aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorSage Weil <sage@newdream.net>2010-10-29 15:37:34 -0400
committerChris Mason <chris.mason@oracle.com>2010-10-29 15:37:34 -0400
commit99d16cbcaf694c803a1b6bf7e851694ffe1d255d (patch)
treedacf7f8eca75dc0d92566b5d3876f3efe8196e35 /fs
parentfccdae435c1b295cca546f23f6f43126a28ffac3 (diff)
Btrfs: fix deadlock in btrfs_commit_transaction
We calculate timeout (either 1 or MAX_SCHEDULE_TIMEOUT) based on whether num_writers > 1 or should_grow at the top of the loop. Then, much much later, we wait for that timeout if either num_writers or should_grow is true. However, it's possible for a racing process (calling btrfs_end_transaction()) to decrement num_writers such that we wait forever instead of for 1. Fix this by deciding how long to wait when we wait. Include a smp_mb() before checking if the waitqueue is active to ensure the num_writers is visible. Signed-off-by: Sage Weil <sage@newdream.net> Signed-off-by: Chris Mason <chris.mason@oracle.com>
Diffstat (limited to 'fs')
-rw-r--r--fs/btrfs/transaction.c13
1 files changed, 5 insertions, 8 deletions
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 325d9a5f0128..700dc4b34ada 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -402,6 +402,7 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
402 WARN_ON(cur_trans->num_writers < 1); 402 WARN_ON(cur_trans->num_writers < 1);
403 cur_trans->num_writers--; 403 cur_trans->num_writers--;
404 404
405 smp_mb();
405 if (waitqueue_active(&cur_trans->writer_wait)) 406 if (waitqueue_active(&cur_trans->writer_wait))
406 wake_up(&cur_trans->writer_wait); 407 wake_up(&cur_trans->writer_wait);
407 put_transaction(cur_trans); 408 put_transaction(cur_trans);
@@ -1010,7 +1011,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1010 struct btrfs_root *root) 1011 struct btrfs_root *root)
1011{ 1012{
1012 unsigned long joined = 0; 1013 unsigned long joined = 0;
1013 unsigned long timeout = 1;
1014 struct btrfs_transaction *cur_trans; 1014 struct btrfs_transaction *cur_trans;
1015 struct btrfs_transaction *prev_trans = NULL; 1015 struct btrfs_transaction *prev_trans = NULL;
1016 DEFINE_WAIT(wait); 1016 DEFINE_WAIT(wait);
@@ -1081,11 +1081,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1081 snap_pending = 1; 1081 snap_pending = 1;
1082 1082
1083 WARN_ON(cur_trans != trans->transaction); 1083 WARN_ON(cur_trans != trans->transaction);
1084 if (cur_trans->num_writers > 1)
1085 timeout = MAX_SCHEDULE_TIMEOUT;
1086 else if (should_grow)
1087 timeout = 1;
1088
1089 mutex_unlock(&root->fs_info->trans_mutex); 1084 mutex_unlock(&root->fs_info->trans_mutex);
1090 1085
1091 if (flush_on_commit || snap_pending) { 1086 if (flush_on_commit || snap_pending) {
@@ -1107,8 +1102,10 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1107 TASK_UNINTERRUPTIBLE); 1102 TASK_UNINTERRUPTIBLE);
1108 1103
1109 smp_mb(); 1104 smp_mb();
1110 if (cur_trans->num_writers > 1 || should_grow) 1105 if (cur_trans->num_writers > 1)
1111 schedule_timeout(timeout); 1106 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1107 else if (should_grow)
1108 schedule_timeout(1);
1112 1109
1113 mutex_lock(&root->fs_info->trans_mutex); 1110 mutex_lock(&root->fs_info->trans_mutex);
1114 finish_wait(&cur_trans->writer_wait, &wait); 1111 finish_wait(&cur_trans->writer_wait, &wait);