summaryrefslogtreecommitdiff
path: root/fs/btrfs/ordered-data.c
diff options
context:
space:
mode:
authorJosef Bacik <jbacik@fb.com>2015-09-24 23:17:39 +0300
committerChris Mason <clm@fb.com>2015-10-22 04:51:40 +0300
commit161c3549b45aeef05451b6822d8aaaf39c7bedce (patch)
treef14c534323f7b8d6bf181c9610df66313358360b /fs/btrfs/ordered-data.c
parenta408365c62762c30419018587cffd2b89836434e (diff)
downloadlinux-161c3549b45aeef05451b6822d8aaaf39c7bedce.tar.xz
Btrfs: change how we wait for pending ordered extents
We have a mechanism to make sure we don't lose updates for ordered extents that were logged in the transaction that is currently running. We add the ordered extent to a transaction list and then the transaction waits on all the ordered extents in that list. However are substantially large file systems this list can be extremely large, and can give us soft lockups, since the ordered extents don't remove themselves from the list when they do complete. To fix this we simply add a counter to the transaction that is incremented any time we have a logged extent that needs to be completed in the current transaction. Then when the ordered extent finally completes it decrements the per transaction counter and wakes up the transaction if we are the last ones. This will eliminate the softlockup. Thanks, Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Chris Mason <clm@fb.com>
Diffstat (limited to 'fs/btrfs/ordered-data.c')
-rw-r--r--fs/btrfs/ordered-data.c64
1 files changed, 50 insertions, 14 deletions
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 071005f008c1..8c27292ea9ea 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -490,15 +490,16 @@ void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
spin_lock_irq(&log->log_extents_lock[index]);
while (!list_empty(&log->logged_list[index])) {
+ struct inode *inode;
ordered = list_first_entry(&log->logged_list[index],
struct btrfs_ordered_extent,
log_list);
list_del_init(&ordered->log_list);
+ inode = ordered->inode;
spin_unlock_irq(&log->log_extents_lock[index]);
if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
!test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
- struct inode *inode = ordered->inode;
u64 start = ordered->file_offset;
u64 end = ordered->file_offset + ordered->len - 1;
@@ -509,20 +510,25 @@ void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
&ordered->flags));
/*
- * If our ordered extent completed it means it updated the
- * fs/subvol and csum trees already, so no need to make the
- * current transaction's commit wait for it, as we end up
- * holding memory unnecessarily and delaying the inode's iput
- * until the transaction commit (we schedule an iput for the
- * inode when the ordered extent's refcount drops to 0), which
- * prevents it from being evictable until the transaction
- * commits.
+ * In order to keep us from losing our ordered extent
+ * information when committing the transaction we have to make
+ * sure that any logged extents are completed when we go to
+ * commit the transaction. To do this we simply increase the
+ * current transactions pending_ordered counter and decrement it
+ * when the ordered extent completes.
*/
- if (test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags))
- btrfs_put_ordered_extent(ordered);
- else
- list_add_tail(&ordered->trans_list, &trans->ordered);
-
+ if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
+ struct btrfs_ordered_inode_tree *tree;
+
+ tree = &BTRFS_I(inode)->ordered_tree;
+ spin_lock_irq(&tree->lock);
+ if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
+ set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
+ atomic_inc(&trans->transaction->pending_ordered);
+ }
+ spin_unlock_irq(&tree->lock);
+ }
+ btrfs_put_ordered_extent(ordered);
spin_lock_irq(&log->log_extents_lock[index]);
}
spin_unlock_irq(&log->log_extents_lock[index]);
@@ -584,6 +590,7 @@ void btrfs_remove_ordered_extent(struct inode *inode,
struct btrfs_ordered_inode_tree *tree;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct rb_node *node;
+ bool dec_pending_ordered = false;
tree = &BTRFS_I(inode)->ordered_tree;
spin_lock_irq(&tree->lock);
@@ -593,8 +600,37 @@ void btrfs_remove_ordered_extent(struct inode *inode,
if (tree->last == node)
tree->last = NULL;
set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
+ if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
+ dec_pending_ordered = true;
spin_unlock_irq(&tree->lock);
+ /*
+ * The current running transaction is waiting on us, we need to let it
+ * know that we're complete and wake it up.
+ */
+ if (dec_pending_ordered) {
+ struct btrfs_transaction *trans;
+
+ /*
+ * The checks for trans are just a formality, it should be set,
+ * but if it isn't we don't want to deref/assert under the spin
+ * lock, so be nice and check if trans is set, but ASSERT() so
+ * if it isn't set a developer will notice.
+ */
+ spin_lock(&root->fs_info->trans_lock);
+ trans = root->fs_info->running_transaction;
+ if (trans)
+ atomic_inc(&trans->use_count);
+ spin_unlock(&root->fs_info->trans_lock);
+
+ ASSERT(trans);
+ if (trans) {
+ if (atomic_dec_and_test(&trans->pending_ordered))
+ wake_up(&trans->pending_wait);
+ btrfs_put_transaction(trans);
+ }
+ }
+
spin_lock(&root->ordered_extent_lock);
list_del_init(&entry->root_extent_list);
root->nr_ordered_extents--;