blob: bc7081f1fbe80dd5add381c5746b1bbc74e4883a [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;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001032 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001033
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;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001046 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001047 if (*logical < le32_to_cpu(ex->ee_block)) {
1048 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1049 while (--depth >= 0) {
1050 ix = path[depth].p_idx;
1051 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1052 }
1053 return 0;
1054 }
1055
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001056 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
Alex Tomas1988b512008-01-28 23:58:27 -05001057
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001058 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1059 *phys = ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001060 return 0;
1061}
1062
1063/*
1064 * search the closest allocated block to the right for *logical
1065 * and returns it at @logical + it's physical address at @phys
1066 * if *logical is the smallest allocated block, the function
1067 * returns 0 at @phys
1068 * return value contains 0 (success) or error code
1069 */
1070int
1071ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1072 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1073{
1074 struct buffer_head *bh = NULL;
1075 struct ext4_extent_header *eh;
1076 struct ext4_extent_idx *ix;
1077 struct ext4_extent *ex;
1078 ext4_fsblk_t block;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001079 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001080
1081 BUG_ON(path == NULL);
1082 depth = path->p_depth;
1083 *phys = 0;
1084
1085 if (depth == 0 && path->p_ext == NULL)
1086 return 0;
1087
1088 /* usually extent in the path covers blocks smaller
1089 * then *logical, but it can be that extent is the
1090 * first one in the file */
1091
1092 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001093 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001094 if (*logical < le32_to_cpu(ex->ee_block)) {
1095 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1096 while (--depth >= 0) {
1097 ix = path[depth].p_idx;
1098 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1099 }
1100 *logical = le32_to_cpu(ex->ee_block);
1101 *phys = ext_pblock(ex);
1102 return 0;
1103 }
1104
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001105 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
Alex Tomas1988b512008-01-28 23:58:27 -05001106
1107 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1108 /* next allocated block in this leaf */
1109 ex++;
1110 *logical = le32_to_cpu(ex->ee_block);
1111 *phys = ext_pblock(ex);
1112 return 0;
1113 }
1114
1115 /* go up and search for index to the right */
1116 while (--depth >= 0) {
1117 ix = path[depth].p_idx;
1118 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1119 break;
1120 }
1121
1122 if (depth < 0) {
1123 /* we've gone up to the root and
1124 * found no index to the right */
1125 return 0;
1126 }
1127
1128 /* we've found index to the right, let's
1129 * follow it and find the closest allocated
1130 * block to the right */
1131 ix++;
1132 block = idx_pblock(ix);
1133 while (++depth < path->p_depth) {
1134 bh = sb_bread(inode->i_sb, block);
1135 if (bh == NULL)
1136 return -EIO;
1137 eh = ext_block_hdr(bh);
1138 if (ext4_ext_check_header(inode, eh, depth)) {
1139 put_bh(bh);
1140 return -EIO;
1141 }
1142 ix = EXT_FIRST_INDEX(eh);
1143 block = idx_pblock(ix);
1144 put_bh(bh);
1145 }
1146
1147 bh = sb_bread(inode->i_sb, block);
1148 if (bh == NULL)
1149 return -EIO;
1150 eh = ext_block_hdr(bh);
1151 if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1152 put_bh(bh);
1153 return -EIO;
1154 }
1155 ex = EXT_FIRST_EXTENT(eh);
1156 *logical = le32_to_cpu(ex->ee_block);
1157 *phys = ext_pblock(ex);
1158 put_bh(bh);
1159 return 0;
1160
1161}
1162
1163/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001164 * ext4_ext_next_allocated_block:
1165 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1166 * NOTE: it considers block number from index entry as
1167 * allocated block. Thus, index entries have to be consistent
1168 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001169 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001170static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001171ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1172{
1173 int depth;
1174
1175 BUG_ON(path == NULL);
1176 depth = path->p_depth;
1177
1178 if (depth == 0 && path->p_ext == NULL)
1179 return EXT_MAX_BLOCK;
1180
1181 while (depth >= 0) {
1182 if (depth == path->p_depth) {
1183 /* leaf */
1184 if (path[depth].p_ext !=
1185 EXT_LAST_EXTENT(path[depth].p_hdr))
1186 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1187 } else {
1188 /* index */
1189 if (path[depth].p_idx !=
1190 EXT_LAST_INDEX(path[depth].p_hdr))
1191 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1192 }
1193 depth--;
1194 }
1195
1196 return EXT_MAX_BLOCK;
1197}
1198
1199/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001200 * ext4_ext_next_leaf_block:
Alex Tomasa86c6182006-10-11 01:21:03 -07001201 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1202 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001203static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
Andrew Morton63f57932006-10-11 01:21:24 -07001204 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001205{
1206 int depth;
1207
1208 BUG_ON(path == NULL);
1209 depth = path->p_depth;
1210
1211 /* zero-tree has no leaf blocks at all */
1212 if (depth == 0)
1213 return EXT_MAX_BLOCK;
1214
1215 /* go to index block */
1216 depth--;
1217
1218 while (depth >= 0) {
1219 if (path[depth].p_idx !=
1220 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001221 return (ext4_lblk_t)
1222 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001223 depth--;
1224 }
1225
1226 return EXT_MAX_BLOCK;
1227}
1228
1229/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001230 * ext4_ext_correct_indexes:
1231 * if leaf gets modified and modified extent is first in the leaf,
1232 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001233 * TODO: do we need to correct tree in all cases?
1234 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001235static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001236 struct ext4_ext_path *path)
1237{
1238 struct ext4_extent_header *eh;
1239 int depth = ext_depth(inode);
1240 struct ext4_extent *ex;
1241 __le32 border;
1242 int k, err = 0;
1243
1244 eh = path[depth].p_hdr;
1245 ex = path[depth].p_ext;
1246 BUG_ON(ex == NULL);
1247 BUG_ON(eh == NULL);
1248
1249 if (depth == 0) {
1250 /* there is no tree at all */
1251 return 0;
1252 }
1253
1254 if (ex != EXT_FIRST_EXTENT(eh)) {
1255 /* we correct tree if first leaf got modified only */
1256 return 0;
1257 }
1258
1259 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001260 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001261 */
1262 k = depth - 1;
1263 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001264 err = ext4_ext_get_access(handle, inode, path + k);
1265 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001266 return err;
1267 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001268 err = ext4_ext_dirty(handle, inode, path + k);
1269 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001270 return err;
1271
1272 while (k--) {
1273 /* change all left-side indexes */
1274 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1275 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001276 err = ext4_ext_get_access(handle, inode, path + k);
1277 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001278 break;
1279 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001280 err = ext4_ext_dirty(handle, inode, path + k);
1281 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001282 break;
1283 }
1284
1285 return err;
1286}
1287
Avantika Mathur09b88252006-12-06 20:41:36 -08001288static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001289ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1290 struct ext4_extent *ex2)
1291{
Amit Arora749269f2007-07-18 09:02:56 -04001292 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001293
1294 /*
1295 * Make sure that either both extents are uninitialized, or
1296 * both are _not_.
1297 */
1298 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1299 return 0;
1300
Amit Arora749269f2007-07-18 09:02:56 -04001301 if (ext4_ext_is_uninitialized(ex1))
1302 max_len = EXT_UNINIT_MAX_LEN;
1303 else
1304 max_len = EXT_INIT_MAX_LEN;
1305
Amit Aroraa2df2a62007-07-17 21:42:41 -04001306 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1307 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1308
1309 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001310 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001311 return 0;
1312
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001313 /*
1314 * To allow future support for preallocated extents to be added
1315 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001316 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001317 */
Amit Arora749269f2007-07-18 09:02:56 -04001318 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001319 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001320#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001321 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001322 return 0;
1323#endif
1324
Amit Aroraa2df2a62007-07-17 21:42:41 -04001325 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001326 return 1;
1327 return 0;
1328}
1329
1330/*
Amit Arora56055d32007-07-17 21:42:38 -04001331 * This function tries to merge the "ex" extent to the next extent in the tree.
1332 * It always tries to merge towards right. If you want to merge towards
1333 * left, pass "ex - 1" as argument instead of "ex".
1334 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1335 * 1 if they got merged.
1336 */
1337int ext4_ext_try_to_merge(struct inode *inode,
1338 struct ext4_ext_path *path,
1339 struct ext4_extent *ex)
1340{
1341 struct ext4_extent_header *eh;
1342 unsigned int depth, len;
1343 int merge_done = 0;
1344 int uninitialized = 0;
1345
1346 depth = ext_depth(inode);
1347 BUG_ON(path[depth].p_hdr == NULL);
1348 eh = path[depth].p_hdr;
1349
1350 while (ex < EXT_LAST_EXTENT(eh)) {
1351 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1352 break;
1353 /* merge with next extent! */
1354 if (ext4_ext_is_uninitialized(ex))
1355 uninitialized = 1;
1356 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1357 + ext4_ext_get_actual_len(ex + 1));
1358 if (uninitialized)
1359 ext4_ext_mark_uninitialized(ex);
1360
1361 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1362 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1363 * sizeof(struct ext4_extent);
1364 memmove(ex + 1, ex + 2, len);
1365 }
1366 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
1367 merge_done = 1;
1368 WARN_ON(eh->eh_entries == 0);
1369 if (!eh->eh_entries)
1370 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1371 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1372 }
1373
1374 return merge_done;
1375}
1376
1377/*
Amit Arora25d14f92007-05-24 13:04:13 -04001378 * check if a portion of the "newext" extent overlaps with an
1379 * existing extent.
1380 *
1381 * If there is an overlap discovered, it updates the length of the newext
1382 * such that there will be no overlap, and then returns 1.
1383 * If there is no overlap found, it returns 0.
1384 */
1385unsigned int ext4_ext_check_overlap(struct inode *inode,
1386 struct ext4_extent *newext,
1387 struct ext4_ext_path *path)
1388{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001389 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001390 unsigned int depth, len1;
1391 unsigned int ret = 0;
1392
1393 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001394 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001395 depth = ext_depth(inode);
1396 if (!path[depth].p_ext)
1397 goto out;
1398 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1399
1400 /*
1401 * get the next allocated block if the extent in the path
1402 * is before the requested block(s)
1403 */
1404 if (b2 < b1) {
1405 b2 = ext4_ext_next_allocated_block(path);
1406 if (b2 == EXT_MAX_BLOCK)
1407 goto out;
1408 }
1409
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001410 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001411 if (b1 + len1 < b1) {
1412 len1 = EXT_MAX_BLOCK - b1;
1413 newext->ee_len = cpu_to_le16(len1);
1414 ret = 1;
1415 }
1416
1417 /* check for overlap */
1418 if (b1 + len1 > b2) {
1419 newext->ee_len = cpu_to_le16(b2 - b1);
1420 ret = 1;
1421 }
1422out:
1423 return ret;
1424}
1425
1426/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001427 * ext4_ext_insert_extent:
1428 * tries to merge requsted extent into the existing extent or
1429 * inserts requested extent as new one into the tree,
1430 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001431 */
1432int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1433 struct ext4_ext_path *path,
1434 struct ext4_extent *newext)
1435{
1436 struct ext4_extent_header * eh;
1437 struct ext4_extent *ex, *fex;
1438 struct ext4_extent *nearex; /* nearest extent */
1439 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001440 int depth, len, err;
1441 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001442 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001443
Amit Aroraa2df2a62007-07-17 21:42:41 -04001444 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001445 depth = ext_depth(inode);
1446 ex = path[depth].p_ext;
1447 BUG_ON(path[depth].p_hdr == NULL);
1448
1449 /* try to insert block into found extent and return */
1450 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming Cao2ae02102006-10-11 01:21:11 -07001451 ext_debug("append %d block to %d:%d (from %llu)\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001452 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001453 le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001454 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001455 err = ext4_ext_get_access(handle, inode, path + depth);
1456 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001457 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001458
1459 /*
1460 * ext4_can_extents_be_merged should have checked that either
1461 * both extents are uninitialized, or both aren't. Thus we
1462 * need to check only one of them here.
1463 */
1464 if (ext4_ext_is_uninitialized(ex))
1465 uninitialized = 1;
1466 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1467 + ext4_ext_get_actual_len(newext));
1468 if (uninitialized)
1469 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001470 eh = path[depth].p_hdr;
1471 nearex = ex;
1472 goto merge;
1473 }
1474
1475repeat:
1476 depth = ext_depth(inode);
1477 eh = path[depth].p_hdr;
1478 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1479 goto has_space;
1480
1481 /* probably next leaf has space for us? */
1482 fex = EXT_LAST_EXTENT(eh);
1483 next = ext4_ext_next_leaf_block(inode, path);
1484 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1485 && next != EXT_MAX_BLOCK) {
1486 ext_debug("next leaf block - %d\n", next);
1487 BUG_ON(npath != NULL);
1488 npath = ext4_ext_find_extent(inode, next, NULL);
1489 if (IS_ERR(npath))
1490 return PTR_ERR(npath);
1491 BUG_ON(npath->p_depth != path->p_depth);
1492 eh = npath[depth].p_hdr;
1493 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1494 ext_debug("next leaf isnt full(%d)\n",
1495 le16_to_cpu(eh->eh_entries));
1496 path = npath;
1497 goto repeat;
1498 }
1499 ext_debug("next leaf has no free space(%d,%d)\n",
1500 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1501 }
1502
1503 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001504 * There is no free space in the found leaf.
1505 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001506 */
1507 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1508 if (err)
1509 goto cleanup;
1510 depth = ext_depth(inode);
1511 eh = path[depth].p_hdr;
1512
1513has_space:
1514 nearex = path[depth].p_ext;
1515
Avantika Mathur7e028972006-12-06 20:41:33 -08001516 err = ext4_ext_get_access(handle, inode, path + depth);
1517 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001518 goto cleanup;
1519
1520 if (!nearex) {
1521 /* there is no extent in this leaf, create first one */
Mingming Cao2ae02102006-10-11 01:21:11 -07001522 ext_debug("first extent in the leaf: %d:%llu:%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001523 le32_to_cpu(newext->ee_block),
1524 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001525 ext4_ext_get_actual_len(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001526 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1527 } else if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001528 > le32_to_cpu(nearex->ee_block)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001529/* BUG_ON(newext->ee_block == nearex->ee_block); */
1530 if (nearex != EXT_LAST_EXTENT(eh)) {
1531 len = EXT_MAX_EXTENT(eh) - nearex;
1532 len = (len - 1) * sizeof(struct ext4_extent);
1533 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001534 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001535 "move %d from 0x%p to 0x%p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001536 le32_to_cpu(newext->ee_block),
1537 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001538 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001539 nearex, len, nearex + 1, nearex + 2);
1540 memmove(nearex + 2, nearex + 1, len);
1541 }
1542 path[depth].p_ext = nearex + 1;
1543 } else {
1544 BUG_ON(newext->ee_block == nearex->ee_block);
1545 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1546 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001547 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001548 "move %d from 0x%p to 0x%p\n",
1549 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001550 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001551 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001552 nearex, len, nearex + 1, nearex + 2);
1553 memmove(nearex + 1, nearex, len);
1554 path[depth].p_ext = nearex;
1555 }
1556
1557 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1558 nearex = path[depth].p_ext;
1559 nearex->ee_block = newext->ee_block;
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -04001560 ext4_ext_store_pblock(nearex, ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001561 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001562
1563merge:
1564 /* try to merge extents to the right */
Amit Arora56055d32007-07-17 21:42:38 -04001565 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001566
1567 /* try to merge extents to the left */
1568
1569 /* time to correct all indexes above */
1570 err = ext4_ext_correct_indexes(handle, inode, path);
1571 if (err)
1572 goto cleanup;
1573
1574 err = ext4_ext_dirty(handle, inode, path + depth);
1575
1576cleanup:
1577 if (npath) {
1578 ext4_ext_drop_refs(npath);
1579 kfree(npath);
1580 }
1581 ext4_ext_tree_changed(inode);
1582 ext4_ext_invalidate_cache(inode);
1583 return err;
1584}
1585
Avantika Mathur09b88252006-12-06 20:41:36 -08001586static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001587ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Mingming Caodd545672007-07-31 00:37:46 -07001588 __u32 len, ext4_fsblk_t start, int type)
Alex Tomasa86c6182006-10-11 01:21:03 -07001589{
1590 struct ext4_ext_cache *cex;
1591 BUG_ON(len == 0);
1592 cex = &EXT4_I(inode)->i_cached_extent;
1593 cex->ec_type = type;
1594 cex->ec_block = block;
1595 cex->ec_len = len;
1596 cex->ec_start = start;
1597}
1598
1599/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001600 * ext4_ext_put_gap_in_cache:
1601 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001602 * and cache this gap
1603 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001604static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001605ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001606 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001607{
1608 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001609 unsigned long len;
1610 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001611 struct ext4_extent *ex;
1612
1613 ex = path[depth].p_ext;
1614 if (ex == NULL) {
1615 /* there is no extent yet, so gap is [0;-] */
1616 lblock = 0;
1617 len = EXT_MAX_BLOCK;
1618 ext_debug("cache gap(whole file):");
1619 } else if (block < le32_to_cpu(ex->ee_block)) {
1620 lblock = block;
1621 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05001622 ext_debug("cache gap(before): %u [%u:%u]",
1623 block,
1624 le32_to_cpu(ex->ee_block),
1625 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001626 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001627 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001628 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001629 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001630 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001631
1632 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05001633 ext_debug("cache gap(after): [%u:%u] %u",
1634 le32_to_cpu(ex->ee_block),
1635 ext4_ext_get_actual_len(ex),
1636 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001637 BUG_ON(next == lblock);
1638 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001639 } else {
1640 lblock = len = 0;
1641 BUG();
1642 }
1643
Eric Sandeenbba90742008-01-28 23:58:27 -05001644 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001645 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1646}
1647
Avantika Mathur09b88252006-12-06 20:41:36 -08001648static int
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001649ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
Alex Tomasa86c6182006-10-11 01:21:03 -07001650 struct ext4_extent *ex)
1651{
1652 struct ext4_ext_cache *cex;
1653
1654 cex = &EXT4_I(inode)->i_cached_extent;
1655
1656 /* has cache valid data? */
1657 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1658 return EXT4_EXT_CACHE_NO;
1659
1660 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1661 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1662 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001663 ex->ee_block = cpu_to_le32(cex->ec_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001664 ext4_ext_store_pblock(ex, cex->ec_start);
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001665 ex->ee_len = cpu_to_le16(cex->ec_len);
Eric Sandeenbba90742008-01-28 23:58:27 -05001666 ext_debug("%u cached by %u:%u:%llu\n",
1667 block,
1668 cex->ec_block, cex->ec_len, cex->ec_start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001669 return cex->ec_type;
1670 }
1671
1672 /* not in cache */
1673 return EXT4_EXT_CACHE_NO;
1674}
1675
1676/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001677 * ext4_ext_rm_idx:
1678 * removes index from the index block.
1679 * It's used in truncate case only, thus all requests are for
1680 * last index in the block only.
Alex Tomasa86c6182006-10-11 01:21:03 -07001681 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001682static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001683 struct ext4_ext_path *path)
1684{
1685 struct buffer_head *bh;
1686 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001687 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07001688
1689 /* free index block */
1690 path--;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001691 leaf = idx_pblock(path->p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -07001692 BUG_ON(path->p_hdr->eh_entries == 0);
Avantika Mathur7e028972006-12-06 20:41:33 -08001693 err = ext4_ext_get_access(handle, inode, path);
1694 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001695 return err;
1696 path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
Avantika Mathur7e028972006-12-06 20:41:33 -08001697 err = ext4_ext_dirty(handle, inode, path);
1698 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001699 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07001700 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Alex Tomasa86c6182006-10-11 01:21:03 -07001701 bh = sb_find_get_block(inode->i_sb, leaf);
1702 ext4_forget(handle, 1, inode, bh, leaf);
Alex Tomasc9de5602008-01-29 00:19:52 -05001703 ext4_free_blocks(handle, inode, leaf, 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001704 return err;
1705}
1706
1707/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001708 * ext4_ext_calc_credits_for_insert:
1709 * This routine returns max. credits that the extent tree can consume.
Alex Tomasa86c6182006-10-11 01:21:03 -07001710 * It should be OK for low-performance paths like ->writepage()
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001711 * To allow many writing processes to fit into a single transaction,
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001712 * the caller should calculate credits under i_data_sem and
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001713 * pass the actual path.
Alex Tomasa86c6182006-10-11 01:21:03 -07001714 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001715int ext4_ext_calc_credits_for_insert(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001716 struct ext4_ext_path *path)
1717{
1718 int depth, needed;
1719
1720 if (path) {
1721 /* probably there is space in leaf? */
1722 depth = ext_depth(inode);
1723 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1724 < le16_to_cpu(path[depth].p_hdr->eh_max))
1725 return 1;
1726 }
1727
1728 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001729 * given 32-bit logical block (4294967296 blocks), max. tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001730 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001731 * Let's also add one more level for imbalance.
Alex Tomasa86c6182006-10-11 01:21:03 -07001732 */
1733 depth = 5;
1734
1735 /* allocation of new data block(s) */
1736 needed = 2;
1737
1738 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001739 * tree can be full, so it would need to grow in depth:
Johann Lombardifeb18922006-12-06 20:40:46 -08001740 * we need one credit to modify old root, credits for
1741 * new root will be added in split accounting
Alex Tomasa86c6182006-10-11 01:21:03 -07001742 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001743 needed += 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001744
1745 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001746 * Index split can happen, we would need:
Alex Tomasa86c6182006-10-11 01:21:03 -07001747 * allocate intermediate indexes (bitmap + group)
1748 * + change two blocks at each level, but root (already included)
1749 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001750 needed += (depth * 2) + (depth * 2);
Alex Tomasa86c6182006-10-11 01:21:03 -07001751
1752 /* any allocation modifies superblock */
1753 needed += 1;
1754
1755 return needed;
1756}
1757
1758static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1759 struct ext4_extent *ex,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001760 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07001761{
1762 struct buffer_head *bh;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001763 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasc9de5602008-01-29 00:19:52 -05001764 int i, metadata = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001765
Alex Tomasc9de5602008-01-29 00:19:52 -05001766 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1767 metadata = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001768#ifdef EXTENTS_STATS
1769 {
1770 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07001771 spin_lock(&sbi->s_ext_stats_lock);
1772 sbi->s_ext_blocks += ee_len;
1773 sbi->s_ext_extents++;
1774 if (ee_len < sbi->s_ext_min)
1775 sbi->s_ext_min = ee_len;
1776 if (ee_len > sbi->s_ext_max)
1777 sbi->s_ext_max = ee_len;
1778 if (ext_depth(inode) > sbi->s_depth_max)
1779 sbi->s_depth_max = ext_depth(inode);
1780 spin_unlock(&sbi->s_ext_stats_lock);
1781 }
1782#endif
1783 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001784 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001785 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001786 ext4_lblk_t num;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001787 ext4_fsblk_t start;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001788
Amit Aroraa2df2a62007-07-17 21:42:41 -04001789 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1790 start = ext_pblock(ex) + ee_len - num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001791 ext_debug("free last %u blocks starting %llu\n", num, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001792 for (i = 0; i < num; i++) {
1793 bh = sb_find_get_block(inode->i_sb, start + i);
1794 ext4_forget(handle, 0, inode, bh, start + i);
1795 }
Alex Tomasc9de5602008-01-29 00:19:52 -05001796 ext4_free_blocks(handle, inode, start, num, metadata);
Alex Tomasa86c6182006-10-11 01:21:03 -07001797 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001798 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001799 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001800 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001801 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001802 printk(KERN_INFO "strange request: removal(2) "
1803 "%u-%u from %u:%u\n",
1804 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001805 }
1806 return 0;
1807}
1808
1809static int
1810ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001811 struct ext4_ext_path *path, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001812{
1813 int err = 0, correct_index = 0;
1814 int depth = ext_depth(inode), credits;
1815 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001816 ext4_lblk_t a, b, block;
1817 unsigned num;
1818 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07001819 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001820 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001821 struct ext4_extent *ex;
1822
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001823 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001824 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001825 if (!path[depth].p_hdr)
1826 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1827 eh = path[depth].p_hdr;
1828 BUG_ON(eh == NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001829
1830 /* find where to start removing */
1831 ex = EXT_LAST_EXTENT(eh);
1832
1833 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001834 if (ext4_ext_is_uninitialized(ex))
1835 uninitialized = 1;
1836 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001837
1838 while (ex >= EXT_FIRST_EXTENT(eh) &&
1839 ex_ee_block + ex_ee_len > start) {
1840 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1841 path[depth].p_ext = ex;
1842
1843 a = ex_ee_block > start ? ex_ee_block : start;
1844 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1845 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1846
1847 ext_debug(" border %u:%u\n", a, b);
1848
1849 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1850 block = 0;
1851 num = 0;
1852 BUG();
1853 } else if (a != ex_ee_block) {
1854 /* remove tail of the extent */
1855 block = ex_ee_block;
1856 num = a - block;
1857 } else if (b != ex_ee_block + ex_ee_len - 1) {
1858 /* remove head of the extent */
1859 block = a;
1860 num = b - a;
1861 /* there is no "make a hole" API yet */
1862 BUG();
1863 } else {
1864 /* remove whole extent: excellent! */
1865 block = ex_ee_block;
1866 num = 0;
1867 BUG_ON(a != ex_ee_block);
1868 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1869 }
1870
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001871 /* at present, extent can't cross block group: */
Alex Tomasa86c6182006-10-11 01:21:03 -07001872 /* leaf + bitmap + group desc + sb + inode */
1873 credits = 5;
1874 if (ex == EXT_FIRST_EXTENT(eh)) {
1875 correct_index = 1;
1876 credits += (ext_depth(inode)) + 1;
1877 }
1878#ifdef CONFIG_QUOTA
1879 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1880#endif
1881
1882 handle = ext4_ext_journal_restart(handle, credits);
1883 if (IS_ERR(handle)) {
1884 err = PTR_ERR(handle);
1885 goto out;
1886 }
1887
1888 err = ext4_ext_get_access(handle, inode, path + depth);
1889 if (err)
1890 goto out;
1891
1892 err = ext4_remove_blocks(handle, inode, ex, a, b);
1893 if (err)
1894 goto out;
1895
1896 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001897 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001898 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001899 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1900 }
1901
1902 ex->ee_block = cpu_to_le32(block);
1903 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04001904 /*
1905 * Do not mark uninitialized if all the blocks in the
1906 * extent have been removed.
1907 */
1908 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001909 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001910
1911 err = ext4_ext_dirty(handle, inode, path + depth);
1912 if (err)
1913 goto out;
1914
Mingming Cao2ae02102006-10-11 01:21:11 -07001915 ext_debug("new extent: %u:%u:%llu\n", block, num,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001916 ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001917 ex--;
1918 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001919 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001920 }
1921
1922 if (correct_index && eh->eh_entries)
1923 err = ext4_ext_correct_indexes(handle, inode, path);
1924
1925 /* if this leaf is free, then we should
1926 * remove it from index block above */
1927 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1928 err = ext4_ext_rm_idx(handle, inode, path + depth);
1929
1930out:
1931 return err;
1932}
1933
1934/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001935 * ext4_ext_more_to_rm:
1936 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07001937 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001938static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001939ext4_ext_more_to_rm(struct ext4_ext_path *path)
1940{
1941 BUG_ON(path->p_idx == NULL);
1942
1943 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1944 return 0;
1945
1946 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001947 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07001948 * so we have to consider current index for truncation
1949 */
1950 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1951 return 0;
1952 return 1;
1953}
1954
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001955static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001956{
1957 struct super_block *sb = inode->i_sb;
1958 int depth = ext_depth(inode);
1959 struct ext4_ext_path *path;
1960 handle_t *handle;
1961 int i = 0, err = 0;
1962
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001963 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001964
1965 /* probably first extent we're gonna free will be last in block */
1966 handle = ext4_journal_start(inode, depth + 1);
1967 if (IS_ERR(handle))
1968 return PTR_ERR(handle);
1969
1970 ext4_ext_invalidate_cache(inode);
1971
1972 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001973 * We start scanning from right side, freeing all the blocks
1974 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07001975 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -08001976 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001977 if (path == NULL) {
1978 ext4_journal_stop(handle);
1979 return -ENOMEM;
1980 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001981 path[0].p_hdr = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001982 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001983 err = -EIO;
1984 goto out;
1985 }
1986 path[0].p_depth = depth;
1987
1988 while (i >= 0 && err == 0) {
1989 if (i == depth) {
1990 /* this is leaf block */
1991 err = ext4_ext_rm_leaf(handle, inode, path, start);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001992 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07001993 brelse(path[i].p_bh);
1994 path[i].p_bh = NULL;
1995 i--;
1996 continue;
1997 }
1998
1999 /* this is index block */
2000 if (!path[i].p_hdr) {
2001 ext_debug("initialize header\n");
2002 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002003 }
2004
Alex Tomasa86c6182006-10-11 01:21:03 -07002005 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002006 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002007 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2008 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2009 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2010 path[i].p_hdr,
2011 le16_to_cpu(path[i].p_hdr->eh_entries));
2012 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002013 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002014 path[i].p_idx--;
2015 }
2016
2017 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2018 i, EXT_FIRST_INDEX(path[i].p_hdr),
2019 path[i].p_idx);
2020 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002021 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002022 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002023 ext_debug("move to level %d (block %llu)\n",
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002024 i + 1, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002025 memset(path + i + 1, 0, sizeof(*path));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002026 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2027 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002028 /* should we reset i_size? */
2029 err = -EIO;
2030 break;
2031 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002032 if (WARN_ON(i + 1 > depth)) {
2033 err = -EIO;
2034 break;
2035 }
2036 if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2037 depth - i - 1)) {
2038 err = -EIO;
2039 break;
2040 }
2041 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002042
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002043 /* save actual number of indexes since this
2044 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002045 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2046 i++;
2047 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002048 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002049 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002050 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002051 * handle must be already prepared by the
2052 * truncatei_leaf() */
2053 err = ext4_ext_rm_idx(handle, inode, path + i);
2054 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002055 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002056 brelse(path[i].p_bh);
2057 path[i].p_bh = NULL;
2058 i--;
2059 ext_debug("return to level %d\n", i);
2060 }
2061 }
2062
2063 /* TODO: flexible tree reduction should be here */
2064 if (path->p_hdr->eh_entries == 0) {
2065 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002066 * truncate to zero freed all the tree,
2067 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002068 */
2069 err = ext4_ext_get_access(handle, inode, path);
2070 if (err == 0) {
2071 ext_inode_hdr(inode)->eh_depth = 0;
2072 ext_inode_hdr(inode)->eh_max =
2073 cpu_to_le16(ext4_ext_space_root(inode));
2074 err = ext4_ext_dirty(handle, inode, path);
2075 }
2076 }
2077out:
2078 ext4_ext_tree_changed(inode);
2079 ext4_ext_drop_refs(path);
2080 kfree(path);
2081 ext4_journal_stop(handle);
2082
2083 return err;
2084}
2085
2086/*
2087 * called at mount time
2088 */
2089void ext4_ext_init(struct super_block *sb)
2090{
2091 /*
2092 * possible initialization would be here
2093 */
2094
2095 if (test_opt(sb, EXTENTS)) {
2096 printk("EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002097#ifdef AGGRESSIVE_TEST
2098 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07002099#endif
2100#ifdef CHECK_BINSEARCH
2101 printk(", check binsearch");
2102#endif
2103#ifdef EXTENTS_STATS
2104 printk(", stats");
2105#endif
2106 printk("\n");
2107#ifdef EXTENTS_STATS
2108 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2109 EXT4_SB(sb)->s_ext_min = 1 << 30;
2110 EXT4_SB(sb)->s_ext_max = 0;
2111#endif
2112 }
2113}
2114
2115/*
2116 * called at umount time
2117 */
2118void ext4_ext_release(struct super_block *sb)
2119{
2120 if (!test_opt(sb, EXTENTS))
2121 return;
2122
2123#ifdef EXTENTS_STATS
2124 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2125 struct ext4_sb_info *sbi = EXT4_SB(sb);
2126 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2127 sbi->s_ext_blocks, sbi->s_ext_extents,
2128 sbi->s_ext_blocks / sbi->s_ext_extents);
2129 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2130 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2131 }
2132#endif
2133}
2134
Amit Arora56055d32007-07-17 21:42:38 -04002135/*
2136 * This function is called by ext4_ext_get_blocks() if someone tries to write
2137 * to an uninitialized extent. It may result in splitting the uninitialized
2138 * extent into multiple extents (upto three - one initialized and two
2139 * uninitialized).
2140 * There are three possibilities:
2141 * a> There is no split required: Entire extent should be initialized
2142 * b> Splits in two extents: Write is happening at either end of the extent
2143 * c> Splits in three extents: Somone is writing in middle of the extent
2144 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002145static int ext4_ext_convert_to_initialized(handle_t *handle,
2146 struct inode *inode,
2147 struct ext4_ext_path *path,
2148 ext4_lblk_t iblock,
2149 unsigned long max_blocks)
Amit Arora56055d32007-07-17 21:42:38 -04002150{
2151 struct ext4_extent *ex, newex;
2152 struct ext4_extent *ex1 = NULL;
2153 struct ext4_extent *ex2 = NULL;
2154 struct ext4_extent *ex3 = NULL;
2155 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002156 ext4_lblk_t ee_block;
2157 unsigned int allocated, ee_len, depth;
Amit Arora56055d32007-07-17 21:42:38 -04002158 ext4_fsblk_t newblock;
2159 int err = 0;
2160 int ret = 0;
2161
2162 depth = ext_depth(inode);
2163 eh = path[depth].p_hdr;
2164 ex = path[depth].p_ext;
2165 ee_block = le32_to_cpu(ex->ee_block);
2166 ee_len = ext4_ext_get_actual_len(ex);
2167 allocated = ee_len - (iblock - ee_block);
2168 newblock = iblock - ee_block + ext_pblock(ex);
2169 ex2 = ex;
2170
2171 /* ex1: ee_block to iblock - 1 : uninitialized */
2172 if (iblock > ee_block) {
2173 ex1 = ex;
2174 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2175 ext4_ext_mark_uninitialized(ex1);
2176 ex2 = &newex;
2177 }
2178 /*
2179 * for sanity, update the length of the ex2 extent before
2180 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2181 * overlap of blocks.
2182 */
2183 if (!ex1 && allocated > max_blocks)
2184 ex2->ee_len = cpu_to_le16(max_blocks);
2185 /* ex3: to ee_block + ee_len : uninitialised */
2186 if (allocated > max_blocks) {
2187 unsigned int newdepth;
2188 ex3 = &newex;
2189 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2190 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2191 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2192 ext4_ext_mark_uninitialized(ex3);
2193 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2194 if (err)
2195 goto out;
2196 /*
2197 * The depth, and hence eh & ex might change
2198 * as part of the insert above.
2199 */
2200 newdepth = ext_depth(inode);
2201 if (newdepth != depth) {
2202 depth = newdepth;
2203 path = ext4_ext_find_extent(inode, iblock, NULL);
2204 if (IS_ERR(path)) {
2205 err = PTR_ERR(path);
2206 path = NULL;
2207 goto out;
2208 }
2209 eh = path[depth].p_hdr;
2210 ex = path[depth].p_ext;
2211 if (ex2 != &newex)
2212 ex2 = ex;
2213 }
2214 allocated = max_blocks;
2215 }
2216 /*
2217 * If there was a change of depth as part of the
2218 * insertion of ex3 above, we need to update the length
2219 * of the ex1 extent again here
2220 */
2221 if (ex1 && ex1 != ex) {
2222 ex1 = ex;
2223 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2224 ext4_ext_mark_uninitialized(ex1);
2225 ex2 = &newex;
2226 }
2227 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2228 ex2->ee_block = cpu_to_le32(iblock);
Amit Arora56055d32007-07-17 21:42:38 -04002229 ext4_ext_store_pblock(ex2, newblock);
2230 ex2->ee_len = cpu_to_le16(allocated);
2231 if (ex2 != ex)
2232 goto insert;
2233 err = ext4_ext_get_access(handle, inode, path + depth);
2234 if (err)
2235 goto out;
2236 /*
2237 * New (initialized) extent starts from the first block
2238 * in the current extent. i.e., ex2 == ex
2239 * We have to see if it can be merged with the extent
2240 * on the left.
2241 */
2242 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2243 /*
2244 * To merge left, pass "ex2 - 1" to try_to_merge(),
2245 * since it merges towards right _only_.
2246 */
2247 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2248 if (ret) {
2249 err = ext4_ext_correct_indexes(handle, inode, path);
2250 if (err)
2251 goto out;
2252 depth = ext_depth(inode);
2253 ex2--;
2254 }
2255 }
2256 /*
2257 * Try to Merge towards right. This might be required
2258 * only when the whole extent is being written to.
2259 * i.e. ex2 == ex and ex3 == NULL.
2260 */
2261 if (!ex3) {
2262 ret = ext4_ext_try_to_merge(inode, path, ex2);
2263 if (ret) {
2264 err = ext4_ext_correct_indexes(handle, inode, path);
2265 if (err)
2266 goto out;
2267 }
2268 }
2269 /* Mark modified extent as dirty */
2270 err = ext4_ext_dirty(handle, inode, path + depth);
2271 goto out;
2272insert:
2273 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2274out:
2275 return err ? err : allocated;
2276}
2277
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002278/*
2279 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002280 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2281 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002282 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002283int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002284 ext4_lblk_t iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002285 unsigned long max_blocks, struct buffer_head *bh_result,
2286 int create, int extend_disksize)
2287{
2288 struct ext4_ext_path *path = NULL;
Amit Arora56055d32007-07-17 21:42:38 -04002289 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002290 struct ext4_extent newex, *ex;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002291 ext4_fsblk_t goal, newblock;
Amit Arora56055d32007-07-17 21:42:38 -04002292 int err = 0, depth, ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002293 unsigned long allocated = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05002294 struct ext4_allocation_request ar;
Alex Tomasa86c6182006-10-11 01:21:03 -07002295
2296 __clear_bit(BH_New, &bh_result->b_state);
Eric Sandeenbba90742008-01-28 23:58:27 -05002297 ext_debug("blocks %u/%lu requested for inode %u\n",
2298 iblock, max_blocks, inode->i_ino);
Alex Tomasa86c6182006-10-11 01:21:03 -07002299
2300 /* check in cache */
Avantika Mathur7e028972006-12-06 20:41:33 -08002301 goal = ext4_ext_in_cache(inode, iblock, &newex);
2302 if (goal) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002303 if (goal == EXT4_EXT_CACHE_GAP) {
2304 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002305 /*
2306 * block isn't allocated yet and
2307 * user doesn't want to allocate it
2308 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002309 goto out2;
2310 }
2311 /* we should allocate requested block */
2312 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2313 /* block is already allocated */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002314 newblock = iblock
2315 - le32_to_cpu(newex.ee_block)
2316 + ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002317 /* number of remaining blocks in the extent */
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002318 allocated = ext4_ext_get_actual_len(&newex) -
Alex Tomasa86c6182006-10-11 01:21:03 -07002319 (iblock - le32_to_cpu(newex.ee_block));
2320 goto out;
2321 } else {
2322 BUG();
2323 }
2324 }
2325
2326 /* find extent for this block */
2327 path = ext4_ext_find_extent(inode, iblock, NULL);
2328 if (IS_ERR(path)) {
2329 err = PTR_ERR(path);
2330 path = NULL;
2331 goto out2;
2332 }
2333
2334 depth = ext_depth(inode);
2335
2336 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002337 * consistent leaf must not be empty;
2338 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07002339 * this is why assert can't be put in ext4_ext_find_extent()
2340 */
2341 BUG_ON(path[depth].p_ext == NULL && depth != 0);
Amit Arora56055d32007-07-17 21:42:38 -04002342 eh = path[depth].p_hdr;
Alex Tomasa86c6182006-10-11 01:21:03 -07002343
Avantika Mathur7e028972006-12-06 20:41:33 -08002344 ex = path[depth].p_ext;
2345 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002346 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002347 ext4_fsblk_t ee_start = ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002348 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002349
2350 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002351 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04002352 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002353 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002354 ee_len = ext4_ext_get_actual_len(ex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002355 /* if found extent covers block, simply return it */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002356 if (iblock >= ee_block && iblock < ee_block + ee_len) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002357 newblock = iblock - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002358 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002359 allocated = ee_len - (iblock - ee_block);
Eric Sandeenbba90742008-01-28 23:58:27 -05002360 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002361 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04002362
Amit Aroraa2df2a62007-07-17 21:42:41 -04002363 /* Do not put uninitialized extent in the cache */
Amit Arora56055d32007-07-17 21:42:38 -04002364 if (!ext4_ext_is_uninitialized(ex)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002365 ext4_ext_put_in_cache(inode, ee_block,
2366 ee_len, ee_start,
2367 EXT4_EXT_CACHE_EXTENT);
Amit Arora56055d32007-07-17 21:42:38 -04002368 goto out;
2369 }
2370 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2371 goto out;
2372 if (!create)
2373 goto out2;
2374
2375 ret = ext4_ext_convert_to_initialized(handle, inode,
2376 path, iblock,
2377 max_blocks);
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002378 if (ret <= 0) {
2379 err = ret;
Amit Arora56055d32007-07-17 21:42:38 -04002380 goto out2;
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002381 } else
Amit Arora56055d32007-07-17 21:42:38 -04002382 allocated = ret;
2383 goto outnew;
Alex Tomasa86c6182006-10-11 01:21:03 -07002384 }
2385 }
2386
2387 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002388 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07002389 * we couldn't try to create block if create flag is zero
2390 */
2391 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002392 /*
2393 * put just found gap into cache to speed up
2394 * subsequent requests
2395 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002396 ext4_ext_put_gap_in_cache(inode, path, iblock);
2397 goto out2;
2398 }
2399 /*
Andrew Morton63f57932006-10-11 01:21:24 -07002400 * Okay, we need to do block allocation. Lazily initialize the block
2401 * allocation info here if necessary.
2402 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002403 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2404 ext4_init_block_alloc_info(inode);
2405
Alex Tomasc9de5602008-01-29 00:19:52 -05002406 /* find neighbour allocated blocks */
2407 ar.lleft = iblock;
2408 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2409 if (err)
2410 goto out2;
2411 ar.lright = iblock;
2412 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2413 if (err)
2414 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04002415
Amit Arora749269f2007-07-18 09:02:56 -04002416 /*
2417 * See if request is beyond maximum number of blocks we can have in
2418 * a single extent. For an initialized extent this limit is
2419 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2420 * EXT_UNINIT_MAX_LEN.
2421 */
2422 if (max_blocks > EXT_INIT_MAX_LEN &&
2423 create != EXT4_CREATE_UNINITIALIZED_EXT)
2424 max_blocks = EXT_INIT_MAX_LEN;
2425 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2426 create == EXT4_CREATE_UNINITIALIZED_EXT)
2427 max_blocks = EXT_UNINIT_MAX_LEN;
2428
Amit Arora25d14f92007-05-24 13:04:13 -04002429 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2430 newex.ee_block = cpu_to_le32(iblock);
2431 newex.ee_len = cpu_to_le16(max_blocks);
2432 err = ext4_ext_check_overlap(inode, &newex, path);
2433 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002434 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04002435 else
2436 allocated = max_blocks;
Alex Tomasc9de5602008-01-29 00:19:52 -05002437
2438 /* allocate new block */
2439 ar.inode = inode;
2440 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2441 ar.logical = iblock;
2442 ar.len = allocated;
2443 if (S_ISREG(inode->i_mode))
2444 ar.flags = EXT4_MB_HINT_DATA;
2445 else
2446 /* disable in-core preallocation for non-regular files */
2447 ar.flags = 0;
2448 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07002449 if (!newblock)
2450 goto out2;
Mingming Cao2ae02102006-10-11 01:21:11 -07002451 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002452 goal, newblock, allocated);
2453
2454 /* try to insert new extent into found leaf and return */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002455 ext4_ext_store_pblock(&newex, newblock);
Alex Tomasc9de5602008-01-29 00:19:52 -05002456 newex.ee_len = cpu_to_le16(ar.len);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002457 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2458 ext4_ext_mark_uninitialized(&newex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002459 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Alex Tomas315054f2007-05-24 13:04:25 -04002460 if (err) {
2461 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05002462 /* not a good idea to call discard here directly,
2463 * but otherwise we'd need to call it every free() */
2464 ext4_mb_discard_inode_preallocations(inode);
Alex Tomas315054f2007-05-24 13:04:25 -04002465 ext4_free_blocks(handle, inode, ext_pblock(&newex),
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002466 ext4_ext_get_actual_len(&newex), 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002467 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04002468 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002469
2470 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2471 EXT4_I(inode)->i_disksize = inode->i_size;
2472
2473 /* previous routine could use block we allocated */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002474 newblock = ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002475 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora56055d32007-07-17 21:42:38 -04002476outnew:
Alex Tomasa86c6182006-10-11 01:21:03 -07002477 __set_bit(BH_New, &bh_result->b_state);
2478
Amit Aroraa2df2a62007-07-17 21:42:41 -04002479 /* Cache only when it is _not_ an uninitialized extent */
2480 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2481 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2482 EXT4_EXT_CACHE_EXTENT);
Alex Tomasa86c6182006-10-11 01:21:03 -07002483out:
2484 if (allocated > max_blocks)
2485 allocated = max_blocks;
2486 ext4_ext_show_leaf(inode, path);
2487 __set_bit(BH_Mapped, &bh_result->b_state);
2488 bh_result->b_bdev = inode->i_sb->s_bdev;
2489 bh_result->b_blocknr = newblock;
2490out2:
2491 if (path) {
2492 ext4_ext_drop_refs(path);
2493 kfree(path);
2494 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002495 return err ? err : allocated;
2496}
2497
2498void ext4_ext_truncate(struct inode * inode, struct page *page)
2499{
2500 struct address_space *mapping = inode->i_mapping;
2501 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002502 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002503 handle_t *handle;
2504 int err = 0;
2505
2506 /*
2507 * probably first extent we're gonna free will be last in block
2508 */
2509 err = ext4_writepage_trans_blocks(inode) + 3;
2510 handle = ext4_journal_start(inode, err);
2511 if (IS_ERR(handle)) {
2512 if (page) {
2513 clear_highpage(page);
2514 flush_dcache_page(page);
2515 unlock_page(page);
2516 page_cache_release(page);
2517 }
2518 return;
2519 }
2520
2521 if (page)
2522 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2523
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002524 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07002525 ext4_ext_invalidate_cache(inode);
2526
Alex Tomasc9de5602008-01-29 00:19:52 -05002527 ext4_mb_discard_inode_preallocations(inode);
2528
Alex Tomasa86c6182006-10-11 01:21:03 -07002529 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002530 * TODO: optimization is possible here.
2531 * Probably we need not scan at all,
2532 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07002533 */
2534 if (ext4_orphan_add(handle, inode))
2535 goto out_stop;
2536
2537 /* we have to know where to truncate from in crash case */
2538 EXT4_I(inode)->i_disksize = inode->i_size;
2539 ext4_mark_inode_dirty(handle, inode);
2540
2541 last_block = (inode->i_size + sb->s_blocksize - 1)
2542 >> EXT4_BLOCK_SIZE_BITS(sb);
2543 err = ext4_ext_remove_space(inode, last_block);
2544
2545 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04002546 * transaction synchronous.
2547 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002548 if (IS_SYNC(inode))
2549 handle->h_sync = 1;
2550
2551out_stop:
2552 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002553 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07002554 * then we need to clear up the orphan record which we created above.
2555 * However, if this was a real unlink then we were called by
2556 * ext4_delete_inode(), and we allow that function to clean up the
2557 * orphan info for us.
2558 */
2559 if (inode->i_nlink)
2560 ext4_orphan_del(handle, inode);
2561
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002562 up_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07002563 ext4_journal_stop(handle);
2564}
2565
2566/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002567 * ext4_ext_writepage_trans_blocks:
2568 * calculate max number of blocks we could modify
Alex Tomasa86c6182006-10-11 01:21:03 -07002569 * in order to allocate new block for an inode
2570 */
2571int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2572{
2573 int needed;
2574
2575 needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2576
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002577 /* caller wants to allocate num blocks, but note it includes sb */
Alex Tomasa86c6182006-10-11 01:21:03 -07002578 needed = needed * num - (num - 1);
2579
2580#ifdef CONFIG_QUOTA
2581 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2582#endif
2583
2584 return needed;
2585}
Amit Aroraa2df2a62007-07-17 21:42:41 -04002586
2587/*
2588 * preallocate space for a file. This implements ext4's fallocate inode
2589 * operation, which gets called from sys_fallocate system call.
2590 * For block-mapped files, posix_fallocate should fall back to the method
2591 * of writing zeroes to the required new blocks (the same behavior which is
2592 * expected for file systems which do not support fallocate() system call).
2593 */
2594long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2595{
2596 handle_t *handle;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002597 ext4_lblk_t block;
2598 unsigned long max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002599 ext4_fsblk_t nblocks = 0;
2600 int ret = 0;
2601 int ret2 = 0;
2602 int retries = 0;
2603 struct buffer_head map_bh;
2604 unsigned int credits, blkbits = inode->i_blkbits;
2605
2606 /*
2607 * currently supporting (pre)allocate mode for extent-based
2608 * files _only_
2609 */
2610 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2611 return -EOPNOTSUPP;
2612
2613 /* preallocation to directories is currently not supported */
2614 if (S_ISDIR(inode->i_mode))
2615 return -ENODEV;
2616
2617 block = offset >> blkbits;
2618 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2619 - block;
2620
2621 /*
2622 * credits to insert 1 extent into extent tree + buffers to be able to
2623 * modify 1 super block, 1 block bitmap and 1 group descriptor.
2624 */
2625 credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002626 down_write((&EXT4_I(inode)->i_data_sem));
Amit Aroraa2df2a62007-07-17 21:42:41 -04002627retry:
2628 while (ret >= 0 && ret < max_blocks) {
2629 block = block + ret;
2630 max_blocks = max_blocks - ret;
2631 handle = ext4_journal_start(inode, credits);
2632 if (IS_ERR(handle)) {
2633 ret = PTR_ERR(handle);
2634 break;
2635 }
2636
2637 ret = ext4_ext_get_blocks(handle, inode, block,
2638 max_blocks, &map_bh,
2639 EXT4_CREATE_UNINITIALIZED_EXT, 0);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05002640 WARN_ON(ret <= 0);
2641 if (ret <= 0) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002642 ext4_error(inode->i_sb, "ext4_fallocate",
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05002643 "ext4_ext_get_blocks returned error: "
2644 "inode#%lu, block=%u, max_blocks=%lu",
2645 inode->i_ino, block, max_blocks);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002646 ret = -EIO;
2647 ext4_mark_inode_dirty(handle, inode);
2648 ret2 = ext4_journal_stop(handle);
2649 break;
2650 }
2651 if (ret > 0) {
2652 /* check wrap through sign-bit/zero here */
2653 if ((block + ret) < 0 || (block + ret) < block) {
2654 ret = -EIO;
2655 ext4_mark_inode_dirty(handle, inode);
2656 ret2 = ext4_journal_stop(handle);
2657 break;
2658 }
2659 if (buffer_new(&map_bh) && ((block + ret) >
2660 (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits)
2661 >> blkbits)))
2662 nblocks = nblocks + ret;
2663 }
2664
2665 /* Update ctime if new blocks get allocated */
2666 if (nblocks) {
2667 struct timespec now;
2668
2669 now = current_fs_time(inode->i_sb);
2670 if (!timespec_equal(&inode->i_ctime, &now))
2671 inode->i_ctime = now;
2672 }
2673
2674 ext4_mark_inode_dirty(handle, inode);
2675 ret2 = ext4_journal_stop(handle);
2676 if (ret2)
2677 break;
2678 }
2679
2680 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2681 goto retry;
2682
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002683 up_write((&EXT4_I(inode)->i_data_sem));
Amit Aroraa2df2a62007-07-17 21:42:41 -04002684 /*
2685 * Time to update the file size.
2686 * Update only when preallocation was requested beyond the file size.
2687 */
2688 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2689 (offset + len) > i_size_read(inode)) {
2690 if (ret > 0) {
2691 /*
2692 * if no error, we assume preallocation succeeded
2693 * completely
2694 */
2695 mutex_lock(&inode->i_mutex);
2696 i_size_write(inode, offset + len);
2697 EXT4_I(inode)->i_disksize = i_size_read(inode);
2698 mutex_unlock(&inode->i_mutex);
2699 } else if (ret < 0 && nblocks) {
2700 /* Handle partial allocation scenario */
2701 loff_t newsize;
2702
2703 mutex_lock(&inode->i_mutex);
2704 newsize = (nblocks << blkbits) + i_size_read(inode);
2705 i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits));
2706 EXT4_I(inode)->i_disksize = i_size_read(inode);
2707 mutex_unlock(&inode->i_mutex);
2708 }
2709 }
2710
2711 return ret > 0 ? ret2 : ret;
2712}