reiser4 crash

reiser4 crash

[Andrew Benton]

I've been using reiser4 for a couple of months. A couple of days ago I
did something stupid with Abiword, the disk started thrashing and as the
system crashed it left this on the screen

------------[ cut here ]------------
kernel BUG at fs/reiser4/plugin/file/tail_conversion.c:58!
invalid operand:0000 [#1]
SMP
Modules linked in: nvidia
CPU:    0
EIP:    0060:[<c020144e>]    Tainted: P      VLI
EFLAGS: 00010282   (2.6.10-rc2-bk13)
EIP is at get_nonexclusive_access+0x29/0x33
eax: df033eac   ebx: df032000   ecx: d96fee80   edx: de5a70dc
esi: 00000000   edi: de5a70dc   ebp: de5a7144   esp:df033d84
ds: 007b   es: 007b   ss: 0068
Process syslogd (pid: 1488, threadinfo=df032000 task=df7df020)
Stack: c0200633 de5a70dc 0804dcc7 00000001 00000001 df033bdc 00000000
df032000
        00000000 00000001 00000001 00000001 00000000 00000001 00000001
c13a0a40
        de91d144 de91d10c c0200183 df033ddc 0804b000 c01c0492 d96fee80
00101cca
Call Trace:
[<c0200633>] write_unix_file+0x2a1/0x45b
[<c0200183>] unix_file_filemap_nopage+0xaa/0xcf
[<c01c0492>] atom_should_commit+0x5a/0x60
[<c01bd241>] init_context+0x75/0xb4
[<c01d3f46>] reiser4_write+0x8f/0xfc
[<c0261e36>] copy_from_user+0x42/0x70
[<c01524b0>] do_readv_writev+0x170/0x29a
[<c01d3eb7>] reiser4_write+0x0/0xfc
[<c037515f>] sys_recv+0x37/0x3b
[<c015268e>] vfs_writev+0x58/0x5c
[<c0152763>] sys_writev+0x51/0x80
[<c010264d>] sysenter_past_esp+0x52/0x71
Code: 00 c3 b8 00 e0 ff ff 21 e0 8b 00 8b 80 d4 04 00 00 8b 54 24 04 8b 
40 44 8b 48 0c 85 c9 75 0c 89 d0 f0 ff 00 0f 88 b8 11 00 00 c3 <0f> 0b 
3a 00 e4 58 43 c0 eb ea 8b 44 24 04 ba ff ff ff ff f0 0f

To me this looked very similar to 
http://www.uwsg.iu.edu/hypermail/linux/kernel/0411.3/0493.html so I 
applied Vladimir Saveliev's patch, recompiled and the patch seems to 
have done the trick.

# This is a BitKeeper generated diff -Nru style patch.
#
# ChangeSet
# 2004/11/23 14:07:34+03:00 vs@xxxxxxxxxxxxxxxxxxxxx
# read_unix_file: missing calls to txn_restart() are added
#
# plugin/file/file.c
# 2004/11/23 14:07:31+03:00 vs@xxxxxxxxxxxxxxxxxxxxx +4 -1
# read_unix_file: missing calls to txn_restart() are added
#
diff -Nru a/plugin/file/file.c b/plugin/file/file.c
--- a/plugin/file/file.c 2004-11-23 14:45:34 +03:00
+++ b/plugin/file/file.c 2004-11-23 14:45:34 +03:00
@@ -1741,6 +1741,8 @@
while (left > 0) {
size_t to_read;

+ txn_restart_current();
+
size = i_size_read(inode);
if (*off >= size)
/* position to read from is past the end of file */
@@ -1774,7 +1776,6 @@
if (user_space)
reiser4_put_user_pages(pages, nr_pages);
drop_nonexclusive_access(uf_info);
- txn_restart_current();

if (read < 0) {
result = read;
@@ -1974,6 +1975,8 @@

drop_nonexclusive_access(unix_file_inode_data(inode));
up_read(&reiser4_inode_data(inode)->coc_sem);
+
+ txn_restart_current();

reiser4_exit_context(&ctx);
return page;

Now if I give Abiword further abuse, memory fills, the disk thrashes but 
when the swap partition is about half full Abiword caves and the system 
recovers.

Re: reiser4 crash

[Frank Denis (Jedi/Sector One)]

On Sat, Dec 04, 2004 at 11:46:50AM +0000, Andrew Benton wrote:
> I've been using reiser4 for a couple of months. A couple of days ago I
> did something stupid with Abiword, the disk started thrashing and as the
> system crashed it left this on the screen

  You need that patch.

diff -urN linux-2.6.10-rc2-mm4/fs/reiser4/plugin/file/file.c /usr/src/linux-2.6.10-rc2-mm4-jedi1/fs/reiser4/plugin/file/file.c
--- linux-2.6.10-rc2-mm4/fs/reiser4/plugin/file/file.c	2004-12-03 22:31:54.502630648 +0100
+++ /usr/src/linux-2.6.10-rc2-mm4-jedi1/fs/reiser4/plugin/file/file.c	2004-12-03 22:17:11.113926144 +0100
@@ -1741,6 +1741,8 @@
 	while (left > 0) {
 		size_t to_read;		
 
+		txn_restart_current();
+
 		size = i_size_read(inode);
 		if (*off >= size)
 			/* position to read from is past the end of file */
@@ -1774,7 +1776,6 @@
 		if (user_space)
 			reiser4_put_user_pages(pages, nr_pages);
 		drop_nonexclusive_access(uf_info);
-		txn_restart_current();
 
 		if (read < 0) {
 			result = read;
@@ -1975,6 +1976,8 @@
 	drop_nonexclusive_access(unix_file_inode_data(inode));
 	up_read(&reiser4_inode_data(inode)->coc_sem);
 
+	txn_restart_current();
+
 	reiser4_exit_context(&ctx);
 	return page;
 }

--
Frank - my stupid blog: http://00f.net

Re: reiser4 crash

[Frank Denis (Jedi/Sector One)]

On Sat, Dec 04, 2004 at 11:46:50AM +0000, Andrew Benton wrote:
> To me this looked very similar to 
> http://www.uwsg.iu.edu/hypermail/linux/kernel/0411.3/0493.html

  Oops I didn't read the end of your mail :(
  
--
Frank - my stupid blog: http://00f.net

Re: reiser4 crash

[Mark Nipper]

[-- Attachment #1: Type: text/plain, Size: 1396 bytes --]

On 04 Dec 2004, Andrew Benton wrote:
> I've been using reiser4 for a couple of months. A couple of days ago I
> did something stupid with Abiword, the disk started thrashing and as the
> system crashed it left this on the screen
<snipped>

	Which if you are trying to patch this against the current
Namesys 2.6.10-rc1 patches requires Saveliev's previous patch
from a couple of weeks ago which I've also attached to this
thread now as I'm in the middle of upgrading to 2.6.10-rc3 and
had to go looking for the previous patch myself.  :)

-- 
Mark Nipper                                                e-contacts:
4475 Carter Creek Parkway                           nipsy@bitgnome.net
Apartment 724                               http://nipsy.bitgnome.net/
Bryan, Texas, 77802-4481           AIM/Yahoo: texasnipsy ICQ: 66971617
(979)575-3193                                      MSN: nipsy@tamu.edu

-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GG/IT d- s++:+ a- C++$ UBL++++$ P--->+++ L+++$ !E---
W++(--) N+ o K++ w(---) O++ M V(--) PS+++(+) PE(--)
Y+ PGP t+ 5 X R tv b+++@ DI+(++) D+ G e h r++ y+(**)
------END GEEK CODE BLOCK------

---begin random quote of the moment---
"I know the forces of spontaneous, emergent life are stronger
than the forces of evil, repression and death, and the forces of
death will destroy themselves."
 -- William S. Burroughs
----end random quote of the moment----

[-- Attachment #2: 1.1755 --]
[-- Type: text/plain, Size: 994 bytes --]

# This is a BitKeeper generated diff -Nru style patch.
#
# ChangeSet
#   2004/11/15 16:23:47+03:00 vs@tribesman.namesys.com 
#   unix_file_filemap_nopage: missing context creation is added
# 
# plugin/file/file.c
#   2004/11/15 16:23:45+03:00 vs@tribesman.namesys.com +5 -1
#   unix_file_filemap_nopage: missing context creation is added
# 
diff -Nru a/plugin/file/file.c b/plugin/file/file.c
--- a/plugin/file/file.c	2004-11-17 09:36:11 +03:00
+++ b/plugin/file/file.c	2004-11-17 09:36:11 +03:00
@@ -1961,8 +1961,10 @@
 {
 	struct page *page;
 	struct inode *inode;
-
+	reiser4_context ctx;
+	
 	inode = area->vm_file->f_dentry->d_inode;
+	init_context(&ctx, inode->i_sb);
 
 	/* block filemap_nopage if copy on capture is processing with a node of this file */
 	down_read(&reiser4_inode_data(inode)->coc_sem);
@@ -1972,6 +1974,8 @@
 
 	drop_nonexclusive_access(unix_file_inode_data(inode));
 	up_read(&reiser4_inode_data(inode)->coc_sem);
+
+	reiser4_exit_context(&ctx);
 	return page;
 }
 

Re: reiser4 crash

[Andrew Benton]

Mark Nipper wrote:

> <snipped>
> 
> 	Which if you are trying to patch this against the current
> Namesys 2.6.10-rc1 patches requires Saveliev's previous patch
> from a couple of weeks ago

Yes, that patch is in the current 2.6.10-rc2-mm4 patch. The fix, 
however, isn't.