linux-kernel - Re: [PATCH 1/2] ext4: Fix use-after-free about sbi->s_mmp

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 6 Jul 2021 13:11:37 +0200
From:   Jan Kara <jack@...e.cz>
To:     Theodore Ts'o <tytso@....edu>
Cc:     Jan Kara <jack@...e.cz>, Ye Bin <yebin10@...wei.com>,
        adilger.kernel@...ger.ca, linux-ext4@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/2] ext4: Fix use-after-free about sbi->s_mmp_tsk

On Mon 05-07-21 16:35:28, Theodore Ts'o wrote:
> On Mon, Jul 05, 2021 at 01:15:48PM +0200, Jan Kara wrote:
> > 
> > That being said for this scheme spinlock is enough, you don't need a mutex
> > for s_mmp_lock.
> 
> I think we can solve this without using using either a spinlock or a
> mutex, and it's a smaller and simpler patch as a result.  (This is the
> -v2 version of this patch, which removes an unused label compared to
> the earlier version.)

Yeah, what you suggest is probably simpler. Some comments below.

> From 22ebc97aac75e27a5fd11acdb2bc3030d1da58d1 Mon Sep 17 00:00:00 2001
> From: Theodore Ts'o <tytso@....edu>
> Date: Fri, 2 Jul 2021 12:45:02 -0400
> Subject: [PATCH] ext4: fix possible UAF when remounting r/o a mmp-protected file system
> 
> After commit 618f003199c6 ("ext4: fix memory leak in
> ext4_fill_super"), after the file system is remounted read-only, there
> is a race where the kmmpd thread can exit, causing sbi->s_mmp_tsk to
> point at freed memory, which the call to ext4_stop_mmpd() can trip
> over.
> 
> Fix this by only allowing kmmpd() to exit when it is stopped via
> ext4_stop_mmpd().
> 
> Link: https://lore.kernel.org/r/e525c0bf7b18da426bb3d3dd63830a3f85218a9e.1625244710.git.tytso@mit.edu
> Reported-by: Ye Bin <yebin10@...wei.com>
> Bug-Report-Link: <20210629143603.2166962-1-yebin10@...wei.com>
> Signed-off-by: Theodore Ts'o <tytso@....edu>
> ---
>  fs/ext4/mmp.c   | 33 +++++++++++++++++----------------
>  fs/ext4/super.c |  6 +++++-
>  2 files changed, 22 insertions(+), 17 deletions(-)
> 
> diff --git a/fs/ext4/mmp.c b/fs/ext4/mmp.c
> index 6cb598b549ca..1e95cee3d8b7 100644
> --- a/fs/ext4/mmp.c
> +++ b/fs/ext4/mmp.c
> @@ -157,6 +157,17 @@ static int kmmpd(void *data)
>  	       sizeof(mmp->mmp_nodename));
>  
>  	while (!kthread_should_stop()) {
> +		if (!(le32_to_cpu(es->s_feature_incompat) &
> +		    EXT4_FEATURE_INCOMPAT_MMP)) {

We can probably use ext4_has_feature_mmp() macro when changing this?

> +			ext4_warning(sb, "kmmpd being stopped since MMP feature"
> +				     " has been disabled.");
> +			goto wait_to_exit;
> +		}
> +		if (sb_rdonly(sb)) {
> +			if (!kthread_should_stop())
> +				schedule_timeout_interruptible(HZ);

Cannot this effectively block remount RO for 1s when we wait for kmmpd to
exit? I think doing 'break' when we detected RO super is fine. We'll write
the mmp block and then wait for kthread_should_stop() condition as in any
other abort case. Am I missing something?

> +			continue;
> +		}
>  		if (++seq > EXT4_MMP_SEQ_MAX)
>  			seq = 1;
>  
> @@ -177,16 +188,6 @@ static int kmmpd(void *data)
>  			failed_writes++;
>  		}
>  
> -		if (!(le32_to_cpu(es->s_feature_incompat) &
> -		    EXT4_FEATURE_INCOMPAT_MMP)) {
> -			ext4_warning(sb, "kmmpd being stopped since MMP feature"
> -				     " has been disabled.");
> -			goto exit_thread;
> -		}
> -
> -		if (sb_rdonly(sb))
> -			break;
> -
>  		diff = jiffies - last_update_time;
>  		if (diff < mmp_update_interval * HZ)
>  			schedule_timeout_interruptible(mmp_update_interval *
> @@ -207,7 +208,7 @@ static int kmmpd(void *data)
>  				ext4_error_err(sb, -retval,
>  					       "error reading MMP data: %d",
>  					       retval);
> -				goto exit_thread;
> +				goto wait_to_exit;
>  			}
>  
>  			mmp_check = (struct mmp_struct *)(bh_check->b_data);
> @@ -221,7 +222,7 @@ static int kmmpd(void *data)
>  				ext4_error_err(sb, EBUSY, "abort");
>  				put_bh(bh_check);
>  				retval = -EBUSY;
> -				goto exit_thread;
> +				goto wait_to_exit;
>  			}
>  			put_bh(bh_check);
>  		}
> @@ -242,9 +243,11 @@ static int kmmpd(void *data)
>  	mmp->mmp_seq = cpu_to_le32(EXT4_MMP_SEQ_CLEAN);
>  	mmp->mmp_time = cpu_to_le64(ktime_get_real_seconds());
>  
> -	retval = write_mmp_block(sb, bh);
> +	return write_mmp_block(sb, bh);
>  
> -exit_thread:
> +wait_to_exit:
> +	while (!kthread_should_stop())
> +		schedule();

This makes me a bit nervous that we could unnecessarily burn CPU for
potentially a long time (e.g. if somebody uses tune2fs to disable MMP, we
would be sitting in this loop until the fs in remounted / unmounted). So
maybe we should have something like:

	while (!kthread_should_stop()) {
		set_task_state(TASK_INTERRUPTIBLE);
		if (!kthread_should_stop())
			schedule();
	}

This should safely synchronize with (and not miss wakeup from)
kthread_stop() since that first sets KTHREAD_SHOULD_STOP and after that
calls wake_up_process().

								Honza
-- 
Jan Kara <jack@...e.com>
SUSE Labs, CR