2003-09-23 05:26:14

by Peter Lieverdink

[permalink] [raw]
Subject: cryptoloop hard lockup 2.6.0-test5-mm4

I just had a play with an article from an old LinuxJournal, to see if I
could set up a cryptoloop homedir on my Debian unstable box. Found making
an fs on the loop device locks the kernel hard. (need reset button) When
using loopback without crypto it all works fine.

Ayone else have the same experience?

Using util-linux 2.12, aes (same prob with twofish, blowfish).

Crash dump is attached, as is kernel config. When not running uml-switch,
the crash still occurs but doesn't print the VM: messages. (I don't have a
swap partition anyway, so what's that all about?)

- Peter.


Attachments:
crash.txt (0.98 kB)
config-2.6.0-test5-mm4 (27.25 kB)
Download all attachments

2003-09-23 13:36:37

by Thiago Rondon

[permalink] [raw]
Subject: Re: cryptoloop hard lockup 2.6.0-test5-mm4

Hi Peter,

Can you please test that?

--- mm/swapfile.c.orig 2003-09-23 09:34:16.000000000 -0400
+++ mm/swapfile.c 2003-09-23 09:34:39.000000000 -0400
@@ -1223,7 +1223,7 @@
unsigned int type;
int i, prev;
int error;
- static int least_priority;
+ static int least_priority = 0;
union swap_header *swap_header = 0;
int swap_header_version;
int nr_good_pages = 0;


On Tue, Sep 23, 2003 at 03:25:47PM +1000, Peter Lieverdink wrote:
> I just had a play with an article from an old LinuxJournal, to see if I
> could set up a cryptoloop homedir on my Debian unstable box. Found making
> an fs on the loop device locks the kernel hard. (need reset button) When
> using loopback without crypto it all works fine.
>
> Ayone else have the same experience?
>
> Using util-linux 2.12, aes (same prob with twofish, blowfish).
>
> Crash dump is attached, as is kernel config. When not running uml-switch,
> the crash still occurs but doesn't print the VM: messages. (I don't have a
> swap partition anyway, so what's that all about?)
>
> - Peter.

> kahlua:~# modprobe cryptoloop
> loop: loaded (max 8 devices)
> kahlua:~# modprobe aes
> kahlua:~# losetup -e aes -k 128 /dev/loop0 /home/cafuego.img
> kahlua:~# mkfs -t ext2 /dev/loop0
> mke2fs 1.35-WIP (21-Aug-2003)
> Filesystem label=
> OS type: Linux
> Block size=1024 (log=0)
> Fragment size=1024 (log=0)
> 131072 inodes, 524288 blocks
> 26214 blocks (5.00%) reserved for the super user
> First data block=1
> 64 block groups
> 8192 blocks per group, 8192 fragments per group
> 2048 inodes per group
> Superblock backups stored on blocks:
> 8193, 24577, 40961, 57345, 73729, 204801, 221185, 401409
>
> Writing inode tables: done
> Writing superblocks and filesystem accounting information: do_wp_page: bogus page at address 0804d056
> VM: killing process uml_switch
> swap_dup: Bad swap file entry 90885fdb
> VM: killing process uml_switch
> swap_dup: Bad swap file entry 286be58f
> VM: killing process uml_switch
> Debug: sleeping fucntion called from invalid context at include/linux/rwsem.h:43
> in_atomic():0, irqs_disabled():1
> Call Trace:



--
-Thiago Rondon,
MAKETRADEFAIR.ORG

2003-11-08 21:15:07

by Florian Weimer

[permalink] [raw]
Subject: Re: cryptoloop hard lockup 2.6.0-test5-mm4

Thiago Rondon wrote:

> - static int least_priority;
> + static int least_priority = 0;

This change has no effect whatsoever (objects with static storage
duration are always initialized).

Any other ideas? I stumbled across this problem using vanilla
2.6.0-test9.