]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
xfs: move xfs_buf_iowait out of (__)xfs_buf_submit
authorChristoph Hellwig <hch@lst.de>
Mon, 13 Jan 2025 14:12:08 +0000 (15:12 +0100)
committerCarlos Maiolino <cem@kernel.org>
Tue, 14 Jan 2025 10:38:14 +0000 (11:38 +0100)
There is no good reason to pass a bool argument to wait for a buffer when
the callers that want that can easily just wait themselves.

This means the wait moves out of the extra hold of the buffer, but as the
callers of synchronous buffer I/O need to hold a reference anyway that is
perfectly fine.

Because all async buffer submitters ignore the error return value, and
the synchronous ones catch the error condition through b_error and
xfs_buf_iowait this also means the new xfs_buf_submit doesn't have to
return an error code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Acked-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Carlos Maiolino <cem@kernel.org>
fs/xfs/xfs_buf.c

index 5702cad9ccc9ed30e165b25b16c9e8443d3b38b9..5abada2b4a4a822310f4026da692d213320f5e97 100644 (file)
@@ -53,14 +53,8 @@ struct kmem_cache *xfs_buf_cache;
  *       b_lock (trylock due to inversion)
  */
 
-static int __xfs_buf_submit(struct xfs_buf *bp, bool wait);
-
-static inline int
-xfs_buf_submit(
-       struct xfs_buf          *bp)
-{
-       return __xfs_buf_submit(bp, !(bp->b_flags & XBF_ASYNC));
-}
+static void xfs_buf_submit(struct xfs_buf *bp);
+static int xfs_buf_iowait(struct xfs_buf *bp);
 
 static inline bool xfs_buf_is_uncached(struct xfs_buf *bp)
 {
@@ -804,7 +798,10 @@ _xfs_buf_read(
        bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD | XBF_DONE);
        bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
 
-       return xfs_buf_submit(bp);
+       xfs_buf_submit(bp);
+       if (flags & XBF_ASYNC)
+               return 0;
+       return xfs_buf_iowait(bp);
 }
 
 /*
@@ -980,8 +977,8 @@ xfs_buf_read_uncached(
        bp->b_ops = ops;
 
        xfs_buf_submit(bp);
-       if (bp->b_error) {
-               error = bp->b_error;
+       error = xfs_buf_iowait(bp);
+       if (error) {
                xfs_buf_relse(bp);
                return error;
        }
@@ -1483,7 +1480,8 @@ xfs_bwrite(
        bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
                         XBF_DONE);
 
-       error = xfs_buf_submit(bp);
+       xfs_buf_submit(bp);
+       error = xfs_buf_iowait(bp);
        if (error)
                xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
        return error;
@@ -1698,13 +1696,10 @@ xfs_buf_iowait(
  * safe to reference the buffer after a call to this function unless the caller
  * holds an additional reference itself.
  */
-static int
-__xfs_buf_submit(
-       struct xfs_buf  *bp,
-       bool            wait)
+static void
+xfs_buf_submit(
+       struct xfs_buf  *bp)
 {
-       int             error = 0;
-
        trace_xfs_buf_submit(bp, _RET_IP_);
 
        ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
@@ -1724,10 +1719,9 @@ __xfs_buf_submit(
         * state here rather than mount state to avoid corrupting the log tail
         * on shutdown.
         */
-       if (bp->b_mount->m_log &&
-           xlog_is_shutdown(bp->b_mount->m_log)) {
+       if (bp->b_mount->m_log && xlog_is_shutdown(bp->b_mount->m_log)) {
                xfs_buf_ioend_fail(bp);
-               return -EIO;
+               return;
        }
 
        /*
@@ -1765,16 +1759,12 @@ __xfs_buf_submit(
                        xfs_buf_ioend_async(bp);
        }
 
-       if (wait)
-               error = xfs_buf_iowait(bp);
-
        /*
         * Release the hold that keeps the buffer referenced for the entire
         * I/O. Note that if the buffer is async, it is not safe to reference
         * after this release.
         */
        xfs_buf_rele(bp);
-       return error;
 }
 
 void *
@@ -2323,7 +2313,7 @@ xfs_buf_delwri_submit_buffers(
                        bp->b_flags |= XBF_ASYNC;
                        xfs_buf_list_del(bp);
                }
-               __xfs_buf_submit(bp, false);
+               xfs_buf_submit(bp);
        }
        blk_finish_plug(&plug);