Btrfs: fix deadlock in wait_for_more_refs

Commit a168650c introduced a waiting mechanism to prevent busy waiting in
btrfs_run_delayed_refs. This can deadlock with btrfs_run_ordered_operations,
where a tree_mod_seq is held while waiting for the io to complete, while
the end_io calls btrfs_run_delayed_refs.
This whole mechanism is unnecessary. If not enough runnable refs are
available to satisfy count, just return as count is more like a guideline
than a strict requirement.
In case we have to run all refs, commit transaction makes sure that no
other threads are working in the transaction anymore, so we just assert
here that no refs are blocked.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>

1 files changed, 21 insertions, 56 deletions

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 45c69c4184c..d3df65f83b5 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2318,12 +2318,6 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
                 ref->in_tree = 0;
                 rb_erase(&ref->rb_node, &delayed_refs->root);
                 delayed_refs->num_entries--;
-                /*
-                 * we modified num_entries, but as we're currently running
-                 * delayed refs, skip
-                 *     wake_up(&delayed_refs->seq_wait);
-                 * here.
-                 */
                 spin_unlock(&delayed_refs->lock);
 
                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
@@ -2350,22 +2344,6 @@ next:
         return count;
 }
 
-static void wait_for_more_refs(struct btrfs_fs_info *fs_info,
-                               struct btrfs_delayed_ref_root *delayed_refs,
-                               unsigned long num_refs,
-                               struct list_head *first_seq)
-{
-        spin_unlock(&delayed_refs->lock);
-        pr_debug("waiting for more refs (num %ld, first %p)\n",
-                 num_refs, first_seq);
-        wait_event(fs_info->tree_mod_seq_wait,
-                   num_refs != delayed_refs->num_entries ||
-                   fs_info->tree_mod_seq_list.next != first_seq);
-        pr_debug("done waiting for more refs (num %ld, first %p)\n",
-                 delayed_refs->num_entries, fs_info->tree_mod_seq_list.next);
-        spin_lock(&delayed_refs->lock);
-}
-
 #ifdef SCRAMBLE_DELAYED_REFS
 /*
  * Normally delayed refs get processed in ascending bytenr order. This
@@ -2460,13 +2438,11 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
         struct btrfs_delayed_ref_root *delayed_refs;
         struct btrfs_delayed_ref_node *ref;
         struct list_head cluster;
-        struct list_head *first_seq = NULL;
         int ret;
         u64 delayed_start;
         int run_all = count == (unsigned long)-1;
         int run_most = 0;
-        unsigned long num_refs = 0;
-        int consider_waiting;
+        int loops;
 
         /* We'll clean this up in btrfs_cleanup_transaction */
         if (trans->aborted)
@@ -2484,7 +2460,7 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
         delayed_refs = &trans->transaction->delayed_refs;
         INIT_LIST_HEAD(&cluster);
 again:
-        consider_waiting = 0;
+        loops = 0;
         spin_lock(&delayed_refs->lock);
 
 #ifdef SCRAMBLE_DELAYED_REFS
@@ -2512,31 +2488,6 @@ again:
                 if (ret)
                         break;
 
-                if (delayed_start >= delayed_refs->run_delayed_start) {
-                        if (consider_waiting == 0) {
-                                /*
-                                 * btrfs_find_ref_cluster looped. let's do one
-                                 * more cycle. if we don't run any delayed ref
-                                 * during that cycle (because we can't because
-                                 * all of them are blocked) and if the number of
-                                 * refs doesn't change, we avoid busy waiting.
-                                 */
-                                consider_waiting = 1;
-                                num_refs = delayed_refs->num_entries;
-                                first_seq = root->fs_info->tree_mod_seq_list.next;
-                        } else {
-                                wait_for_more_refs(root->fs_info, delayed_refs,
-                                                   num_refs, first_seq);
-                                /*
-                                 * after waiting, things have changed. we
-                                 * dropped the lock and someone else might have
-                                 * run some refs, built new clusters and so on.
-                                 * therefore, we restart staleness detection.
-                                 */
-                                consider_waiting = 0;
-                        }
-                }
-
                 ret = run_clustered_refs(trans, root, &cluster);
                 if (ret < 0) {
                         spin_unlock(&delayed_refs->lock);
@@ -2549,9 +2500,26 @@ again:
                 if (count == 0)
                         break;
 
-                if (ret || delayed_refs->run_delayed_start == 0) {
+                if (delayed_start >= delayed_refs->run_delayed_start) {
+                        if (loops == 0) {
+                                /*
+                                 * btrfs_find_ref_cluster looped. let's do one
+                                 * more cycle. if we don't run any delayed ref
+                                 * during that cycle (because we can't because
+                                 * all of them are blocked), bail out.
+                                 */
+                                loops = 1;
+                        } else {
+                                /*
+                                 * no runnable refs left, stop trying
+                                 */
+                                BUG_ON(run_all);
+                                break;
+                        }
+                }
+                if (ret) {
                         /* refs were run, let's reset staleness detection */
-                        consider_waiting = 0;
+                        loops = 0;
                 }
         }
 
@@ -5296,9 +5264,6 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
         rb_erase(&head->node.rb_node, &delayed_refs->root);
 
         delayed_refs->num_entries--;
-        smp_mb();
-        if (waitqueue_active(&root->fs_info->tree_mod_seq_wait))
-                wake_up(&root->fs_info->tree_mod_seq_wait);
 
         /*
          * we don't take a ref on the node because we're removing it from the