ext4: add ext4_split_extent_at() and ext4_split_extent()
authorYongqiang Yang <xiaoqiangnk@gmail.com>
Tue, 3 May 2011 16:23:07 +0000 (12:23 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Tue, 3 May 2011 16:23:07 +0000 (12:23 -0400)
Add two functions: ext4_split_extent_at(), which splits an extent into
two extents at given logical block, and ext4_split_extent() which
splits an extent into three extents.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Tested-by: Allison Henderson <achender@linux.vnet.ibm.com>
fs/ext4/extents.c

index 23a98b6f00a237235e6ab03bf8c93a69edd031e5..10317eb1d2a93a5426798961e3ef6c358b61d41c 100644 (file)
@@ -2554,6 +2554,193 @@ static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
        return ret;
 }
 
+/*
+ * used by extent splitting.
+ */
+#define EXT4_EXT_MAY_ZEROOUT   0x1  /* safe to zeroout if split fails \
+                                       due to ENOSPC */
+#define EXT4_EXT_MARK_UNINIT1  0x2  /* mark first half uninitialized */
+#define EXT4_EXT_MARK_UNINIT2  0x4  /* mark second half uninitialized */
+
+/*
+ * ext4_split_extent_at() splits an extent at given block.
+ *
+ * @handle: the journal handle
+ * @inode: the file inode
+ * @path: the path to the extent
+ * @split: the logical block where the extent is splitted.
+ * @split_flags: indicates if the extent could be zeroout if split fails, and
+ *              the states(init or uninit) of new extents.
+ * @flags: flags used to insert new extent to extent tree.
+ *
+ *
+ * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
+ * of which are deterimined by split_flag.
+ *
+ * There are two cases:
+ *  a> the extent are splitted into two extent.
+ *  b> split is not needed, and just mark the extent.
+ *
+ * return 0 on success.
+ */
+static int ext4_split_extent_at(handle_t *handle,
+                            struct inode *inode,
+                            struct ext4_ext_path *path,
+                            ext4_lblk_t split,
+                            int split_flag,
+                            int flags)
+{
+       ext4_fsblk_t newblock;
+       ext4_lblk_t ee_block;
+       struct ext4_extent *ex, newex, orig_ex;
+       struct ext4_extent *ex2 = NULL;
+       unsigned int ee_len, depth;
+       int err = 0;
+
+       ext_debug("ext4_split_extents_at: inode %lu, logical"
+               "block %llu\n", inode->i_ino, (unsigned long long)split);
+
+       ext4_ext_show_leaf(inode, path);
+
+       depth = ext_depth(inode);
+       ex = path[depth].p_ext;
+       ee_block = le32_to_cpu(ex->ee_block);
+       ee_len = ext4_ext_get_actual_len(ex);
+       newblock = split - ee_block + ext4_ext_pblock(ex);
+
+       BUG_ON(split < ee_block || split >= (ee_block + ee_len));
+
+       err = ext4_ext_get_access(handle, inode, path + depth);
+       if (err)
+               goto out;
+
+       if (split == ee_block) {
+               /*
+                * case b: block @split is the block that the extent begins with
+                * then we just change the state of the extent, and splitting
+                * is not needed.
+                */
+               if (split_flag & EXT4_EXT_MARK_UNINIT2)
+                       ext4_ext_mark_uninitialized(ex);
+               else
+                       ext4_ext_mark_initialized(ex);
+
+               if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
+                       ext4_ext_try_to_merge(inode, path, ex);
+
+               err = ext4_ext_dirty(handle, inode, path + depth);
+               goto out;
+       }
+
+       /* case a */
+       memcpy(&orig_ex, ex, sizeof(orig_ex));
+       ex->ee_len = cpu_to_le16(split - ee_block);
+       if (split_flag & EXT4_EXT_MARK_UNINIT1)
+               ext4_ext_mark_uninitialized(ex);
+
+       /*
+        * path may lead to new leaf, not to original leaf any more
+        * after ext4_ext_insert_extent() returns,
+        */
+       err = ext4_ext_dirty(handle, inode, path + depth);
+       if (err)
+               goto fix_extent_len;
+
+       ex2 = &newex;
+       ex2->ee_block = cpu_to_le32(split);
+       ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
+       ext4_ext_store_pblock(ex2, newblock);
+       if (split_flag & EXT4_EXT_MARK_UNINIT2)
+               ext4_ext_mark_uninitialized(ex2);
+
+       err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
+       if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
+               err = ext4_ext_zeroout(inode, &orig_ex);
+               if (err)
+                       goto fix_extent_len;
+               /* update the extent length and mark as initialized */
+               ex->ee_len = cpu_to_le32(ee_len);
+               ext4_ext_try_to_merge(inode, path, ex);
+               err = ext4_ext_dirty(handle, inode, path + depth);
+               goto out;
+       } else if (err)
+               goto fix_extent_len;
+
+out:
+       ext4_ext_show_leaf(inode, path);
+       return err;
+
+fix_extent_len:
+       ex->ee_len = orig_ex.ee_len;
+       ext4_ext_dirty(handle, inode, path + depth);
+       return err;
+}
+
+/*
+ * ext4_split_extents() splits an extent and mark extent which is covered
+ * by @map as split_flags indicates
+ *
+ * It may result in splitting the extent into multiple extents (upto three)
+ * There are three possibilities:
+ *   a> There is no split required
+ *   b> Splits in two extents: Split is happening at either end of the extent
+ *   c> Splits in three extents: Somone is splitting in middle of the extent
+ *
+ */
+static int ext4_split_extent(handle_t *handle,
+                             struct inode *inode,
+                             struct ext4_ext_path *path,
+                             struct ext4_map_blocks *map,
+                             int split_flag,
+                             int flags)
+{
+       ext4_lblk_t ee_block;
+       struct ext4_extent *ex;
+       unsigned int ee_len, depth;
+       int err = 0;
+       int uninitialized;
+       int split_flag1, flags1;
+
+       depth = ext_depth(inode);
+       ex = path[depth].p_ext;
+       ee_block = le32_to_cpu(ex->ee_block);
+       ee_len = ext4_ext_get_actual_len(ex);
+       uninitialized = ext4_ext_is_uninitialized(ex);
+
+       if (map->m_lblk + map->m_len < ee_block + ee_len) {
+               split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
+                             EXT4_EXT_MAY_ZEROOUT : 0;
+               flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
+               if (uninitialized)
+                       split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
+                                      EXT4_EXT_MARK_UNINIT2;
+               err = ext4_split_extent_at(handle, inode, path,
+                               map->m_lblk + map->m_len, split_flag1, flags1);
+       }
+
+       ext4_ext_drop_refs(path);
+       path = ext4_ext_find_extent(inode, map->m_lblk, path);
+       if (IS_ERR(path))
+               return PTR_ERR(path);
+
+       if (map->m_lblk >= ee_block) {
+               split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
+                             EXT4_EXT_MAY_ZEROOUT : 0;
+               if (uninitialized)
+                       split_flag1 |= EXT4_EXT_MARK_UNINIT1;
+               if (split_flag & EXT4_EXT_MARK_UNINIT2)
+                       split_flag1 |= EXT4_EXT_MARK_UNINIT2;
+               err = ext4_split_extent_at(handle, inode, path,
+                               map->m_lblk, split_flag1, flags);
+               if (err)
+                       goto out;
+       }
+
+       ext4_ext_show_leaf(inode, path);
+out:
+       return err ? err : map->m_len;
+}
+
 #define EXT4_EXT_ZERO_LEN 7
 /*
  * This function is called by ext4_ext_map_blocks() if someone tries to write