]> err.no Git - linux-2.6/commitdiff
ocfs2: btree support for removal of arbirtrary extents
authorMark Fasheh <mark.fasheh@oracle.com>
Tue, 3 Jul 2007 20:27:22 +0000 (13:27 -0700)
committerMark Fasheh <mark.fasheh@oracle.com>
Wed, 11 Jul 2007 00:32:05 +0000 (17:32 -0700)
Add code to the btree paths to support the removal of arbitrary regions
within an existing extent. With proper higher level support this can be used
to "punch holes" in a file. Truncate (a special case of hole punching) could
also be converted to use these methods.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
fs/ocfs2/alloc.c

index 2e389831646854bdde11150d41062a5b15d0e094..fac1adb3880af22a40294502bc2fb4518da46ca5 100644 (file)
@@ -4139,6 +4139,373 @@ out:
        return ret;
 }
 
+static int ocfs2_split_tree(struct inode *inode, struct buffer_head *di_bh,
+                           handle_t *handle, struct ocfs2_path *path,
+                           int index, u32 new_range,
+                           struct ocfs2_alloc_context *meta_ac)
+{
+       int ret, depth, credits = handle->h_buffer_credits;
+       struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+       struct buffer_head *last_eb_bh = NULL;
+       struct ocfs2_extent_block *eb;
+       struct ocfs2_extent_list *rightmost_el, *el;
+       struct ocfs2_extent_rec split_rec;
+       struct ocfs2_extent_rec *rec;
+       struct ocfs2_insert_type insert;
+
+       /*
+        * Setup the record to split before we grow the tree.
+        */
+       el = path_leaf_el(path);
+       rec = &el->l_recs[index];
+       ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec);
+
+       depth = path->p_tree_depth;
+       if (depth > 0) {
+               ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
+                                      le64_to_cpu(di->i_last_eb_blk),
+                                      &last_eb_bh, OCFS2_BH_CACHED, inode);
+               if (ret < 0) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+               rightmost_el = &eb->h_list;
+       } else
+               rightmost_el = path_leaf_el(path);
+
+       credits += path->p_tree_depth + ocfs2_extend_meta_needed(di);
+       ret = ocfs2_extend_trans(handle, credits);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
+           le16_to_cpu(rightmost_el->l_count)) {
+               int old_depth = depth;
+
+               ret = ocfs2_grow_tree(inode, handle, di_bh, &depth, &last_eb_bh,
+                                     meta_ac);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               if (old_depth != depth) {
+                       eb = (struct ocfs2_extent_block *)last_eb_bh->b_data;
+                       rightmost_el = &eb->h_list;
+               }
+       }
+
+       memset(&insert, 0, sizeof(struct ocfs2_insert_type));
+       insert.ins_appending = APPEND_NONE;
+       insert.ins_contig = CONTIG_NONE;
+       insert.ins_split = SPLIT_RIGHT;
+       insert.ins_free_records = le16_to_cpu(rightmost_el->l_count)
+               - le16_to_cpu(rightmost_el->l_next_free_rec);
+       insert.ins_tree_depth = depth;
+
+       ret = ocfs2_do_insert_extent(inode, handle, di_bh, &split_rec, &insert);
+       if (ret)
+               mlog_errno(ret);
+
+out:
+       brelse(last_eb_bh);
+       return ret;
+}
+
+static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle,
+                             struct ocfs2_path *path, int index,
+                             struct ocfs2_cached_dealloc_ctxt *dealloc,
+                             u32 cpos, u32 len)
+{
+       int ret;
+       u32 left_cpos, rec_range, trunc_range;
+       int wants_rotate = 0, is_rightmost_tree_rec = 0;
+       struct super_block *sb = inode->i_sb;
+       struct ocfs2_path *left_path = NULL;
+       struct ocfs2_extent_list *el = path_leaf_el(path);
+       struct ocfs2_extent_rec *rec;
+       struct ocfs2_extent_block *eb;
+
+       if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
+               ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               index--;
+       }
+
+       if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
+           path->p_tree_depth) {
+               /*
+                * Check whether this is the rightmost tree record. If
+                * we remove all of this record or part of its right
+                * edge then an update of the record lengths above it
+                * will be required.
+                */
+               eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+               if (eb->h_next_leaf_blk == 0)
+                       is_rightmost_tree_rec = 1;
+       }
+
+       rec = &el->l_recs[index];
+       if (index == 0 && path->p_tree_depth &&
+           le32_to_cpu(rec->e_cpos) == cpos) {
+               /*
+                * Changing the leftmost offset (via partial or whole
+                * record truncate) of an interior (or rightmost) path
+                * means we have to update the subtree that is formed
+                * by this leaf and the one to it's left.
+                *
+                * There are two cases we can skip:
+                *   1) Path is the leftmost one in our inode tree.
+                *   2) The leaf is rightmost and will be empty after
+                *      we remove the extent record - the rotate code
+                *      knows how to update the newly formed edge.
+                */
+
+               ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path,
+                                                   &left_cpos);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
+                       left_path = ocfs2_new_path(path_root_bh(path),
+                                                  path_root_el(path));
+                       if (!left_path) {
+                               ret = -ENOMEM;
+                               mlog_errno(ret);
+                               goto out;
+                       }
+
+                       ret = ocfs2_find_path(inode, left_path, left_cpos);
+                       if (ret) {
+                               mlog_errno(ret);
+                               goto out;
+                       }
+               }
+       }
+
+       ret = ocfs2_extend_rotate_transaction(handle, 0,
+                                             handle->h_buffer_credits,
+                                             path);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_journal_access_path(inode, handle, path);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_journal_access_path(inode, handle, left_path);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+       trunc_range = cpos + len;
+
+       if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
+               int next_free;
+
+               memset(rec, 0, sizeof(*rec));
+               ocfs2_cleanup_merge(el, index);
+               wants_rotate = 1;
+
+               next_free = le16_to_cpu(el->l_next_free_rec);
+               if (is_rightmost_tree_rec && next_free > 1) {
+                       /*
+                        * We skip the edge update if this path will
+                        * be deleted by the rotate code.
+                        */
+                       rec = &el->l_recs[next_free - 1];
+                       ocfs2_adjust_rightmost_records(inode, handle, path,
+                                                      rec);
+               }
+       } else if (le32_to_cpu(rec->e_cpos) == cpos) {
+               /* Remove leftmost portion of the record. */
+               le32_add_cpu(&rec->e_cpos, len);
+               le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
+               le16_add_cpu(&rec->e_leaf_clusters, -len);
+       } else if (rec_range == trunc_range) {
+               /* Remove rightmost portion of the record */
+               le16_add_cpu(&rec->e_leaf_clusters, -len);
+               if (is_rightmost_tree_rec)
+                       ocfs2_adjust_rightmost_records(inode, handle, path, rec);
+       } else {
+               /* Caller should have trapped this. */
+               mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) "
+                    "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno,
+                    le32_to_cpu(rec->e_cpos),
+                    le16_to_cpu(rec->e_leaf_clusters), cpos, len);
+               BUG();
+       }
+
+       if (left_path) {
+               int subtree_index;
+
+               subtree_index = ocfs2_find_subtree_root(inode, left_path, path);
+               ocfs2_complete_edge_insert(inode, handle, left_path, path,
+                                          subtree_index);
+       }
+
+       ocfs2_journal_dirty(handle, path_leaf_bh(path));
+
+       ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+out:
+       ocfs2_free_path(left_path);
+       return ret;
+}
+
+static int ocfs2_remove_extent(struct inode *inode, struct buffer_head *di_bh,
+                              u32 cpos, u32 len, handle_t *handle,
+                              struct ocfs2_alloc_context *meta_ac,
+                              struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+       int ret, index;
+       u32 rec_range, trunc_range;
+       struct ocfs2_extent_rec *rec;
+       struct ocfs2_extent_list *el;
+       struct ocfs2_path *path;
+
+       ocfs2_extent_map_trunc(inode, 0);
+
+       path = ocfs2_new_inode_path(di_bh);
+       if (!path) {
+               ret = -ENOMEM;
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_find_path(inode, path, cpos);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       el = path_leaf_el(path);
+       index = ocfs2_search_extent_list(el, cpos);
+       if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
+               ocfs2_error(inode->i_sb,
+                           "Inode %llu has an extent at cpos %u which can no "
+                           "longer be found.\n",
+                           (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
+               ret = -EROFS;
+               goto out;
+       }
+
+       /*
+        * We have 3 cases of extent removal:
+        *   1) Range covers the entire extent rec
+        *   2) Range begins or ends on one edge of the extent rec
+        *   3) Range is in the middle of the extent rec (no shared edges)
+        *
+        * For case 1 we remove the extent rec and left rotate to
+        * fill the hole.
+        *
+        * For case 2 we just shrink the existing extent rec, with a
+        * tree update if the shrinking edge is also the edge of an
+        * extent block.
+        *
+        * For case 3 we do a right split to turn the extent rec into
+        * something case 2 can handle.
+        */
+       rec = &el->l_recs[index];
+       rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+       trunc_range = cpos + len;
+
+       BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
+
+       mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d "
+            "(cpos %u, len %u)\n",
+            (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index,
+            le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec));
+
+       if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
+               ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
+                                        cpos, len);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+       } else {
+               ret = ocfs2_split_tree(inode, di_bh, handle, path, index,
+                                      trunc_range, meta_ac);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               /*
+                * The split could have manipulated the tree enough to
+                * move the record location, so we have to look for it again.
+                */
+               ocfs2_reinit_path(path, 1);
+
+               ret = ocfs2_find_path(inode, path, cpos);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               el = path_leaf_el(path);
+               index = ocfs2_search_extent_list(el, cpos);
+               if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
+                       ocfs2_error(inode->i_sb,
+                                   "Inode %llu: split at cpos %u lost record.",
+                                   (unsigned long long)OCFS2_I(inode)->ip_blkno,
+                                   cpos);
+                       ret = -EROFS;
+                       goto out;
+               }
+
+               /*
+                * Double check our values here. If anything is fishy,
+                * it's easier to catch it at the top level.
+                */
+               rec = &el->l_recs[index];
+               rec_range = le32_to_cpu(rec->e_cpos) +
+                       ocfs2_rec_clusters(el, rec);
+               if (rec_range != trunc_range) {
+                       ocfs2_error(inode->i_sb,
+                                   "Inode %llu: error after split at cpos %u"
+                                   "trunc len %u, existing record is (%u,%u)",
+                                   (unsigned long long)OCFS2_I(inode)->ip_blkno,
+                                   cpos, len, le32_to_cpu(rec->e_cpos),
+                                   ocfs2_rec_clusters(el, rec));
+                       ret = -EROFS;
+                       goto out;
+               }
+
+               ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
+                                        cpos, len);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+       }
+
+out:
+       ocfs2_free_path(path);
+       return ret;
+}
+
 static inline int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
 {
        struct buffer_head *tl_bh = osb->osb_tl_bh;