blob: ce9aa5860569a3120c0ecf6ba16fd20f72bb2218 [file] [log] [blame]
Alex Tomasa86c6182006-10-11 01:21:03 -07001/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
21 */
22
23/*
24 * Extents support for EXT4
25 *
26 * TODO:
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
30 */
31
32#include <linux/module.h>
33#include <linux/fs.h>
34#include <linux/time.h>
35#include <linux/ext4_jbd2.h>
Mingming Caocd02ff02007-10-16 18:38:25 -040036#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070037#include <linux/highuid.h>
38#include <linux/pagemap.h>
39#include <linux/quotaops.h>
40#include <linux/string.h>
41#include <linux/slab.h>
Amit Aroraa2df2a62007-07-17 21:42:41 -040042#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070043#include <linux/ext4_fs_extents.h>
44#include <asm/uaccess.h>
45
46
Randy Dunlapd0d856e2006-10-11 01:21:07 -070047/*
48 * ext_pblock:
49 * combine low and high parts of physical block number into ext4_fsblk_t
50 */
Avantika Mathur09b88252006-12-06 20:41:36 -080051static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070052{
53 ext4_fsblk_t block;
54
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040055 block = le32_to_cpu(ex->ee_start_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070056 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070057 return block;
58}
59
Randy Dunlapd0d856e2006-10-11 01:21:07 -070060/*
61 * idx_pblock:
62 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
63 */
Aneesh Kumar K.Vc14c6fd2008-01-28 23:58:26 -050064ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070065{
66 ext4_fsblk_t block;
67
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040068 block = le32_to_cpu(ix->ei_leaf_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070069 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070070 return block;
71}
72
Randy Dunlapd0d856e2006-10-11 01:21:07 -070073/*
74 * ext4_ext_store_pblock:
75 * stores a large physical block number into an extent struct,
76 * breaking it into parts
77 */
Aneesh Kumar K.Vc14c6fd2008-01-28 23:58:26 -050078void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070079{
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040080 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070081 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070082}
83
Randy Dunlapd0d856e2006-10-11 01:21:07 -070084/*
85 * ext4_idx_store_pblock:
86 * stores a large physical block number into an index struct,
87 * breaking it into parts
88 */
Avantika Mathur09b88252006-12-06 20:41:36 -080089static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070090{
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040091 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070092 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070093}
94
Alex Tomasa86c6182006-10-11 01:21:03 -070095static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
96{
97 int err;
98
99 if (handle->h_buffer_credits > needed)
100 return handle;
101 if (!ext4_journal_extend(handle, needed))
102 return handle;
103 err = ext4_journal_restart(handle, needed);
104
105 return handle;
106}
107
108/*
109 * could return:
110 * - EROFS
111 * - ENOMEM
112 */
113static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
114 struct ext4_ext_path *path)
115{
116 if (path->p_bh) {
117 /* path points to block */
118 return ext4_journal_get_write_access(handle, path->p_bh);
119 }
120 /* path points to leaf/index in inode body */
121 /* we use in-core data, no need to protect them */
122 return 0;
123}
124
125/*
126 * could return:
127 * - EROFS
128 * - ENOMEM
129 * - EIO
130 */
131static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
132 struct ext4_ext_path *path)
133{
134 int err;
135 if (path->p_bh) {
136 /* path points to block */
137 err = ext4_journal_dirty_metadata(handle, path->p_bh);
138 } else {
139 /* path points to leaf/index in inode body */
140 err = ext4_mark_inode_dirty(handle, inode);
141 }
142 return err;
143}
144
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700145static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700146 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500147 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700148{
149 struct ext4_inode_info *ei = EXT4_I(inode);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700150 ext4_fsblk_t bg_start;
151 ext4_grpblk_t colour;
Alex Tomasa86c6182006-10-11 01:21:03 -0700152 int depth;
153
154 if (path) {
155 struct ext4_extent *ex;
156 depth = path->p_depth;
157
158 /* try to predict block placement */
Avantika Mathur7e028972006-12-06 20:41:33 -0800159 ex = path[depth].p_ext;
160 if (ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700161 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700162
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700163 /* it looks like index is empty;
164 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700165 if (path[depth].p_bh)
166 return path[depth].p_bh->b_blocknr;
167 }
168
169 /* OK. use inode's group */
170 bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
171 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
172 colour = (current->pid % 16) *
173 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
174 return bg_start + colour + block;
175}
176
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700177static ext4_fsblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -0700178ext4_ext_new_block(handle_t *handle, struct inode *inode,
179 struct ext4_ext_path *path,
180 struct ext4_extent *ex, int *err)
181{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700182 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700183
184 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
185 newblock = ext4_new_block(handle, inode, goal, err);
186 return newblock;
187}
188
Avantika Mathur09b88252006-12-06 20:41:36 -0800189static int ext4_ext_space_block(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700190{
191 int size;
192
193 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
194 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100195#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700196 if (size > 6)
197 size = 6;
198#endif
199 return size;
200}
201
Avantika Mathur09b88252006-12-06 20:41:36 -0800202static int ext4_ext_space_block_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700203{
204 int size;
205
206 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
207 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100208#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700209 if (size > 5)
210 size = 5;
211#endif
212 return size;
213}
214
Avantika Mathur09b88252006-12-06 20:41:36 -0800215static int ext4_ext_space_root(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700216{
217 int size;
218
219 size = sizeof(EXT4_I(inode)->i_data);
220 size -= sizeof(struct ext4_extent_header);
221 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100222#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700223 if (size > 3)
224 size = 3;
225#endif
226 return size;
227}
228
Avantika Mathur09b88252006-12-06 20:41:36 -0800229static int ext4_ext_space_root_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700230{
231 int size;
232
233 size = sizeof(EXT4_I(inode)->i_data);
234 size -= sizeof(struct ext4_extent_header);
235 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100236#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700237 if (size > 4)
238 size = 4;
239#endif
240 return size;
241}
242
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400243static int
244ext4_ext_max_entries(struct inode *inode, int depth)
245{
246 int max;
247
248 if (depth == ext_depth(inode)) {
249 if (depth == 0)
250 max = ext4_ext_space_root(inode);
251 else
252 max = ext4_ext_space_root_idx(inode);
253 } else {
254 if (depth == 0)
255 max = ext4_ext_space_block(inode);
256 else
257 max = ext4_ext_space_block_idx(inode);
258 }
259
260 return max;
261}
262
263static int __ext4_ext_check_header(const char *function, struct inode *inode,
264 struct ext4_extent_header *eh,
265 int depth)
266{
267 const char *error_msg;
268 int max = 0;
269
270 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
271 error_msg = "invalid magic";
272 goto corrupted;
273 }
274 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
275 error_msg = "unexpected eh_depth";
276 goto corrupted;
277 }
278 if (unlikely(eh->eh_max == 0)) {
279 error_msg = "invalid eh_max";
280 goto corrupted;
281 }
282 max = ext4_ext_max_entries(inode, depth);
283 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
284 error_msg = "too large eh_max";
285 goto corrupted;
286 }
287 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
288 error_msg = "invalid eh_entries";
289 goto corrupted;
290 }
291 return 0;
292
293corrupted:
294 ext4_error(inode->i_sb, function,
295 "bad header in inode #%lu: %s - magic %x, "
296 "entries %u, max %u(%u), depth %u(%u)",
297 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
298 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
299 max, le16_to_cpu(eh->eh_depth), depth);
300
301 return -EIO;
302}
303
304#define ext4_ext_check_header(inode, eh, depth) \
305 __ext4_ext_check_header(__FUNCTION__, inode, eh, depth)
306
Alex Tomasa86c6182006-10-11 01:21:03 -0700307#ifdef EXT_DEBUG
308static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
309{
310 int k, l = path->p_depth;
311
312 ext_debug("path:");
313 for (k = 0; k <= l; k++, path++) {
314 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700315 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700316 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700317 } else if (path->p_ext) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700318 ext_debug(" %d:%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700319 le32_to_cpu(path->p_ext->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400320 ext4_ext_get_actual_len(path->p_ext),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700321 ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700322 } else
323 ext_debug(" []");
324 }
325 ext_debug("\n");
326}
327
328static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
329{
330 int depth = ext_depth(inode);
331 struct ext4_extent_header *eh;
332 struct ext4_extent *ex;
333 int i;
334
335 if (!path)
336 return;
337
338 eh = path[depth].p_hdr;
339 ex = EXT_FIRST_EXTENT(eh);
340
341 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700342 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400343 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700344 }
345 ext_debug("\n");
346}
347#else
348#define ext4_ext_show_path(inode,path)
349#define ext4_ext_show_leaf(inode,path)
350#endif
351
352static void ext4_ext_drop_refs(struct ext4_ext_path *path)
353{
354 int depth = path->p_depth;
355 int i;
356
357 for (i = 0; i <= depth; i++, path++)
358 if (path->p_bh) {
359 brelse(path->p_bh);
360 path->p_bh = NULL;
361 }
362}
363
364/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700365 * ext4_ext_binsearch_idx:
366 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400367 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700368 */
369static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500370ext4_ext_binsearch_idx(struct inode *inode,
371 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700372{
373 struct ext4_extent_header *eh = path->p_hdr;
374 struct ext4_extent_idx *r, *l, *m;
375
Alex Tomasa86c6182006-10-11 01:21:03 -0700376
Eric Sandeenbba90742008-01-28 23:58:27 -0500377 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700378
379 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400380 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700381 while (l <= r) {
382 m = l + (r - l) / 2;
383 if (block < le32_to_cpu(m->ei_block))
384 r = m - 1;
385 else
386 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400387 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
388 m, le32_to_cpu(m->ei_block),
389 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700390 }
391
392 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700393 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400394 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700395
396#ifdef CHECK_BINSEARCH
397 {
398 struct ext4_extent_idx *chix, *ix;
399 int k;
400
401 chix = ix = EXT_FIRST_INDEX(eh);
402 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
403 if (k != 0 &&
404 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
405 printk("k=%d, ix=0x%p, first=0x%p\n", k,
406 ix, EXT_FIRST_INDEX(eh));
407 printk("%u <= %u\n",
408 le32_to_cpu(ix->ei_block),
409 le32_to_cpu(ix[-1].ei_block));
410 }
411 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400412 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700413 if (block < le32_to_cpu(ix->ei_block))
414 break;
415 chix = ix;
416 }
417 BUG_ON(chix != path->p_idx);
418 }
419#endif
420
421}
422
423/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700424 * ext4_ext_binsearch:
425 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400426 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700427 */
428static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500429ext4_ext_binsearch(struct inode *inode,
430 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700431{
432 struct ext4_extent_header *eh = path->p_hdr;
433 struct ext4_extent *r, *l, *m;
434
Alex Tomasa86c6182006-10-11 01:21:03 -0700435 if (eh->eh_entries == 0) {
436 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700437 * this leaf is empty:
438 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700439 */
440 return;
441 }
442
Eric Sandeenbba90742008-01-28 23:58:27 -0500443 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700444
445 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400446 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700447
448 while (l <= r) {
449 m = l + (r - l) / 2;
450 if (block < le32_to_cpu(m->ee_block))
451 r = m - 1;
452 else
453 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400454 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
455 m, le32_to_cpu(m->ee_block),
456 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700457 }
458
459 path->p_ext = l - 1;
Mingming Cao2ae02102006-10-11 01:21:11 -0700460 ext_debug(" -> %d:%llu:%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400461 le32_to_cpu(path->p_ext->ee_block),
462 ext_pblock(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400463 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700464
465#ifdef CHECK_BINSEARCH
466 {
467 struct ext4_extent *chex, *ex;
468 int k;
469
470 chex = ex = EXT_FIRST_EXTENT(eh);
471 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
472 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400473 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700474 if (block < le32_to_cpu(ex->ee_block))
475 break;
476 chex = ex;
477 }
478 BUG_ON(chex != path->p_ext);
479 }
480#endif
481
482}
483
484int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
485{
486 struct ext4_extent_header *eh;
487
488 eh = ext_inode_hdr(inode);
489 eh->eh_depth = 0;
490 eh->eh_entries = 0;
491 eh->eh_magic = EXT4_EXT_MAGIC;
492 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
493 ext4_mark_inode_dirty(handle, inode);
494 ext4_ext_invalidate_cache(inode);
495 return 0;
496}
497
498struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500499ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
500 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700501{
502 struct ext4_extent_header *eh;
503 struct buffer_head *bh;
504 short int depth, i, ppos = 0, alloc = 0;
505
506 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400507 depth = ext_depth(inode);
508 if (ext4_ext_check_header(inode, eh, depth))
Alex Tomasa86c6182006-10-11 01:21:03 -0700509 return ERR_PTR(-EIO);
510
Alex Tomasa86c6182006-10-11 01:21:03 -0700511
512 /* account possible depth increase */
513 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800514 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700515 GFP_NOFS);
516 if (!path)
517 return ERR_PTR(-ENOMEM);
518 alloc = 1;
519 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700520 path[0].p_hdr = eh;
521
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400522 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700523 /* walk through the tree */
524 while (i) {
525 ext_debug("depth %d: num %d, max %d\n",
526 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400527
Alex Tomasa86c6182006-10-11 01:21:03 -0700528 ext4_ext_binsearch_idx(inode, path + ppos, block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700529 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700530 path[ppos].p_depth = i;
531 path[ppos].p_ext = NULL;
532
533 bh = sb_bread(inode->i_sb, path[ppos].p_block);
534 if (!bh)
535 goto err;
536
537 eh = ext_block_hdr(bh);
538 ppos++;
539 BUG_ON(ppos > depth);
540 path[ppos].p_bh = bh;
541 path[ppos].p_hdr = eh;
542 i--;
543
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400544 if (ext4_ext_check_header(inode, eh, i))
Alex Tomasa86c6182006-10-11 01:21:03 -0700545 goto err;
546 }
547
548 path[ppos].p_depth = i;
549 path[ppos].p_hdr = eh;
550 path[ppos].p_ext = NULL;
551 path[ppos].p_idx = NULL;
552
Alex Tomasa86c6182006-10-11 01:21:03 -0700553 /* find extent */
554 ext4_ext_binsearch(inode, path + ppos, block);
555
556 ext4_ext_show_path(inode, path);
557
558 return path;
559
560err:
561 ext4_ext_drop_refs(path);
562 if (alloc)
563 kfree(path);
564 return ERR_PTR(-EIO);
565}
566
567/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700568 * ext4_ext_insert_index:
569 * insert new index [@logical;@ptr] into the block at @curp;
570 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700571 */
572static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
573 struct ext4_ext_path *curp,
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700574 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700575{
576 struct ext4_extent_idx *ix;
577 int len, err;
578
Avantika Mathur7e028972006-12-06 20:41:33 -0800579 err = ext4_ext_get_access(handle, inode, curp);
580 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700581 return err;
582
583 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
584 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
585 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
586 /* insert after */
587 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
588 len = (len - 1) * sizeof(struct ext4_extent_idx);
589 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400590 ext_debug("insert new index %d after: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700591 "move %d from 0x%p to 0x%p\n",
592 logical, ptr, len,
593 (curp->p_idx + 1), (curp->p_idx + 2));
594 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
595 }
596 ix = curp->p_idx + 1;
597 } else {
598 /* insert before */
599 len = len * sizeof(struct ext4_extent_idx);
600 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400601 ext_debug("insert new index %d before: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700602 "move %d from 0x%p to 0x%p\n",
603 logical, ptr, len,
604 curp->p_idx, (curp->p_idx + 1));
605 memmove(curp->p_idx + 1, curp->p_idx, len);
606 ix = curp->p_idx;
607 }
608
609 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700610 ext4_idx_store_pblock(ix, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700611 curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
612
613 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400614 > le16_to_cpu(curp->p_hdr->eh_max));
Alex Tomasa86c6182006-10-11 01:21:03 -0700615 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
616
617 err = ext4_ext_dirty(handle, inode, curp);
618 ext4_std_error(inode->i_sb, err);
619
620 return err;
621}
622
623/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700624 * ext4_ext_split:
625 * inserts new subtree into the path, using free index entry
626 * at depth @at:
627 * - allocates all needed blocks (new leaf and all intermediate index blocks)
628 * - makes decision where to split
629 * - moves remaining extents and index entries (right to the split point)
630 * into the newly allocated blocks
631 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700632 */
633static int ext4_ext_split(handle_t *handle, struct inode *inode,
634 struct ext4_ext_path *path,
635 struct ext4_extent *newext, int at)
636{
637 struct buffer_head *bh = NULL;
638 int depth = ext_depth(inode);
639 struct ext4_extent_header *neh;
640 struct ext4_extent_idx *fidx;
641 struct ext4_extent *ex;
642 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700643 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700644 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700645 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700646 int err = 0;
647
648 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700649 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700650
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700651 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700652 * border from split point */
653 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
654 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
655 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700656 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700657 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400658 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700659 } else {
660 border = newext->ee_block;
661 ext_debug("leaf will be added."
662 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400663 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700664 }
665
666 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700667 * If error occurs, then we break processing
668 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700669 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700670 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700671 */
672
673 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700674 * Get array to track all allocated blocks.
675 * We need this to handle errors and free blocks
676 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700677 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800678 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700679 if (!ablocks)
680 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700681
682 /* allocate all needed blocks */
683 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
684 for (a = 0; a < depth - at; a++) {
685 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
686 if (newblock == 0)
687 goto cleanup;
688 ablocks[a] = newblock;
689 }
690
691 /* initialize new leaf */
692 newblock = ablocks[--a];
693 BUG_ON(newblock == 0);
694 bh = sb_getblk(inode->i_sb, newblock);
695 if (!bh) {
696 err = -EIO;
697 goto cleanup;
698 }
699 lock_buffer(bh);
700
Avantika Mathur7e028972006-12-06 20:41:33 -0800701 err = ext4_journal_get_create_access(handle, bh);
702 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700703 goto cleanup;
704
705 neh = ext_block_hdr(bh);
706 neh->eh_entries = 0;
707 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
708 neh->eh_magic = EXT4_EXT_MAGIC;
709 neh->eh_depth = 0;
710 ex = EXT_FIRST_EXTENT(neh);
711
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700712 /* move remainder of path[depth] to the new leaf */
Alex Tomasa86c6182006-10-11 01:21:03 -0700713 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
714 /* start copy from next extent */
715 /* TODO: we could do it by single memmove */
716 m = 0;
717 path[depth].p_ext++;
718 while (path[depth].p_ext <=
719 EXT_MAX_EXTENT(path[depth].p_hdr)) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700720 ext_debug("move %d:%llu:%d in new leaf %llu\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400721 le32_to_cpu(path[depth].p_ext->ee_block),
722 ext_pblock(path[depth].p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400723 ext4_ext_get_actual_len(path[depth].p_ext),
Alex Tomasa86c6182006-10-11 01:21:03 -0700724 newblock);
725 /*memmove(ex++, path[depth].p_ext++,
726 sizeof(struct ext4_extent));
727 neh->eh_entries++;*/
728 path[depth].p_ext++;
729 m++;
730 }
731 if (m) {
732 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
733 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
734 }
735
736 set_buffer_uptodate(bh);
737 unlock_buffer(bh);
738
Avantika Mathur7e028972006-12-06 20:41:33 -0800739 err = ext4_journal_dirty_metadata(handle, bh);
740 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700741 goto cleanup;
742 brelse(bh);
743 bh = NULL;
744
745 /* correct old leaf */
746 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -0800747 err = ext4_ext_get_access(handle, inode, path + depth);
748 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700749 goto cleanup;
750 path[depth].p_hdr->eh_entries =
751 cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
Avantika Mathur7e028972006-12-06 20:41:33 -0800752 err = ext4_ext_dirty(handle, inode, path + depth);
753 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700754 goto cleanup;
755
756 }
757
758 /* create intermediate indexes */
759 k = depth - at - 1;
760 BUG_ON(k < 0);
761 if (k)
762 ext_debug("create %d intermediate indices\n", k);
763 /* insert new index into current index block */
764 /* current depth stored in i var */
765 i = depth - 1;
766 while (k--) {
767 oldblock = newblock;
768 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -0500769 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700770 if (!bh) {
771 err = -EIO;
772 goto cleanup;
773 }
774 lock_buffer(bh);
775
Avantika Mathur7e028972006-12-06 20:41:33 -0800776 err = ext4_journal_get_create_access(handle, bh);
777 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700778 goto cleanup;
779
780 neh = ext_block_hdr(bh);
781 neh->eh_entries = cpu_to_le16(1);
782 neh->eh_magic = EXT4_EXT_MAGIC;
783 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
784 neh->eh_depth = cpu_to_le16(depth - i);
785 fidx = EXT_FIRST_INDEX(neh);
786 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700787 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700788
Eric Sandeenbba90742008-01-28 23:58:27 -0500789 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
790 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700791 /* copy indexes */
792 m = 0;
793 path[i].p_idx++;
794
795 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
796 EXT_MAX_INDEX(path[i].p_hdr));
797 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
798 EXT_LAST_INDEX(path[i].p_hdr));
799 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400800 ext_debug("%d: move %d:%llu in new index %llu\n", i,
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400801 le32_to_cpu(path[i].p_idx->ei_block),
802 idx_pblock(path[i].p_idx),
803 newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700804 /*memmove(++fidx, path[i].p_idx++,
805 sizeof(struct ext4_extent_idx));
806 neh->eh_entries++;
807 BUG_ON(neh->eh_entries > neh->eh_max);*/
808 path[i].p_idx++;
809 m++;
810 }
811 if (m) {
812 memmove(++fidx, path[i].p_idx - m,
813 sizeof(struct ext4_extent_idx) * m);
814 neh->eh_entries =
815 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
816 }
817 set_buffer_uptodate(bh);
818 unlock_buffer(bh);
819
Avantika Mathur7e028972006-12-06 20:41:33 -0800820 err = ext4_journal_dirty_metadata(handle, bh);
821 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700822 goto cleanup;
823 brelse(bh);
824 bh = NULL;
825
826 /* correct old index */
827 if (m) {
828 err = ext4_ext_get_access(handle, inode, path + i);
829 if (err)
830 goto cleanup;
831 path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
832 err = ext4_ext_dirty(handle, inode, path + i);
833 if (err)
834 goto cleanup;
835 }
836
837 i--;
838 }
839
840 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -0700841 err = ext4_ext_insert_index(handle, inode, path + at,
842 le32_to_cpu(border), newblock);
843
844cleanup:
845 if (bh) {
846 if (buffer_locked(bh))
847 unlock_buffer(bh);
848 brelse(bh);
849 }
850
851 if (err) {
852 /* free all allocated blocks in error case */
853 for (i = 0; i < depth; i++) {
854 if (!ablocks[i])
855 continue;
Alex Tomasc9de5602008-01-29 00:19:52 -0500856 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700857 }
858 }
859 kfree(ablocks);
860
861 return err;
862}
863
864/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700865 * ext4_ext_grow_indepth:
866 * implements tree growing procedure:
867 * - allocates new block
868 * - moves top-level data (index block or leaf) into the new block
869 * - initializes new top-level, creating index that points to the
870 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -0700871 */
872static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
873 struct ext4_ext_path *path,
874 struct ext4_extent *newext)
875{
876 struct ext4_ext_path *curp = path;
877 struct ext4_extent_header *neh;
878 struct ext4_extent_idx *fidx;
879 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700880 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700881 int err = 0;
882
883 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
884 if (newblock == 0)
885 return err;
886
887 bh = sb_getblk(inode->i_sb, newblock);
888 if (!bh) {
889 err = -EIO;
890 ext4_std_error(inode->i_sb, err);
891 return err;
892 }
893 lock_buffer(bh);
894
Avantika Mathur7e028972006-12-06 20:41:33 -0800895 err = ext4_journal_get_create_access(handle, bh);
896 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -0700897 unlock_buffer(bh);
898 goto out;
899 }
900
901 /* move top-level index/leaf into new block */
902 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
903
904 /* set size of new block */
905 neh = ext_block_hdr(bh);
906 /* old root could have indexes or leaves
907 * so calculate e_max right way */
908 if (ext_depth(inode))
909 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
910 else
911 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
912 neh->eh_magic = EXT4_EXT_MAGIC;
913 set_buffer_uptodate(bh);
914 unlock_buffer(bh);
915
Avantika Mathur7e028972006-12-06 20:41:33 -0800916 err = ext4_journal_dirty_metadata(handle, bh);
917 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700918 goto out;
919
920 /* create index in new top-level index: num,max,pointer */
Avantika Mathur7e028972006-12-06 20:41:33 -0800921 err = ext4_ext_get_access(handle, inode, curp);
922 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700923 goto out;
924
925 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
926 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
927 curp->p_hdr->eh_entries = cpu_to_le16(1);
928 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400929
930 if (path[0].p_hdr->eh_depth)
931 curp->p_idx->ei_block =
932 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
933 else
934 curp->p_idx->ei_block =
935 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700936 ext4_idx_store_pblock(curp->p_idx, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700937
938 neh = ext_inode_hdr(inode);
939 fidx = EXT_FIRST_INDEX(neh);
Mingming Cao2ae02102006-10-11 01:21:11 -0700940 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700941 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700942 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700943
944 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
945 err = ext4_ext_dirty(handle, inode, curp);
946out:
947 brelse(bh);
948
949 return err;
950}
951
952/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700953 * ext4_ext_create_new_leaf:
954 * finds empty index and adds new leaf.
955 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -0700956 */
957static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
958 struct ext4_ext_path *path,
959 struct ext4_extent *newext)
960{
961 struct ext4_ext_path *curp;
962 int depth, i, err = 0;
963
964repeat:
965 i = depth = ext_depth(inode);
966
967 /* walk up to the tree and look for free index entry */
968 curp = path + depth;
969 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
970 i--;
971 curp--;
972 }
973
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700974 /* we use already allocated block for index block,
975 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -0700976 if (EXT_HAS_FREE_INDEX(curp)) {
977 /* if we found index with free entry, then use that
978 * entry: create all needed subtree and add new leaf */
979 err = ext4_ext_split(handle, inode, path, newext, i);
980
981 /* refill path */
982 ext4_ext_drop_refs(path);
983 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500984 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
985 path);
Alex Tomasa86c6182006-10-11 01:21:03 -0700986 if (IS_ERR(path))
987 err = PTR_ERR(path);
988 } else {
989 /* tree is full, time to grow in depth */
990 err = ext4_ext_grow_indepth(handle, inode, path, newext);
991 if (err)
992 goto out;
993
994 /* refill path */
995 ext4_ext_drop_refs(path);
996 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500997 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
998 path);
Alex Tomasa86c6182006-10-11 01:21:03 -0700999 if (IS_ERR(path)) {
1000 err = PTR_ERR(path);
1001 goto out;
1002 }
1003
1004 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001005 * only first (depth 0 -> 1) produces free space;
1006 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001007 */
1008 depth = ext_depth(inode);
1009 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001010 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001011 goto repeat;
1012 }
1013 }
1014
1015out:
1016 return err;
1017}
1018
1019/*
Alex Tomas1988b512008-01-28 23:58:27 -05001020 * search the closest allocated block to the left for *logical
1021 * and returns it at @logical + it's physical address at @phys
1022 * if *logical is the smallest allocated block, the function
1023 * returns 0 at @phys
1024 * return value contains 0 (success) or error code
1025 */
1026int
1027ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1028 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1029{
1030 struct ext4_extent_idx *ix;
1031 struct ext4_extent *ex;
1032 int depth;
1033
1034 BUG_ON(path == NULL);
1035 depth = path->p_depth;
1036 *phys = 0;
1037
1038 if (depth == 0 && path->p_ext == NULL)
1039 return 0;
1040
1041 /* usually extent in the path covers blocks smaller
1042 * then *logical, but it can be that extent is the
1043 * first one in the file */
1044
1045 ex = path[depth].p_ext;
1046 if (*logical < le32_to_cpu(ex->ee_block)) {
1047 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1048 while (--depth >= 0) {
1049 ix = path[depth].p_idx;
1050 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1051 }
1052 return 0;
1053 }
1054
1055 BUG_ON(*logical < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
1056
1057 *logical = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1;
1058 *phys = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - 1;
1059 return 0;
1060}
1061
1062/*
1063 * search the closest allocated block to the right for *logical
1064 * and returns it at @logical + it's physical address at @phys
1065 * if *logical is the smallest allocated block, the function
1066 * returns 0 at @phys
1067 * return value contains 0 (success) or error code
1068 */
1069int
1070ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1071 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1072{
1073 struct buffer_head *bh = NULL;
1074 struct ext4_extent_header *eh;
1075 struct ext4_extent_idx *ix;
1076 struct ext4_extent *ex;
1077 ext4_fsblk_t block;
1078 int depth;
1079
1080 BUG_ON(path == NULL);
1081 depth = path->p_depth;
1082 *phys = 0;
1083
1084 if (depth == 0 && path->p_ext == NULL)
1085 return 0;
1086
1087 /* usually extent in the path covers blocks smaller
1088 * then *logical, but it can be that extent is the
1089 * first one in the file */
1090
1091 ex = path[depth].p_ext;
1092 if (*logical < le32_to_cpu(ex->ee_block)) {
1093 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1094 while (--depth >= 0) {
1095 ix = path[depth].p_idx;
1096 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1097 }
1098 *logical = le32_to_cpu(ex->ee_block);
1099 *phys = ext_pblock(ex);
1100 return 0;
1101 }
1102
1103 BUG_ON(*logical < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
1104
1105 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1106 /* next allocated block in this leaf */
1107 ex++;
1108 *logical = le32_to_cpu(ex->ee_block);
1109 *phys = ext_pblock(ex);
1110 return 0;
1111 }
1112
1113 /* go up and search for index to the right */
1114 while (--depth >= 0) {
1115 ix = path[depth].p_idx;
1116 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1117 break;
1118 }
1119
1120 if (depth < 0) {
1121 /* we've gone up to the root and
1122 * found no index to the right */
1123 return 0;
1124 }
1125
1126 /* we've found index to the right, let's
1127 * follow it and find the closest allocated
1128 * block to the right */
1129 ix++;
1130 block = idx_pblock(ix);
1131 while (++depth < path->p_depth) {
1132 bh = sb_bread(inode->i_sb, block);
1133 if (bh == NULL)
1134 return -EIO;
1135 eh = ext_block_hdr(bh);
1136 if (ext4_ext_check_header(inode, eh, depth)) {
1137 put_bh(bh);
1138 return -EIO;
1139 }
1140 ix = EXT_FIRST_INDEX(eh);
1141 block = idx_pblock(ix);
1142 put_bh(bh);
1143 }
1144
1145 bh = sb_bread(inode->i_sb, block);
1146 if (bh == NULL)
1147 return -EIO;
1148 eh = ext_block_hdr(bh);
1149 if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1150 put_bh(bh);
1151 return -EIO;
1152 }
1153 ex = EXT_FIRST_EXTENT(eh);
1154 *logical = le32_to_cpu(ex->ee_block);
1155 *phys = ext_pblock(ex);
1156 put_bh(bh);
1157 return 0;
1158
1159}
1160
1161/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001162 * ext4_ext_next_allocated_block:
1163 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1164 * NOTE: it considers block number from index entry as
1165 * allocated block. Thus, index entries have to be consistent
1166 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001167 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001168static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001169ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1170{
1171 int depth;
1172
1173 BUG_ON(path == NULL);
1174 depth = path->p_depth;
1175
1176 if (depth == 0 && path->p_ext == NULL)
1177 return EXT_MAX_BLOCK;
1178
1179 while (depth >= 0) {
1180 if (depth == path->p_depth) {
1181 /* leaf */
1182 if (path[depth].p_ext !=
1183 EXT_LAST_EXTENT(path[depth].p_hdr))
1184 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1185 } else {
1186 /* index */
1187 if (path[depth].p_idx !=
1188 EXT_LAST_INDEX(path[depth].p_hdr))
1189 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1190 }
1191 depth--;
1192 }
1193
1194 return EXT_MAX_BLOCK;
1195}
1196
1197/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001198 * ext4_ext_next_leaf_block:
Alex Tomasa86c6182006-10-11 01:21:03 -07001199 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1200 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001201static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
Andrew Morton63f57932006-10-11 01:21:24 -07001202 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001203{
1204 int depth;
1205
1206 BUG_ON(path == NULL);
1207 depth = path->p_depth;
1208
1209 /* zero-tree has no leaf blocks at all */
1210 if (depth == 0)
1211 return EXT_MAX_BLOCK;
1212
1213 /* go to index block */
1214 depth--;
1215
1216 while (depth >= 0) {
1217 if (path[depth].p_idx !=
1218 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001219 return (ext4_lblk_t)
1220 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001221 depth--;
1222 }
1223
1224 return EXT_MAX_BLOCK;
1225}
1226
1227/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001228 * ext4_ext_correct_indexes:
1229 * if leaf gets modified and modified extent is first in the leaf,
1230 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001231 * TODO: do we need to correct tree in all cases?
1232 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001233static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001234 struct ext4_ext_path *path)
1235{
1236 struct ext4_extent_header *eh;
1237 int depth = ext_depth(inode);
1238 struct ext4_extent *ex;
1239 __le32 border;
1240 int k, err = 0;
1241
1242 eh = path[depth].p_hdr;
1243 ex = path[depth].p_ext;
1244 BUG_ON(ex == NULL);
1245 BUG_ON(eh == NULL);
1246
1247 if (depth == 0) {
1248 /* there is no tree at all */
1249 return 0;
1250 }
1251
1252 if (ex != EXT_FIRST_EXTENT(eh)) {
1253 /* we correct tree if first leaf got modified only */
1254 return 0;
1255 }
1256
1257 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001258 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001259 */
1260 k = depth - 1;
1261 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001262 err = ext4_ext_get_access(handle, inode, path + k);
1263 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001264 return err;
1265 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001266 err = ext4_ext_dirty(handle, inode, path + k);
1267 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001268 return err;
1269
1270 while (k--) {
1271 /* change all left-side indexes */
1272 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1273 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001274 err = ext4_ext_get_access(handle, inode, path + k);
1275 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001276 break;
1277 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001278 err = ext4_ext_dirty(handle, inode, path + k);
1279 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001280 break;
1281 }
1282
1283 return err;
1284}
1285
Avantika Mathur09b88252006-12-06 20:41:36 -08001286static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001287ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1288 struct ext4_extent *ex2)
1289{
Amit Arora749269f2007-07-18 09:02:56 -04001290 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001291
1292 /*
1293 * Make sure that either both extents are uninitialized, or
1294 * both are _not_.
1295 */
1296 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1297 return 0;
1298
Amit Arora749269f2007-07-18 09:02:56 -04001299 if (ext4_ext_is_uninitialized(ex1))
1300 max_len = EXT_UNINIT_MAX_LEN;
1301 else
1302 max_len = EXT_INIT_MAX_LEN;
1303
Amit Aroraa2df2a62007-07-17 21:42:41 -04001304 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1305 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1306
1307 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001308 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001309 return 0;
1310
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001311 /*
1312 * To allow future support for preallocated extents to be added
1313 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001314 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001315 */
Amit Arora749269f2007-07-18 09:02:56 -04001316 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001317 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001318#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -07001319 if (le16_to_cpu(ex1->ee_len) >= 4)
1320 return 0;
1321#endif
1322
Amit Aroraa2df2a62007-07-17 21:42:41 -04001323 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001324 return 1;
1325 return 0;
1326}
1327
1328/*
Amit Arora56055d32007-07-17 21:42:38 -04001329 * This function tries to merge the "ex" extent to the next extent in the tree.
1330 * It always tries to merge towards right. If you want to merge towards
1331 * left, pass "ex - 1" as argument instead of "ex".
1332 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1333 * 1 if they got merged.
1334 */
1335int ext4_ext_try_to_merge(struct inode *inode,
1336 struct ext4_ext_path *path,
1337 struct ext4_extent *ex)
1338{
1339 struct ext4_extent_header *eh;
1340 unsigned int depth, len;
1341 int merge_done = 0;
1342 int uninitialized = 0;
1343
1344 depth = ext_depth(inode);
1345 BUG_ON(path[depth].p_hdr == NULL);
1346 eh = path[depth].p_hdr;
1347
1348 while (ex < EXT_LAST_EXTENT(eh)) {
1349 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1350 break;
1351 /* merge with next extent! */
1352 if (ext4_ext_is_uninitialized(ex))
1353 uninitialized = 1;
1354 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1355 + ext4_ext_get_actual_len(ex + 1));
1356 if (uninitialized)
1357 ext4_ext_mark_uninitialized(ex);
1358
1359 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1360 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1361 * sizeof(struct ext4_extent);
1362 memmove(ex + 1, ex + 2, len);
1363 }
1364 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
1365 merge_done = 1;
1366 WARN_ON(eh->eh_entries == 0);
1367 if (!eh->eh_entries)
1368 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1369 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1370 }
1371
1372 return merge_done;
1373}
1374
1375/*
Amit Arora25d14f92007-05-24 13:04:13 -04001376 * check if a portion of the "newext" extent overlaps with an
1377 * existing extent.
1378 *
1379 * If there is an overlap discovered, it updates the length of the newext
1380 * such that there will be no overlap, and then returns 1.
1381 * If there is no overlap found, it returns 0.
1382 */
1383unsigned int ext4_ext_check_overlap(struct inode *inode,
1384 struct ext4_extent *newext,
1385 struct ext4_ext_path *path)
1386{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001387 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001388 unsigned int depth, len1;
1389 unsigned int ret = 0;
1390
1391 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001392 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001393 depth = ext_depth(inode);
1394 if (!path[depth].p_ext)
1395 goto out;
1396 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1397
1398 /*
1399 * get the next allocated block if the extent in the path
1400 * is before the requested block(s)
1401 */
1402 if (b2 < b1) {
1403 b2 = ext4_ext_next_allocated_block(path);
1404 if (b2 == EXT_MAX_BLOCK)
1405 goto out;
1406 }
1407
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001408 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001409 if (b1 + len1 < b1) {
1410 len1 = EXT_MAX_BLOCK - b1;
1411 newext->ee_len = cpu_to_le16(len1);
1412 ret = 1;
1413 }
1414
1415 /* check for overlap */
1416 if (b1 + len1 > b2) {
1417 newext->ee_len = cpu_to_le16(b2 - b1);
1418 ret = 1;
1419 }
1420out:
1421 return ret;
1422}
1423
1424/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001425 * ext4_ext_insert_extent:
1426 * tries to merge requsted extent into the existing extent or
1427 * inserts requested extent as new one into the tree,
1428 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001429 */
1430int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1431 struct ext4_ext_path *path,
1432 struct ext4_extent *newext)
1433{
1434 struct ext4_extent_header * eh;
1435 struct ext4_extent *ex, *fex;
1436 struct ext4_extent *nearex; /* nearest extent */
1437 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001438 int depth, len, err;
1439 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001440 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001441
Amit Aroraa2df2a62007-07-17 21:42:41 -04001442 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001443 depth = ext_depth(inode);
1444 ex = path[depth].p_ext;
1445 BUG_ON(path[depth].p_hdr == NULL);
1446
1447 /* try to insert block into found extent and return */
1448 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming Cao2ae02102006-10-11 01:21:11 -07001449 ext_debug("append %d block to %d:%d (from %llu)\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001450 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001451 le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001452 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001453 err = ext4_ext_get_access(handle, inode, path + depth);
1454 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001455 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001456
1457 /*
1458 * ext4_can_extents_be_merged should have checked that either
1459 * both extents are uninitialized, or both aren't. Thus we
1460 * need to check only one of them here.
1461 */
1462 if (ext4_ext_is_uninitialized(ex))
1463 uninitialized = 1;
1464 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1465 + ext4_ext_get_actual_len(newext));
1466 if (uninitialized)
1467 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001468 eh = path[depth].p_hdr;
1469 nearex = ex;
1470 goto merge;
1471 }
1472
1473repeat:
1474 depth = ext_depth(inode);
1475 eh = path[depth].p_hdr;
1476 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1477 goto has_space;
1478
1479 /* probably next leaf has space for us? */
1480 fex = EXT_LAST_EXTENT(eh);
1481 next = ext4_ext_next_leaf_block(inode, path);
1482 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1483 && next != EXT_MAX_BLOCK) {
1484 ext_debug("next leaf block - %d\n", next);
1485 BUG_ON(npath != NULL);
1486 npath = ext4_ext_find_extent(inode, next, NULL);
1487 if (IS_ERR(npath))
1488 return PTR_ERR(npath);
1489 BUG_ON(npath->p_depth != path->p_depth);
1490 eh = npath[depth].p_hdr;
1491 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1492 ext_debug("next leaf isnt full(%d)\n",
1493 le16_to_cpu(eh->eh_entries));
1494 path = npath;
1495 goto repeat;
1496 }
1497 ext_debug("next leaf has no free space(%d,%d)\n",
1498 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1499 }
1500
1501 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001502 * There is no free space in the found leaf.
1503 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001504 */
1505 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1506 if (err)
1507 goto cleanup;
1508 depth = ext_depth(inode);
1509 eh = path[depth].p_hdr;
1510
1511has_space:
1512 nearex = path[depth].p_ext;
1513
Avantika Mathur7e028972006-12-06 20:41:33 -08001514 err = ext4_ext_get_access(handle, inode, path + depth);
1515 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001516 goto cleanup;
1517
1518 if (!nearex) {
1519 /* there is no extent in this leaf, create first one */
Mingming Cao2ae02102006-10-11 01:21:11 -07001520 ext_debug("first extent in the leaf: %d:%llu:%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001521 le32_to_cpu(newext->ee_block),
1522 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001523 ext4_ext_get_actual_len(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001524 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1525 } else if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001526 > le32_to_cpu(nearex->ee_block)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001527/* BUG_ON(newext->ee_block == nearex->ee_block); */
1528 if (nearex != EXT_LAST_EXTENT(eh)) {
1529 len = EXT_MAX_EXTENT(eh) - nearex;
1530 len = (len - 1) * sizeof(struct ext4_extent);
1531 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001532 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001533 "move %d from 0x%p to 0x%p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001534 le32_to_cpu(newext->ee_block),
1535 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001536 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001537 nearex, len, nearex + 1, nearex + 2);
1538 memmove(nearex + 2, nearex + 1, len);
1539 }
1540 path[depth].p_ext = nearex + 1;
1541 } else {
1542 BUG_ON(newext->ee_block == nearex->ee_block);
1543 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1544 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001545 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001546 "move %d from 0x%p to 0x%p\n",
1547 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001548 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001549 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001550 nearex, len, nearex + 1, nearex + 2);
1551 memmove(nearex + 1, nearex, len);
1552 path[depth].p_ext = nearex;
1553 }
1554
1555 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1556 nearex = path[depth].p_ext;
1557 nearex->ee_block = newext->ee_block;
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -04001558 ext4_ext_store_pblock(nearex, ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001559 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001560
1561merge:
1562 /* try to merge extents to the right */
Amit Arora56055d32007-07-17 21:42:38 -04001563 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001564
1565 /* try to merge extents to the left */
1566
1567 /* time to correct all indexes above */
1568 err = ext4_ext_correct_indexes(handle, inode, path);
1569 if (err)
1570 goto cleanup;
1571
1572 err = ext4_ext_dirty(handle, inode, path + depth);
1573
1574cleanup:
1575 if (npath) {
1576 ext4_ext_drop_refs(npath);
1577 kfree(npath);
1578 }
1579 ext4_ext_tree_changed(inode);
1580 ext4_ext_invalidate_cache(inode);
1581 return err;
1582}
1583
Avantika Mathur09b88252006-12-06 20:41:36 -08001584static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001585ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Mingming Caodd545672007-07-31 00:37:46 -07001586 __u32 len, ext4_fsblk_t start, int type)
Alex Tomasa86c6182006-10-11 01:21:03 -07001587{
1588 struct ext4_ext_cache *cex;
1589 BUG_ON(len == 0);
1590 cex = &EXT4_I(inode)->i_cached_extent;
1591 cex->ec_type = type;
1592 cex->ec_block = block;
1593 cex->ec_len = len;
1594 cex->ec_start = start;
1595}
1596
1597/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001598 * ext4_ext_put_gap_in_cache:
1599 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001600 * and cache this gap
1601 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001602static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001603ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001604 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001605{
1606 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001607 unsigned long len;
1608 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001609 struct ext4_extent *ex;
1610
1611 ex = path[depth].p_ext;
1612 if (ex == NULL) {
1613 /* there is no extent yet, so gap is [0;-] */
1614 lblock = 0;
1615 len = EXT_MAX_BLOCK;
1616 ext_debug("cache gap(whole file):");
1617 } else if (block < le32_to_cpu(ex->ee_block)) {
1618 lblock = block;
1619 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05001620 ext_debug("cache gap(before): %u [%u:%u]",
1621 block,
1622 le32_to_cpu(ex->ee_block),
1623 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001624 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001625 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001626 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001627 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001628 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001629
1630 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05001631 ext_debug("cache gap(after): [%u:%u] %u",
1632 le32_to_cpu(ex->ee_block),
1633 ext4_ext_get_actual_len(ex),
1634 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001635 BUG_ON(next == lblock);
1636 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001637 } else {
1638 lblock = len = 0;
1639 BUG();
1640 }
1641
Eric Sandeenbba90742008-01-28 23:58:27 -05001642 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001643 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1644}
1645
Avantika Mathur09b88252006-12-06 20:41:36 -08001646static int
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001647ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
Alex Tomasa86c6182006-10-11 01:21:03 -07001648 struct ext4_extent *ex)
1649{
1650 struct ext4_ext_cache *cex;
1651
1652 cex = &EXT4_I(inode)->i_cached_extent;
1653
1654 /* has cache valid data? */
1655 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1656 return EXT4_EXT_CACHE_NO;
1657
1658 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1659 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1660 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001661 ex->ee_block = cpu_to_le32(cex->ec_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001662 ext4_ext_store_pblock(ex, cex->ec_start);
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001663 ex->ee_len = cpu_to_le16(cex->ec_len);
Eric Sandeenbba90742008-01-28 23:58:27 -05001664 ext_debug("%u cached by %u:%u:%llu\n",
1665 block,
1666 cex->ec_block, cex->ec_len, cex->ec_start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001667 return cex->ec_type;
1668 }
1669
1670 /* not in cache */
1671 return EXT4_EXT_CACHE_NO;
1672}
1673
1674/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001675 * ext4_ext_rm_idx:
1676 * removes index from the index block.
1677 * It's used in truncate case only, thus all requests are for
1678 * last index in the block only.
Alex Tomasa86c6182006-10-11 01:21:03 -07001679 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001680static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001681 struct ext4_ext_path *path)
1682{
1683 struct buffer_head *bh;
1684 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001685 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07001686
1687 /* free index block */
1688 path--;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001689 leaf = idx_pblock(path->p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -07001690 BUG_ON(path->p_hdr->eh_entries == 0);
Avantika Mathur7e028972006-12-06 20:41:33 -08001691 err = ext4_ext_get_access(handle, inode, path);
1692 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001693 return err;
1694 path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
Avantika Mathur7e028972006-12-06 20:41:33 -08001695 err = ext4_ext_dirty(handle, inode, path);
1696 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001697 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07001698 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Alex Tomasa86c6182006-10-11 01:21:03 -07001699 bh = sb_find_get_block(inode->i_sb, leaf);
1700 ext4_forget(handle, 1, inode, bh, leaf);
Alex Tomasc9de5602008-01-29 00:19:52 -05001701 ext4_free_blocks(handle, inode, leaf, 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001702 return err;
1703}
1704
1705/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001706 * ext4_ext_calc_credits_for_insert:
1707 * This routine returns max. credits that the extent tree can consume.
Alex Tomasa86c6182006-10-11 01:21:03 -07001708 * It should be OK for low-performance paths like ->writepage()
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001709 * To allow many writing processes to fit into a single transaction,
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001710 * the caller should calculate credits under i_data_sem and
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001711 * pass the actual path.
Alex Tomasa86c6182006-10-11 01:21:03 -07001712 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001713int ext4_ext_calc_credits_for_insert(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001714 struct ext4_ext_path *path)
1715{
1716 int depth, needed;
1717
1718 if (path) {
1719 /* probably there is space in leaf? */
1720 depth = ext_depth(inode);
1721 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1722 < le16_to_cpu(path[depth].p_hdr->eh_max))
1723 return 1;
1724 }
1725
1726 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001727 * given 32-bit logical block (4294967296 blocks), max. tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001728 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001729 * Let's also add one more level for imbalance.
Alex Tomasa86c6182006-10-11 01:21:03 -07001730 */
1731 depth = 5;
1732
1733 /* allocation of new data block(s) */
1734 needed = 2;
1735
1736 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001737 * tree can be full, so it would need to grow in depth:
Johann Lombardifeb18922006-12-06 20:40:46 -08001738 * we need one credit to modify old root, credits for
1739 * new root will be added in split accounting
Alex Tomasa86c6182006-10-11 01:21:03 -07001740 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001741 needed += 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001742
1743 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001744 * Index split can happen, we would need:
Alex Tomasa86c6182006-10-11 01:21:03 -07001745 * allocate intermediate indexes (bitmap + group)
1746 * + change two blocks at each level, but root (already included)
1747 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001748 needed += (depth * 2) + (depth * 2);
Alex Tomasa86c6182006-10-11 01:21:03 -07001749
1750 /* any allocation modifies superblock */
1751 needed += 1;
1752
1753 return needed;
1754}
1755
1756static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1757 struct ext4_extent *ex,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001758 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07001759{
1760 struct buffer_head *bh;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001761 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasc9de5602008-01-29 00:19:52 -05001762 int i, metadata = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001763
Alex Tomasc9de5602008-01-29 00:19:52 -05001764 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1765 metadata = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001766#ifdef EXTENTS_STATS
1767 {
1768 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07001769 spin_lock(&sbi->s_ext_stats_lock);
1770 sbi->s_ext_blocks += ee_len;
1771 sbi->s_ext_extents++;
1772 if (ee_len < sbi->s_ext_min)
1773 sbi->s_ext_min = ee_len;
1774 if (ee_len > sbi->s_ext_max)
1775 sbi->s_ext_max = ee_len;
1776 if (ext_depth(inode) > sbi->s_depth_max)
1777 sbi->s_depth_max = ext_depth(inode);
1778 spin_unlock(&sbi->s_ext_stats_lock);
1779 }
1780#endif
1781 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001782 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001783 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001784 ext4_lblk_t num;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001785 ext4_fsblk_t start;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001786
Amit Aroraa2df2a62007-07-17 21:42:41 -04001787 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1788 start = ext_pblock(ex) + ee_len - num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001789 ext_debug("free last %u blocks starting %llu\n", num, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001790 for (i = 0; i < num; i++) {
1791 bh = sb_find_get_block(inode->i_sb, start + i);
1792 ext4_forget(handle, 0, inode, bh, start + i);
1793 }
Alex Tomasc9de5602008-01-29 00:19:52 -05001794 ext4_free_blocks(handle, inode, start, num, metadata);
Alex Tomasa86c6182006-10-11 01:21:03 -07001795 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001796 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001797 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001798 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001799 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001800 printk(KERN_INFO "strange request: removal(2) "
1801 "%u-%u from %u:%u\n",
1802 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001803 }
1804 return 0;
1805}
1806
1807static int
1808ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001809 struct ext4_ext_path *path, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001810{
1811 int err = 0, correct_index = 0;
1812 int depth = ext_depth(inode), credits;
1813 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001814 ext4_lblk_t a, b, block;
1815 unsigned num;
1816 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07001817 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001818 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001819 struct ext4_extent *ex;
1820
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001821 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001822 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001823 if (!path[depth].p_hdr)
1824 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1825 eh = path[depth].p_hdr;
1826 BUG_ON(eh == NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001827
1828 /* find where to start removing */
1829 ex = EXT_LAST_EXTENT(eh);
1830
1831 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001832 if (ext4_ext_is_uninitialized(ex))
1833 uninitialized = 1;
1834 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001835
1836 while (ex >= EXT_FIRST_EXTENT(eh) &&
1837 ex_ee_block + ex_ee_len > start) {
1838 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1839 path[depth].p_ext = ex;
1840
1841 a = ex_ee_block > start ? ex_ee_block : start;
1842 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1843 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1844
1845 ext_debug(" border %u:%u\n", a, b);
1846
1847 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1848 block = 0;
1849 num = 0;
1850 BUG();
1851 } else if (a != ex_ee_block) {
1852 /* remove tail of the extent */
1853 block = ex_ee_block;
1854 num = a - block;
1855 } else if (b != ex_ee_block + ex_ee_len - 1) {
1856 /* remove head of the extent */
1857 block = a;
1858 num = b - a;
1859 /* there is no "make a hole" API yet */
1860 BUG();
1861 } else {
1862 /* remove whole extent: excellent! */
1863 block = ex_ee_block;
1864 num = 0;
1865 BUG_ON(a != ex_ee_block);
1866 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1867 }
1868
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001869 /* at present, extent can't cross block group: */
Alex Tomasa86c6182006-10-11 01:21:03 -07001870 /* leaf + bitmap + group desc + sb + inode */
1871 credits = 5;
1872 if (ex == EXT_FIRST_EXTENT(eh)) {
1873 correct_index = 1;
1874 credits += (ext_depth(inode)) + 1;
1875 }
1876#ifdef CONFIG_QUOTA
1877 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1878#endif
1879
1880 handle = ext4_ext_journal_restart(handle, credits);
1881 if (IS_ERR(handle)) {
1882 err = PTR_ERR(handle);
1883 goto out;
1884 }
1885
1886 err = ext4_ext_get_access(handle, inode, path + depth);
1887 if (err)
1888 goto out;
1889
1890 err = ext4_remove_blocks(handle, inode, ex, a, b);
1891 if (err)
1892 goto out;
1893
1894 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001895 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001896 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001897 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1898 }
1899
1900 ex->ee_block = cpu_to_le32(block);
1901 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04001902 /*
1903 * Do not mark uninitialized if all the blocks in the
1904 * extent have been removed.
1905 */
1906 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001907 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001908
1909 err = ext4_ext_dirty(handle, inode, path + depth);
1910 if (err)
1911 goto out;
1912
Mingming Cao2ae02102006-10-11 01:21:11 -07001913 ext_debug("new extent: %u:%u:%llu\n", block, num,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001914 ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001915 ex--;
1916 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001917 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001918 }
1919
1920 if (correct_index && eh->eh_entries)
1921 err = ext4_ext_correct_indexes(handle, inode, path);
1922
1923 /* if this leaf is free, then we should
1924 * remove it from index block above */
1925 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1926 err = ext4_ext_rm_idx(handle, inode, path + depth);
1927
1928out:
1929 return err;
1930}
1931
1932/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001933 * ext4_ext_more_to_rm:
1934 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07001935 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001936static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001937ext4_ext_more_to_rm(struct ext4_ext_path *path)
1938{
1939 BUG_ON(path->p_idx == NULL);
1940
1941 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1942 return 0;
1943
1944 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001945 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07001946 * so we have to consider current index for truncation
1947 */
1948 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1949 return 0;
1950 return 1;
1951}
1952
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001953static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001954{
1955 struct super_block *sb = inode->i_sb;
1956 int depth = ext_depth(inode);
1957 struct ext4_ext_path *path;
1958 handle_t *handle;
1959 int i = 0, err = 0;
1960
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001961 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001962
1963 /* probably first extent we're gonna free will be last in block */
1964 handle = ext4_journal_start(inode, depth + 1);
1965 if (IS_ERR(handle))
1966 return PTR_ERR(handle);
1967
1968 ext4_ext_invalidate_cache(inode);
1969
1970 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001971 * We start scanning from right side, freeing all the blocks
1972 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07001973 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -08001974 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001975 if (path == NULL) {
1976 ext4_journal_stop(handle);
1977 return -ENOMEM;
1978 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001979 path[0].p_hdr = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001980 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001981 err = -EIO;
1982 goto out;
1983 }
1984 path[0].p_depth = depth;
1985
1986 while (i >= 0 && err == 0) {
1987 if (i == depth) {
1988 /* this is leaf block */
1989 err = ext4_ext_rm_leaf(handle, inode, path, start);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001990 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07001991 brelse(path[i].p_bh);
1992 path[i].p_bh = NULL;
1993 i--;
1994 continue;
1995 }
1996
1997 /* this is index block */
1998 if (!path[i].p_hdr) {
1999 ext_debug("initialize header\n");
2000 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002001 }
2002
Alex Tomasa86c6182006-10-11 01:21:03 -07002003 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002004 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002005 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2006 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2007 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2008 path[i].p_hdr,
2009 le16_to_cpu(path[i].p_hdr->eh_entries));
2010 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002011 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002012 path[i].p_idx--;
2013 }
2014
2015 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2016 i, EXT_FIRST_INDEX(path[i].p_hdr),
2017 path[i].p_idx);
2018 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002019 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002020 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002021 ext_debug("move to level %d (block %llu)\n",
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002022 i + 1, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002023 memset(path + i + 1, 0, sizeof(*path));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002024 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2025 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002026 /* should we reset i_size? */
2027 err = -EIO;
2028 break;
2029 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002030 if (WARN_ON(i + 1 > depth)) {
2031 err = -EIO;
2032 break;
2033 }
2034 if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2035 depth - i - 1)) {
2036 err = -EIO;
2037 break;
2038 }
2039 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002040
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002041 /* save actual number of indexes since this
2042 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002043 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2044 i++;
2045 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002046 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002047 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002048 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002049 * handle must be already prepared by the
2050 * truncatei_leaf() */
2051 err = ext4_ext_rm_idx(handle, inode, path + i);
2052 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002053 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002054 brelse(path[i].p_bh);
2055 path[i].p_bh = NULL;
2056 i--;
2057 ext_debug("return to level %d\n", i);
2058 }
2059 }
2060
2061 /* TODO: flexible tree reduction should be here */
2062 if (path->p_hdr->eh_entries == 0) {
2063 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002064 * truncate to zero freed all the tree,
2065 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002066 */
2067 err = ext4_ext_get_access(handle, inode, path);
2068 if (err == 0) {
2069 ext_inode_hdr(inode)->eh_depth = 0;
2070 ext_inode_hdr(inode)->eh_max =
2071 cpu_to_le16(ext4_ext_space_root(inode));
2072 err = ext4_ext_dirty(handle, inode, path);
2073 }
2074 }
2075out:
2076 ext4_ext_tree_changed(inode);
2077 ext4_ext_drop_refs(path);
2078 kfree(path);
2079 ext4_journal_stop(handle);
2080
2081 return err;
2082}
2083
2084/*
2085 * called at mount time
2086 */
2087void ext4_ext_init(struct super_block *sb)
2088{
2089 /*
2090 * possible initialization would be here
2091 */
2092
2093 if (test_opt(sb, EXTENTS)) {
2094 printk("EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002095#ifdef AGGRESSIVE_TEST
2096 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07002097#endif
2098#ifdef CHECK_BINSEARCH
2099 printk(", check binsearch");
2100#endif
2101#ifdef EXTENTS_STATS
2102 printk(", stats");
2103#endif
2104 printk("\n");
2105#ifdef EXTENTS_STATS
2106 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2107 EXT4_SB(sb)->s_ext_min = 1 << 30;
2108 EXT4_SB(sb)->s_ext_max = 0;
2109#endif
2110 }
2111}
2112
2113/*
2114 * called at umount time
2115 */
2116void ext4_ext_release(struct super_block *sb)
2117{
2118 if (!test_opt(sb, EXTENTS))
2119 return;
2120
2121#ifdef EXTENTS_STATS
2122 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2123 struct ext4_sb_info *sbi = EXT4_SB(sb);
2124 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2125 sbi->s_ext_blocks, sbi->s_ext_extents,
2126 sbi->s_ext_blocks / sbi->s_ext_extents);
2127 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2128 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2129 }
2130#endif
2131}
2132
Amit Arora56055d32007-07-17 21:42:38 -04002133/*
2134 * This function is called by ext4_ext_get_blocks() if someone tries to write
2135 * to an uninitialized extent. It may result in splitting the uninitialized
2136 * extent into multiple extents (upto three - one initialized and two
2137 * uninitialized).
2138 * There are three possibilities:
2139 * a> There is no split required: Entire extent should be initialized
2140 * b> Splits in two extents: Write is happening at either end of the extent
2141 * c> Splits in three extents: Somone is writing in middle of the extent
2142 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002143static int ext4_ext_convert_to_initialized(handle_t *handle,
2144 struct inode *inode,
2145 struct ext4_ext_path *path,
2146 ext4_lblk_t iblock,
2147 unsigned long max_blocks)
Amit Arora56055d32007-07-17 21:42:38 -04002148{
2149 struct ext4_extent *ex, newex;
2150 struct ext4_extent *ex1 = NULL;
2151 struct ext4_extent *ex2 = NULL;
2152 struct ext4_extent *ex3 = NULL;
2153 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002154 ext4_lblk_t ee_block;
2155 unsigned int allocated, ee_len, depth;
Amit Arora56055d32007-07-17 21:42:38 -04002156 ext4_fsblk_t newblock;
2157 int err = 0;
2158 int ret = 0;
2159
2160 depth = ext_depth(inode);
2161 eh = path[depth].p_hdr;
2162 ex = path[depth].p_ext;
2163 ee_block = le32_to_cpu(ex->ee_block);
2164 ee_len = ext4_ext_get_actual_len(ex);
2165 allocated = ee_len - (iblock - ee_block);
2166 newblock = iblock - ee_block + ext_pblock(ex);
2167 ex2 = ex;
2168
2169 /* ex1: ee_block to iblock - 1 : uninitialized */
2170 if (iblock > ee_block) {
2171 ex1 = ex;
2172 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2173 ext4_ext_mark_uninitialized(ex1);
2174 ex2 = &newex;
2175 }
2176 /*
2177 * for sanity, update the length of the ex2 extent before
2178 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2179 * overlap of blocks.
2180 */
2181 if (!ex1 && allocated > max_blocks)
2182 ex2->ee_len = cpu_to_le16(max_blocks);
2183 /* ex3: to ee_block + ee_len : uninitialised */
2184 if (allocated > max_blocks) {
2185 unsigned int newdepth;
2186 ex3 = &newex;
2187 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2188 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2189 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2190 ext4_ext_mark_uninitialized(ex3);
2191 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2192 if (err)
2193 goto out;
2194 /*
2195 * The depth, and hence eh & ex might change
2196 * as part of the insert above.
2197 */
2198 newdepth = ext_depth(inode);
2199 if (newdepth != depth) {
2200 depth = newdepth;
2201 path = ext4_ext_find_extent(inode, iblock, NULL);
2202 if (IS_ERR(path)) {
2203 err = PTR_ERR(path);
2204 path = NULL;
2205 goto out;
2206 }
2207 eh = path[depth].p_hdr;
2208 ex = path[depth].p_ext;
2209 if (ex2 != &newex)
2210 ex2 = ex;
2211 }
2212 allocated = max_blocks;
2213 }
2214 /*
2215 * If there was a change of depth as part of the
2216 * insertion of ex3 above, we need to update the length
2217 * of the ex1 extent again here
2218 */
2219 if (ex1 && ex1 != ex) {
2220 ex1 = ex;
2221 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2222 ext4_ext_mark_uninitialized(ex1);
2223 ex2 = &newex;
2224 }
2225 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2226 ex2->ee_block = cpu_to_le32(iblock);
Amit Arora56055d32007-07-17 21:42:38 -04002227 ext4_ext_store_pblock(ex2, newblock);
2228 ex2->ee_len = cpu_to_le16(allocated);
2229 if (ex2 != ex)
2230 goto insert;
2231 err = ext4_ext_get_access(handle, inode, path + depth);
2232 if (err)
2233 goto out;
2234 /*
2235 * New (initialized) extent starts from the first block
2236 * in the current extent. i.e., ex2 == ex
2237 * We have to see if it can be merged with the extent
2238 * on the left.
2239 */
2240 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2241 /*
2242 * To merge left, pass "ex2 - 1" to try_to_merge(),
2243 * since it merges towards right _only_.
2244 */
2245 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2246 if (ret) {
2247 err = ext4_ext_correct_indexes(handle, inode, path);
2248 if (err)
2249 goto out;
2250 depth = ext_depth(inode);
2251 ex2--;
2252 }
2253 }
2254 /*
2255 * Try to Merge towards right. This might be required
2256 * only when the whole extent is being written to.
2257 * i.e. ex2 == ex and ex3 == NULL.
2258 */
2259 if (!ex3) {
2260 ret = ext4_ext_try_to_merge(inode, path, ex2);
2261 if (ret) {
2262 err = ext4_ext_correct_indexes(handle, inode, path);
2263 if (err)
2264 goto out;
2265 }
2266 }
2267 /* Mark modified extent as dirty */
2268 err = ext4_ext_dirty(handle, inode, path + depth);
2269 goto out;
2270insert:
2271 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2272out:
2273 return err ? err : allocated;
2274}
2275
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002276/*
2277 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002278 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2279 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002280 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002281int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002282 ext4_lblk_t iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002283 unsigned long max_blocks, struct buffer_head *bh_result,
2284 int create, int extend_disksize)
2285{
2286 struct ext4_ext_path *path = NULL;
Amit Arora56055d32007-07-17 21:42:38 -04002287 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002288 struct ext4_extent newex, *ex;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002289 ext4_fsblk_t goal, newblock;
Amit Arora56055d32007-07-17 21:42:38 -04002290 int err = 0, depth, ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002291 unsigned long allocated = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05002292 struct ext4_allocation_request ar;
Alex Tomasa86c6182006-10-11 01:21:03 -07002293
2294 __clear_bit(BH_New, &bh_result->b_state);
Eric Sandeenbba90742008-01-28 23:58:27 -05002295 ext_debug("blocks %u/%lu requested for inode %u\n",
2296 iblock, max_blocks, inode->i_ino);
Alex Tomasa86c6182006-10-11 01:21:03 -07002297
2298 /* check in cache */
Avantika Mathur7e028972006-12-06 20:41:33 -08002299 goal = ext4_ext_in_cache(inode, iblock, &newex);
2300 if (goal) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002301 if (goal == EXT4_EXT_CACHE_GAP) {
2302 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002303 /*
2304 * block isn't allocated yet and
2305 * user doesn't want to allocate it
2306 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002307 goto out2;
2308 }
2309 /* we should allocate requested block */
2310 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2311 /* block is already allocated */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002312 newblock = iblock
2313 - le32_to_cpu(newex.ee_block)
2314 + ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002315 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002316 allocated = le16_to_cpu(newex.ee_len) -
2317 (iblock - le32_to_cpu(newex.ee_block));
2318 goto out;
2319 } else {
2320 BUG();
2321 }
2322 }
2323
2324 /* find extent for this block */
2325 path = ext4_ext_find_extent(inode, iblock, NULL);
2326 if (IS_ERR(path)) {
2327 err = PTR_ERR(path);
2328 path = NULL;
2329 goto out2;
2330 }
2331
2332 depth = ext_depth(inode);
2333
2334 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002335 * consistent leaf must not be empty;
2336 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07002337 * this is why assert can't be put in ext4_ext_find_extent()
2338 */
2339 BUG_ON(path[depth].p_ext == NULL && depth != 0);
Amit Arora56055d32007-07-17 21:42:38 -04002340 eh = path[depth].p_hdr;
Alex Tomasa86c6182006-10-11 01:21:03 -07002341
Avantika Mathur7e028972006-12-06 20:41:33 -08002342 ex = path[depth].p_ext;
2343 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002344 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002345 ext4_fsblk_t ee_start = ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002346 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002347
2348 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002349 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04002350 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002351 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002352 ee_len = ext4_ext_get_actual_len(ex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002353 /* if found extent covers block, simply return it */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002354 if (iblock >= ee_block && iblock < ee_block + ee_len) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002355 newblock = iblock - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002356 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002357 allocated = ee_len - (iblock - ee_block);
Eric Sandeenbba90742008-01-28 23:58:27 -05002358 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002359 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04002360
Amit Aroraa2df2a62007-07-17 21:42:41 -04002361 /* Do not put uninitialized extent in the cache */
Amit Arora56055d32007-07-17 21:42:38 -04002362 if (!ext4_ext_is_uninitialized(ex)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002363 ext4_ext_put_in_cache(inode, ee_block,
2364 ee_len, ee_start,
2365 EXT4_EXT_CACHE_EXTENT);
Amit Arora56055d32007-07-17 21:42:38 -04002366 goto out;
2367 }
2368 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2369 goto out;
2370 if (!create)
2371 goto out2;
2372
2373 ret = ext4_ext_convert_to_initialized(handle, inode,
2374 path, iblock,
2375 max_blocks);
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002376 if (ret <= 0) {
2377 err = ret;
Amit Arora56055d32007-07-17 21:42:38 -04002378 goto out2;
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002379 } else
Amit Arora56055d32007-07-17 21:42:38 -04002380 allocated = ret;
2381 goto outnew;
Alex Tomasa86c6182006-10-11 01:21:03 -07002382 }
2383 }
2384
2385 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002386 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07002387 * we couldn't try to create block if create flag is zero
2388 */
2389 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002390 /*
2391 * put just found gap into cache to speed up
2392 * subsequent requests
2393 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002394 ext4_ext_put_gap_in_cache(inode, path, iblock);
2395 goto out2;
2396 }
2397 /*
Andrew Morton63f57932006-10-11 01:21:24 -07002398 * Okay, we need to do block allocation. Lazily initialize the block
2399 * allocation info here if necessary.
2400 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002401 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2402 ext4_init_block_alloc_info(inode);
2403
Alex Tomasc9de5602008-01-29 00:19:52 -05002404 /* find neighbour allocated blocks */
2405 ar.lleft = iblock;
2406 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2407 if (err)
2408 goto out2;
2409 ar.lright = iblock;
2410 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2411 if (err)
2412 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04002413
Amit Arora749269f2007-07-18 09:02:56 -04002414 /*
2415 * See if request is beyond maximum number of blocks we can have in
2416 * a single extent. For an initialized extent this limit is
2417 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2418 * EXT_UNINIT_MAX_LEN.
2419 */
2420 if (max_blocks > EXT_INIT_MAX_LEN &&
2421 create != EXT4_CREATE_UNINITIALIZED_EXT)
2422 max_blocks = EXT_INIT_MAX_LEN;
2423 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2424 create == EXT4_CREATE_UNINITIALIZED_EXT)
2425 max_blocks = EXT_UNINIT_MAX_LEN;
2426
Amit Arora25d14f92007-05-24 13:04:13 -04002427 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2428 newex.ee_block = cpu_to_le32(iblock);
2429 newex.ee_len = cpu_to_le16(max_blocks);
2430 err = ext4_ext_check_overlap(inode, &newex, path);
2431 if (err)
2432 allocated = le16_to_cpu(newex.ee_len);
2433 else
2434 allocated = max_blocks;
Alex Tomasc9de5602008-01-29 00:19:52 -05002435
2436 /* allocate new block */
2437 ar.inode = inode;
2438 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2439 ar.logical = iblock;
2440 ar.len = allocated;
2441 if (S_ISREG(inode->i_mode))
2442 ar.flags = EXT4_MB_HINT_DATA;
2443 else
2444 /* disable in-core preallocation for non-regular files */
2445 ar.flags = 0;
2446 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07002447 if (!newblock)
2448 goto out2;
Mingming Cao2ae02102006-10-11 01:21:11 -07002449 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002450 goal, newblock, allocated);
2451
2452 /* try to insert new extent into found leaf and return */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002453 ext4_ext_store_pblock(&newex, newblock);
Alex Tomasc9de5602008-01-29 00:19:52 -05002454 newex.ee_len = cpu_to_le16(ar.len);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002455 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2456 ext4_ext_mark_uninitialized(&newex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002457 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Alex Tomas315054f2007-05-24 13:04:25 -04002458 if (err) {
2459 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05002460 /* not a good idea to call discard here directly,
2461 * but otherwise we'd need to call it every free() */
2462 ext4_mb_discard_inode_preallocations(inode);
Alex Tomas315054f2007-05-24 13:04:25 -04002463 ext4_free_blocks(handle, inode, ext_pblock(&newex),
Alex Tomasc9de5602008-01-29 00:19:52 -05002464 le16_to_cpu(newex.ee_len), 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002465 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04002466 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002467
2468 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2469 EXT4_I(inode)->i_disksize = inode->i_size;
2470
2471 /* previous routine could use block we allocated */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002472 newblock = ext_pblock(&newex);
Alex Tomasc9de5602008-01-29 00:19:52 -05002473 allocated = le16_to_cpu(newex.ee_len);
Amit Arora56055d32007-07-17 21:42:38 -04002474outnew:
Alex Tomasa86c6182006-10-11 01:21:03 -07002475 __set_bit(BH_New, &bh_result->b_state);
2476
Amit Aroraa2df2a62007-07-17 21:42:41 -04002477 /* Cache only when it is _not_ an uninitialized extent */
2478 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2479 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2480 EXT4_EXT_CACHE_EXTENT);
Alex Tomasa86c6182006-10-11 01:21:03 -07002481out:
2482 if (allocated > max_blocks)
2483 allocated = max_blocks;
2484 ext4_ext_show_leaf(inode, path);
2485 __set_bit(BH_Mapped, &bh_result->b_state);
2486 bh_result->b_bdev = inode->i_sb->s_bdev;
2487 bh_result->b_blocknr = newblock;
2488out2:
2489 if (path) {
2490 ext4_ext_drop_refs(path);
2491 kfree(path);
2492 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002493 return err ? err : allocated;
2494}
2495
2496void ext4_ext_truncate(struct inode * inode, struct page *page)
2497{
2498 struct address_space *mapping = inode->i_mapping;
2499 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002500 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002501 handle_t *handle;
2502 int err = 0;
2503
2504 /*
2505 * probably first extent we're gonna free will be last in block
2506 */
2507 err = ext4_writepage_trans_blocks(inode) + 3;
2508 handle = ext4_journal_start(inode, err);
2509 if (IS_ERR(handle)) {
2510 if (page) {
2511 clear_highpage(page);
2512 flush_dcache_page(page);
2513 unlock_page(page);
2514 page_cache_release(page);
2515 }
2516 return;
2517 }
2518
2519 if (page)
2520 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2521
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002522 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07002523 ext4_ext_invalidate_cache(inode);
2524
Alex Tomasc9de5602008-01-29 00:19:52 -05002525 ext4_mb_discard_inode_preallocations(inode);
2526
Alex Tomasa86c6182006-10-11 01:21:03 -07002527 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002528 * TODO: optimization is possible here.
2529 * Probably we need not scan at all,
2530 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07002531 */
2532 if (ext4_orphan_add(handle, inode))
2533 goto out_stop;
2534
2535 /* we have to know where to truncate from in crash case */
2536 EXT4_I(inode)->i_disksize = inode->i_size;
2537 ext4_mark_inode_dirty(handle, inode);
2538
2539 last_block = (inode->i_size + sb->s_blocksize - 1)
2540 >> EXT4_BLOCK_SIZE_BITS(sb);
2541 err = ext4_ext_remove_space(inode, last_block);
2542
2543 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04002544 * transaction synchronous.
2545 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002546 if (IS_SYNC(inode))
2547 handle->h_sync = 1;
2548
2549out_stop:
2550 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002551 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07002552 * then we need to clear up the orphan record which we created above.
2553 * However, if this was a real unlink then we were called by
2554 * ext4_delete_inode(), and we allow that function to clean up the
2555 * orphan info for us.
2556 */
2557 if (inode->i_nlink)
2558 ext4_orphan_del(handle, inode);
2559
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002560 up_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07002561 ext4_journal_stop(handle);
2562}
2563
2564/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002565 * ext4_ext_writepage_trans_blocks:
2566 * calculate max number of blocks we could modify
Alex Tomasa86c6182006-10-11 01:21:03 -07002567 * in order to allocate new block for an inode
2568 */
2569int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2570{
2571 int needed;
2572
2573 needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2574
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002575 /* caller wants to allocate num blocks, but note it includes sb */
Alex Tomasa86c6182006-10-11 01:21:03 -07002576 needed = needed * num - (num - 1);
2577
2578#ifdef CONFIG_QUOTA
2579 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2580#endif
2581
2582 return needed;
2583}
Amit Aroraa2df2a62007-07-17 21:42:41 -04002584
2585/*
2586 * preallocate space for a file. This implements ext4's fallocate inode
2587 * operation, which gets called from sys_fallocate system call.
2588 * For block-mapped files, posix_fallocate should fall back to the method
2589 * of writing zeroes to the required new blocks (the same behavior which is
2590 * expected for file systems which do not support fallocate() system call).
2591 */
2592long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2593{
2594 handle_t *handle;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002595 ext4_lblk_t block;
2596 unsigned long max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002597 ext4_fsblk_t nblocks = 0;
2598 int ret = 0;
2599 int ret2 = 0;
2600 int retries = 0;
2601 struct buffer_head map_bh;
2602 unsigned int credits, blkbits = inode->i_blkbits;
2603
2604 /*
2605 * currently supporting (pre)allocate mode for extent-based
2606 * files _only_
2607 */
2608 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2609 return -EOPNOTSUPP;
2610
2611 /* preallocation to directories is currently not supported */
2612 if (S_ISDIR(inode->i_mode))
2613 return -ENODEV;
2614
2615 block = offset >> blkbits;
2616 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2617 - block;
2618
2619 /*
2620 * credits to insert 1 extent into extent tree + buffers to be able to
2621 * modify 1 super block, 1 block bitmap and 1 group descriptor.
2622 */
2623 credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002624 down_write((&EXT4_I(inode)->i_data_sem));
Amit Aroraa2df2a62007-07-17 21:42:41 -04002625retry:
2626 while (ret >= 0 && ret < max_blocks) {
2627 block = block + ret;
2628 max_blocks = max_blocks - ret;
2629 handle = ext4_journal_start(inode, credits);
2630 if (IS_ERR(handle)) {
2631 ret = PTR_ERR(handle);
2632 break;
2633 }
2634
2635 ret = ext4_ext_get_blocks(handle, inode, block,
2636 max_blocks, &map_bh,
2637 EXT4_CREATE_UNINITIALIZED_EXT, 0);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05002638 WARN_ON(ret <= 0);
2639 if (ret <= 0) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002640 ext4_error(inode->i_sb, "ext4_fallocate",
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05002641 "ext4_ext_get_blocks returned error: "
2642 "inode#%lu, block=%u, max_blocks=%lu",
2643 inode->i_ino, block, max_blocks);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002644 ret = -EIO;
2645 ext4_mark_inode_dirty(handle, inode);
2646 ret2 = ext4_journal_stop(handle);
2647 break;
2648 }
2649 if (ret > 0) {
2650 /* check wrap through sign-bit/zero here */
2651 if ((block + ret) < 0 || (block + ret) < block) {
2652 ret = -EIO;
2653 ext4_mark_inode_dirty(handle, inode);
2654 ret2 = ext4_journal_stop(handle);
2655 break;
2656 }
2657 if (buffer_new(&map_bh) && ((block + ret) >
2658 (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits)
2659 >> blkbits)))
2660 nblocks = nblocks + ret;
2661 }
2662
2663 /* Update ctime if new blocks get allocated */
2664 if (nblocks) {
2665 struct timespec now;
2666
2667 now = current_fs_time(inode->i_sb);
2668 if (!timespec_equal(&inode->i_ctime, &now))
2669 inode->i_ctime = now;
2670 }
2671
2672 ext4_mark_inode_dirty(handle, inode);
2673 ret2 = ext4_journal_stop(handle);
2674 if (ret2)
2675 break;
2676 }
2677
2678 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2679 goto retry;
2680
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002681 up_write((&EXT4_I(inode)->i_data_sem));
Amit Aroraa2df2a62007-07-17 21:42:41 -04002682 /*
2683 * Time to update the file size.
2684 * Update only when preallocation was requested beyond the file size.
2685 */
2686 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2687 (offset + len) > i_size_read(inode)) {
2688 if (ret > 0) {
2689 /*
2690 * if no error, we assume preallocation succeeded
2691 * completely
2692 */
2693 mutex_lock(&inode->i_mutex);
2694 i_size_write(inode, offset + len);
2695 EXT4_I(inode)->i_disksize = i_size_read(inode);
2696 mutex_unlock(&inode->i_mutex);
2697 } else if (ret < 0 && nblocks) {
2698 /* Handle partial allocation scenario */
2699 loff_t newsize;
2700
2701 mutex_lock(&inode->i_mutex);
2702 newsize = (nblocks << blkbits) + i_size_read(inode);
2703 i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits));
2704 EXT4_I(inode)->i_disksize = i_size_read(inode);
2705 mutex_unlock(&inode->i_mutex);
2706 }
2707 }
2708
2709 return ret > 0 ? ret2 : ret;
2710}