
1. I'm using distcc to speed things up. Without offloading the compiling
work, my laptop has lasted ~3.5hrs before a panic. My desktop with
distcc configured usually only lasts a few minutes.
2. My local storage is encrypted through dm-crypt, but I've also tried
over open-iscsi and got the same results.
Running fsck.reiser4 before and after the panic doesn't show any complaints.
The partition is heavily used. I'm not aware of any other problem.
Vanilla-2.6.21.6 (kernel.org) with reiser4-2.6.21-path (namesys.com).
Not that I understood the code, but why is it an assertion at all?
Couldn't one just use an empty hint if the current one is invalid?
Sure, it is possible to not use it at all. But if the current one is valid,
it would be nice to use it to avoid tree traversal with waiting for
possible locks, etc..
Thanks,
Edward.
Update hint when reading tails
Signed-off-by: Edward Shishkin <edward@xxxxxxxxxxx>
--- linux-2.6.22-rc6-mm1/fs/reiser4/plugin/item/tail.c.orig
+++ linux-2.6.22-rc6-mm1/fs/reiser4/plugin/item/tail.c
@@ -758,7 +758,7 @@
coord->unit_pos--;
coord->between = AFTER_UNIT;
}
-
+ reiser4_set_hint(hint, &f->key, ZNODE_READ_LOCK);
return 0;
}
