Re: [PATCH v2 1/1] jbd2: gracefully abort on checkpointing state corruptions

[Jan Kara]

On Tue 10-03-26 21:15:48, Milos Nikic wrote:
> This patch targets two internal state machine invariants in checkpoint.c
> residing inside functions that natively return integer error codes.
> 
> - In jbd2_cleanup_journal_tail(): A blocknr of 0 indicates a severely
> corrupted journal superblock. Replaced the J_ASSERT with a WARN_ON_ONCE
> and a graceful journal abort, returning -EFSCORRUPTED.
> 
> - In jbd2_log_do_checkpoint(): Replaced the J_ASSERT_BH checking for
> an unexpected buffer_jwrite state. If the warning triggers, we
> explicitly drop the just-taken get_bh() reference and call __flush_batch()
> to safely clean up any previously queued buffers in the j_chkpt_bhs array,
> preventing a memory leak before returning -EFSCORRUPTED.
> 
> Signed-off-by: Milos Nikic <nikic.milos@xxxxxxxxx>
> Reviewed-by: Andreas Dilger <adilger@xxxxxxxxx>
> Reviewed-by: Zhang Yi <yi.zhang@xxxxxxxxxx>
> Reviewed-by: Baokun Li <libaokun@xxxxxxxxxxxxxxxxx>

Looks good to me. Feel free to add:

Reviewed-by: Jan Kara <jack@xxxxxxx>

								Honza

> ---
>  Changes in v2:
> 
>     Replaced the -EUCLEAN error code with -EFSCORRUPTED to better align with ext4/jbd2 semantics for on-disk metadata inconsistencies (per Baokun's review).
> 
>     Reordered the error path in jbd2_log_do_checkpoint() so that jbd2_journal_abort() is called after __flush_batch(). This ensures cleanly batched buffers are logically flushed before the journal kill switch is flipped.
> 
>     Collected Reviewed-by tags from Andreas Dilger, Zhang Yi, and Baokun Li.
> 
> Changes in v1:
> 
>     Initial implementation converting J_ASSERTs in jbd2_cleanup_journal_tail() and jbd2_log_do_checkpoint() to WARN_ON_ONCE and graceful journal aborts.
> 
>  fs/jbd2/checkpoint.c | 15 +++++++++++++--
>  1 file changed, 13 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
> index de89c5bef607..1508e2f54462 100644
> --- a/fs/jbd2/checkpoint.c
> +++ b/fs/jbd2/checkpoint.c
> @@ -267,7 +267,15 @@ int jbd2_log_do_checkpoint(journal_t *journal)
>  			 */
>  			BUFFER_TRACE(bh, "queue");
>  			get_bh(bh);
> -			J_ASSERT_BH(bh, !buffer_jwrite(bh));
> +			if (WARN_ON_ONCE(buffer_jwrite(bh))) {
> +				put_bh(bh); /* drop the ref we just took */
> +				spin_unlock(&journal->j_list_lock);
> +				/* Clean up any previously batched buffers */
> +				if (batch_count)
> +					__flush_batch(journal, &batch_count);
> +				jbd2_journal_abort(journal, -EFSCORRUPTED);
> +				return -EFSCORRUPTED;
> +			}
>  			journal->j_chkpt_bhs[batch_count++] = bh;
>  			transaction->t_chp_stats.cs_written++;
>  			transaction->t_checkpoint_list = jh->b_cpnext;
> @@ -325,7 +333,10 @@ int jbd2_cleanup_journal_tail(journal_t *journal)
>  
>  	if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr))
>  		return 1;
> -	J_ASSERT(blocknr != 0);
> +	if (WARN_ON_ONCE(blocknr == 0)) {
> +		jbd2_journal_abort(journal, -EFSCORRUPTED);
> +		return -EFSCORRUPTED;
> +	}
>  
>  	/*
>  	 * We need to make sure that any blocks that were recently written out
> -- 
> 2.53.0
> 
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR