ext4: unlock unused_pages timely when doing writeback

In mpage_add_bh_to_extent(), when accumulated extents length is greater
than MAX_WRITEPAGES_EXTENT_LEN or buffer head's b_stat is not equal, we
will not continue to search unmapped area for this page, but note this
page is locked, and will only be unlocked in mpage_release_unused_pages()
after ext4_io_submit, if io also is throttled by blk-throttle or similar
io qos, we will hold this page locked for a while, it's unnecessary.

I think the best fix is to refactor mpage_add_bh_to_extent() to let it
return some hints whether to unlock this page, but given that we will
improve dioread_nolock later, we can let it done later, so currently
the simple fix would just call mpage_release_unused_pages() before
ext4_io_submit().

Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

1 files changed, 5 insertions, 4 deletions

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index dd4641f3cdb9..9c3402ec85bc 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2829,12 +2829,12 @@ retry:
                 goto unplug;
         }
         ret = mpage_prepare_extent_to_map(&mpd);
+        /* Unlock pages we didn't use */
+        mpage_release_unused_pages(&mpd, false);
         /* Submit prepared bio */
         ext4_io_submit(&mpd.io_submit);
         ext4_put_io_end_defer(mpd.io_submit.io_end);
         mpd.io_submit.io_end = NULL;
-        /* Unlock pages we didn't use */
-        mpage_release_unused_pages(&mpd, false);
         if (ret < 0)
                 goto unplug;
 
@@ -2902,10 +2902,11 @@ retry:
                         handle = NULL;
                         mpd.do_map = 0;
                 }
-                /* Submit prepared bio */
-                ext4_io_submit(&mpd.io_submit);
                 /* Unlock pages we didn't use */
                 mpage_release_unused_pages(&mpd, give_up_on_write);
+                /* Submit prepared bio */
+                ext4_io_submit(&mpd.io_submit);
+
                 /*
                  * Drop our io_end reference we got from init. We have
                  * to be careful and use deferred io_end finishing if