ocfs2: Fix extent lookup to return true size of holes
[linux-2.6.git] / fs / ocfs2 / extent_map.c
blobf35e04f27f3230149ca98e985f8ea541a2c81b7f
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * extent_map.c
6 * Block/Cluster mapping functions
8 * Copyright (C) 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License, version 2, as published by the Free Software Foundation.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
19 * You should have received a copy of the GNU General Public
20 * License along with this program; if not, write to the
21 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
22 * Boston, MA 021110-1307, USA.
25 #include <linux/fs.h>
26 #include <linux/init.h>
27 #include <linux/types.h>
29 #define MLOG_MASK_PREFIX ML_EXTENT_MAP
30 #include <cluster/masklog.h>
32 #include "ocfs2.h"
34 #include "alloc.h"
35 #include "extent_map.h"
36 #include "inode.h"
37 #include "super.h"
39 #include "buffer_head_io.h"
42 * Return the 1st index within el which contains an extent start
43 * larger than v_cluster.
45 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
46 u32 v_cluster)
48 int i;
49 struct ocfs2_extent_rec *rec;
51 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
52 rec = &el->l_recs[i];
54 if (v_cluster < le32_to_cpu(rec->e_cpos))
55 break;
58 return i;
62 * Figure out the size of a hole which starts at v_cluster within the given
63 * extent list.
65 * If there is no more allocation past v_cluster, we return the maximum
66 * cluster size minus v_cluster.
68 * If we have in-inode extents, then el points to the dinode list and
69 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
70 * containing el.
72 static int ocfs2_figure_hole_clusters(struct inode *inode,
73 struct ocfs2_extent_list *el,
74 struct buffer_head *eb_bh,
75 u32 v_cluster,
76 u32 *num_clusters)
78 int ret, i;
79 struct buffer_head *next_eb_bh = NULL;
80 struct ocfs2_extent_block *eb, *next_eb;
82 i = ocfs2_search_for_hole_index(el, v_cluster);
84 if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
85 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
88 * Check the next leaf for any extents.
91 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
92 goto no_more_extents;
94 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
95 le64_to_cpu(eb->h_next_leaf_blk),
96 &next_eb_bh, OCFS2_BH_CACHED, inode);
97 if (ret) {
98 mlog_errno(ret);
99 goto out;
101 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
103 if (!OCFS2_IS_VALID_EXTENT_BLOCK(next_eb)) {
104 ret = -EROFS;
105 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, next_eb);
106 goto out;
109 el = &next_eb->h_list;
111 i = ocfs2_search_for_hole_index(el, v_cluster);
114 no_more_extents:
115 if (i == le16_to_cpu(el->l_next_free_rec)) {
117 * We're at the end of our existing allocation. Just
118 * return the maximum number of clusters we could
119 * possibly allocate.
121 *num_clusters = UINT_MAX - v_cluster;
122 } else {
123 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
126 ret = 0;
127 out:
128 brelse(next_eb_bh);
129 return ret;
133 * Return the index of the extent record which contains cluster #v_cluster.
134 * -1 is returned if it was not found.
136 * Should work fine on interior and exterior nodes.
138 static int ocfs2_search_extent_list(struct ocfs2_extent_list *el,
139 u32 v_cluster)
141 int ret = -1;
142 int i;
143 struct ocfs2_extent_rec *rec;
144 u32 rec_end, rec_start, clusters;
146 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
147 rec = &el->l_recs[i];
149 rec_start = le32_to_cpu(rec->e_cpos);
150 clusters = ocfs2_rec_clusters(el, rec);
152 rec_end = rec_start + clusters;
154 if (v_cluster >= rec_start && v_cluster < rec_end) {
155 ret = i;
156 break;
160 return ret;
163 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
164 u32 *p_cluster, u32 *num_clusters,
165 unsigned int *extent_flags)
167 int ret, i;
168 unsigned int flags = 0;
169 struct buffer_head *di_bh = NULL;
170 struct buffer_head *eb_bh = NULL;
171 struct ocfs2_dinode *di;
172 struct ocfs2_extent_block *eb;
173 struct ocfs2_extent_list *el;
174 struct ocfs2_extent_rec *rec;
175 u32 coff;
177 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), OCFS2_I(inode)->ip_blkno,
178 &di_bh, OCFS2_BH_CACHED, inode);
179 if (ret) {
180 mlog_errno(ret);
181 goto out;
184 di = (struct ocfs2_dinode *) di_bh->b_data;
185 el = &di->id2.i_list;
187 if (el->l_tree_depth) {
188 ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
189 if (ret) {
190 mlog_errno(ret);
191 goto out;
194 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
195 el = &eb->h_list;
197 if (el->l_tree_depth) {
198 ocfs2_error(inode->i_sb,
199 "Inode %lu has non zero tree depth in "
200 "leaf block %llu\n", inode->i_ino,
201 (unsigned long long)eb_bh->b_blocknr);
202 ret = -EROFS;
203 goto out;
207 i = ocfs2_search_extent_list(el, v_cluster);
208 if (i == -1) {
210 * A hole was found. Return some canned values that
211 * callers can key on. If asked for, num_clusters will
212 * be populated with the size of the hole.
214 *p_cluster = 0;
215 if (num_clusters) {
216 ret = ocfs2_figure_hole_clusters(inode, el, eb_bh,
217 v_cluster,
218 num_clusters);
219 if (ret) {
220 mlog_errno(ret);
221 goto out;
224 } else {
225 rec = &el->l_recs[i];
227 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
229 if (!rec->e_blkno) {
230 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
231 "record (%u, %u, 0)", inode->i_ino,
232 le32_to_cpu(rec->e_cpos),
233 ocfs2_rec_clusters(el, rec));
234 ret = -EROFS;
235 goto out;
238 coff = v_cluster - le32_to_cpu(rec->e_cpos);
240 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
241 le64_to_cpu(rec->e_blkno));
242 *p_cluster = *p_cluster + coff;
244 if (num_clusters)
245 *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
247 flags = rec->e_flags;
250 if (extent_flags)
251 *extent_flags = flags;
253 out:
254 brelse(di_bh);
255 brelse(eb_bh);
256 return ret;
260 * This expects alloc_sem to be held. The allocation cannot change at
261 * all while the map is in the process of being updated.
263 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
264 u64 *ret_count, unsigned int *extent_flags)
266 int ret;
267 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
268 u32 cpos, num_clusters, p_cluster;
269 u64 boff = 0;
271 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
273 ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
274 extent_flags);
275 if (ret) {
276 mlog_errno(ret);
277 goto out;
281 * p_cluster == 0 indicates a hole.
283 if (p_cluster) {
284 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
285 boff += (v_blkno & (u64)(bpc - 1));
288 *p_blkno = boff;
290 if (ret_count) {
291 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
292 *ret_count -= v_blkno & (u64)(bpc - 1);
295 out:
296 return ret;