]> err.no Git - linux-2.6/blobdiff - fs/ext4/inode.c
ext4: Fix lack of credits BUG() when deleting a badly fragmented inode
[linux-2.6] / fs / ext4 / inode.c
index 0fbe678d40bb222ebedd1ae6d2c1d304e03f9bea..2697eaf0368f2cf76d64b936b407f5c4a97840e4 100644 (file)
@@ -191,6 +191,7 @@ static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
 void ext4_delete_inode (struct inode * inode)
 {
        handle_t *handle;
+       int err;
 
        if (ext4_should_order_data(inode))
                ext4_begin_ordered_truncate(inode, 0);
@@ -199,8 +200,9 @@ void ext4_delete_inode (struct inode * inode)
        if (is_bad_inode(inode))
                goto no_delete;
 
-       handle = start_transaction(inode);
+       handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
        if (IS_ERR(handle)) {
+               ext4_std_error(inode->i_sb, PTR_ERR(handle));
                /*
                 * If we're going to skip the normal cleanup, we still need to
                 * make sure that the in-core orphan linked list is properly
@@ -213,8 +215,34 @@ void ext4_delete_inode (struct inode * inode)
        if (IS_SYNC(inode))
                handle->h_sync = 1;
        inode->i_size = 0;
+       err = ext4_mark_inode_dirty(handle, inode);
+       if (err) {
+               ext4_warning(inode->i_sb, __func__,
+                            "couldn't mark inode dirty (err %d)", err);
+               goto stop_handle;
+       }
        if (inode->i_blocks)
                ext4_truncate(inode);
+
+       /*
+        * ext4_ext_truncate() doesn't reserve any slop when it
+        * restarts journal transactions; therefore there may not be
+        * enough credits left in the handle to remove the inode from
+        * the orphan list and set the dtime field.
+        */
+       if (handle->h_buffer_credits < 3) {
+               err = ext4_journal_extend(handle, 3);
+               if (err > 0)
+                       err = ext4_journal_restart(handle, 3);
+               if (err != 0) {
+                       ext4_warning(inode->i_sb, __func__,
+                                    "couldn't extend journal (err %d)", err);
+               stop_handle:
+                       ext4_journal_stop(handle);
+                       goto no_delete;
+               }
+       }
+
        /*
         * Kill off the orphan record which ext4_truncate created.
         * AKPM: I think this can be inside the above `if'.
@@ -2280,8 +2308,11 @@ retry:
        }
 
        page = __grab_cache_page(mapping, index);
-       if (!page)
-               return -ENOMEM;
+       if (!page) {
+               ext4_journal_stop(handle);
+               ret = -ENOMEM;
+               goto out;
+       }
        *pagep = page;
 
        ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
@@ -2806,59 +2837,63 @@ static int ext4_journalled_set_page_dirty(struct page *page)
 }
 
 static const struct address_space_operations ext4_ordered_aops = {
-       .readpage       = ext4_readpage,
-       .readpages      = ext4_readpages,
-       .writepage      = ext4_normal_writepage,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext4_write_begin,
-       .write_end      = ext4_ordered_write_end,
-       .bmap           = ext4_bmap,
-       .invalidatepage = ext4_invalidatepage,
-       .releasepage    = ext4_releasepage,
-       .direct_IO      = ext4_direct_IO,
-       .migratepage    = buffer_migrate_page,
+       .readpage               = ext4_readpage,
+       .readpages              = ext4_readpages,
+       .writepage              = ext4_normal_writepage,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext4_write_begin,
+       .write_end              = ext4_ordered_write_end,
+       .bmap                   = ext4_bmap,
+       .invalidatepage         = ext4_invalidatepage,
+       .releasepage            = ext4_releasepage,
+       .direct_IO              = ext4_direct_IO,
+       .migratepage            = buffer_migrate_page,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 static const struct address_space_operations ext4_writeback_aops = {
-       .readpage       = ext4_readpage,
-       .readpages      = ext4_readpages,
-       .writepage      = ext4_normal_writepage,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext4_write_begin,
-       .write_end      = ext4_writeback_write_end,
-       .bmap           = ext4_bmap,
-       .invalidatepage = ext4_invalidatepage,
-       .releasepage    = ext4_releasepage,
-       .direct_IO      = ext4_direct_IO,
-       .migratepage    = buffer_migrate_page,
+       .readpage               = ext4_readpage,
+       .readpages              = ext4_readpages,
+       .writepage              = ext4_normal_writepage,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext4_write_begin,
+       .write_end              = ext4_writeback_write_end,
+       .bmap                   = ext4_bmap,
+       .invalidatepage         = ext4_invalidatepage,
+       .releasepage            = ext4_releasepage,
+       .direct_IO              = ext4_direct_IO,
+       .migratepage            = buffer_migrate_page,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 static const struct address_space_operations ext4_journalled_aops = {
-       .readpage       = ext4_readpage,
-       .readpages      = ext4_readpages,
-       .writepage      = ext4_journalled_writepage,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext4_write_begin,
-       .write_end      = ext4_journalled_write_end,
-       .set_page_dirty = ext4_journalled_set_page_dirty,
-       .bmap           = ext4_bmap,
-       .invalidatepage = ext4_invalidatepage,
-       .releasepage    = ext4_releasepage,
+       .readpage               = ext4_readpage,
+       .readpages              = ext4_readpages,
+       .writepage              = ext4_journalled_writepage,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext4_write_begin,
+       .write_end              = ext4_journalled_write_end,
+       .set_page_dirty         = ext4_journalled_set_page_dirty,
+       .bmap                   = ext4_bmap,
+       .invalidatepage         = ext4_invalidatepage,
+       .releasepage            = ext4_releasepage,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 static const struct address_space_operations ext4_da_aops = {
-       .readpage       = ext4_readpage,
-       .readpages      = ext4_readpages,
-       .writepage      = ext4_da_writepage,
-       .writepages     = ext4_da_writepages,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext4_da_write_begin,
-       .write_end      = ext4_da_write_end,
-       .bmap           = ext4_bmap,
-       .invalidatepage = ext4_da_invalidatepage,
-       .releasepage    = ext4_releasepage,
-       .direct_IO      = ext4_direct_IO,
-       .migratepage    = buffer_migrate_page,
+       .readpage               = ext4_readpage,
+       .readpages              = ext4_readpages,
+       .writepage              = ext4_da_writepage,
+       .writepages             = ext4_da_writepages,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext4_da_write_begin,
+       .write_end              = ext4_da_write_end,
+       .bmap                   = ext4_bmap,
+       .invalidatepage         = ext4_da_invalidatepage,
+       .releasepage            = ext4_releasepage,
+       .direct_IO              = ext4_direct_IO,
+       .migratepage            = buffer_migrate_page,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 void ext4_set_aops(struct inode *inode)
@@ -3586,6 +3621,16 @@ static int __ext4_get_inode_loc(struct inode *inode,
        }
        if (!buffer_uptodate(bh)) {
                lock_buffer(bh);
+
+               /*
+                * If the buffer has the write error flag, we have failed
+                * to write out another inode in the same block.  In this
+                * case, we don't have to read the block because we may
+                * read the old inode data successfully.
+                */
+               if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
+                       set_buffer_uptodate(bh);
+
                if (buffer_uptodate(bh)) {
                        /* someone brought it uptodate while we waited */
                        unlock_buffer(bh);
@@ -4231,6 +4276,32 @@ err_out:
        return error;
 }
 
+int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
+                struct kstat *stat)
+{
+       struct inode *inode;
+       unsigned long delalloc_blocks;
+
+       inode = dentry->d_inode;
+       generic_fillattr(inode, stat);
+
+       /*
+        * We can't update i_blocks if the block allocation is delayed
+        * otherwise in the case of system crash before the real block
+        * allocation is done, we will have i_blocks inconsistent with
+        * on-disk file blocks.
+        * We always keep i_blocks updated together with real
+        * allocation. But to not confuse with user, stat
+        * will return the blocks that include the delayed allocation
+        * blocks for this file.
+        */
+       spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
+       delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
+       spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
+
+       stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
+       return 0;
+}
 
 /*
  * How many blocks doth make a writepage()?