xfs: reduce ilock hold times in xfs_file_aio_write_checks
authorChristoph Hellwig <hch@infradead.org>
Tue, 27 Mar 2012 14:34:47 +0000 (10:34 -0400)
committerBen Myers <bpm@sgi.com>
Mon, 14 May 2012 21:20:17 +0000 (16:20 -0500)
We do not need the ilock for generic_write_checks and the i_size_read,
which are protected by i_mutex and/or iolock, so reduce the ilock
critical section to just the call to xfs_zero_eof.

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>
fs/xfs/xfs_file.c

index 54a67dd9ac0a5fbe5a7caf4271d14c67798fd872..3537c8d0af48e7e175dca6880d5df6f1ae71dcfc 100644 (file)
@@ -593,35 +593,31 @@ xfs_file_aio_write_checks(
        struct xfs_inode        *ip = XFS_I(inode);
        int                     error = 0;
 
-       xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
 restart:
        error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
-       if (error) {
-               xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
+       if (error)
                return error;
-       }
 
        /*
         * If the offset is beyond the size of the file, we need to zero any
         * blocks that fall between the existing EOF and the start of this
         * write.  If zeroing is needed and we are currently holding the
-        * iolock shared, we need to update it to exclusive which involves
-        * dropping all locks and relocking to maintain correct locking order.
-        * If we do this, restart the function to ensure all checks and values
-        * are still valid.
+        * iolock shared, we need to update it to exclusive which implies
+        * having to redo all checks before.
         */
        if (*pos > i_size_read(inode)) {
                if (*iolock == XFS_IOLOCK_SHARED) {
-                       xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
+                       xfs_rw_iunlock(ip, *iolock);
                        *iolock = XFS_IOLOCK_EXCL;
-                       xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
+                       xfs_rw_ilock(ip, *iolock);
                        goto restart;
                }
+               xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
                error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
+               xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
+               if (error)
+                       return error;
        }
-       xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
-       if (error)
-               return error;
 
        /*
         * Updating the timestamps will grab the ilock again from
@@ -638,7 +634,6 @@ restart:
         * people from modifying setuid and setgid binaries.
         */
        return file_remove_suid(file);
-
 }
 
 /*