ext4: use ext4_da_writepages() for all modes
authorTheodore Ts'o <tytso@mit.edu>
Thu, 6 Jun 2013 18:00:46 +0000 (14:00 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Thu, 6 Jun 2013 18:00:46 +0000 (14:00 -0400)
Rename ext4_da_writepages() to ext4_writepages() and use it for all
modes.  We still need to iterate over all the pages in the case of
data=journalling, but in the case of nodelalloc/data=ordered (which is
what file systems mounted using ext3 backwards compatibility will use)
this will allow us to use a much more efficient I/O submission path.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/ext4/inode.c
include/trace/events/ext4.h

index 442c5d2..0db830d 100644 (file)
@@ -1799,7 +1799,7 @@ out:
  * lock so we have to do some magic.
  *
  * This function can get called via...
- *   - ext4_da_writepages after taking page lock (have journal handle)
+ *   - ext4_writepages after taking page lock (have journal handle)
  *   - journal_submit_inode_data_buffers (no journal handle)
  *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
  *   - grab_page_cache when doing write_begin (have journal handle)
@@ -2217,7 +2217,7 @@ static int mpage_map_and_submit_extent(handle_t *handle,
 
 /*
  * Calculate the total number of credits to reserve for one writepages
- * iteration. This is called from ext4_da_writepages(). We map an extent of
+ * iteration. This is called from ext4_writepages(). We map an extent of
  * upto MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
  * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
  * bpp - 1 blocks in bpp different extents.
@@ -2349,8 +2349,17 @@ out:
        return err;
 }
 
-static int ext4_da_writepages(struct address_space *mapping,
-                             struct writeback_control *wbc)
+static int __writepage(struct page *page, struct writeback_control *wbc,
+                      void *data)
+{
+       struct address_space *mapping = data;
+       int ret = ext4_writepage(page, wbc);
+       mapping_set_error(mapping, ret);
+       return ret;
+}
+
+static int ext4_writepages(struct address_space *mapping,
+                          struct writeback_control *wbc)
 {
        pgoff_t writeback_index = 0;
        long nr_to_write = wbc->nr_to_write;
@@ -2364,7 +2373,7 @@ static int ext4_da_writepages(struct address_space *mapping,
        bool done;
        struct blk_plug plug;
 
-       trace_ext4_da_writepages(inode, wbc);
+       trace_ext4_writepages(inode, wbc);
 
        /*
         * No pages to write? This is mainly a kludge to avoid starting
@@ -2374,13 +2383,23 @@ static int ext4_da_writepages(struct address_space *mapping,
        if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
                return 0;
 
+       if (ext4_should_journal_data(inode)) {
+               struct blk_plug plug;
+               int ret;
+
+               blk_start_plug(&plug);
+               ret = write_cache_pages(mapping, wbc, __writepage, mapping);
+               blk_finish_plug(&plug);
+               return ret;
+       }
+
        /*
         * If the filesystem has aborted, it is read-only, so return
         * right away instead of dumping stack traces later on that
         * will obscure the real source of the problem.  We test
         * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
         * the latter could be true if the filesystem is mounted
-        * read-only, and in that case, ext4_da_writepages should
+        * read-only, and in that case, ext4_writepages should
         * *never* be called, so if that ever happens, we would want
         * the stack trace.
         */
@@ -2520,8 +2539,8 @@ retry:
                mapping->writeback_index = mpd.first_page;
 
 out_writepages:
-       trace_ext4_da_writepages_result(inode, wbc, ret,
-                                       nr_to_write - wbc->nr_to_write);
+       trace_ext4_writepages_result(inode, wbc, ret,
+                                    nr_to_write - wbc->nr_to_write);
        return ret;
 }
 
@@ -2769,7 +2788,7 @@ int ext4_alloc_da_blocks(struct inode *inode)
         * laptop_mode, not even desirable).  However, to do otherwise
         * would require replicating code paths in:
         *
-        * ext4_da_writepages() ->
+        * ext4_writepages() ->
         *    write_cache_pages() ---> (via passed in callback function)
         *        __mpage_da_writepage() -->
         *           mpage_add_bh_to_extent()
@@ -3213,6 +3232,7 @@ static const struct address_space_operations ext4_aops = {
        .readpage               = ext4_readpage,
        .readpages              = ext4_readpages,
        .writepage              = ext4_writepage,
+       .writepages             = ext4_writepages,
        .write_begin            = ext4_write_begin,
        .write_end              = ext4_write_end,
        .bmap                   = ext4_bmap,
@@ -3228,6 +3248,7 @@ static const struct address_space_operations ext4_journalled_aops = {
        .readpage               = ext4_readpage,
        .readpages              = ext4_readpages,
        .writepage              = ext4_writepage,
+       .writepages             = ext4_writepages,
        .write_begin            = ext4_write_begin,
        .write_end              = ext4_journalled_write_end,
        .set_page_dirty         = ext4_journalled_set_page_dirty,
@@ -3243,7 +3264,7 @@ static const struct address_space_operations ext4_da_aops = {
        .readpage               = ext4_readpage,
        .readpages              = ext4_readpages,
        .writepage              = ext4_writepage,
-       .writepages             = ext4_da_writepages,
+       .writepages             = ext4_writepages,
        .write_begin            = ext4_da_write_begin,
        .write_end              = ext4_da_write_end,
        .bmap                   = ext4_bmap,
index 832a412..72f523e 100644 (file)
@@ -281,7 +281,7 @@ DEFINE_EVENT(ext4__write_end, ext4_da_write_end,
        TP_ARGS(inode, pos, len, copied)
 );
 
-TRACE_EVENT(ext4_da_writepages,
+TRACE_EVENT(ext4_writepages,
        TP_PROTO(struct inode *inode, struct writeback_control *wbc),
 
        TP_ARGS(inode, wbc),
@@ -379,7 +379,7 @@ TRACE_EVENT(ext4_da_write_pages_extent,
                  __entry->flags)
 );
 
-TRACE_EVENT(ext4_da_writepages_result,
+TRACE_EVENT(ext4_writepages_result,
        TP_PROTO(struct inode *inode, struct writeback_control *wbc,
                        int ret, int pages_written),