Message-ID: <03aa0da9-7719-40f1-88f9-d0531c9146af@oracle.com>
Date: Mon, 24 Feb 2025 08:56:18 -0600
From: Dave Kleikamp <dave.kleikamp@...cle.com>
To: Edward Adam Davis <eadavis@...com>
Cc: jfs-discussion@...ts.sourceforge.net, linux-kernel@...r.kernel.org,
        syzbot+355da3b3a74881008e8f@...kaller.appspotmail.com,
        syzkaller-bugs@...glegroups.com
Subject: Re: [PATCH V3] jfs: Prevent copying of nlink with value 0 from disk
 inode

On 2/21/25 6:16PM, Edward Adam Davis wrote:
> On Thu, 20 Feb 2025 17:28:49 -0600, Dave Kleikamp wrote:
>> On 2/20/25 5:22PM, Edward Adam Davis wrote:
>>> On Thu, 20 Feb 2025 10:15:04 -0600, Dave Kleikamp wrote:
>>>>> syzbot report a deadlock in diFree. [1]
>>>>>
>>>>> When calling "ioctl$LOOP_SET_STATUS64", the offset value passed in is 4,
>>>>> which does not match the mounted loop device, causing the mapping of the
>>>>> mounted loop device to be invalidated.
>>>>>
>>>>> When creating the directory and creating the inode of iag in diReadSpecial(),
>>>>> read the page of fixed disk inode (AIT) in raw mode in read_metapage(), the
>>>>> metapage data it returns is corrupted, which causes the nlink value of 0 to be
>>>>> assigned to the iag inode when executing copy_from_dinode(), which ultimately
>>>>> causes a deadlock when entering diFree().
>>>>>
>>>>> To avoid this, first check the nlink value of dinode before setting iag inode.
>>>>>
>>>>> [1]
>>>>> WARNING: possible recursive locking detected
>>>>> 6.12.0-rc7-syzkaller-00212-g4a5df3796467 #0 Not tainted
>>>>> --------------------------------------------
>>>>> syz-executor301/5309 is trying to acquire lock:
>>>>> ffff888044548920 (&(imap->im_aglock[index])){+.+.}-{3:3}, at: diFree+0x37c/0x2fb0 fs/jfs/jfs_imap.c:889
>>>>>
>>>>> but task is already holding lock:
>>>>> ffff888044548920 (&(imap->im_aglock[index])){+.+.}-{3:3}, at: diAlloc+0x1b6/0x1630
>>>>>
>>>>> other info that might help us debug this:
>>>>>     Possible unsafe locking scenario:
>>>>>
>>>>>           CPU0
>>>>>           ----
>>>>>      lock(&(imap->im_aglock[index]));
>>>>>      lock(&(imap->im_aglock[index]));
>>>>>
>>>>>     *** DEADLOCK ***
>>>>>
>>>>>     May be due to missing lock nesting notation
>>>>>
>>>>> 5 locks held by syz-executor301/5309:
>>>>>     #0: ffff8880422a4420 (sb_writers#9){.+.+}-{0:0}, at: mnt_want_write+0x3f/0x90 fs/namespace.c:515
>>>>>     #1: ffff88804755b390 (&type->i_mutex_dir_key#6/1){+.+.}-{3:3}, at: inode_lock_nested include/linux/fs.h:850 [inline]
>>>>>     #1: ffff88804755b390 (&type->i_mutex_dir_key#6/1){+.+.}-{3:3}, at: filename_create+0x260/0x540 fs/namei.c:4026
>>>>>     #2: ffff888044548920 (&(imap->im_aglock[index])){+.+.}-{3:3}, at: diAlloc+0x1b6/0x1630
>>>>>     #3: ffff888044548890 (&imap->im_freelock){+.+.}-{3:3}, at: diNewIAG fs/jfs/jfs_imap.c:2460 [inline]
>>>>>     #3: ffff888044548890 (&imap->im_freelock){+.+.}-{3:3}, at: diAllocExt fs/jfs/jfs_imap.c:1905 [inline]
>>>>>     #3: ffff888044548890 (&imap->im_freelock){+.+.}-{3:3}, at: diAllocAG+0x4b7/0x1e50 fs/jfs/jfs_imap.c:1669
>>>>>     #4: ffff88804755a618 (&jfs_ip->rdwrlock/1){++++}-{3:3}, at: diNewIAG fs/jfs/jfs_imap.c:2477 [inline]
>>>>>     #4: ffff88804755a618 (&jfs_ip->rdwrlock/1){++++}-{3:3}, at: diAllocExt fs/jfs/jfs_imap.c:1905 [inline]
>>>>>     #4: ffff88804755a618 (&jfs_ip->rdwrlock/1){++++}-{3:3}, at: diAllocAG+0x869/0x1e50 fs/jfs/jfs_imap.c:1669
>>>>>
>>>>> stack backtrace:
>>>>> CPU: 0 UID: 0 PID: 5309 Comm: syz-executor301 Not tainted 6.12.0-rc7-syzkaller-00212-g4a5df3796467 #0
>>>>> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
>>>>> Call Trace:
>>>>>     <TASK>
>>>>>     __dump_stack lib/dump_stack.c:94 [inline]
>>>>>     dump_stack_lvl+0x241/0x360 lib/dump_stack.c:120
>>>>>     print_deadlock_bug+0x483/0x620 kernel/locking/lockdep.c:3037
>>>>>     check_deadlock kernel/locking/lockdep.c:3089 [inline]
>>>>>     validate_chain+0x15e2/0x5920 kernel/locking/lockdep.c:3891
>>>>>     __lock_acquire+0x1384/0x2050 kernel/locking/lockdep.c:5202
>>>>>     lock_acquire+0x1ed/0x550 kernel/locking/lockdep.c:5825
>>>>>     __mutex_lock_common kernel/locking/mutex.c:608 [inline]
>>>>>     __mutex_lock+0x136/0xd70 kernel/locking/mutex.c:752
>>>>>     diFree+0x37c/0x2fb0 fs/jfs/jfs_imap.c:889
>>>>>     jfs_evict_inode+0x32d/0x440 fs/jfs/inode.c:156
>>>>>     evict+0x4e8/0x9b0 fs/inode.c:725
>>>>>     diFreeSpecial fs/jfs/jfs_imap.c:552 [inline]
>>>>>     duplicateIXtree+0x3c6/0x550 fs/jfs/jfs_imap.c:3022
>>>>>     diNewIAG fs/jfs/jfs_imap.c:2597 [inline]
>>>>>     diAllocExt fs/jfs/jfs_imap.c:1905 [inline]
>>>>>     diAllocAG+0x17dc/0x1e50 fs/jfs/jfs_imap.c:1669
>>>>>     diAlloc+0x1d2/0x1630 fs/jfs/jfs_imap.c:1590
>>>>>     ialloc+0x8f/0x900 fs/jfs/jfs_inode.c:56
>>>>>     jfs_mkdir+0x1c5/0xba0 fs/jfs/namei.c:225
>>>>>     vfs_mkdir+0x2f9/0x4f0 fs/namei.c:4257
>>>>>     do_mkdirat+0x264/0x3a0 fs/namei.c:4280
>>>>>     __do_sys_mkdirat fs/namei.c:4295 [inline]
>>>>>     __se_sys_mkdirat fs/namei.c:4293 [inline]
>>>>>     __x64_sys_mkdirat+0x87/0xa0 fs/namei.c:4293
>>>>>     do_syscall_x64 arch/x86/entry/common.c:52 [inline]
>>>>>     do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
>>>>>     entry_SYSCALL_64_after_hwframe+0x77/0x7f
>>>>
>>>> I'm taking this patch, but making a change. It's a little cleaner to check ip->i_nlink after calling copy_from_dinode.
>>>>
>>>>>
>>>>> Reported-by: syzbot+355da3b3a74881008e8f@...kaller.appspotmail.com
>>>>> Closes: https://syzkaller.appspot.com/bug?extid=355da3b3a74881008e8f
>>>>> Signed-off-by: Edward Adam Davis <eadavis@...com>
>>>>> ---
>>>>> V1 -> V2: if the nlink of disk inode is 0 return -EIO
>>>>> V2 -> V3: move the checking to diReadSpecial
>>>>>
>>>>> ---
>>>>>     fs/jfs/jfs_imap.c | 2 +-
>>>>>     1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/fs/jfs/jfs_imap.c b/fs/jfs/jfs_imap.c
>>>>> index 0cedaccb7218..25bb3485da3b 100644
>>>>> --- a/fs/jfs/jfs_imap.c
>>>>> +++ b/fs/jfs/jfs_imap.c
>>>>> @@ -460,7 +460,7 @@ struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
>>>>>     	dp += inum % 8;		/* 8 inodes per 4K page */
>>>>>     
>>>>>     	/* copy on-disk inode to in-memory inode */
>>>>> -	if ((copy_from_dinode(dp, ip)) != 0) {
>>>>> +	if (!le32_to_cpu(dp->di_nlink) || (copy_from_dinode(dp, ip)) != 0) {
>>>>>     		/* handle bad return by returning NULL for ip */
>>>>>     		set_nlink(ip, 1);	/* Don't want iput() deleting it */
>>>>>     		iput(ip);
>>>>
>>>> My change:
>>>>
>>>> diff --git a/fs/jfs/jfs_imap.c b/fs/jfs/jfs_imap.c
>>>> index 298445f6d3d4..ecb8e05b8b84 100644
>>>> --- a/fs/jfs/jfs_imap.c
>>>> +++ b/fs/jfs/jfs_imap.c
>>>> @@ -456,7 +456,7 @@ struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
>>>>     	dp += inum % 8;		/* 8 inodes per 4K page */
>>>>     
>>>>     	/* copy on-disk inode to in-memory inode */
>>>> -	if ((copy_from_dinode(dp, ip)) != 0) {
>>>> +	if ((copy_from_dinode(dp, ip) != 0) || (ip->i_nlink == 0)) {
>>> This is incorrect. The purpose of adding this check is to prevent copy_from_dinode()
>>> from using dip->i_nlink with a value of 0 to assign to ip.
>>>>     		/* handle bad return by returning NULL for ip */
>>>>     		set_nlink(ip, 1);	/* Don't want iput() deleting it */
>>
>> It will get set  to 1 right here ^^^
> Things that can be determined by "di_nlink" before executing copy_from_dinode(),
> Why let the CPU run copy_from_dinode() for an extra time before checking?
> Isn't this a waste of CPU?

It's an exceptional case. It's very, very unlikely to fail, so the extra 
cpu cycles that are executed in the common case are not a concern.

> 
> BR,
> Edward
> 

Powered by blists - more mailing lists