Cleanup waitqueue_active and barriers

Currently the code assumes that there's an implied barrier by the sequence of code preceding the wakeup, namely the mutex unlock.

As Nikolay pointed out:

I think this is wrong (not your code) but the original assumption that the RELEASE semantics provided by mutex_unlock is sufficient. According to memory-barriers.txt:

Section 'LOCK ACQUISITION FUNCTIONS' states:

(2) RELEASE operation implication:

Memory operations issued before the RELEASE will be completed before the RELEASE operation has completed.

Memory operations issued after the RELEASE *may* be completed before the RELEASE operation has completed.

(I've bolded the may portion)

The example given there:

As an example, consider the following:

    *A = a;
    *B = b;
    ACQUIRE
    *C = c;
    *D = d;
    RELEASE
    *E = e;
    *F = f;

The following sequence of events is acceptable:

    ACQUIRE, {*F,*A}, *E, {*C,*D}, *B, RELEASE

So if we assume that *C is modifying the flag which the waitqueue is checking, and *E is the actual wakeup, then those accesses can be re-ordered...

IMHO this code should be considered broken...
---

To be on the safe side, add the barriers. The synchronization logic around log using the mutexes and several other threads does not make it easy to reason for/against the barrier.

CC Nikolay Borisov <[email protected]>

Link https://lkml.kernel.org/r/[email protected] open here

Signed-off-by David Sterba <[email protected]>

---

 fs/btrfs/tree-log.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

fs/btrfs/tree-log.c +8 -2

index 43758e30aa7a..fa5b3dc5f4d5 100644
@@ -3116,8 +3116,11 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	mutex_unlock(&log_root_tree->log_mutex);
 
 	/*
-	 * The barrier before waitqueue_active is implied by mutex_unlock
+	 * The barrier before waitqueue_active is needed so all the updates
+	 * above are seen by the woken threads. It might not be necessary, but
+	 * proving that seems to be hard.
 	 */
+	smp_mb();
 	if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
 		wake_up(&log_root_tree->log_commit_wait[index2]);
 out:

@@ -3128,8 +3131,11 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	mutex_unlock(&root->log_mutex);
 
 	/*
-	 * The barrier before waitqueue_active is implied by mutex_unlock
+	 * The barrier before waitqueue_active is needed so all the updates
+	 * above are seen by the woken threads. It might not be necessary, but
+	 * proving that seems to be hard.
 	 */
+	smp_mb();
 	if (waitqueue_active(&root->log_commit_wait[index1]))
 		wake_up(&root->log_commit_wait[index1]);
 	return ret;

Use the wrappers and reduce the amount of low-level details about the waitqueue management.

Signed-off-by David Sterba <[email protected]>

---
 fs/btrfs/compression.c   |  7 +------
 fs/btrfs/delayed-inode.c |  9 +++------
 fs/btrfs/dev-replace.c   | 10 ++++------
 fs/btrfs/extent-tree.c   |  7 +------
 fs/btrfs/inode.c         |  9 +++------
 fs/btrfs/locking.c       | 34 +++++++++++-----------------------
 fs/btrfs/ordered-data.c  | 14 ++++----------
 fs/btrfs/transaction.c   |  7 +------
 fs/btrfs/tree-log.c      | 34 ++++++++++++----------------------
 9 files changed, 40 insertions(+), 91 deletions(-)

fs/btrfs/compression.c +1 -6

index 5268c9f85ca7..dcebb91e4e0c 100644
@@ -990,12 +990,7 @@ static void __free_workspace(int type, struct list_head *workspace,
 		btrfs_compress_op[idx]->free_workspace(workspace);
 	atomic_dec(total_ws);
 wake:
-	/*
-	 * Make sure counter is updated before we wake up waiters.
-	 */
-	smp_mb();
-	if (waitqueue_active(ws_wait))
-		wake_up(ws_wait);
+	cond_wake_up(ws_wait);
 }
 
 static void free_workspace(int type, struct list_head *ws)

fs/btrfs/delayed-inode.c +3 -6

index a8d492dbd3e7..fe6caa7e698b 100644
@@ -460,13 +460,10 @@ static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 {
 	int seq = atomic_inc_return(&delayed_root->items_seq);
 
-	/*
-	 * atomic_dec_return implies a barrier for waitqueue_active
-	 */
+	/* atomic_dec_return implies a barrier */
 	if ((atomic_dec_return(&delayed_root->items) <
-	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
-	    waitqueue_active(&delayed_root->wait))
-		wake_up(&delayed_root->wait);
+	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
+		cond_wake_up_nomb(&delayed_root->wait);
 }
 
 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)

fs/btrfs/dev-replace.c +4 -6

index f82be266ba4b..d33ff5b23302 100644
@@ -916,9 +916,9 @@ void btrfs_dev_replace_clear_lock_blocking(
 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 	ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
 	read_lock(&dev_replace->lock);
-	if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
-	    waitqueue_active(&dev_replace->read_lock_wq))
-		wake_up(&dev_replace->read_lock_wq);
+	/* Barrier implied by atomic_dec_and_test */
+	if (atomic_dec_and_test(&dev_replace->blocking_readers))
+		cond_wake_up_nomb(&dev_replace->read_lock_wq);
 }
 
 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)

@@ -929,9 +929,7 @@ void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 {
 	percpu_counter_sub(&fs_info->bio_counter, amount);
-
-	if (waitqueue_active(&fs_info->replace_wait))
-		wake_up(&fs_info->replace_wait);
+	cond_wake_up_nomb(&fs_info->replace_wait);
 }
 
 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)

fs/btrfs/extent-tree.c +1 -6

index 3871658b6ab1..94fc825a126d 100644
@@ -11094,12 +11094,7 @@ int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
 {
 	percpu_counter_dec(&root->subv_writers->counter);
-	/*
-	 * Make sure counter is updated before we wake up waiters.
-	 */
-	smp_mb();
-	if (waitqueue_active(&root->subv_writers->wait))
-		wake_up(&root->subv_writers->wait);
+	cond_wake_up(&root->subv_writers->wait);
 }
 
 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)

fs/btrfs/inode.c +3 -6

index d241285a0d2a..cf6f2815ffee 100644
@@ -1156,13 +1156,10 @@ static noinline void async_cow_submit(struct btrfs_work *work)
 	nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
 		PAGE_SHIFT;
 
-	/*
-	 * atomic_sub_return implies a barrier for waitqueue_active
-	 */
+	/* atomic_sub_return implies a barrier */
 	if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
-	    5 * SZ_1M &&
-	    waitqueue_active(&fs_info->async_submit_wait))
-		wake_up(&fs_info->async_submit_wait);
+	    5 * SZ_1M)
+		cond_wake_up_nomb(&fs_info->async_submit_wait);
 
 	if (async_cow->inode)
 		submit_compressed_extents(async_cow->inode, async_cow);

fs/btrfs/locking.c +11 -23

index e4faefac9d16..1da768e5ef75 100644
@@ -66,22 +66,16 @@ void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
 		write_lock(&eb->lock);
 		WARN_ON(atomic_read(&eb->spinning_writers));
 		atomic_inc(&eb->spinning_writers);
-		/*
-		 * atomic_dec_and_test implies a barrier for waitqueue_active
-		 */
-		if (atomic_dec_and_test(&eb->blocking_writers) &&
-		    waitqueue_active(&eb->write_lock_wq))
-			wake_up(&eb->write_lock_wq);
+		/* atomic_dec_and_test implies a barrier */
+		if (atomic_dec_and_test(&eb->blocking_writers))
+			cond_wake_up_nomb(&eb->write_lock_wq);
 	} else if (rw == BTRFS_READ_LOCK_BLOCKING) {
 		BUG_ON(atomic_read(&eb->blocking_readers) == 0);
 		read_lock(&eb->lock);
 		atomic_inc(&eb->spinning_readers);
-		/*
-		 * atomic_dec_and_test implies a barrier for waitqueue_active
-		 */
-		if (atomic_dec_and_test(&eb->blocking_readers) &&
-		    waitqueue_active(&eb->read_lock_wq))
-			wake_up(&eb->read_lock_wq);
+		/* atomic_dec_and_test implies a barrier */
+		if (atomic_dec_and_test(&eb->blocking_readers))
+			cond_wake_up_nomb(&eb->read_lock_wq);
 	}
 }
 

@@ -221,12 +215,9 @@ void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
 	}
 	btrfs_assert_tree_read_locked(eb);
 	WARN_ON(atomic_read(&eb->blocking_readers) == 0);
-	/*
-	 * atomic_dec_and_test implies a barrier for waitqueue_active
-	 */
-	if (atomic_dec_and_test(&eb->blocking_readers) &&
-	    waitqueue_active(&eb->read_lock_wq))
-		wake_up(&eb->read_lock_wq);
+	/* atomic_dec_and_test implies a barrier */
+	if (atomic_dec_and_test(&eb->blocking_readers))
+		cond_wake_up_nomb(&eb->read_lock_wq);
 	atomic_dec(&eb->read_locks);
 }
 

@@ -275,12 +266,9 @@ void btrfs_tree_unlock(struct extent_buffer *eb)
 	if (blockers) {
 		WARN_ON(atomic_read(&eb->spinning_writers));
 		atomic_dec(&eb->blocking_writers);
-		/*
-		 * Make sure counter is updated before we wake up waiters.
-		 */
+		/* Use the lighter barrier after atomic */
 		smp_mb__after_atomic();
-		if (waitqueue_active(&eb->write_lock_wq))
-			wake_up(&eb->write_lock_wq);
+		cond_wake_up_nomb(&eb->write_lock_wq);
 	} else {
 		WARN_ON(atomic_read(&eb->spinning_writers) != 1);
 		atomic_dec(&eb->spinning_writers);

fs/btrfs/ordered-data.c +4 -10

index 6db8bb2f2c28..2e1a1694a33d 100644
@@ -343,11 +343,8 @@ int btrfs_dec_test_first_ordered_pending(struct inode *inode,
 
 	if (entry->bytes_left == 0) {
 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
-		/*
-		 * Implicit memory barrier after test_and_set_bit
-		 */
-		if (waitqueue_active(&entry->wait))
-			wake_up(&entry->wait);
+		/* test_and_set_bit implies a barrier */
+		cond_wake_up_nomb(&entry->wait);
 	} else {
 		ret = 1;
 	}

@@ -410,11 +407,8 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
 
 	if (entry->bytes_left == 0) {
 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
-		/*
-		 * Implicit memory barrier after test_and_set_bit
-		 */
-		if (waitqueue_active(&entry->wait))
-			wake_up(&entry->wait);
+		/* test_and_set_bit implies a barrier */
+		cond_wake_up_nomb(&entry->wait);
 	} else {
 		ret = 1;
 	}

fs/btrfs/transaction.c +1 -6

index c944b4769e3c..ff841abb756e 100644
@@ -877,12 +877,7 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 	atomic_dec(&cur_trans->num_writers);
 	extwriter_counter_dec(cur_trans, trans->type);
 
-	/*
-	 * Make sure counter is updated before we wake up waiters.
-	 */
-	smp_mb();
-	if (waitqueue_active(&cur_trans->writer_wait))
-		wake_up(&cur_trans->writer_wait);
+	cond_wake_up(&cur_trans->writer_wait);
 	btrfs_put_transaction(cur_trans);
 
 	if (current->journal_info == trans)

fs/btrfs/tree-log.c +12 -22

index fa5b3dc5f4d5..a3ebda24ae12 100644
@@ -222,11 +222,8 @@ int btrfs_pin_log_trans(struct btrfs_root *root)
 void btrfs_end_log_trans(struct btrfs_root *root)
 {
 	if (atomic_dec_and_test(&root->log_writers)) {
-		/*
-		 * Implicit memory barrier after atomic_dec_and_test
-		 */
-		if (waitqueue_active(&root->log_writer_wait))
-			wake_up(&root->log_writer_wait);
+		/* atomic_dec_and_test implies a barrier */
+		cond_wake_up_nomb(&root->log_writer_wait);
 	}
 }
 

@@ -2988,11 +2985,8 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 
 	mutex_lock(&log_root_tree->log_mutex);
 	if (atomic_dec_and_test(&log_root_tree->log_writers)) {
-		/*
-		 * Implicit memory barrier after atomic_dec_and_test
-		 */
-		if (waitqueue_active(&log_root_tree->log_writer_wait))
-			wake_up(&log_root_tree->log_writer_wait);
+		/* atomic_dec_and_test implies a barrier */
+		cond_wake_up_nomb(&log_root_tree->log_writer_wait);
 	}
 
 	if (ret) {

@@ -3116,13 +3110,11 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	mutex_unlock(&log_root_tree->log_mutex);
 
 	/*
-	 * The barrier before waitqueue_active is needed so all the updates
-	 * above are seen by the woken threads. It might not be necessary, but
-	 * proving that seems to be hard.
+	 * The barrier before waitqueue_active (in cond_wake_up) is needed so
+	 * all the updates above are seen by the woken threads. It might not be
+	 * necessary, but proving that seems to be hard.
 	 */
-	smp_mb();
-	if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
-		wake_up(&log_root_tree->log_commit_wait[index2]);
+	cond_wake_up(&log_root_tree->log_commit_wait[index2]);
 out:
 	mutex_lock(&root->log_mutex);
 	btrfs_remove_all_log_ctxs(root, index1, ret);

@@ -3131,13 +3123,11 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
 	mutex_unlock(&root->log_mutex);
 
 	/*
-	 * The barrier before waitqueue_active is needed so all the updates
-	 * above are seen by the woken threads. It might not be necessary, but
-	 * proving that seems to be hard.
+	 * The barrier before waitqueue_active (in cond_wake_up) is needed so
+	 * all the updates above are seen by the woken threads. It might not be
+	 * necessary, but proving that seems to be hard.
 	 */
-	smp_mb();
-	if (waitqueue_active(&root->log_commit_wait[index1]))
-		wake_up(&root->log_commit_wait[index1]);
+	cond_wake_up(&root->log_commit_wait[index1]);
 	return ret;
 }