Centralized Summary

[Peter Grayson]

I have applied the new patch. By the way, the patch does not apply
cleanly to the latest mtd from cvs, but it doesn't take to much to fix
it up (this was true for the previous patch also).

This version of the patch seems to take care of the previous problems I
reported. I can now mount and unmount a previously empty partition just
fine.

There are some new problems that I am running into though. The first
seems benign, but it ruins the CS mount performance. When mounting, I
will occasionally see this message:

jffs2_cs_process_inocache_node: Crc failure on cent_sum inocache:
cache_crc!

It looks like when this happens, the CS is abandoned and the mount time
goes from super-fast to regular-slow. I have found that this error can
be readily reproduced by running the following command-line once or
twice. Seems more likely to happen if there is some data in the
filesystem.

  # mount /mnt/mtdb3 && touch /mnt/mtdb3/foo && umount /mnt/mtdb3

The other problems I am seeing are more severe. A kernel panic
occasionally happens at mount time. There is also a kernel panic that
happens at unmount time. I have included stack traces below. The failure
modes seem consistent. The test I am running to make these problems
happen is to write a summarized jffs2 image to a partition and then
repeatedly mount and unmount the partition (like above, but without the
touch).

# mount /mnt/mtdb3
Oops: kernel access of bad area, sig: 11 [#1]
PREEMPT
NIP: C00B0884 LR: C00B10D8 SP: C3F21C30 REGS: c3f21b80 TRAP: 0300    Not
tainted
MSR: 00029030 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11
DAR: C423FFCC, DSISR: 00800000
TASK = c3eb2ab0[97] 'mount' THREAD: c3f20000
Last syscall: 21
GPR00: 000005B7 C3F21C30 C3EB2AB0 C3EBD600 C3F21C50 C3EBD6C8 00000000
C033A810
GPR08: C3F00000 0033FFCC C3F00000 C423FFCC 39008039 000066A4 00000000
00000000
GPR16: 10079130 10070000 00020000 00000001 0B440000 00000003 C3EEEDE0
C3EEEE0B
GPR24: C01C0000 000003B5 C036964C 00000012 C01C0000 C01DFC4C C01C0010
C3EBD600
NIP [c00b0884] jffs2_cs_process_list+0x38/0x5c
LR [c00b10d8] jffs2_cs_build_filesystem+0x830/0xcd0
Call trace:
 [c00a8238] jffs2_do_mount_fs+0x220/0x8a4
 [c00aaa24] jffs2_do_fill_super+0x1ac/0x2b8
 [c00aae68] jffs2_get_sb_mtd+0xb4/0x13c
 [c00ab0fc] jffs2_get_sb+0x19c/0x1dc
 [c004edbc] do_kern_mount+0x5c/0x124
 [c0067458] do_mount+0x6c0/0x71c
 [c0067938] sys_mount+0x9c/0xf0
 [c0002a20] ret_from_syscall+0x0/0x48

# mount /mnt/mtdb3 && umount /mnt/mtdb3
dumping centsum data
jffs2_flash_writev(): Non-contiguous write to 00000000
kernel BUG in jffs2_flash_writev
at /home/pgrayson/project/realmlinux1/linux-2.6/fs/jffs2/wbuf.c:690!
Oops: Exception in kernel mode, sig: 5 [#1]
PREEMPT
NIP: C00AC5AC LR: C00AC58C SP: C0369CE0 REGS: c0369c30 TRAP: 0700    Not
tainted
MSR: 00029030 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11
TASK = c3eb3190[854] 'umount' THREAD: c0368000
Last syscall: 22
GPR00: 00000000 C0369CE0 C3EB3190 0000003A 00002F4C FFFFFFFF 00002F12
C0190000
GPR08: C0195C43 C01B0000 00000008 C0369C10 52000095 000066A4 00000000
00000000
GPR16: 10000140 C0369DB0 C3EBD114 00000000 00000000 100743A8 FFFFFFFF
00000002
GPR24: C0369D80 00000000 C3FC0000 00000000 00000000 C3880000 00000CB4
C3EBD000
NIP [c00ac5ac] jffs2_flash_writev+0x1cc/0x61c
LR [c00ac58c] jffs2_flash_writev+0x1ac/0x61c
Call trace:
 [c00b024c] jffs2_cs_write_inocache+0xcc/0x174
 [c00b04e4] jffs2_cs_dump_ic+0x1f0/0x284
 [c00ab198] jffs2_put_super+0x5c/0x184
 [c004d828] generic_shutdown_super+0xc8/0x1bc
 [c00ab2d8] jffs2_kill_sb+0x18/0x3c
 [c004d72c] deactivate_super+0xa0/0xd4
 [c0065abc] __mntput+0x30/0x44
 [c0055560] path_release_on_umount+0x4c/0x60
 [c0066390] sys_umount+0x2bc/0x2d4
 [c0002a20] ret_from_syscall+0x0/0x48

Thanks for your efforts.

Pete

On Fri, 2005-05-13 at 10:32 +0200, Ferenc Havasi wrote:
> Hi Pete,
> 
> We updated the summary patches.
> 
> Erase block summary:
> - fix a small bug, may cause problem only in debug mode
> 
> Centralized summary:
> - fix a small bug, may cause problem only in debug mode
> - fix locking problem on empty fs
> - fix bad block size accounting (and counting free space)
> 
> You can download the patches from our site.
> 
> Bye,
> Ferenc
>