[PATCH 1/2 v2] btrfs: fix error handling when run_delayed_extent_op fails

[PATCH 1/2 v2] btrfs: fix error handling when run_delayed_extent_op fails

[jeffm]

From: Jeff Mahoney <jeffm@suse.com>

In __btrfs_run_delayed_refs, the error path when run_delayed_extent_op
fails sets locked_ref->processing = 0 but doesn't re-increment
delayed_refs->num_heads_ready.  As a result, we end up triggering
the WARN_ON in btrfs_select_ref_head.

Fixes: d7df2c796d7 (Btrfs: attach delayed ref updates to delayed ref heads)
Reported-by: Jon Nelson <jnelson-suse@jamponi.net>
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
---
 fs/btrfs/extent-tree.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index db74889..d74adf1 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2576,7 +2576,10 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 					 */
 					if (must_insert_reserved)
 						locked_ref->must_insert_reserved = 1;
+					spin_lock(&delayed_refs->lock);
 					locked_ref->processing = 0;
+					delayed_refs->num_heads_ready++;
+					spin_unlock(&delayed_refs->lock);
 					btrfs_debug(fs_info,
 						    "run_delayed_extent_op returned %d",
 						    ret);
-- 
1.8.5.6

[PATCH 2/2] btrfs: fix locking when we put back a delayed ref that's too new

[jeffm]

From: Jeff Mahoney <jeffm@suse.com>

In __btrfs_run_delayed_refs, when we put back a delayed ref that's too
new, we have already dropped the lock on locked_ref when we set
->processing = 0.

This patch keeps the lock to cover that assignment.

Fixes: d7df2c796d7 (Btrfs: attach delayed ref updates to delayed ref heads)
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
---
 fs/btrfs/extent-tree.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index d74adf1..930ac8e 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2526,11 +2526,11 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 		if (ref && ref->seq &&
 		    btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
 			spin_unlock(&locked_ref->lock);
-			btrfs_delayed_ref_unlock(locked_ref);
 			spin_lock(&delayed_refs->lock);
 			locked_ref->processing = 0;
 			delayed_refs->num_heads_ready++;
 			spin_unlock(&delayed_refs->lock);
+			btrfs_delayed_ref_unlock(locked_ref);
 			locked_ref = NULL;
 			cond_resched();
 			count++;
-- 
1.8.5.6

Re: [PATCH 2/2] btrfs: fix locking when we put back a delayed ref that's too new

[Liu Bo]

On Tue, Dec 20, 2016 at 01:28:28PM -0500, jeffm@suse.com wrote:
> From: Jeff Mahoney <jeffm@suse.com>
> 
> In __btrfs_run_delayed_refs, when we put back a delayed ref that's too
> new, we have already dropped the lock on locked_ref when we set
> ->processing = 0.
> 
> This patch keeps the lock to cover that assignment.
> 
> Fixes: d7df2c796d7 (Btrfs: attach delayed ref updates to delayed ref heads)
> Signed-off-by: Jeff Mahoney <jeffm@suse.com>
> ---
>  fs/btrfs/extent-tree.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> index d74adf1..930ac8e 100644
> --- a/fs/btrfs/extent-tree.c
> +++ b/fs/btrfs/extent-tree.c
> @@ -2526,11 +2526,11 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
>  		if (ref && ref->seq &&
>  		    btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
>  			spin_unlock(&locked_ref->lock);
> -			btrfs_delayed_ref_unlock(locked_ref);
>  			spin_lock(&delayed_refs->lock);
>  			locked_ref->processing = 0;
>  			delayed_refs->num_heads_ready++;
>  			spin_unlock(&delayed_refs->lock);
> +			btrfs_delayed_ref_unlock(locked_ref);

I don't think that this would end up a deadlock as we use mutex_try_lock
for head->mutex everywhere, but I'd rather have it cleaned up.

Reviewed-by: Liu Bo <bo.li.liu@oracle.com>

Thanks,

-liubo
>  			locked_ref = NULL;
>  			cond_resched();
>  			count++;
> -- 
> 1.8.5.6
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH 1/2 v2] btrfs: fix error handling when run_delayed_extent_op fails

[Liu Bo]

On Tue, Dec 20, 2016 at 01:28:27PM -0500, jeffm@suse.com wrote:
> From: Jeff Mahoney <jeffm@suse.com>
> 
> In __btrfs_run_delayed_refs, the error path when run_delayed_extent_op
> fails sets locked_ref->processing = 0 but doesn't re-increment
> delayed_refs->num_heads_ready.  As a result, we end up triggering
> the WARN_ON in btrfs_select_ref_head.
> 
> Fixes: d7df2c796d7 (Btrfs: attach delayed ref updates to delayed ref heads)
> Reported-by: Jon Nelson <jnelson-suse@jamponi.net>
> Signed-off-by: Jeff Mahoney <jeffm@suse.com>
> ---
>  fs/btrfs/extent-tree.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> index db74889..d74adf1 100644
> --- a/fs/btrfs/extent-tree.c
> +++ b/fs/btrfs/extent-tree.c
> @@ -2576,7 +2576,10 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
>  					 */
>  					if (must_insert_reserved)
>  						locked_ref->must_insert_reserved = 1;
> +					spin_lock(&delayed_refs->lock);
>  					locked_ref->processing = 0;
> +					delayed_refs->num_heads_ready++;
> +					spin_unlock(&delayed_refs->lock);

Looks good to me.

Reviewed-by: Liu Bo <bo.li.liu@oracle.com>

Thanks,

-liubo
>  					btrfs_debug(fs_info,
>  						    "run_delayed_extent_op returned %d",
>  						    ret);
> -- 
> 1.8.5.6
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html