]> err.no Git - linux-2.6/blobdiff - fs/ext4/inode.c
Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
[linux-2.6] / fs / ext4 / inode.c
index 8dd22eade42cd356e197887f6fc53a2404e1943d..7e91913e325bb511086c5a2ddb0ddeda0bc28e01 100644 (file)
@@ -1848,29 +1848,53 @@ static void mpage_da_map_blocks(struct mpage_da_data *mpd)
 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
                                   sector_t logical, struct buffer_head *bh)
 {
-       struct buffer_head *lbh = &mpd->lbh;
        sector_t next;
+       size_t b_size = bh->b_size;
+       struct buffer_head *lbh = &mpd->lbh;
+       int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
 
-       next = lbh->b_blocknr + (lbh->b_size >> mpd->inode->i_blkbits);
-
+       /* check if thereserved journal credits might overflow */
+       if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
+               if (nrblocks >= EXT4_MAX_TRANS_DATA) {
+                       /*
+                        * With non-extent format we are limited by the journal
+                        * credit available.  Total credit needed to insert
+                        * nrblocks contiguous blocks is dependent on the
+                        * nrblocks.  So limit nrblocks.
+                        */
+                       goto flush_it;
+               } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
+                               EXT4_MAX_TRANS_DATA) {
+                       /*
+                        * Adding the new buffer_head would make it cross the
+                        * allowed limit for which we have journal credit
+                        * reserved. So limit the new bh->b_size
+                        */
+                       b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
+                                               mpd->inode->i_blkbits;
+                       /* we will do mpage_da_submit_io in the next loop */
+               }
+       }
        /*
         * First block in the extent
         */
        if (lbh->b_size == 0) {
                lbh->b_blocknr = logical;
-               lbh->b_size = bh->b_size;
+               lbh->b_size = b_size;
                lbh->b_state = bh->b_state & BH_FLAGS;
                return;
        }
 
+       next = lbh->b_blocknr + nrblocks;
        /*
         * Can we merge the block to our big extent?
         */
        if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
-               lbh->b_size += bh->b_size;
+               lbh->b_size += b_size;
                return;
        }
 
+flush_it:
        /*
         * We couldn't merge the block to our extent, so we
         * need to flush current  extent and start new one
@@ -2231,28 +2255,39 @@ static int ext4_da_writepage(struct page *page,
 }
 
 /*
- * For now just follow the DIO way to estimate the max credits
- * needed to write out EXT4_MAX_WRITEBACK_PAGES.
- * todo: need to calculate the max credits need for
- * extent based files, currently the DIO credits is based on
- * indirect-blocks mapping way.
- *
- * Probably should have a generic way to calculate credits
- * for DIO, writepages, and truncate
+ * This is called via ext4_da_writepages() to
+ * calulate the total number of credits to reserve to fit
+ * a single extent allocation into a single transaction,
+ * ext4_da_writpeages() will loop calling this before
+ * the block allocation.
  */
-#define EXT4_MAX_WRITEBACK_PAGES      DIO_MAX_BLOCKS
-#define EXT4_MAX_WRITEBACK_CREDITS    25
+
+static int ext4_da_writepages_trans_blocks(struct inode *inode)
+{
+       int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
+
+       /*
+        * With non-extent format the journal credit needed to
+        * insert nrblocks contiguous block is dependent on
+        * number of contiguous block. So we will limit
+        * number of contiguous block to a sane value
+        */
+       if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
+           (max_blocks > EXT4_MAX_TRANS_DATA))
+               max_blocks = EXT4_MAX_TRANS_DATA;
+
+       return ext4_chunk_trans_blocks(inode, max_blocks);
+}
 
 static int ext4_da_writepages(struct address_space *mapping,
                              struct writeback_control *wbc)
 {
-       struct inode *inode = mapping->host;
        handle_t *handle = NULL;
-       int needed_blocks;
-       int ret = 0;
-       long to_write;
        loff_t range_start = 0;
-       long pages_skipped = 0;
+       struct inode *inode = mapping->host;
+       int needed_blocks, ret = 0, nr_to_writebump = 0;
+       long to_write, pages_skipped = 0;
+       struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
 
        /*
         * No pages to write? This is mainly a kludge to avoid starting
@@ -2261,6 +2296,16 @@ static int ext4_da_writepages(struct address_space *mapping,
         */
        if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
                return 0;
+       /*
+        * Make sure nr_to_write is >= sbi->s_mb_stream_request
+        * This make sure small files blocks are allocated in
+        * single attempt. This ensure that small files
+        * get less fragmented.
+        */
+       if (wbc->nr_to_write < sbi->s_mb_stream_request) {
+               nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
+               wbc->nr_to_write = sbi->s_mb_stream_request;
+       }
 
        if (!wbc->range_cyclic)
                /*
@@ -2283,7 +2328,7 @@ restart_loop:
                 * by delalloc
                 */
                BUG_ON(ext4_should_journal_data(inode));
-               needed_blocks = EXT4_DATA_TRANS_BLOCKS(inode->i_sb);
+               needed_blocks = ext4_da_writepages_trans_blocks(inode);
 
                /* start a new transaction*/
                handle = ext4_journal_start(inode, needed_blocks);
@@ -2341,7 +2386,7 @@ restart_loop:
        }
 
 out_writepages:
-       wbc->nr_to_write = to_write;
+       wbc->nr_to_write = to_write - nr_to_writebump;
        wbc->range_start = range_start;
        return ret;
 }
@@ -4461,11 +4506,9 @@ int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
  * the modification of a single pages into a single transaction,
  * which may include multiple chunks of block allocations.
  *
- * This could be called via ext4_write_begin() or later
- * ext4_da_writepages() in delalyed allocation case.
+ * This could be called via ext4_write_begin()
  *
- * In both case it's possible that we could allocating multiple
- * chunks of blocks. We need to consider the worse case, when
+ * We need to consider the worse case, when
  * one new block per extent.
  */
 int ext4_writepage_trans_blocks(struct inode *inode)