xfs: use shared ilock mode for direct IO writes by default

For the direct IO write path, we only really need the ilock to be taken in
exclusive mode during IO submission if we need to do extent allocation
instead of all the time.

Change the block mapping code to take the ilock in shared mode for the
initial block mapping, and only retake it exclusively when we actually
have to perform extent allocations.  We were already dropping the ilock
for the transaction allocation, so this doesn't introduce new race windows.

Based on an earlier patch from Dave Chinner.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>

1 files changed, 26 insertions, 4 deletions

diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index 0dbb9e70fe21..0fd7c2bfa402 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -1146,7 +1146,14 @@ __xfs_get_blocks(
         if (!create && direct && offset >= i_size_read(inode))
                 return 0;
 
-        if (create) {
+        /*
+         * Direct I/O is usually done on preallocated files, so try getting
+         * a block mapping without an exclusive lock first.  For buffered
+         * writes we already have the exclusive iolock anyway, so avoiding
+         * a lock roundtrip here by taking the ilock exclusive from the
+         * beginning is a useful micro optimization.
+         */
+        if (create && !direct) {
                 lockmode = XFS_ILOCK_EXCL;
                 xfs_ilock(ip, lockmode);
         } else {
@@ -1169,22 +1176,37 @@ __xfs_get_blocks(
              (imap.br_startblock == HOLESTARTBLOCK ||
               imap.br_startblock == DELAYSTARTBLOCK))) {
                 if (direct) {
+                        /*
+                         * Drop the ilock in preparation for starting the block
+                         * allocation transaction.  It will be retaken
+                         * exclusively inside xfs_iomap_write_direct for the
+                         * actual allocation.
+                         */
+                        xfs_iunlock(ip, lockmode);
                         error = xfs_iomap_write_direct(ip, offset, size,
                                                        &imap, nimaps);
+                        if (error)
+                                return -error;
                 } else {
+                        /*
+                         * Delalloc reservations do not require a transaction,
+                         * we can go on without dropping the lock here.
+                         */
                         error = xfs_iomap_write_delay(ip, offset, size, &imap);
+                        if (error)
+                                goto out_unlock;
+
+                        xfs_iunlock(ip, lockmode);
                 }
-                if (error)
-                        goto out_unlock;
 
                 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
         } else if (nimaps) {
                 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
+                xfs_iunlock(ip, lockmode);
         } else {
                 trace_xfs_get_blocks_notfound(ip, offset, size);
                 goto out_unlock;
         }
-        xfs_iunlock(ip, lockmode);
 
         if (imap.br_startblock != HOLESTARTBLOCK &&
             imap.br_startblock != DELAYSTARTBLOCK) {