Skip to contentSkip to conversation

Set XIP mount option on ext2 bypass check.

BUG? Yan Zheng 4 participants 6 messages Jun 20, 2007 – Jun 22, 2007

Subsystems (2)
lore
Share & tools
lore

Yan Zheng

author
#0 lore raw

hi, all.

xip mount option can be set on ext2 fs by using remount, even though the fs is used with a block device that does not support the direct_access block device operation or fs block size is not equal to PAGE_SIZE. Is this the expected behaviour ?

Thanks in advance
-----
Best Regards

YZ

author
#1 lore raw

I mount an ext2 fs , then remount it with xip option set. I get message below when do write operation in the fs.

--------
kernel BUG at fs/ext2/xip.c:21!
invalid opcode: 0000 [#1]
SMP
last sysfs file: /class/net/eth0/carrier

Modules linked in: ext2 autofs4 hidp rfcomm l2cap bluetooth sunrpc ipv6 dm_mirro

r dm_multipath video sbs i2c_ec button dock battery ac lp floppy snd_ens1371 gam eport snd_rawmidi snd_ac97_codec pcnet32 ac97_bus snd_seq_dummy snd_seq_oss snd_ seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss snd_pcm snd_time r snd soundcore snd_page_alloc mii serio_raw parport_pc parport pcspkr BusLogic i2c_piix4 i2c_core sg ext3 jbd mbcache squashfs dm_snapshot dm_mod loop sd_mod e hci_hcd uhci_hcd ata_piix ata_generic libata sr_mod scsi_mod cdrom

CPU:    0
EIP:    0060:[<d8cba143>]    Not tainted VLI
EFLAGS: 00010246   (2.6.21-1.3194.fc7 #1)

EIP is at ext2_clear_xip_target+0x1e/0x47 [ext2]
eax: d887da00 ebx: d310e3c0 ecx: d0648bd4 edx: 00001e01
esi: d0648b4c edi: 00000000 ebp: cc863e44 esp: cc863df8
ds: 007b es: 007b fs: 00d8 gs: 0033 ss: 0068
Process dd (pid: 3231, ti=cc863000 task=cd61b710 task.ti=cc863000)

Stack: 000000d8 ffffff10 00000200 00001e01 d8cb5b5b cc863e44 cc863e8c 00000c9f
      cd61b710 00000001 d0648bd4 00000000 cc863e50 00000001 00000001 00000400
      cc863e74 cc863e50 c043c220 d0648b4c 00001e01 00000000 00000f8c 00000e34
Call Trace:
[<d8cb5b5b>] ext2_get_block+0x3f4/0x52b [ext2]
[<c043c220>] clockevents_program_event+0xb2/0xb9
[<c04bed36>] avc_has_perm+0x4e/0x58
[<d8cba05d>] ext2_get_xip_page+0x65/0xde [ext2]
[<c0473972>] xip_file_write+0x232/0x38d
[<c0473740>] xip_file_write+0x0/0x38d
[<c0475d33>] vfs_write+0xa8/0x154
[<c0476342>] sys_write+0x41/0x67
[<c0404f70>] syscall_call+0x7/0xb
=======================
Code: 89 0c 24 e8 88 cd ff ff 83 c4 0c 5b c3 57 53 83 ec 08 8b 80 9c 00 00 00 8b

98 98 00 00 00 8b 43 5c 8b 40 34 8b 78 14 85 ff 75 04 <0f> 0b eb fe 8d 44 24 04 31 c9 c1 e2 03 89 04 24 89 d8 ff d7 85
EIP: [<d8cba143>] ext2_clear_xip_target+0x1e/0x47 [ext2] SS:ESP 0068:cc863df8

#2 lore raw
Quoted 2 lines · Yan Zheng

I mount an ext2 fs , then remount it with xip option set. I get message below when do write operation in the fs.

Ouch. Like on mount, we should refuse -o xip on remount. The patch below fixes this issue.

---

fs/ext2/super.c +3 -0
Index: linux-2.6.22-rc4-mm/fs/ext2/super.c
===================================================================
@@ -1071,6 +1071,9 @@ static int ext2_remount (struct super_bl
 	sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 		((sbi->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
 
+	ext2_xip_verify_sb(sb); /* see if bdev supports xip, unset
+				    EXT2_MOUNT_XIP if not */
+
 	es = sbi->s_es;
 	if (((sbi->s_mount_opt & EXT2_MOUNT_XIP) !=
 	    (old_mount_opt & EXT2_MOUNT_XIP)) &&
#3 lore raw

This is an updated version of my bugfix patch. Yan Zheng pointed out, that ext2_remount lacks checking if -o xip should be enabled or not. This patch checks for presence of direct_access on the backing block device and if the blocksize meets the requirements. Andrew, please consider adding this patch to -mm.

---

fs/ext2/super.c +8 -0
Index: linux-2.6.22-rc4-mm/fs/ext2/super.c
===================================================================
@@ -1071,6 +1071,14 @@ static int ext2_remount (struct super_bl
 	sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 		((sbi->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
 
+	ext2_xip_verify_sb(sb); /* see if bdev supports xip, unset
+				    EXT2_MOUNT_XIP if not */
+
+	if ((ext2_use_xip(sb)) && (sb->s_blocksize != PAGE_SIZE)) {
+		printk("XIP: Unsupported blocksize\n");
+		goto restore_opts; 
+	}
+
 	es = sbi->s_es;
 	if (((sbi->s_mount_opt & EXT2_MOUNT_XIP) !=
 	    (old_mount_opt & EXT2_MOUNT_XIP)) &&
#4 lore raw
Quoted 8 lines · Carsten Otte

This is an updated version of my bugfix patch. Yan Zheng pointed out, that ext2_remount lacks checking if -o xip should be enabled or not. This patch checks for presence of direct_access on the backing block device and if the blocksize meets the requirements. Andrew, please consider adding this patch to -mm.

It looks to me like a local denial of service attack in case of user-mountable ext2 file systems in /etc/fstab.

Shouldn't that make it go into 2.6.22?

	Arnd <><
#5 lore raw

Hi,

Quoted 8 lines · Arnd Bergmann
Quoted 7 lines · Carsten Otte

This is an updated version of my bugfix patch. Yan Zheng pointed out, that ext2_remount lacks checking if -o xip should be enabled or not. This patch checks for presence of direct_access on the backing block device and if the blocksize meets the requirements. Andrew, please consider adding this patch to -mm.

Quoted 6 lines
Quoted 1 line

It looks to me like a local denial of service attack in case of user-mountable ext2 file systems in /etc/fstab.

Shouldn't that make it go into 2.6.22?

I agree. I would go on to suggest that all trivially-triggered oopsen / panics from userspace (even if they require privileges, such as the cat /dev/snapshot == oops issue posted last week) in fact ought to be CVE's, and the corresponding fixes for such issues be considered as candidates for -stable, if applicable to the current stable kernel.

Satyam