Re: [PATCH v4] jbd2: Fix data missing when reusing bh which is ready to be checkpointed

[Jan Kara]

On Tue 10-01-23 09:53:27, Zhihao Cheng wrote:
> Following process will make data lost and could lead to a filesystem
> corrupted problem:
> 
> 1. jh(bh) is inserted into T1->t_checkpoint_list, bh is dirty, and
>    jh->b_transaction = NULL
> 2. T1 is added into journal->j_checkpoint_transactions.
> 3. Get bh prepare to write while doing checkpoing:
>            PA				    PB
>    do_get_write_access             jbd2_log_do_checkpoint
>     spin_lock(&jh->b_state_lock)
>      if (buffer_dirty(bh))
>       clear_buffer_dirty(bh)   // clear buffer dirty
>        set_buffer_jbddirty(bh)
> 				    transaction =
> 				    journal->j_checkpoint_transactions
> 				    jh = transaction->t_checkpoint_list
> 				    if (!buffer_dirty(bh))
> 		                      __jbd2_journal_remove_checkpoint(jh)
> 				      // bh won't be flushed
> 		                    jbd2_cleanup_journal_tail
>     __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved)
> 4. Aborting journal/Power-cut before writing latest bh on journal area.
> 
> In this way we get a corrupted filesystem with bh's data lost.
> 
> Fix it by moving the clearing of buffer_dirty bit just before the call
> to __jbd2_journal_file_buffer(), both bit clearing and jh->b_transaction
> assignment are under journal->j_list_lock locked, so that
> jbd2_log_do_checkpoint() will wait until jh's new transaction fininshed
> even bh is currently not dirty. And journal_shrink_one_cp_list() won't
> remove jh from checkpoint list if the buffer head is reused in
> do_get_write_access().
> 
> Fetch a reproducer in [Link].
> 
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=216898
> Cc: <stable@xxxxxxxxxx>
> Signed-off-by: Zhihao Cheng <chengzhihao1@xxxxxxxxxx>
> Signed-off-by: zhanchengbin <zhanchengbin1@xxxxxxxxxx>
> Suggested-by: Jan Kara <jack@xxxxxxx>

The patch looks good to me! Feel free to add:

Reviewed-by: Jan Kara <jack@xxxxxxx>

								Honza

> ---
>  v1->v2: Adopt Jan's suggestion, move the clearing of buffer_dirty bit
> 	 and __jbd2_journal_file_buffer() inside journal->j_list_lock
> 	 locking area.
>  v2->v3: Remove redundant assertions in in branch 'if (jh->b_transaction)'
>          Add reproducer link in commit message.
>  v3->v4: Rephrase the comment for set_buffer_jbddirty().
>  fs/jbd2/transaction.c | 50 +++++++++++++++++++++++++------------------
>  1 file changed, 29 insertions(+), 21 deletions(-)
> 
> diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
> index 6a404ac1c178..15de1385012e 100644
> --- a/fs/jbd2/transaction.c
> +++ b/fs/jbd2/transaction.c
> @@ -1010,36 +1010,28 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
>  	 * ie. locked but not dirty) or tune2fs (which may actually have
>  	 * the buffer dirtied, ugh.)  */
>  
> -	if (buffer_dirty(bh)) {
> +	if (buffer_dirty(bh) && jh->b_transaction) {
> +		warn_dirty_buffer(bh);
>  		/*
> -		 * First question: is this buffer already part of the current
> -		 * transaction or the existing committing transaction?
> -		 */
> -		if (jh->b_transaction) {
> -			J_ASSERT_JH(jh,
> -				jh->b_transaction == transaction ||
> -				jh->b_transaction ==
> -					journal->j_committing_transaction);
> -			if (jh->b_next_transaction)
> -				J_ASSERT_JH(jh, jh->b_next_transaction ==
> -							transaction);
> -			warn_dirty_buffer(bh);
> -		}
> -		/*
> -		 * In any case we need to clean the dirty flag and we must
> -		 * do it under the buffer lock to be sure we don't race
> -		 * with running write-out.
> +		 * We need to clean the dirty flag and we must do it under the
> +		 * buffer lock to be sure we don't race with running write-out.
>  		 */
>  		JBUFFER_TRACE(jh, "Journalling dirty buffer");
>  		clear_buffer_dirty(bh);
> +		/*
> +		 * The buffer is going to be added to BJ_Reserved list now and
> +		 * nothing guarantees jbd2_journal_dirty_metadata() will be
> +		 * ever called for it. So we need to set jbddirty bit here to
> +		 * make sure the buffer is dirtied and written out when the
> +		 * journaling machinery is done with it.
> +		 */
>  		set_buffer_jbddirty(bh);
>  	}
>  
> -	unlock_buffer(bh);
> -
>  	error = -EROFS;
>  	if (is_handle_aborted(handle)) {
>  		spin_unlock(&jh->b_state_lock);
> +		unlock_buffer(bh);
>  		goto out;
>  	}
>  	error = 0;
> @@ -1049,8 +1041,10 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
>  	 * b_next_transaction points to it
>  	 */
>  	if (jh->b_transaction == transaction ||
> -	    jh->b_next_transaction == transaction)
> +	    jh->b_next_transaction == transaction) {
> +		unlock_buffer(bh);
>  		goto done;
> +	}
>  
>  	/*
>  	 * this is the first time this transaction is touching this buffer,
> @@ -1074,10 +1068,24 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
>  		 */
>  		smp_wmb();
>  		spin_lock(&journal->j_list_lock);
> +		if (test_clear_buffer_dirty(bh)) {
> +			/*
> +			 * Execute buffer dirty clearing and jh->b_transaction
> +			 * assignment under journal->j_list_lock locked to
> +			 * prevent bh being removed from checkpoint list if
> +			 * the buffer is in an intermediate state (not dirty
> +			 * and jh->b_transaction is NULL).
> +			 */
> +			JBUFFER_TRACE(jh, "Journalling dirty buffer");
> +			set_buffer_jbddirty(bh);
> +		}
>  		__jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
>  		spin_unlock(&journal->j_list_lock);
> +		unlock_buffer(bh);
>  		goto done;
>  	}
> +	unlock_buffer(bh);
> +
>  	/*
>  	 * If there is already a copy-out version of this buffer, then we don't
>  	 * need to make another one
> -- 
> 2.31.1
> 
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR