Skip to contentSkip to conversation

BUG in fs/jfs/jfs_xtree.c

PATCH In review Eric Sesterhenn 2 participants 3 messages Dec 18, 2006 – Dec 19, 2006 JFS filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 BUG in fs/jfs/jfs_xtree.c

hi,

while playing around with fsfuzzer, i got the following oops with jfs:

[  851.804875] BUG at fs/jfs/jfs_xtree.c:760 assert(!BT_STACK_FULL(btstack))
[  851.805179] ------------[ cut here ]------------
[  851.805238] kernel BUG at fs/jfs/jfs_xtree.c:760!
[  851.805287] invalid opcode: 0000 [#1]
[  851.805327] PREEMPT 
[  851.805391] Modules linked in:
[  851.805469] CPU:    0
[  851.805474] EIP:    0060:[<c028e57a>]    Not tainted VLI
[  851.805482] EFLAGS: 00010286   (2.6.19-git18 #16)
[  851.805620] EIP is at xtSearch+0x2fa/0x6a0
[  851.805669] eax: 00000050   ebx: 00000001   ecx: cc826000   edx: 00000000
[  851.805720] esi: 00030012   edi: cc826b94   ebp: cc826b58   esp: cc826ab8
[  851.805772] ds: 007b   es: 007b   ss: 0068
[  851.805823] Process mount (pid: 4721, ti=cc826000 task=d1faeb10 task.ti=cc826000)
[  851.805872] Stack: c04fec94 c04feccd 000002f8 c04fecb5 0000f000 00000000 d1faeb10 cc826afc 
[  851.806173]        0000f000 00000000 00000000 c063387c d1faeb10 00000000 00000000 00000000 
[  851.807252]        cc053368 cc826b9c cc053088 ffffffff 00000000 00000000 cc0530d8 cc0531e4 
[  851.807513] Call Trace:
[  851.807593]  [<c01045da>] show_trace_log_lvl+0x1a/0x40
[  851.807697]  [<c01046a9>] show_stack_log_lvl+0xa9/0xe0
[  851.807774]  [<c01048c4>] show_registers+0x1e4/0x360
[  851.807852]  [<c0104b3a>] die+0xfa/0x240
[  851.807925]  [<c0104cf3>] do_trap+0x73/0xa0
[  851.808000]  [<c01056b7>] do_invalid_op+0x97/0xc0
[  851.808077]  [<c04734e4>] error_code+0x74/0x80
[  851.808347]  [<c028fb4a>] xtLookup+0xaa/0x240
[  851.808432]  [<c02a3c95>] metapage_get_blocks+0x95/0xe0
[  851.808512]  [<c02a44ea>] metapage_readpage+0xea/0x160
[  851.808593]  [<c014acf3>] read_cache_page+0x93/0x160
[  851.808676]  [<c02a46f4>] __get_metapage+0x114/0x420
[  851.808754]  [<c029a906>] dbMount+0x46/0x160
[  851.808835]  [<c028e024>] jfs_mount+0x84/0x180
[  851.808913]  [<c028a13c>] jfs_fill_super+0x11c/0x2a0
[  851.808994]  [<c016948c>] get_sb_bdev+0xec/0x140
[  851.809077]  [<c0289c21>] jfs_get_sb+0x21/0x40
[  851.809155]  [<c0168f64>] vfs_kern_mount+0xc4/0x140
[  851.809232]  [<c0169029>] do_kern_mount+0x29/0x60
[  851.809306]  [<c017e76b>] do_mount+0x2ab/0x6e0
[  851.809385]  [<c017ec0d>] sys_mount+0x6d/0xc0
[  851.809460]  [<c0102fd0>] syscall_call+0x7/0xb
[  851.809536]  =======================
[  851.809579] Code: c2 0f 85 20 fe ff ff c7 44 24 0c b5 ec 4f c0 c7 44 24 08
f8 02 00 00 c7 44 24 04 cd ec 4f c0 c7 04 24 94 ec 4f c0 e8 a6 c3 e8 ff <0f>
0b eb fe 89 f6 8b 7d bc f6 47 10 02 0f 84 33 01 00 00 8b 45 
[  851.811210] EIP: [<c028e57a>] xtSearch+0x2fa/0x6a0 SS:ESP 0068:cc826ab8
[  851.811572]  

On a damaged filesystem we might have a full stack and should not progress further, and return instead of calling BUG()

Signed-off-by Eric Sesterhenn
fs/jfs/jfs_xtree.c renamed from fs/jfs/jfs_xtree.c.orig +2 -0
@@ -757,6 +757,8 @@ static int xtSearch(struct inode *ip, s6
 			nsplit = 0;
 
 		/* push (bn, index) of the parent page/entry */
+		if (BT_STACK_FULL(btstack))
+			return -EINVAL;
 		BT_PUSH(btstack, bn, index);
 
 		/* get the child page block number */
#1 lore raw
Quoted 5 lines · Eric Sesterhenn

hi,

while playing around with fsfuzzer, i got the following oops with jfs:

[  851.804875] BUG at fs/jfs/jfs_xtree.c:760 assert(!BT_STACK_FULL(btstack))

...

Quoted 9 lines

On a damaged filesystem we might have a full stack and should not progress further, and return instead of calling BUG()

Signed-off-by Eric Sesterhenn

fs/jfs/jfs_xtree.c static int xtSearch(struct inode *ip, s6 · line 764

Quoted 8 lines
fs/jfs/jfs_xtree.c renamed from fs/jfs/jfs_xtree.c.orig +2 -0
@@ -757,6 +757,8 @@ static int xtSearch(struct inode *ip, s6
 			nsplit = 0;
 
 		/* push (bn, index) of the parent page/entry */
+		if (BT_STACK_FULL(btstack))
+			return -EINVAL;
 		BT_PUSH(btstack, bn, index);
 
 		/* get the child page block number */

I typically return -EIO for metadata-corruption problems. I also want to look at the other places in jfs_xtree.c that call BT_PUSH. I'm on vacation today, so I'll take a closer look at this tomorrow and push it upstream.

Thanks!
Shaggy

Signature · 2 lines
David Kleikamp
IBM Linux Technology Center
#2 lore raw
In reply to Eric Sesterhenn
Quoted 7 lines · Eric Sesterhenn

hi,

while playing around with fsfuzzer, i got the following oops with jfs:

[  851.804875] BUG at fs/jfs/jfs_xtree.c:760 assert(!BT_STACK_FULL(btstack))
[  851.805179] ------------[ cut here ]------------
[  851.805238] kernel BUG at fs/jfs/jfs_xtree.c:760!

...

fs/jfs/jfs_xtree.c static int xtSearch(struct inode *ip, s6 · line 761

Quoted 13 lines

On a damaged filesystem we might have a full stack and should not progress further, and return instead of calling BUG()

Signed-off-by Eric Sesterhenn
fs/jfs/jfs_xtree.c renamed from fs/jfs/jfs_xtree.c.orig +2 -0
@@ -757,6 +757,8 @@ static int xtSearch(struct inode *ip, s6
 			nsplit = 0;
 
 		/* push (bn, index) of the parent page/entry */
+		if (BT_STACK_FULL(btstack))
+			return -EINVAL;

This leaks a reference on an xtree page. I'd also like to mark the superblock dirty here, so that fsck.jfs will fix the file system on the next boot. jfs_error does this as well as the requested error action. The default is to remount read-only.

fs/jfs/jfs_xtree.c static int xtSearch(struct inode *ip, s6 · line 764

Quoted 4 lines
fs/jfs/jfs_xtree.c renamed from fs/jfs/jfs_xtree.c.orig +0 -0
 		BT_PUSH(btstack, bn, index);
 
 		/* get the child page block number */

Do you want to retry your fsfuzzer testing with this patch? I'm going to take a look at the other calls to BT_PUSH in xtTruncate and xtTruncate_pmap, but I'm too busy now, so I'll send you what I have so far (untested).

fs/jfs/jfs_xtree.c +5 -0
index e98eb03..176e984 100644
@@ -757,6 +757,11 @@ static int xtSearch(struct inode *ip, s64 xoff,	s64 *nextp,
 			nsplit = 0;
 
 		/* push (bn, index) of the parent page/entry */
+		if (BT_STACK_FULL(btstack)) {
+			jfs_error(ip->i_sb, "stack overrun in xtSearch!");
+			XT_PUTPAGE(mp);
+			return -EIO;
+		}
 		BT_PUSH(btstack, bn, index);
 
 		/* get the child page block number */
Signature · 2 lines
David Kleikamp
IBM Linux Technology Center