btrfs: raid56: do not park sync writes into nocow runs

An in-place overwrite of a nodatacow file under stripe_alloc parked
like any partial write, but with a waiter behind it and nothing worth
merging: in-place overwrites arrive one fsync at a time, and unlike
the datacow fsync path nothing kicks the parked rbio before the page
writeback wait, so every fsync ate the full sync park deadline.
Measured against stock nodatacow on the same rig, that deadline was
the bulk of a ~3.7ms per-fsync regression.  Skip parking for sync
writes whose stripe belongs to a NOCOW-class run; async writeback
still parks and merges there.

Assisted-by: Claude:claude-fable-5

diff --git a/fs/btrfs/raid56.c b/fs/btrfs/raid56.c
index 6f3a7d2db8c49ef1025224cb0c97906b3ae558d0..9f01a4d5eccd451787916450d4f6d06049ba5376 100644 (file)
--- a/fs/btrfs/raid56.c
+++ b/fs/btrfs/raid56.c
@@ -770,8 +770,11 @@ static bool rbio_try_park(struct btrfs_raid_bio *rbio)
        struct btrfs_fs_info *fs_info = rbio->bioc->fs_info;
        struct btrfs_stripe_hash_table *table = fs_info->stripe_hash_table;
        unsigned int timeout_ms = BTRFS_RBIO_PARK_TIMEOUT_MS;
+       enum btrfs_stripe_run_class class;
 
-       if (!btrfs_stripe_in_open_run(fs_info, rbio->bioc->full_stripe_logical))
+       if (!btrfs_stripe_open_run_class(fs_info,
+                                        rbio->bioc->full_stripe_logical,
+                                        &class))
                return false;
 
        spin_lock(&rbio->bio_list_lock);
@@ -779,6 +782,17 @@ static bool rbio_try_park(struct btrfs_raid_bio *rbio)
                timeout_ms = BTRFS_RBIO_PARK_SYNC_TIMEOUT_MS;
        spin_unlock(&rbio->bio_list_lock);
 
+       /*
+        * A sync write into a nodatacow inode's private run has a waiter
+        * behind it and nothing worth waiting to merge with: in-place
+        * overwrites arrive one fsync at a time, and the plain (unparked)
+        * write path is what stock nodatacow latency looks like.  Park only
+        * async writeback there.
+        */
+       if (class == BTRFS_STRIPE_RUN_NOCOW &&
+           timeout_ms == BTRFS_RBIO_PARK_SYNC_TIMEOUT_MS)
+               return false;
+
        spin_lock(&table->parked_lock);
        set_bit(RBIO_PARKED_BIT, &rbio->flags);
        rbio->park_deadline = jiffies + msecs_to_jiffies(timeout_ms);