ext4: Fix the delalloc writepages to allocate blocks at the right offset.
authorAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Tue, 6 Jan 2009 02:50:43 +0000 (21:50 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Tue, 6 Jan 2009 02:50:43 +0000 (21:50 -0500)
When iterating through the pages which have mapped buffer_heads, we
failed to update the b_state value. This results in allocating blocks
at logical offset 0.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
fs/ext4/inode.c

index ac97348f85b5730c8157ea19913485236c9b764a..c77a7ac753f379efff52fd2283d7592777d07f33 100644 (file)
@@ -1653,35 +1653,39 @@ struct mpage_da_data {
  */
 static int mpage_da_submit_io(struct mpage_da_data *mpd)
 {
-       struct address_space *mapping = mpd->inode->i_mapping;
-       int ret = 0, err, nr_pages, i;
-       unsigned long index, end;
-       struct pagevec pvec;
        long pages_skipped;
+       struct pagevec pvec;
+       unsigned long index, end;
+       int ret = 0, err, nr_pages, i;
+       struct inode *inode = mpd->inode;
+       struct address_space *mapping = inode->i_mapping;
 
        BUG_ON(mpd->next_page <= mpd->first_page);
-       pagevec_init(&pvec, 0);
+       /*
+        * We need to start from the first_page to the next_page - 1
+        * to make sure we also write the mapped dirty buffer_heads.
+        * If we look at mpd->lbh.b_blocknr we would only be looking
+        * at the currently mapped buffer_heads.
+        */
        index = mpd->first_page;
        end = mpd->next_page - 1;
 
+       pagevec_init(&pvec, 0);
        while (index <= end) {
-               /*
-                * We can use PAGECACHE_TAG_DIRTY lookup here because
-                * even though we have cleared the dirty flag on the page
-                * We still keep the page in the radix tree with tag
-                * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
-                * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
-                * which is called via the below writepage callback.
-                */
-               nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
-                                       PAGECACHE_TAG_DIRTY,
-                                       min(end - index,
-                                       (pgoff_t)PAGEVEC_SIZE-1) + 1);
+               nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
                if (nr_pages == 0)
                        break;
                for (i = 0; i < nr_pages; i++) {
                        struct page *page = pvec.pages[i];
 
+                       index = page->index;
+                       if (index > end)
+                               break;
+                       index++;
+
+                       BUG_ON(!PageLocked(page));
+                       BUG_ON(PageWriteback(page));
+
                        pages_skipped = mpd->wbc->pages_skipped;
                        err = mapping->a_ops->writepage(page, mpd->wbc);
                        if (!err && (pages_skipped == mpd->wbc->pages_skipped))
@@ -2095,11 +2099,29 @@ static int __mpage_da_writepage(struct page *page,
                bh = head;
                do {
                        BUG_ON(buffer_locked(bh));
+                       /*
+                        * We need to try to allocate
+                        * unmapped blocks in the same page.
+                        * Otherwise we won't make progress
+                        * with the page in ext4_da_writepage
+                        */
                        if (buffer_dirty(bh) &&
                                (!buffer_mapped(bh) || buffer_delay(bh))) {
                                mpage_add_bh_to_extent(mpd, logical, bh);
                                if (mpd->io_done)
                                        return MPAGE_DA_EXTENT_TAIL;
+                       } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
+                               /*
+                                * mapped dirty buffer. We need to update
+                                * the b_state because we look at
+                                * b_state in mpage_da_map_blocks. We don't
+                                * update b_size because if we find an
+                                * unmapped buffer_head later we need to
+                                * use the b_state flag of that buffer_head.
+                                */
+                               if (mpd->lbh.b_size == 0)
+                                       mpd->lbh.b_state =
+                                               bh->b_state & BH_FLAGS;
                        }
                        logical++;
                } while ((bh = bh->b_this_page) != head);