Re: [BUG] NULL pointer dereference with corrupted image

[Eric Sesterhenn <eric.sesterhenn@lsexperts.de>]

On Wed, 2009-07-15 at 11:54 +0200, Eric Sesterhenn wrote:
> hi,
> 
> mounting the image at http://www.cccmz.de/~snakebyte/ext4.56.img.bz2
> listing the directories and touching a files produces a null pointer
> dereference with -rc3 (verified with -rc2 and current -git). I tested
> this on different boxes with the same result
> 
> root@whiterabbit:~/testing/fsfuzzer-0.6-lmh# mount -o loop
> cfs/ext4.56.img -t ext4dev /media/test/
> root@whiterabbit:~/testing/fsfuzzer-0.6-lmh# ls /media/test 
> lost+found
> root@whiterabbit:~/testing/fsfuzzer-0.6-lmh# touch /media/test/file
> Killed
> 
> If the ls is 
> 
> [29436.010601] EXT4-fs (/dev/loop0): Update your userspace programs to
> mount using ext4
> [29436.010803] EXT4-fs (/dev/loop0): ext4dev backwards compatibility
> will go away by 2.6.31
> [29436.028613] EXT4-fs (loop0): no journal
> [29436.028833] EXT4-fs (loop0): delayed allocation enabled
> [29436.056982] EXT4-fs: mballoc enabled
> [29436.061698] EXT4-fs (loop0): mounted filesystem without journal
> [29436.189556] EXT4-fs error (device loop0): ext4_claim_inode: reserved
> inode or inode > inodes count - block_group = 0, inode=10
> [29436.189835] EXT4-fs (loop0): Remounting filesystem read-only
> [29436.202154] BUG: unable to handle kernel NULL pointer dereference at
> (null)
> [29436.202335] IP: [<c025977d>] ext4_decode_error+0x6d/0x90
> [29436.202540] *pde = 00000000 
> [29436.202643] Oops: 0000 [#1] PREEMPT 
> [29436.202835] last sysfs file: /sys/block/hdc/size
> [29436.202920] Modules linked in:
> [29436.203057] 
> [29436.203057] Pid: 5197, comm: touch Not tainted
> (2.6.31-rc3-23244-g6847e15 #29) System Name
> [29436.203057] EIP: 0060:[<c025977d>] EFLAGS: 00010282 CPU: 0
> [29436.203057] EIP is at ext4_decode_error+0x6d/0x90
> [29436.203057] EAX: 00000000 EBX: c6555d58 ECX: c0a70886 EDX: ffffffe2
> [29436.203057] ESI: c08fd80e EDI: c282cce4 EBP: c6555d40 ESP: c6555d2c
> [29436.203057]  DS: 007b ES: 007b FS: 0000 GS: 00e0 SS: 0068
> [29436.203057] Process touch (pid: 5197, ti=c6555000 task=c44d41a0
> task.ti=c6555000)
> [29436.203057] Stack:
> [29436.203057]  00000001 00000000 c0154fa2 c44d41a0 c9826e10 c6555d74
> c025a948 00000001
> [29436.203057] <0> 0000045b c44d41a0 0000045b c282cd98 c0252430 c282cce4
> c6555d74 bede78d2
> [29436.203057] <0> c282cd98 ffffffe2 c6555d94 c0252563 c0caefe0 c08fdba0
> c0caefe0 c6555d94
> [29436.203057] Call Trace:
> [29436.203057]  [<c0154fa2>] ? mark_held_locks+0x52/0x70
> [29436.203057]  [<c025a948>] ? __ext4_std_error+0x28/0x90
> [29436.203057]  [<c0252430>] ? ext4_delete_inode+0x0/0x230
> [29436.203057]  [<c0252563>] ? ext4_delete_inode+0x133/0x230
> [29436.203057]  [<c0252430>] ? ext4_delete_inode+0x0/0x230
> [29436.203057]  [<c01d2976>] ? generic_delete_inode+0x96/0x140
> [29436.203057]  [<c01d2b57>] ? generic_drop_inode+0x137/0x1b0
> [29436.203057]  [<c01d2197>] ? iput+0x47/0x50
> [29436.203057]  [<c024a904>] ? ext4_new_inode+0xa64/0xe70
> [29436.203057]  [<c01cfc0b>] ? d_rehash+0x3b/0x50
> [29436.203057]  [<c0255881>] ? ext4_create+0xb1/0x120
> [29436.203057]  [<c01c9096>] ? vfs_create+0xb6/0x170
> [29436.203057]  [<c01cbf18>] ? do_filp_open+0x5d8/0x820
> [29436.203057]  [<c08e2377>] ? _spin_unlock+0x27/0x50
> [29436.203057]  [<c01be6bf>] ? do_sys_open+0x4f/0x100
> [29436.203057]  [<c0102d9a>] ? sysenter_exit+0xf/0x1a
> [29436.203057]  [<c01be7d9>] ? sys_open+0x29/0x40
> [29436.203057]  [<c0102d6b>] ? sysenter_do_call+0x12/0x32
> [29436.203057] Code: 6a 82 35 00 89 d9 85 c0 78 33 8d 74 26 00 83 c4 10
> 89 c8 5b 5d c3 85 c0 74 16 8b 80 d0 02 00 00 b9 86 08 a7 c0 8b 80 30 01
> 00 00 <f6> 00 02 74 de 83 c4 10 b9 a3 08 a7 c0 5b 89 c8 5d c3 31 c9 eb 
> [29436.203057] EIP: [<c025977d>] ext4_decode_error+0x6d/0x90 SS:ESP
> 0068:c6555d2c
> [29436.203057] CR2: 0000000000000000
> [29436.284374] ---[ end trace a3b07131cd2de900 ]---
> 
> 
> 0xc025977d is in ext4_decode_error (fs/ext4/super.c:347).
> 342			break;
> 343		case -ENOMEM:
> 344			errstr = "Out of memory";
> 345			break;
> 346		case -EROFS:
> 347			if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
> 348				errstr = "Journal has aborted";
> 349			else
> 350				errstr = "Readonly filesystem";
> 351			break;
> 
> 0xc0252563 is in ext4_delete_inode (fs/ext4/inode.c:225).
> 220			/*
> 221			 * If we're going to skip the normal cleanup, we still need to
> 222			 * make sure that the in-core orphan linked list is properly
> 223			 * cleaned up.
> 224			 */
> 225			ext4_orphan_del(NULL, inode);
> 226			goto no_delete;
> 227		}
> 228	
> 229		if (IS_SYNC(inode))

The following patch fixes the issue.

If an error happens ext4_decode_error() misses a check if there is a 
valid journal. If there is no journal we dereference a NULL pointer,
when we try to print the matching error string.

Signed-off-by: Eric Sesterhenn <eric.sesterhenn@lsexperts.>

--- linux/fs/ext4/super.c.orig	2009-07-20 12:42:33.000000000 +0200
+++ linux/fs/ext4/super.c	2009-07-20 12:44:00.000000000 +0200
@@ -344,7 +344,8 @@ static const char *ext4_decode_error(str
 		errstr = "Out of memory";
 		break;
 	case -EROFS:
-		if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
+		if (!sb || !EXT4_SB(sb)->s_journal || 
+		    EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 			errstr = "Journal has aborted";
 		else
 			errstr = "Readonly filesystem";