Skip to contentSkip to conversation

bfs: fix null-ptr-deref and possible warning in bfs_move_block() func

PATCH Superseded [email protected] 2 participants 5 messages Jul 11, 2024

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover bfs: fix null-ptr-deref and possible warning in bfs_move_block() func

https://syzkaller.appspot.com/bug?extid=d98fd19acd08b36ff422

[PATCHv2 fs/bfs 1/2] bfs: prevent null pointer dereference in bfs_move_block()
---
v2: corrected the commit message and explicitly initialized
the return variable with zero (Markus Elfring)
---
[PATCHv2 fs/bfs 2/2] bfs: add buffer_uptodate check before mark_buffer_dirty
---
v2: renamed the return variable err -> ret (Markus Elfring)
---

#1 lore raw

PATCH v2 1/2 bfs: prevent null pointer dereference in bfs_move_block()

From: Vasiliy Kovalev <[email protected]>

Detect a failed sb_getblk() call (before copying data) so that null pointer dereferences should not happen any more.

Found by Syzkaller:

KASAN: null-ptr-deref in range [0x0000000000000028-0x000000000000002f]
CPU: 1 PID: 1069 Comm: mark_buffer_dir Tainted: G W 6.10.0-un-def-alt0.rc7
RIP: 0010:bfs_get_block+0x3ab/0xe80 [bfs]
Call Trace:
<TASK>
? show_regs+0x8d/0xa0
? die_addr+0x50/0xd0
? exc_general_protection+0x148/0x220
? asm_exc_general_protection+0x22/0x30
? bfs_get_block+0x3ab/0xe80 [bfs]
? bfs_get_block+0x370/0xe80 [bfs]
? __pfx_bfs_get_block+0x10/0x10 [bfs]
__block_write_begin_int+0x4ae/0x16a0
? __pfx_bfs_get_block+0x10/0x10 [bfs]
? __pfx___block_write_begin_int+0x10/0x10
block_write_begin+0xb5/0x410
? __pfx_bfs_get_block+0x10/0x10 [bfs]
bfs_write_begin+0x32/0xe0 [bfs]
generic_perform_write+0x265/0x610
? __pfx_generic_perform_write+0x10/0x10
? generic_write_checks+0x323/0x4a0
? __pfx_generic_file_write_iter+0x10/0x10
__generic_file_write_iter+0x16a/0x1b0
generic_file_write_iter+0xf0/0x360
? __pfx_generic_file_write_iter+0x10/0x10
vfs_write+0x670/0x1120
? __pfx_vfs_write+0x10/0x10
ksys_write+0x127/0x260
? __pfx_ksys_write+0x10/0x10
do_syscall_64+0x9f/0x190
? __ct_user_enter+0x74/0xc0
? syscall_exit_to_user_mode+0xbb/0x1d0
? do_syscall_64+0xab/0x190
? ct_kernel_exit.isra.0+0xbb/0xe0
? __ct_user_enter+0x74/0xc0
? syscall_exit_to_user_mode+0xbb/0x1d0
? do_syscall_64+0xab/0x190
? ct_kernel_exit.isra.0+0xbb/0xe0
? clear_bhb_loop+0x45/0xa0
? clear_bhb_loop+0x45/0xa0
? clear_bhb_loop+0x45/0xa0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
RIP: 0033:0x7f2bc708ed29
Fixes 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by [email protected]
---
 fs/bfs/file.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
fs/bfs/file.c +8 -2
index 0dceefc54b48a..e99dc8ace2027 100644
@@ -34,16 +34,22 @@ static int bfs_move_block(unsigned long from, unsigned long to,
 					struct super_block *sb)
 {
 	struct buffer_head *bh, *new;
+	int ret = 0;
 
 	bh = sb_bread(sb, from);
 	if (!bh)
 		return -EIO;
 	new = sb_getblk(sb, to);
+	if (unlikely(!new)) {
+		ret = -EIO;
+		goto out_err_new;
+	}
 	memcpy(new->b_data, bh->b_data, bh->b_size);
 	mark_buffer_dirty(new);
-	bforget(bh);
 	brelse(new);
-	return 0;
+out_err_new:
+	bforget(bh);
+	return ret;
 }
 
 static int bfs_move_blocks(struct super_block *sb, unsigned long start,
Signature · 1 line
2.33.8
#2 lore raw

PATCH v2 2/2 bfs: add buffer_uptodate check before mark_buffer_dirty call

In reply to [email protected]

From: Vasiliy Kovalev <[email protected]>

Add a check in bfs_move_block to ensure the new buffer is up-to-date (buffer_uptodate) before calling mark_buffer_dirty.

Found by Syzkaller:

WARNING: CPU: 1 PID: 1046 at fs/buffer.c:1183 mark_buffer_dirty+0x394/0x3f0
CPU: 1 PID: 1046 Comm: mark_buffer_dir Not tainted 6.10.0-un-def-alt0.rc7.kasan
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-alt1 04/01/2014
RIP: 0010:mark_buffer_dirty+0x394/0x3f0
Call Trace:
<TASK>
? show_regs+0x8d/0xa0
? __warn+0xe6/0x380
? mark_buffer_dirty+0x394/0x3f0
? report_bug+0x348/0x480
? handle_bug+0x60/0xc0
? exc_invalid_op+0x13/0x50
? asm_exc_invalid_op+0x16/0x20
? mark_buffer_dirty+0x394/0x3f0
? mark_buffer_dirty+0x394/0x3f0
bfs_get_block+0x3ec/0xe80 [bfs]
? __pfx_bfs_get_block+0x10/0x10 [bfs]
__block_write_begin_int+0x4ae/0x16a0
? __pfx_bfs_get_block+0x10/0x10 [bfs]
? __pfx___block_write_begin_int+0x10/0x10
block_write_begin+0xb5/0x410
? __pfx_bfs_get_block+0x10/0x10 [bfs]
bfs_write_begin+0x32/0xe0 [bfs]
generic_perform_write+0x265/0x610
? __pfx_generic_perform_write+0x10/0x10
? generic_write_checks+0x323/0x4a0
? __pfx_generic_file_write_iter+0x10/0x10
__generic_file_write_iter+0x16a/0x1b0
generic_file_write_iter+0xf0/0x360
? __pfx_generic_file_write_iter+0x10/0x10
vfs_write+0x670/0x1120
? __pfx_vfs_write+0x10/0x10
ksys_write+0x127/0x260
? __pfx_ksys_write+0x10/0x10
do_syscall_64+0x9f/0x190
? do_syscall_64+0xab/0x190
? syscall_exit_to_user_mode+0xbb/0x1d0
? do_syscall_64+0xab/0x190
? lock_release+0x241/0x730
? __ct_user_enter+0xb3/0xc0
? __pfx_lock_release+0x10/0x10
? get_vtime_delta+0x116/0x270
? ct_kernel_exit.isra.0+0xbb/0xe0
? __ct_user_enter+0x74/0xc0
? syscall_exit_to_user_mode+0xbb/0x1d0
? do_syscall_64+0xab/0x190
? do_syscall_64+0xab/0x190
? ct_kernel_exit.isra.0+0xbb/0xe0
? clear_bhb_loop+0x45/0xa0
? clear_bhb_loop+0x45/0xa0
? clear_bhb_loop+0x45/0xa0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
RIP: 0033:0x7f5bb79a4d2
---
 fs/bfs/file.c | 5 +++++
 1 file changed, 5 insertions(+)
fs/bfs/file.c +5 -0
index e99dc8ace2027..9599b41cbe91b 100644
@@ -44,8 +44,13 @@ static int bfs_move_block(unsigned long from, unsigned long to,
 		ret = -EIO;
 		goto out_err_new;
 	}
+	if (!buffer_uptodate(new)) {
+		ret = -EIO;
+		goto out_err;
+	}
 	memcpy(new->b_data, bh->b_data, bh->b_size);
 	mark_buffer_dirty(new);
+out_err:
 	brelse(new);
 out_err_new:
 	bforget(bh);
Signature · 1 line
2.33.8
#3 lore raw
In reply to [email protected]

…

Quoted 1 line

[PATCHv2 fs/bfs 1/2] bfs: prevent null pointer dereference in bfs_move_block()

…

I find it usually helpful to separate the version identifier from the previous key word.

How do you think about to improve the outline another bit (also for the cover letter)?

Regards,
Markus

#4 lore raw
Quoted 7 lines · Markus Elfring

…

Quoted 1 line

[PATCHv2 fs/bfs 1/2] bfs: prevent null pointer dereference in bfs_move_block()

…

I find it usually helpful to separate the version identifier from the previous key word.

How do you think about to improve the outline another bit (also for the cover letter)?

I will take your recommendation into account when submitting the next versions, if there are any comments on the patches themselves.

Quoted 2 lines

Regards,
Markus

Signature · 2 lines
Thanks,
Vasiliy Kovalev