projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ext4: fix for fiemap last-block test
[deliverable/linux.git] / fs / ext4 / extents.c
CommitLineData
a86c6181
AT		 1/*
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>
cd02ff0b 35#include <linux/jbd2.h>
a86c6181
AT		 36#include <linux/highuid.h>
37#include <linux/pagemap.h>
38#include <linux/quotaops.h>
39#include <linux/string.h>
40#include <linux/slab.h>
a2df2a63 41#include <linux/falloc.h>
a86c6181 42#include <asm/uaccess.h>
6873fa0d 43#include <linux/fiemap.h>
3dcf5451
CH		 44#include "ext4_jbd2.h"
45#include "ext4_extents.h"
a86c6181
AT		 46
47
d0d856e8
RD		 48/*
49 * ext_pblock:
50 * combine low and high parts of physical block number into ext4_fsblk_t
51 */
09b88252 52static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
f65e6fba
AT		 53{
54 ext4_fsblk_t block;
55
b377611d 56 block = le32_to_cpu(ex->ee_start_lo);
9b8f1f01 57 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
f65e6fba
AT		 58 return block;
59}
60
d0d856e8
RD		 61/*
62 * idx_pblock:
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 */
c14c6fd5 65ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
f65e6fba
AT		 66{
67 ext4_fsblk_t block;
68
d8dd0b45 69 block = le32_to_cpu(ix->ei_leaf_lo);
9b8f1f01 70 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
f65e6fba
AT		 71 return block;
72}
73
d0d856e8
RD		 74/*
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
78 */
c14c6fd5 79void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
f65e6fba 80{
b377611d 81 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
9b8f1f01 82 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
f65e6fba
AT		 83}
84
d0d856e8
RD		 85/*
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
89 */
09b88252 90static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
f65e6fba 91{
d8dd0b45 92 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
9b8f1f01 93 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
f65e6fba
AT		 94}
95
9102e4fa 96static int ext4_ext_journal_restart(handle_t *handle, int needed)
a86c6181
AT		 97{
98 int err;
99
0390131b
FM		 100 if (!ext4_handle_valid(handle))
101 return 0;
a86c6181 102 if (handle->h_buffer_credits > needed)
9102e4fa
SF		 103 return 0;
104 err = ext4_journal_extend(handle, needed);
0123c939 105 if (err <= 0)
9102e4fa
SF		 106 return err;
107 return ext4_journal_restart(handle, needed);
a86c6181
AT		 108}
109
110/*
111 * could return:
112 * - EROFS
113 * - ENOMEM
114 */
115static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
116 struct ext4_ext_path *path)
117{
118 if (path->p_bh) {
119 /* path points to block */
120 return ext4_journal_get_write_access(handle, path->p_bh);
121 }
122 /* path points to leaf/index in inode body */
123 /* we use in-core data, no need to protect them */
124 return 0;
125}
126
127/*
128 * could return:
129 * - EROFS
130 * - ENOMEM
131 * - EIO
132 */
133static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
134 struct ext4_ext_path *path)
135{
136 int err;
137 if (path->p_bh) {
138 /* path points to block */
0390131b 139 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
a86c6181
AT		 140 } else {
141 /* path points to leaf/index in inode body */
142 err = ext4_mark_inode_dirty(handle, inode);
143 }
144 return err;
145}
146
f65e6fba 147static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
a86c6181 148 struct ext4_ext_path *path,
725d26d3 149 ext4_lblk_t block)
a86c6181
AT		 150{
151 struct ext4_inode_info *ei = EXT4_I(inode);
f65e6fba 152 ext4_fsblk_t bg_start;
74d3487f 153 ext4_fsblk_t last_block;
f65e6fba 154 ext4_grpblk_t colour;
a4912123
TT		 155 ext4_group_t block_group;
156 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
a86c6181
AT		 157 int depth;
158
159 if (path) {
160 struct ext4_extent *ex;
161 depth = path->p_depth;
162
163 /* try to predict block placement */
7e028976
AM		 164 ex = path[depth].p_ext;
165 if (ex)
f65e6fba 166 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
a86c6181 167
d0d856e8
RD		 168 /* it looks like index is empty;
169 * try to find starting block from index itself */
a86c6181
AT		 170 if (path[depth].p_bh)
171 return path[depth].p_bh->b_blocknr;
172 }
173
174 /* OK. use inode's group */
a4912123
TT		 175 block_group = ei->i_block_group;
176 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
177 /*
178 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
179 * block groups per flexgroup, reserve the first block
180 * group for directories and special files. Regular
181 * files will start at the second block group. This
182 * tends to speed up directory access and improves
183 * fsck times.
184 */
185 block_group &= ~(flex_size-1);
186 if (S_ISREG(inode->i_mode))
187 block_group++;
188 }
189 bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
a86c6181 190 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
74d3487f
VC		 191 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
192
a4912123
TT		 193 /*
194 * If we are doing delayed allocation, we don't need take
195 * colour into account.
196 */
197 if (test_opt(inode->i_sb, DELALLOC))
198 return bg_start;
199
74d3487f
VC		 200 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
201 colour = (current->pid % 16) *
a86c6181 202 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
74d3487f
VC		 203 else
204 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
a86c6181
AT		 205 return bg_start + colour + block;
206}
207
654b4908
AK		 208/*
209 * Allocation for a meta data block
210 */
f65e6fba 211static ext4_fsblk_t
654b4908 212ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
a86c6181
AT		 213 struct ext4_ext_path *path,
214 struct ext4_extent *ex, int *err)
215{
f65e6fba 216 ext4_fsblk_t goal, newblock;
a86c6181
AT		 217
218 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
97df5d15 219 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
a86c6181
AT		 220 return newblock;
221}
222
09b88252 223static int ext4_ext_space_block(struct inode *inode)
a86c6181
AT		 224{
225 int size;
226
227 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
228 / sizeof(struct ext4_extent);
bbf2f9fb 229#ifdef AGGRESSIVE_TEST
a86c6181
AT		 230 if (size > 6)
231 size = 6;
232#endif
233 return size;
234}
235
09b88252 236static int ext4_ext_space_block_idx(struct inode *inode)
a86c6181
AT		 237{
238 int size;
239
240 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
241 / sizeof(struct ext4_extent_idx);
bbf2f9fb 242#ifdef AGGRESSIVE_TEST
a86c6181
AT		 243 if (size > 5)
244 size = 5;
245#endif
246 return size;
247}
248
09b88252 249static int ext4_ext_space_root(struct inode *inode)
a86c6181
AT		 250{
251 int size;
252
253 size = sizeof(EXT4_I(inode)->i_data);
254 size -= sizeof(struct ext4_extent_header);
255 size /= sizeof(struct ext4_extent);
bbf2f9fb 256#ifdef AGGRESSIVE_TEST
a86c6181
AT		 257 if (size > 3)
258 size = 3;
259#endif
260 return size;
261}
262
09b88252 263static int ext4_ext_space_root_idx(struct inode *inode)
a86c6181
AT		 264{
265 int size;
266
267 size = sizeof(EXT4_I(inode)->i_data);
268 size -= sizeof(struct ext4_extent_header);
269 size /= sizeof(struct ext4_extent_idx);
bbf2f9fb 270#ifdef AGGRESSIVE_TEST
a86c6181
AT		 271 if (size > 4)
272 size = 4;
273#endif
274 return size;
275}
276
d2a17637
MC		 277/*
278 * Calculate the number of metadata blocks needed
279 * to allocate @blocks
280 * Worse case is one block per extent
281 */
282int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
283{
284 int lcap, icap, rcap, leafs, idxs, num;
285 int newextents = blocks;
286
287 rcap = ext4_ext_space_root_idx(inode);
288 lcap = ext4_ext_space_block(inode);
289 icap = ext4_ext_space_block_idx(inode);
290
291 /* number of new leaf blocks needed */
292 num = leafs = (newextents + lcap - 1) / lcap;
293
294 /*
295 * Worse case, we need separate index block(s)
296 * to link all new leaf blocks
297 */
298 idxs = (leafs + icap - 1) / icap;
299 do {
300 num += idxs;
301 idxs = (idxs + icap - 1) / icap;
302 } while (idxs > rcap);
303
304 return num;
305}
306
c29c0ae7
AT		 307static int
308ext4_ext_max_entries(struct inode *inode, int depth)
309{
310 int max;
311
312 if (depth == ext_depth(inode)) {
313 if (depth == 0)
314 max = ext4_ext_space_root(inode);
315 else
316 max = ext4_ext_space_root_idx(inode);
317 } else {
318 if (depth == 0)
319 max = ext4_ext_space_block(inode);
320 else
321 max = ext4_ext_space_block_idx(inode);
322 }
323
324 return max;
325}
326
56b19868
AK		 327static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
328{
e84a26ce 329 ext4_fsblk_t block = ext_pblock(ext), valid_block;
56b19868
AK		 330 int len = ext4_ext_get_actual_len(ext);
331 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
e84a26ce
TT		 332
333 valid_block = le32_to_cpu(es->s_first_data_block) +
334 EXT4_SB(inode->i_sb)->s_gdb_count;
335 if (unlikely(block <= valid_block ||
336 ((block + len) > ext4_blocks_count(es))))
56b19868
AK		 337 return 0;
338 else
339 return 1;
340}
341
342static int ext4_valid_extent_idx(struct inode *inode,
343 struct ext4_extent_idx *ext_idx)
344{
e84a26ce 345 ext4_fsblk_t block = idx_pblock(ext_idx), valid_block;
56b19868 346 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
e84a26ce
TT		 347
348 valid_block = le32_to_cpu(es->s_first_data_block) +
349 EXT4_SB(inode->i_sb)->s_gdb_count;
350 if (unlikely(block <= valid_block ||
351 (block >= ext4_blocks_count(es))))
56b19868
AK		 352 return 0;
353 else
354 return 1;
355}
356
357static int ext4_valid_extent_entries(struct inode *inode,
358 struct ext4_extent_header *eh,
359 int depth)
360{
361 struct ext4_extent *ext;
362 struct ext4_extent_idx *ext_idx;
363 unsigned short entries;
364 if (eh->eh_entries == 0)
365 return 1;
366
367 entries = le16_to_cpu(eh->eh_entries);
368
369 if (depth == 0) {
370 /* leaf entries */
371 ext = EXT_FIRST_EXTENT(eh);
372 while (entries) {
373 if (!ext4_valid_extent(inode, ext))
374 return 0;
375 ext++;
376 entries--;
377 }
378 } else {
379 ext_idx = EXT_FIRST_INDEX(eh);
380 while (entries) {
381 if (!ext4_valid_extent_idx(inode, ext_idx))
382 return 0;
383 ext_idx++;
384 entries--;
385 }
386 }
387 return 1;
388}
389
390static int __ext4_ext_check(const char *function, struct inode *inode,
c29c0ae7
AT		 391 struct ext4_extent_header *eh,
392 int depth)
393{
394 const char *error_msg;
395 int max = 0;
396
397 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
398 error_msg = "invalid magic";
399 goto corrupted;
400 }
401 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
402 error_msg = "unexpected eh_depth";
403 goto corrupted;
404 }
405 if (unlikely(eh->eh_max == 0)) {
406 error_msg = "invalid eh_max";
407 goto corrupted;
408 }
409 max = ext4_ext_max_entries(inode, depth);
410 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
411 error_msg = "too large eh_max";
412 goto corrupted;
413 }
414 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
415 error_msg = "invalid eh_entries";
416 goto corrupted;
417 }
56b19868
AK		 418 if (!ext4_valid_extent_entries(inode, eh, depth)) {
419 error_msg = "invalid extent entries";
420 goto corrupted;
421 }
c29c0ae7
AT		 422 return 0;
423
424corrupted:
425 ext4_error(inode->i_sb, function,
56b19868 426 "bad header/extent in inode #%lu: %s - magic %x, "
c29c0ae7
AT		 427 "entries %u, max %u(%u), depth %u(%u)",
428 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
429 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
430 max, le16_to_cpu(eh->eh_depth), depth);
431
432 return -EIO;
433}
434
56b19868
AK		 435#define ext4_ext_check(inode, eh, depth) \
436 __ext4_ext_check(__func__, inode, eh, depth)
c29c0ae7 437
7a262f7c
AK		 438int ext4_ext_check_inode(struct inode *inode)
439{
440 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
441}
442
a86c6181
AT		 443#ifdef EXT_DEBUG
444static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
445{
446 int k, l = path->p_depth;
447
448 ext_debug("path:");
449 for (k = 0; k <= l; k++, path++) {
450 if (path->p_idx) {
2ae02107 451 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
f65e6fba 452 idx_pblock(path->p_idx));
a86c6181 453 } else if (path->p_ext) {
2ae02107 454 ext_debug(" %d:%d:%llu ",
a86c6181 455 le32_to_cpu(path->p_ext->ee_block),
a2df2a63 456 ext4_ext_get_actual_len(path->p_ext),
f65e6fba 457 ext_pblock(path->p_ext));
a86c6181
AT		 458 } else
459 ext_debug(" []");
460 }
461 ext_debug("\n");
462}
463
464static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
465{
466 int depth = ext_depth(inode);
467 struct ext4_extent_header *eh;
468 struct ext4_extent *ex;
469 int i;
470
471 if (!path)
472 return;
473
474 eh = path[depth].p_hdr;
475 ex = EXT_FIRST_EXTENT(eh);
476
477 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
2ae02107 478 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
a2df2a63 479 ext4_ext_get_actual_len(ex), ext_pblock(ex));
a86c6181
AT		 480 }
481 ext_debug("\n");
482}
483#else
af5bc92d
TT		 484#define ext4_ext_show_path(inode, path)
485#define ext4_ext_show_leaf(inode, path)
a86c6181
AT		 486#endif
487
b35905c1 488void ext4_ext_drop_refs(struct ext4_ext_path *path)
a86c6181
AT		 489{
490 int depth = path->p_depth;
491 int i;
492
493 for (i = 0; i <= depth; i++, path++)
494 if (path->p_bh) {
495 brelse(path->p_bh);
496 path->p_bh = NULL;
497 }
498}
499
500/*
d0d856e8
RD		 501 * ext4_ext_binsearch_idx:
502 * binary search for the closest index of the given block
c29c0ae7 503 * the header must be checked before calling this
a86c6181
AT		 504 */
505static void
725d26d3
AK		 506ext4_ext_binsearch_idx(struct inode *inode,
507 struct ext4_ext_path *path, ext4_lblk_t block)
a86c6181
AT		 508{
509 struct ext4_extent_header *eh = path->p_hdr;
510 struct ext4_extent_idx *r, *l, *m;
511
a86c6181 512
bba90743 513 ext_debug("binsearch for %u(idx): ", block);
a86c6181
AT		 514
515 l = EXT_FIRST_INDEX(eh) + 1;
e9f410b1 516 r = EXT_LAST_INDEX(eh);
a86c6181
AT		 517 while (l <= r) {
518 m = l + (r - l) / 2;
519 if (block < le32_to_cpu(m->ei_block))
520 r = m - 1;
521 else
522 l = m + 1;
26d535ed
DM		 523 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
524 m, le32_to_cpu(m->ei_block),
525 r, le32_to_cpu(r->ei_block));
a86c6181
AT		 526 }
527
528 path->p_idx = l - 1;
f65e6fba 529 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
26d535ed 530 idx_pblock(path->p_idx));
a86c6181
AT		 531
532#ifdef CHECK_BINSEARCH
533 {
534 struct ext4_extent_idx *chix, *ix;
535 int k;
536
537 chix = ix = EXT_FIRST_INDEX(eh);
538 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
539 if (k != 0 &&
540 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
4776004f
TT		 541 printk(KERN_DEBUG "k=%d, ix=0x%p, "
542 "first=0x%p\n", k,
543 ix, EXT_FIRST_INDEX(eh));
544 printk(KERN_DEBUG "%u <= %u\n",
a86c6181
AT		 545 le32_to_cpu(ix->ei_block),
546 le32_to_cpu(ix[-1].ei_block));
547 }
548 BUG_ON(k && le32_to_cpu(ix->ei_block)
8c55e204 549 <= le32_to_cpu(ix[-1].ei_block));
a86c6181
AT		 550 if (block < le32_to_cpu(ix->ei_block))
551 break;
552 chix = ix;
553 }
554 BUG_ON(chix != path->p_idx);
555 }
556#endif
557
558}
559
560/*
d0d856e8
RD		 561 * ext4_ext_binsearch:
562 * binary search for closest extent of the given block
c29c0ae7 563 * the header must be checked before calling this
a86c6181
AT		 564 */
565static void
725d26d3
AK		 566ext4_ext_binsearch(struct inode *inode,
567 struct ext4_ext_path *path, ext4_lblk_t block)
a86c6181
AT		 568{
569 struct ext4_extent_header *eh = path->p_hdr;
570 struct ext4_extent *r, *l, *m;
571
a86c6181
AT		 572 if (eh->eh_entries == 0) {
573 /*
d0d856e8
RD		 574 * this leaf is empty:
575 * we get such a leaf in split/add case
a86c6181
AT		 576 */
577 return;
578 }
579
bba90743 580 ext_debug("binsearch for %u: ", block);
a86c6181
AT		 581
582 l = EXT_FIRST_EXTENT(eh) + 1;
e9f410b1 583 r = EXT_LAST_EXTENT(eh);
a86c6181
AT		 584
585 while (l <= r) {
586 m = l + (r - l) / 2;
587 if (block < le32_to_cpu(m->ee_block))
588 r = m - 1;
589 else
590 l = m + 1;
26d535ed
DM		 591 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
592 m, le32_to_cpu(m->ee_block),
593 r, le32_to_cpu(r->ee_block));
a86c6181
AT		 594 }
595
596 path->p_ext = l - 1;
2ae02107 597 ext_debug(" -> %d:%llu:%d ",
8c55e204
DK		 598 le32_to_cpu(path->p_ext->ee_block),
599 ext_pblock(path->p_ext),
a2df2a63 600 ext4_ext_get_actual_len(path->p_ext));
a86c6181
AT		 601
602#ifdef CHECK_BINSEARCH
603 {
604 struct ext4_extent *chex, *ex;
605 int k;
606
607 chex = ex = EXT_FIRST_EXTENT(eh);
608 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
609 BUG_ON(k && le32_to_cpu(ex->ee_block)
8c55e204 610 <= le32_to_cpu(ex[-1].ee_block));
a86c6181
AT		 611 if (block < le32_to_cpu(ex->ee_block))
612 break;
613 chex = ex;
614 }
615 BUG_ON(chex != path->p_ext);
616 }
617#endif
618
619}
620
621int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
622{
623 struct ext4_extent_header *eh;
624
625 eh = ext_inode_hdr(inode);
626 eh->eh_depth = 0;
627 eh->eh_entries = 0;
628 eh->eh_magic = EXT4_EXT_MAGIC;
629 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
630 ext4_mark_inode_dirty(handle, inode);
631 ext4_ext_invalidate_cache(inode);
632 return 0;
633}
634
635struct ext4_ext_path *
725d26d3
AK		 636ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
637 struct ext4_ext_path *path)
a86c6181
AT		 638{
639 struct ext4_extent_header *eh;
640 struct buffer_head *bh;
641 short int depth, i, ppos = 0, alloc = 0;
642
643 eh = ext_inode_hdr(inode);
c29c0ae7 644 depth = ext_depth(inode);
a86c6181
AT		 645
646 /* account possible depth increase */
647 if (!path) {
5d4958f9 648 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
a86c6181
AT		 649 GFP_NOFS);
650 if (!path)
651 return ERR_PTR(-ENOMEM);
652 alloc = 1;
653 }
a86c6181 654 path[0].p_hdr = eh;
1973adcb 655 path[0].p_bh = NULL;
a86c6181 656
c29c0ae7 657 i = depth;
a86c6181
AT		 658 /* walk through the tree */
659 while (i) {
7a262f7c
AK		 660 int need_to_validate = 0;
661
a86c6181
AT		 662 ext_debug("depth %d: num %d, max %d\n",
663 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
c29c0ae7 664
a86c6181 665 ext4_ext_binsearch_idx(inode, path + ppos, block);
f65e6fba 666 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
a86c6181
AT		 667 path[ppos].p_depth = i;
668 path[ppos].p_ext = NULL;
669
7a262f7c
AK		 670 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
671 if (unlikely(!bh))
a86c6181 672 goto err;
7a262f7c
AK		 673 if (!bh_uptodate_or_lock(bh)) {
674 if (bh_submit_read(bh) < 0) {
675 put_bh(bh);
676 goto err;
677 }
678 /* validate the extent entries */
679 need_to_validate = 1;
680 }
a86c6181
AT		 681 eh = ext_block_hdr(bh);
682 ppos++;
683 BUG_ON(ppos > depth);
684 path[ppos].p_bh = bh;
685 path[ppos].p_hdr = eh;
686 i--;
687
7a262f7c 688 if (need_to_validate && ext4_ext_check(inode, eh, i))
a86c6181
AT		 689 goto err;
690 }
691
692 path[ppos].p_depth = i;
a86c6181
AT		 693 path[ppos].p_ext = NULL;
694 path[ppos].p_idx = NULL;
695
a86c6181
AT		 696 /* find extent */
697 ext4_ext_binsearch(inode, path + ppos, block);
1973adcb
SF		 698 /* if not an empty leaf */
699 if (path[ppos].p_ext)
700 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
a86c6181
AT		 701
702 ext4_ext_show_path(inode, path);
703
704 return path;
705
706err:
707 ext4_ext_drop_refs(path);
708 if (alloc)
709 kfree(path);
710 return ERR_PTR(-EIO);
711}
712
713/*
d0d856e8
RD		 714 * ext4_ext_insert_index:
715 * insert new index [@logical;@ptr] into the block at @curp;
716 * check where to insert: before @curp or after @curp
a86c6181
AT		 717 */
718static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
719 struct ext4_ext_path *curp,
f65e6fba 720 int logical, ext4_fsblk_t ptr)
a86c6181
AT		 721{
722 struct ext4_extent_idx *ix;
723 int len, err;
724
7e028976
AM		 725 err = ext4_ext_get_access(handle, inode, curp);
726 if (err)
a86c6181
AT		 727 return err;
728
729 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
730 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
731 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
732 /* insert after */
733 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
734 len = (len - 1) * sizeof(struct ext4_extent_idx);
735 len = len < 0 ? 0 : len;
26d535ed 736 ext_debug("insert new index %d after: %llu. "
a86c6181
AT		 737 "move %d from 0x%p to 0x%p\n",
738 logical, ptr, len,
739 (curp->p_idx + 1), (curp->p_idx + 2));
740 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
741 }
742 ix = curp->p_idx + 1;
743 } else {
744 /* insert before */
745 len = len * sizeof(struct ext4_extent_idx);
746 len = len < 0 ? 0 : len;
26d535ed 747 ext_debug("insert new index %d before: %llu. "
a86c6181
AT		 748 "move %d from 0x%p to 0x%p\n",
749 logical, ptr, len,
750 curp->p_idx, (curp->p_idx + 1));
751 memmove(curp->p_idx + 1, curp->p_idx, len);
752 ix = curp->p_idx;
753 }
754
755 ix->ei_block = cpu_to_le32(logical);
f65e6fba 756 ext4_idx_store_pblock(ix, ptr);
e8546d06 757 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
a86c6181
AT		 758
759 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
8c55e204 760 > le16_to_cpu(curp->p_hdr->eh_max));
a86c6181
AT		 761 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
762
763 err = ext4_ext_dirty(handle, inode, curp);
764 ext4_std_error(inode->i_sb, err);
765
766 return err;
767}
768
769/*
d0d856e8
RD		 770 * ext4_ext_split:
771 * inserts new subtree into the path, using free index entry
772 * at depth @at:
773 * - allocates all needed blocks (new leaf and all intermediate index blocks)
774 * - makes decision where to split
775 * - moves remaining extents and index entries (right to the split point)
776 * into the newly allocated blocks
777 * - initializes subtree
a86c6181
AT		 778 */
779static int ext4_ext_split(handle_t *handle, struct inode *inode,
780 struct ext4_ext_path *path,
781 struct ext4_extent *newext, int at)
782{
783 struct buffer_head *bh = NULL;
784 int depth = ext_depth(inode);
785 struct ext4_extent_header *neh;
786 struct ext4_extent_idx *fidx;
787 struct ext4_extent *ex;
788 int i = at, k, m, a;
f65e6fba 789 ext4_fsblk_t newblock, oldblock;
a86c6181 790 __le32 border;
f65e6fba 791 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
a86c6181
AT		 792 int err = 0;
793
794 /* make decision: where to split? */
d0d856e8 795 /* FIXME: now decision is simplest: at current extent */
a86c6181 796
d0d856e8 797 /* if current leaf will be split, then we should use
a86c6181
AT		 798 * border from split point */
799 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
800 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
801 border = path[depth].p_ext[1].ee_block;
d0d856e8 802 ext_debug("leaf will be split."
a86c6181 803 " next leaf starts at %d\n",
8c55e204 804 le32_to_cpu(border));
a86c6181
AT		 805 } else {
806 border = newext->ee_block;
807 ext_debug("leaf will be added."
808 " next leaf starts at %d\n",
8c55e204 809 le32_to_cpu(border));
a86c6181
AT		 810 }
811
812 /*
d0d856e8
RD		 813 * If error occurs, then we break processing
814 * and mark filesystem read-only. index won't
a86c6181 815 * be inserted and tree will be in consistent
d0d856e8 816 * state. Next mount will repair buffers too.
a86c6181
AT		 817 */
818
819 /*
d0d856e8
RD		 820 * Get array to track all allocated blocks.
821 * We need this to handle errors and free blocks
822 * upon them.
a86c6181 823 */
5d4958f9 824 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
a86c6181
AT		 825 if (!ablocks)
826 return -ENOMEM;
a86c6181
AT		 827
828 /* allocate all needed blocks */
829 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
830 for (a = 0; a < depth - at; a++) {
654b4908
AK		 831 newblock = ext4_ext_new_meta_block(handle, inode, path,
832 newext, &err);
a86c6181
AT		 833 if (newblock == 0)
834 goto cleanup;
835 ablocks[a] = newblock;
836 }
837
838 /* initialize new leaf */
839 newblock = ablocks[--a];
840 BUG_ON(newblock == 0);
841 bh = sb_getblk(inode->i_sb, newblock);
842 if (!bh) {
843 err = -EIO;
844 goto cleanup;
845 }
846 lock_buffer(bh);
847
7e028976
AM		 848 err = ext4_journal_get_create_access(handle, bh);
849 if (err)
a86c6181
AT		 850 goto cleanup;
851
852 neh = ext_block_hdr(bh);
853 neh->eh_entries = 0;
854 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
855 neh->eh_magic = EXT4_EXT_MAGIC;
856 neh->eh_depth = 0;
857 ex = EXT_FIRST_EXTENT(neh);
858
d0d856e8 859 /* move remainder of path[depth] to the new leaf */
a86c6181
AT		 860 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
861 /* start copy from next extent */
862 /* TODO: we could do it by single memmove */
863 m = 0;
864 path[depth].p_ext++;
865 while (path[depth].p_ext <=
866 EXT_MAX_EXTENT(path[depth].p_hdr)) {
2ae02107 867 ext_debug("move %d:%llu:%d in new leaf %llu\n",
8c55e204
DK		 868 le32_to_cpu(path[depth].p_ext->ee_block),
869 ext_pblock(path[depth].p_ext),
a2df2a63 870 ext4_ext_get_actual_len(path[depth].p_ext),
a86c6181
AT		 871 newblock);
872 /*memmove(ex++, path[depth].p_ext++,
873 sizeof(struct ext4_extent));
874 neh->eh_entries++;*/
875 path[depth].p_ext++;
876 m++;
877 }
878 if (m) {
879 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
e8546d06 880 le16_add_cpu(&neh->eh_entries, m);
a86c6181
AT		 881 }
882
883 set_buffer_uptodate(bh);
884 unlock_buffer(bh);
885
0390131b 886 err = ext4_handle_dirty_metadata(handle, inode, bh);
7e028976 887 if (err)
a86c6181
AT		 888 goto cleanup;
889 brelse(bh);
890 bh = NULL;
891
892 /* correct old leaf */
893 if (m) {
7e028976
AM		 894 err = ext4_ext_get_access(handle, inode, path + depth);
895 if (err)
a86c6181 896 goto cleanup;
e8546d06 897 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
7e028976
AM		 898 err = ext4_ext_dirty(handle, inode, path + depth);
899 if (err)
a86c6181
AT		 900 goto cleanup;
901
902 }
903
904 /* create intermediate indexes */
905 k = depth - at - 1;
906 BUG_ON(k < 0);
907 if (k)
908 ext_debug("create %d intermediate indices\n", k);
909 /* insert new index into current index block */
910 /* current depth stored in i var */
911 i = depth - 1;
912 while (k--) {
913 oldblock = newblock;
914 newblock = ablocks[--a];
bba90743 915 bh = sb_getblk(inode->i_sb, newblock);
a86c6181
AT		 916 if (!bh) {
917 err = -EIO;
918 goto cleanup;
919 }
920 lock_buffer(bh);
921
7e028976
AM		 922 err = ext4_journal_get_create_access(handle, bh);
923 if (err)
a86c6181
AT		 924 goto cleanup;
925
926 neh = ext_block_hdr(bh);
927 neh->eh_entries = cpu_to_le16(1);
928 neh->eh_magic = EXT4_EXT_MAGIC;
929 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
930 neh->eh_depth = cpu_to_le16(depth - i);
931 fidx = EXT_FIRST_INDEX(neh);
932 fidx->ei_block = border;
f65e6fba 933 ext4_idx_store_pblock(fidx, oldblock);
a86c6181 934
bba90743
ES		 935 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
936 i, newblock, le32_to_cpu(border), oldblock);
a86c6181
AT		 937 /* copy indexes */
938 m = 0;
939 path[i].p_idx++;
940
941 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
942 EXT_MAX_INDEX(path[i].p_hdr));
943 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
944 EXT_LAST_INDEX(path[i].p_hdr));
945 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
26d535ed 946 ext_debug("%d: move %d:%llu in new index %llu\n", i,
8c55e204
DK		 947 le32_to_cpu(path[i].p_idx->ei_block),
948 idx_pblock(path[i].p_idx),
949 newblock);
a86c6181
AT		 950 /*memmove(++fidx, path[i].p_idx++,
951 sizeof(struct ext4_extent_idx));
952 neh->eh_entries++;
953 BUG_ON(neh->eh_entries > neh->eh_max);*/
954 path[i].p_idx++;
955 m++;
956 }
957 if (m) {
958 memmove(++fidx, path[i].p_idx - m,
959 sizeof(struct ext4_extent_idx) * m);
e8546d06 960 le16_add_cpu(&neh->eh_entries, m);
a86c6181
AT		 961 }
962 set_buffer_uptodate(bh);
963 unlock_buffer(bh);
964
0390131b 965 err = ext4_handle_dirty_metadata(handle, inode, bh);
7e028976 966 if (err)
a86c6181
AT		 967 goto cleanup;
968 brelse(bh);
969 bh = NULL;
970
971 /* correct old index */
972 if (m) {
973 err = ext4_ext_get_access(handle, inode, path + i);
974 if (err)
975 goto cleanup;
e8546d06 976 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
a86c6181
AT		 977 err = ext4_ext_dirty(handle, inode, path + i);
978 if (err)
979 goto cleanup;
980 }
981
982 i--;
983 }
984
985 /* insert new index */
a86c6181
AT		 986 err = ext4_ext_insert_index(handle, inode, path + at,
987 le32_to_cpu(border), newblock);
988
989cleanup:
990 if (bh) {
991 if (buffer_locked(bh))
992 unlock_buffer(bh);
993 brelse(bh);
994 }
995
996 if (err) {
997 /* free all allocated blocks in error case */
998 for (i = 0; i < depth; i++) {
999 if (!ablocks[i])
1000 continue;
c9de560d 1001 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
a86c6181
AT		 1002 }
1003 }
1004 kfree(ablocks);
1005
1006 return err;
1007}
1008
1009/*
d0d856e8
RD		 1010 * ext4_ext_grow_indepth:
1011 * implements tree growing procedure:
1012 * - allocates new block
1013 * - moves top-level data (index block or leaf) into the new block
1014 * - initializes new top-level, creating index that points to the
1015 * just created block
a86c6181
AT		 1016 */
1017static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1018 struct ext4_ext_path *path,
1019 struct ext4_extent *newext)
1020{
1021 struct ext4_ext_path *curp = path;
1022 struct ext4_extent_header *neh;
1023 struct ext4_extent_idx *fidx;
1024 struct buffer_head *bh;
f65e6fba 1025 ext4_fsblk_t newblock;
a86c6181
AT		 1026 int err = 0;
1027
654b4908 1028 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
a86c6181
AT		 1029 if (newblock == 0)
1030 return err;
1031
1032 bh = sb_getblk(inode->i_sb, newblock);
1033 if (!bh) {
1034 err = -EIO;
1035 ext4_std_error(inode->i_sb, err);
1036 return err;
1037 }
1038 lock_buffer(bh);
1039
7e028976
AM		 1040 err = ext4_journal_get_create_access(handle, bh);
1041 if (err) {
a86c6181
AT		 1042 unlock_buffer(bh);
1043 goto out;
1044 }
1045
1046 /* move top-level index/leaf into new block */
1047 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1048
1049 /* set size of new block */
1050 neh = ext_block_hdr(bh);
1051 /* old root could have indexes or leaves
1052 * so calculate e_max right way */
1053 if (ext_depth(inode))
1054 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
1055 else
1056 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
1057 neh->eh_magic = EXT4_EXT_MAGIC;
1058 set_buffer_uptodate(bh);
1059 unlock_buffer(bh);
1060
0390131b 1061 err = ext4_handle_dirty_metadata(handle, inode, bh);
7e028976 1062 if (err)
a86c6181
AT		 1063 goto out;
1064
1065 /* create index in new top-level index: num,max,pointer */
7e028976
AM		 1066 err = ext4_ext_get_access(handle, inode, curp);
1067 if (err)
a86c6181
AT		 1068 goto out;
1069
1070 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1071 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
1072 curp->p_hdr->eh_entries = cpu_to_le16(1);
1073 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
e9f410b1
DM		 1074
1075 if (path[0].p_hdr->eh_depth)
1076 curp->p_idx->ei_block =
1077 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1078 else
1079 curp->p_idx->ei_block =
1080 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
f65e6fba 1081 ext4_idx_store_pblock(curp->p_idx, newblock);
a86c6181
AT		 1082
1083 neh = ext_inode_hdr(inode);
1084 fidx = EXT_FIRST_INDEX(neh);
2ae02107 1085 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
a86c6181 1086 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
f65e6fba 1087 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
a86c6181
AT		 1088
1089 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1090 err = ext4_ext_dirty(handle, inode, curp);
1091out:
1092 brelse(bh);
1093
1094 return err;
1095}
1096
1097/*
d0d856e8
RD		 1098 * ext4_ext_create_new_leaf:
1099 * finds empty index and adds new leaf.
1100 * if no free index is found, then it requests in-depth growing.
a86c6181
AT		 1101 */
1102static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1103 struct ext4_ext_path *path,
1104 struct ext4_extent *newext)
1105{
1106 struct ext4_ext_path *curp;
1107 int depth, i, err = 0;
1108
1109repeat:
1110 i = depth = ext_depth(inode);
1111
1112 /* walk up to the tree and look for free index entry */
1113 curp = path + depth;
1114 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1115 i--;
1116 curp--;
1117 }
1118
d0d856e8
RD		 1119 /* we use already allocated block for index block,
1120 * so subsequent data blocks should be contiguous */
a86c6181
AT		 1121 if (EXT_HAS_FREE_INDEX(curp)) {
1122 /* if we found index with free entry, then use that
1123 * entry: create all needed subtree and add new leaf */
1124 err = ext4_ext_split(handle, inode, path, newext, i);
787e0981
SF		 1125 if (err)
1126 goto out;
a86c6181
AT		 1127
1128 /* refill path */
1129 ext4_ext_drop_refs(path);
1130 path = ext4_ext_find_extent(inode,
725d26d3
AK		 1131 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1132 path);
a86c6181
AT		 1133 if (IS_ERR(path))
1134 err = PTR_ERR(path);
1135 } else {
1136 /* tree is full, time to grow in depth */
1137 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1138 if (err)
1139 goto out;
1140
1141 /* refill path */
1142 ext4_ext_drop_refs(path);
1143 path = ext4_ext_find_extent(inode,
725d26d3
AK		 1144 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1145 path);
a86c6181
AT		 1146 if (IS_ERR(path)) {
1147 err = PTR_ERR(path);
1148 goto out;
1149 }
1150
1151 /*
d0d856e8
RD		 1152 * only first (depth 0 -> 1) produces free space;
1153 * in all other cases we have to split the grown tree
a86c6181
AT		 1154 */
1155 depth = ext_depth(inode);
1156 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
d0d856e8 1157 /* now we need to split */
a86c6181
AT		 1158 goto repeat;
1159 }
1160 }
1161
1162out:
1163 return err;
1164}
1165
1988b51e
AT		 1166/*
1167 * search the closest allocated block to the left for *logical
1168 * and returns it at @logical + it's physical address at @phys
1169 * if *logical is the smallest allocated block, the function
1170 * returns 0 at @phys
1171 * return value contains 0 (success) or error code
1172 */
1173int
1174ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1175 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1176{
1177 struct ext4_extent_idx *ix;
1178 struct ext4_extent *ex;
b939e376 1179 int depth, ee_len;
1988b51e
AT		 1180
1181 BUG_ON(path == NULL);
1182 depth = path->p_depth;
1183 *phys = 0;
1184
1185 if (depth == 0 && path->p_ext == NULL)
1186 return 0;
1187
1188 /* usually extent in the path covers blocks smaller
1189 * then *logical, but it can be that extent is the
1190 * first one in the file */
1191
1192 ex = path[depth].p_ext;
b939e376 1193 ee_len = ext4_ext_get_actual_len(ex);
1988b51e
AT		 1194 if (*logical < le32_to_cpu(ex->ee_block)) {
1195 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1196 while (--depth >= 0) {
1197 ix = path[depth].p_idx;
1198 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1199 }
1200 return 0;
1201 }
1202
b939e376 1203 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1988b51e 1204
b939e376
AK		 1205 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1206 *phys = ext_pblock(ex) + ee_len - 1;
1988b51e
AT		 1207 return 0;
1208}
1209
1210/*
1211 * search the closest allocated block to the right for *logical
1212 * and returns it at @logical + it's physical address at @phys
1213 * if *logical is the smallest allocated block, the function
1214 * returns 0 at @phys
1215 * return value contains 0 (success) or error code
1216 */
1217int
1218ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1219 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1220{
1221 struct buffer_head *bh = NULL;
1222 struct ext4_extent_header *eh;
1223 struct ext4_extent_idx *ix;
1224 struct ext4_extent *ex;
1225 ext4_fsblk_t block;
395a87bf
ES		 1226 int depth; /* Note, NOT eh_depth; depth from top of tree */
1227 int ee_len;
1988b51e
AT		 1228
1229 BUG_ON(path == NULL);
1230 depth = path->p_depth;
1231 *phys = 0;
1232
1233 if (depth == 0 && path->p_ext == NULL)
1234 return 0;
1235
1236 /* usually extent in the path covers blocks smaller
1237 * then *logical, but it can be that extent is the
1238 * first one in the file */
1239
1240 ex = path[depth].p_ext;
b939e376 1241 ee_len = ext4_ext_get_actual_len(ex);
1988b51e
AT		 1242 if (*logical < le32_to_cpu(ex->ee_block)) {
1243 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1244 while (--depth >= 0) {
1245 ix = path[depth].p_idx;
1246 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1247 }
1248 *logical = le32_to_cpu(ex->ee_block);
1249 *phys = ext_pblock(ex);
1250 return 0;
1251 }
1252
b939e376 1253 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1988b51e
AT		 1254
1255 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1256 /* next allocated block in this leaf */
1257 ex++;
1258 *logical = le32_to_cpu(ex->ee_block);
1259 *phys = ext_pblock(ex);
1260 return 0;
1261 }
1262
1263 /* go up and search for index to the right */
1264 while (--depth >= 0) {
1265 ix = path[depth].p_idx;
1266 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
25f1ee3a 1267 goto got_index;
1988b51e
AT		 1268 }
1269
25f1ee3a
WF		 1270 /* we've gone up to the root and found no index to the right */
1271 return 0;
1988b51e 1272
25f1ee3a 1273got_index:
1988b51e
AT		 1274 /* we've found index to the right, let's
1275 * follow it and find the closest allocated
1276 * block to the right */
1277 ix++;
1278 block = idx_pblock(ix);
1279 while (++depth < path->p_depth) {
1280 bh = sb_bread(inode->i_sb, block);
1281 if (bh == NULL)
1282 return -EIO;
1283 eh = ext_block_hdr(bh);
395a87bf 1284 /* subtract from p_depth to get proper eh_depth */
56b19868 1285 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1988b51e
AT		 1286 put_bh(bh);
1287 return -EIO;
1288 }
1289 ix = EXT_FIRST_INDEX(eh);
1290 block = idx_pblock(ix);
1291 put_bh(bh);
1292 }
1293
1294 bh = sb_bread(inode->i_sb, block);
1295 if (bh == NULL)
1296 return -EIO;
1297 eh = ext_block_hdr(bh);
56b19868 1298 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1988b51e
AT		 1299 put_bh(bh);
1300 return -EIO;
1301 }
1302 ex = EXT_FIRST_EXTENT(eh);
1303 *logical = le32_to_cpu(ex->ee_block);
1304 *phys = ext_pblock(ex);
1305 put_bh(bh);
1306 return 0;
1988b51e
AT		 1307}
1308
a86c6181 1309/*
d0d856e8
RD		 1310 * ext4_ext_next_allocated_block:
1311 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1312 * NOTE: it considers block number from index entry as
1313 * allocated block. Thus, index entries have to be consistent
1314 * with leaves.
a86c6181 1315 */
725d26d3 1316static ext4_lblk_t
a86c6181
AT		 1317ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1318{
1319 int depth;
1320
1321 BUG_ON(path == NULL);
1322 depth = path->p_depth;
1323
1324 if (depth == 0 && path->p_ext == NULL)
1325 return EXT_MAX_BLOCK;
1326
1327 while (depth >= 0) {
1328 if (depth == path->p_depth) {
1329 /* leaf */
1330 if (path[depth].p_ext !=
1331 EXT_LAST_EXTENT(path[depth].p_hdr))
1332 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1333 } else {
1334 /* index */
1335 if (path[depth].p_idx !=
1336 EXT_LAST_INDEX(path[depth].p_hdr))
1337 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1338 }
1339 depth--;
1340 }
1341
1342 return EXT_MAX_BLOCK;
1343}
1344
1345/*
d0d856e8 1346 * ext4_ext_next_leaf_block:
a86c6181
AT		 1347 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1348 */
725d26d3 1349static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
63f57933 1350 struct ext4_ext_path *path)
a86c6181
AT		 1351{
1352 int depth;
1353
1354 BUG_ON(path == NULL);
1355 depth = path->p_depth;
1356
1357 /* zero-tree has no leaf blocks at all */
1358 if (depth == 0)
1359 return EXT_MAX_BLOCK;
1360
1361 /* go to index block */
1362 depth--;
1363
1364 while (depth >= 0) {
1365 if (path[depth].p_idx !=
1366 EXT_LAST_INDEX(path[depth].p_hdr))
725d26d3
AK		 1367 return (ext4_lblk_t)
1368 le32_to_cpu(path[depth].p_idx[1].ei_block);
a86c6181
AT		 1369 depth--;
1370 }
1371
1372 return EXT_MAX_BLOCK;
1373}
1374
1375/*
d0d856e8
RD		 1376 * ext4_ext_correct_indexes:
1377 * if leaf gets modified and modified extent is first in the leaf,
1378 * then we have to correct all indexes above.
a86c6181
AT		 1379 * TODO: do we need to correct tree in all cases?
1380 */
1d03ec98 1381static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
a86c6181
AT		 1382 struct ext4_ext_path *path)
1383{
1384 struct ext4_extent_header *eh;
1385 int depth = ext_depth(inode);
1386 struct ext4_extent *ex;
1387 __le32 border;
1388 int k, err = 0;
1389
1390 eh = path[depth].p_hdr;
1391 ex = path[depth].p_ext;
1392 BUG_ON(ex == NULL);
1393 BUG_ON(eh == NULL);
1394
1395 if (depth == 0) {
1396 /* there is no tree at all */
1397 return 0;
1398 }
1399
1400 if (ex != EXT_FIRST_EXTENT(eh)) {
1401 /* we correct tree if first leaf got modified only */
1402 return 0;
1403 }
1404
1405 /*
d0d856e8 1406 * TODO: we need correction if border is smaller than current one
a86c6181
AT		 1407 */
1408 k = depth - 1;
1409 border = path[depth].p_ext->ee_block;
7e028976
AM		 1410 err = ext4_ext_get_access(handle, inode, path + k);
1411 if (err)
a86c6181
AT		 1412 return err;
1413 path[k].p_idx->ei_block = border;
7e028976
AM		 1414 err = ext4_ext_dirty(handle, inode, path + k);
1415 if (err)
a86c6181
AT		 1416 return err;
1417
1418 while (k--) {
1419 /* change all left-side indexes */
1420 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1421 break;
7e028976
AM		 1422 err = ext4_ext_get_access(handle, inode, path + k);
1423 if (err)
a86c6181
AT		 1424 break;
1425 path[k].p_idx->ei_block = border;
7e028976
AM		 1426 err = ext4_ext_dirty(handle, inode, path + k);
1427 if (err)
a86c6181
AT		 1428 break;
1429 }
1430
1431 return err;
1432}
1433
09b88252 1434static int
a86c6181
AT		 1435ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1436 struct ext4_extent *ex2)
1437{
749269fa 1438 unsigned short ext1_ee_len, ext2_ee_len, max_len;
a2df2a63
AA		 1439
1440 /*
1441 * Make sure that either both extents are uninitialized, or
1442 * both are _not_.
1443 */
1444 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1445 return 0;
1446
749269fa
AA		 1447 if (ext4_ext_is_uninitialized(ex1))
1448 max_len = EXT_UNINIT_MAX_LEN;
1449 else
1450 max_len = EXT_INIT_MAX_LEN;
1451
a2df2a63
AA		 1452 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1453 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1454
1455 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
63f57933 1456 le32_to_cpu(ex2->ee_block))
a86c6181
AT		 1457 return 0;
1458
471d4011
SB		 1459 /*
1460 * To allow future support for preallocated extents to be added
1461 * as an RO_COMPAT feature, refuse to merge to extents if
d0d856e8 1462 * this can result in the top bit of ee_len being set.
471d4011 1463 */
749269fa 1464 if (ext1_ee_len + ext2_ee_len > max_len)
471d4011 1465 return 0;
bbf2f9fb 1466#ifdef AGGRESSIVE_TEST
b939e376 1467 if (ext1_ee_len >= 4)
a86c6181
AT		 1468 return 0;
1469#endif
1470
a2df2a63 1471 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
a86c6181
AT		 1472 return 1;
1473 return 0;
1474}
1475
56055d3a
AA		 1476/*
1477 * This function tries to merge the "ex" extent to the next extent in the tree.
1478 * It always tries to merge towards right. If you want to merge towards
1479 * left, pass "ex - 1" as argument instead of "ex".
1480 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1481 * 1 if they got merged.
1482 */
1483int ext4_ext_try_to_merge(struct inode *inode,
1484 struct ext4_ext_path *path,
1485 struct ext4_extent *ex)
1486{
1487 struct ext4_extent_header *eh;
1488 unsigned int depth, len;
1489 int merge_done = 0;
1490 int uninitialized = 0;
1491
1492 depth = ext_depth(inode);
1493 BUG_ON(path[depth].p_hdr == NULL);
1494 eh = path[depth].p_hdr;
1495
1496 while (ex < EXT_LAST_EXTENT(eh)) {
1497 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1498 break;
1499 /* merge with next extent! */
1500 if (ext4_ext_is_uninitialized(ex))
1501 uninitialized = 1;
1502 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1503 + ext4_ext_get_actual_len(ex + 1));
1504 if (uninitialized)
1505 ext4_ext_mark_uninitialized(ex);
1506
1507 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1508 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1509 * sizeof(struct ext4_extent);
1510 memmove(ex + 1, ex + 2, len);
1511 }
e8546d06 1512 le16_add_cpu(&eh->eh_entries, -1);
56055d3a
AA		 1513 merge_done = 1;
1514 WARN_ON(eh->eh_entries == 0);
1515 if (!eh->eh_entries)
1516 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1517 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1518 }
1519
1520 return merge_done;
1521}
1522
25d14f98
AA		 1523/*
1524 * check if a portion of the "newext" extent overlaps with an
1525 * existing extent.
1526 *
1527 * If there is an overlap discovered, it updates the length of the newext
1528 * such that there will be no overlap, and then returns 1.
1529 * If there is no overlap found, it returns 0.
1530 */
1531unsigned int ext4_ext_check_overlap(struct inode *inode,
1532 struct ext4_extent *newext,
1533 struct ext4_ext_path *path)
1534{
725d26d3 1535 ext4_lblk_t b1, b2;
25d14f98
AA		 1536 unsigned int depth, len1;
1537 unsigned int ret = 0;
1538
1539 b1 = le32_to_cpu(newext->ee_block);
a2df2a63 1540 len1 = ext4_ext_get_actual_len(newext);
25d14f98
AA		 1541 depth = ext_depth(inode);
1542 if (!path[depth].p_ext)
1543 goto out;
1544 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1545
1546 /*
1547 * get the next allocated block if the extent in the path
2b2d6d01 1548 * is before the requested block(s)
25d14f98
AA		 1549 */
1550 if (b2 < b1) {
1551 b2 = ext4_ext_next_allocated_block(path);
1552 if (b2 == EXT_MAX_BLOCK)
1553 goto out;
1554 }
1555
725d26d3 1556 /* check for wrap through zero on extent logical start block*/
25d14f98
AA		 1557 if (b1 + len1 < b1) {
1558 len1 = EXT_MAX_BLOCK - b1;
1559 newext->ee_len = cpu_to_le16(len1);
1560 ret = 1;
1561 }
1562
1563 /* check for overlap */
1564 if (b1 + len1 > b2) {
1565 newext->ee_len = cpu_to_le16(b2 - b1);
1566 ret = 1;
1567 }
1568out:
1569 return ret;
1570}
1571
a86c6181 1572/*
d0d856e8
RD		 1573 * ext4_ext_insert_extent:
1574 * tries to merge requsted extent into the existing extent or
1575 * inserts requested extent as new one into the tree,
1576 * creating new leaf in the no-space case.
a86c6181
AT		 1577 */
1578int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1579 struct ext4_ext_path *path,
1580 struct ext4_extent *newext)
1581{
af5bc92d 1582 struct ext4_extent_header *eh;
a86c6181
AT		 1583 struct ext4_extent *ex, *fex;
1584 struct ext4_extent *nearex; /* nearest extent */
1585 struct ext4_ext_path *npath = NULL;
725d26d3
AK		 1586 int depth, len, err;
1587 ext4_lblk_t next;
a2df2a63 1588 unsigned uninitialized = 0;
a86c6181 1589
a2df2a63 1590 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
a86c6181
AT		 1591 depth = ext_depth(inode);
1592 ex = path[depth].p_ext;
1593 BUG_ON(path[depth].p_hdr == NULL);
1594
1595 /* try to insert block into found extent and return */
1596 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
2ae02107 1597 ext_debug("append %d block to %d:%d (from %llu)\n",
a2df2a63 1598 ext4_ext_get_actual_len(newext),
a86c6181 1599 le32_to_cpu(ex->ee_block),
a2df2a63 1600 ext4_ext_get_actual_len(ex), ext_pblock(ex));
7e028976
AM		 1601 err = ext4_ext_get_access(handle, inode, path + depth);
1602 if (err)
a86c6181 1603 return err;
a2df2a63
AA		 1604
1605 /*
1606 * ext4_can_extents_be_merged should have checked that either
1607 * both extents are uninitialized, or both aren't. Thus we
1608 * need to check only one of them here.
1609 */
1610 if (ext4_ext_is_uninitialized(ex))
1611 uninitialized = 1;
1612 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1613 + ext4_ext_get_actual_len(newext));
1614 if (uninitialized)
1615 ext4_ext_mark_uninitialized(ex);
a86c6181
AT		 1616 eh = path[depth].p_hdr;
1617 nearex = ex;
1618 goto merge;
1619 }
1620
1621repeat:
1622 depth = ext_depth(inode);
1623 eh = path[depth].p_hdr;
1624 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1625 goto has_space;
1626
1627 /* probably next leaf has space for us? */
1628 fex = EXT_LAST_EXTENT(eh);
1629 next = ext4_ext_next_leaf_block(inode, path);
1630 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1631 && next != EXT_MAX_BLOCK) {
1632 ext_debug("next leaf block - %d\n", next);
1633 BUG_ON(npath != NULL);
1634 npath = ext4_ext_find_extent(inode, next, NULL);
1635 if (IS_ERR(npath))
1636 return PTR_ERR(npath);
1637 BUG_ON(npath->p_depth != path->p_depth);
1638 eh = npath[depth].p_hdr;
1639 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1640 ext_debug("next leaf isnt full(%d)\n",
1641 le16_to_cpu(eh->eh_entries));
1642 path = npath;
1643 goto repeat;
1644 }
1645 ext_debug("next leaf has no free space(%d,%d)\n",
1646 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1647 }
1648
1649 /*
d0d856e8
RD		 1650 * There is no free space in the found leaf.
1651 * We're gonna add a new leaf in the tree.
a86c6181
AT		 1652 */
1653 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1654 if (err)
1655 goto cleanup;
1656 depth = ext_depth(inode);
1657 eh = path[depth].p_hdr;
1658
1659has_space:
1660 nearex = path[depth].p_ext;
1661
7e028976
AM		 1662 err = ext4_ext_get_access(handle, inode, path + depth);
1663 if (err)
a86c6181
AT		 1664 goto cleanup;
1665
1666 if (!nearex) {
1667 /* there is no extent in this leaf, create first one */
2ae02107 1668 ext_debug("first extent in the leaf: %d:%llu:%d\n",
8c55e204
DK		 1669 le32_to_cpu(newext->ee_block),
1670 ext_pblock(newext),
a2df2a63 1671 ext4_ext_get_actual_len(newext));
a86c6181
AT		 1672 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1673 } else if (le32_to_cpu(newext->ee_block)
8c55e204 1674 > le32_to_cpu(nearex->ee_block)) {
a86c6181
AT		 1675/* BUG_ON(newext->ee_block == nearex->ee_block); */
1676 if (nearex != EXT_LAST_EXTENT(eh)) {
1677 len = EXT_MAX_EXTENT(eh) - nearex;
1678 len = (len - 1) * sizeof(struct ext4_extent);
1679 len = len < 0 ? 0 : len;
2ae02107 1680 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
a86c6181 1681 "move %d from 0x%p to 0x%p\n",
8c55e204
DK		 1682 le32_to_cpu(newext->ee_block),
1683 ext_pblock(newext),
a2df2a63 1684 ext4_ext_get_actual_len(newext),
a86c6181
AT		 1685 nearex, len, nearex + 1, nearex + 2);
1686 memmove(nearex + 2, nearex + 1, len);
1687 }
1688 path[depth].p_ext = nearex + 1;
1689 } else {
1690 BUG_ON(newext->ee_block == nearex->ee_block);
1691 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1692 len = len < 0 ? 0 : len;
2ae02107 1693 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
a86c6181
AT		 1694 "move %d from 0x%p to 0x%p\n",
1695 le32_to_cpu(newext->ee_block),
f65e6fba 1696 ext_pblock(newext),
a2df2a63 1697 ext4_ext_get_actual_len(newext),
a86c6181
AT		 1698 nearex, len, nearex + 1, nearex + 2);
1699 memmove(nearex + 1, nearex, len);
1700 path[depth].p_ext = nearex;
1701 }
1702
e8546d06 1703 le16_add_cpu(&eh->eh_entries, 1);
a86c6181
AT		 1704 nearex = path[depth].p_ext;
1705 nearex->ee_block = newext->ee_block;
b377611d 1706 ext4_ext_store_pblock(nearex, ext_pblock(newext));
a86c6181 1707 nearex->ee_len = newext->ee_len;
a86c6181
AT		 1708
1709merge:
1710 /* try to merge extents to the right */
56055d3a 1711 ext4_ext_try_to_merge(inode, path, nearex);
a86c6181
AT		 1712
1713 /* try to merge extents to the left */
1714
1715 /* time to correct all indexes above */
1716 err = ext4_ext_correct_indexes(handle, inode, path);
1717 if (err)
1718 goto cleanup;
1719
1720 err = ext4_ext_dirty(handle, inode, path + depth);
1721
1722cleanup:
1723 if (npath) {
1724 ext4_ext_drop_refs(npath);
1725 kfree(npath);
1726 }
a86c6181
AT		 1727 ext4_ext_invalidate_cache(inode);
1728 return err;
1729}
1730
6873fa0d
ES		 1731int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1732 ext4_lblk_t num, ext_prepare_callback func,
1733 void *cbdata)
1734{
1735 struct ext4_ext_path *path = NULL;
1736 struct ext4_ext_cache cbex;
1737 struct ext4_extent *ex;
1738 ext4_lblk_t next, start = 0, end = 0;
1739 ext4_lblk_t last = block + num;
1740 int depth, exists, err = 0;
1741
1742 BUG_ON(func == NULL);
1743 BUG_ON(inode == NULL);
1744
1745 while (block < last && block != EXT_MAX_BLOCK) {
1746 num = last - block;
1747 /* find extent for this block */
1748 path = ext4_ext_find_extent(inode, block, path);
1749 if (IS_ERR(path)) {
1750 err = PTR_ERR(path);
1751 path = NULL;
1752 break;
1753 }
1754
1755 depth = ext_depth(inode);
1756 BUG_ON(path[depth].p_hdr == NULL);
1757 ex = path[depth].p_ext;
1758 next = ext4_ext_next_allocated_block(path);
1759
1760 exists = 0;
1761 if (!ex) {
1762 /* there is no extent yet, so try to allocate
1763 * all requested space */
1764 start = block;
1765 end = block + num;
1766 } else if (le32_to_cpu(ex->ee_block) > block) {
1767 /* need to allocate space before found extent */
1768 start = block;
1769 end = le32_to_cpu(ex->ee_block);
1770 if (block + num < end)
1771 end = block + num;
1772 } else if (block >= le32_to_cpu(ex->ee_block)
1773 + ext4_ext_get_actual_len(ex)) {
1774 /* need to allocate space after found extent */
1775 start = block;
1776 end = block + num;
1777 if (end >= next)
1778 end = next;
1779 } else if (block >= le32_to_cpu(ex->ee_block)) {
1780 /*
1781 * some part of requested space is covered
1782 * by found extent
1783 */
1784 start = block;
1785 end = le32_to_cpu(ex->ee_block)
1786 + ext4_ext_get_actual_len(ex);
1787 if (block + num < end)
1788 end = block + num;
1789 exists = 1;
1790 } else {
1791 BUG();
1792 }
1793 BUG_ON(end <= start);
1794
1795 if (!exists) {
1796 cbex.ec_block = start;
1797 cbex.ec_len = end - start;
1798 cbex.ec_start = 0;
1799 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1800 } else {
1801 cbex.ec_block = le32_to_cpu(ex->ee_block);
1802 cbex.ec_len = ext4_ext_get_actual_len(ex);
1803 cbex.ec_start = ext_pblock(ex);
1804 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1805 }
1806
1807 BUG_ON(cbex.ec_len == 0);
1808 err = func(inode, path, &cbex, ex, cbdata);
1809 ext4_ext_drop_refs(path);
1810
1811 if (err < 0)
1812 break;
1813
1814 if (err == EXT_REPEAT)
1815 continue;
1816 else if (err == EXT_BREAK) {
1817 err = 0;
1818 break;
1819 }
1820
1821 if (ext_depth(inode) != depth) {
1822 /* depth was changed. we have to realloc path */
1823 kfree(path);
1824 path = NULL;
1825 }
1826
1827 block = cbex.ec_block + cbex.ec_len;
1828 }
1829
1830 if (path) {
1831 ext4_ext_drop_refs(path);
1832 kfree(path);
1833 }
1834
1835 return err;
1836}
1837
09b88252 1838static void
725d26d3 1839ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
dd54567a 1840 __u32 len, ext4_fsblk_t start, int type)
a86c6181
AT		 1841{
1842 struct ext4_ext_cache *cex;
1843 BUG_ON(len == 0);
2ec0ae3a 1844 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
a86c6181
AT		 1845 cex = &EXT4_I(inode)->i_cached_extent;
1846 cex->ec_type = type;
1847 cex->ec_block = block;
1848 cex->ec_len = len;
1849 cex->ec_start = start;
2ec0ae3a 1850 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
a86c6181
AT		 1851}
1852
1853/*
d0d856e8
RD		 1854 * ext4_ext_put_gap_in_cache:
1855 * calculate boundaries of the gap that the requested block fits into
a86c6181
AT		 1856 * and cache this gap
1857 */
09b88252 1858static void
a86c6181 1859ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
725d26d3 1860 ext4_lblk_t block)
a86c6181
AT		 1861{
1862 int depth = ext_depth(inode);
725d26d3
AK		 1863 unsigned long len;
1864 ext4_lblk_t lblock;
a86c6181
AT		 1865 struct ext4_extent *ex;
1866
1867 ex = path[depth].p_ext;
1868 if (ex == NULL) {
1869 /* there is no extent yet, so gap is [0;-] */
1870 lblock = 0;
1871 len = EXT_MAX_BLOCK;
1872 ext_debug("cache gap(whole file):");
1873 } else if (block < le32_to_cpu(ex->ee_block)) {
1874 lblock = block;
1875 len = le32_to_cpu(ex->ee_block) - block;
bba90743
ES		 1876 ext_debug("cache gap(before): %u [%u:%u]",
1877 block,
1878 le32_to_cpu(ex->ee_block),
1879 ext4_ext_get_actual_len(ex));
a86c6181 1880 } else if (block >= le32_to_cpu(ex->ee_block)
a2df2a63 1881 + ext4_ext_get_actual_len(ex)) {
725d26d3 1882 ext4_lblk_t next;
8c55e204 1883 lblock = le32_to_cpu(ex->ee_block)
a2df2a63 1884 + ext4_ext_get_actual_len(ex);
725d26d3
AK		 1885
1886 next = ext4_ext_next_allocated_block(path);
bba90743
ES		 1887 ext_debug("cache gap(after): [%u:%u] %u",
1888 le32_to_cpu(ex->ee_block),
1889 ext4_ext_get_actual_len(ex),
1890 block);
725d26d3
AK		 1891 BUG_ON(next == lblock);
1892 len = next - lblock;
a86c6181
AT		 1893 } else {
1894 lblock = len = 0;
1895 BUG();
1896 }
1897
bba90743 1898 ext_debug(" -> %u:%lu\n", lblock, len);
a86c6181
AT		 1899 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1900}
1901
09b88252 1902static int
725d26d3 1903ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
a86c6181
AT		 1904 struct ext4_extent *ex)
1905{
1906 struct ext4_ext_cache *cex;
2ec0ae3a 1907 int ret = EXT4_EXT_CACHE_NO;
a86c6181 1908
2ec0ae3a
TT		 1909 /*
1910 * We borrow i_block_reservation_lock to protect i_cached_extent
1911 */
1912 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
a86c6181
AT		 1913 cex = &EXT4_I(inode)->i_cached_extent;
1914
1915 /* has cache valid data? */
1916 if (cex->ec_type == EXT4_EXT_CACHE_NO)
2ec0ae3a 1917 goto errout;
a86c6181
AT		 1918
1919 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1920 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1921 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
8c55e204 1922 ex->ee_block = cpu_to_le32(cex->ec_block);
f65e6fba 1923 ext4_ext_store_pblock(ex, cex->ec_start);
8c55e204 1924 ex->ee_len = cpu_to_le16(cex->ec_len);
bba90743
ES		 1925 ext_debug("%u cached by %u:%u:%llu\n",
1926 block,
1927 cex->ec_block, cex->ec_len, cex->ec_start);
2ec0ae3a 1928 ret = cex->ec_type;
a86c6181 1929 }
2ec0ae3a
TT		 1930errout:
1931 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1932 return ret;
a86c6181
AT		 1933}
1934
1935/*
d0d856e8
RD		 1936 * ext4_ext_rm_idx:
1937 * removes index from the index block.
1938 * It's used in truncate case only, thus all requests are for
1939 * last index in the block only.
a86c6181 1940 */
1d03ec98 1941static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
a86c6181
AT		 1942 struct ext4_ext_path *path)
1943{
1944 struct buffer_head *bh;
1945 int err;
f65e6fba 1946 ext4_fsblk_t leaf;
a86c6181
AT		 1947
1948 /* free index block */
1949 path--;
f65e6fba 1950 leaf = idx_pblock(path->p_idx);
a86c6181 1951 BUG_ON(path->p_hdr->eh_entries == 0);
7e028976
AM		 1952 err = ext4_ext_get_access(handle, inode, path);
1953 if (err)
a86c6181 1954 return err;
e8546d06 1955 le16_add_cpu(&path->p_hdr->eh_entries, -1);
7e028976
AM		 1956 err = ext4_ext_dirty(handle, inode, path);
1957 if (err)
a86c6181 1958 return err;
2ae02107 1959 ext_debug("index is empty, remove it, free block %llu\n", leaf);
a86c6181
AT		 1960 bh = sb_find_get_block(inode->i_sb, leaf);
1961 ext4_forget(handle, 1, inode, bh, leaf);
c9de560d 1962 ext4_free_blocks(handle, inode, leaf, 1, 1);
a86c6181
AT		 1963 return err;
1964}
1965
1966/*
ee12b630
MC		 1967 * ext4_ext_calc_credits_for_single_extent:
1968 * This routine returns max. credits that needed to insert an extent
1969 * to the extent tree.
1970 * When pass the actual path, the caller should calculate credits
1971 * under i_data_sem.
a86c6181 1972 */
525f4ed8 1973int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
a86c6181
AT		 1974 struct ext4_ext_path *path)
1975{
a86c6181 1976 if (path) {
ee12b630 1977 int depth = ext_depth(inode);
f3bd1f3f 1978 int ret = 0;
ee12b630 1979
a86c6181 1980 /* probably there is space in leaf? */
a86c6181 1981 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
ee12b630 1982 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
a86c6181 1983
ee12b630
MC		 1984 /*
1985 * There are some space in the leaf tree, no
1986 * need to account for leaf block credit
1987 *
1988 * bitmaps and block group descriptor blocks
1989 * and other metadat blocks still need to be
1990 * accounted.
1991 */
525f4ed8 1992 /* 1 bitmap, 1 block group descriptor */
ee12b630
MC		 1993 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1994 }
1995 }
a86c6181 1996
525f4ed8 1997 return ext4_chunk_trans_blocks(inode, nrblocks);
ee12b630 1998}
a86c6181 1999
ee12b630
MC		 2000/*
2001 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2002 *
2003 * if nrblocks are fit in a single extent (chunk flag is 1), then
2004 * in the worse case, each tree level index/leaf need to be changed
2005 * if the tree split due to insert a new extent, then the old tree
2006 * index/leaf need to be updated too
2007 *
2008 * If the nrblocks are discontiguous, they could cause
2009 * the whole tree split more than once, but this is really rare.
2010 */
525f4ed8 2011int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
ee12b630
MC		 2012{
2013 int index;
2014 int depth = ext_depth(inode);
a86c6181 2015
ee12b630
MC		 2016 if (chunk)
2017 index = depth * 2;
2018 else
2019 index = depth * 3;
a86c6181 2020
ee12b630 2021 return index;
a86c6181
AT		 2022}
2023
2024static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2025 struct ext4_extent *ex,
725d26d3 2026 ext4_lblk_t from, ext4_lblk_t to)
a86c6181
AT		 2027{
2028 struct buffer_head *bh;
a2df2a63 2029 unsigned short ee_len = ext4_ext_get_actual_len(ex);
c9de560d 2030 int i, metadata = 0;
a86c6181 2031
c9de560d
AT		 2032 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2033 metadata = 1;
a86c6181
AT		 2034#ifdef EXTENTS_STATS
2035 {
2036 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
a86c6181
AT		 2037 spin_lock(&sbi->s_ext_stats_lock);
2038 sbi->s_ext_blocks += ee_len;
2039 sbi->s_ext_extents++;
2040 if (ee_len < sbi->s_ext_min)
2041 sbi->s_ext_min = ee_len;
2042 if (ee_len > sbi->s_ext_max)
2043 sbi->s_ext_max = ee_len;
2044 if (ext_depth(inode) > sbi->s_depth_max)
2045 sbi->s_depth_max = ext_depth(inode);
2046 spin_unlock(&sbi->s_ext_stats_lock);
2047 }
2048#endif
2049 if (from >= le32_to_cpu(ex->ee_block)
a2df2a63 2050 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
a86c6181 2051 /* tail removal */
725d26d3 2052 ext4_lblk_t num;
f65e6fba 2053 ext4_fsblk_t start;
725d26d3 2054
a2df2a63
AA		 2055 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2056 start = ext_pblock(ex) + ee_len - num;
725d26d3 2057 ext_debug("free last %u blocks starting %llu\n", num, start);
a86c6181
AT		 2058 for (i = 0; i < num; i++) {
2059 bh = sb_find_get_block(inode->i_sb, start + i);
2060 ext4_forget(handle, 0, inode, bh, start + i);
2061 }
c9de560d 2062 ext4_free_blocks(handle, inode, start, num, metadata);
a86c6181 2063 } else if (from == le32_to_cpu(ex->ee_block)
a2df2a63 2064 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
725d26d3 2065 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
a2df2a63 2066 from, to, le32_to_cpu(ex->ee_block), ee_len);
a86c6181 2067 } else {
725d26d3
AK		 2068 printk(KERN_INFO "strange request: removal(2) "
2069 "%u-%u from %u:%u\n",
2070 from, to, le32_to_cpu(ex->ee_block), ee_len);
a86c6181
AT		 2071 }
2072 return 0;
2073}
2074
2075static int
2076ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
725d26d3 2077 struct ext4_ext_path *path, ext4_lblk_t start)
a86c6181
AT		 2078{
2079 int err = 0, correct_index = 0;
2080 int depth = ext_depth(inode), credits;
2081 struct ext4_extent_header *eh;
725d26d3
AK		 2082 ext4_lblk_t a, b, block;
2083 unsigned num;
2084 ext4_lblk_t ex_ee_block;
a86c6181 2085 unsigned short ex_ee_len;
a2df2a63 2086 unsigned uninitialized = 0;
a86c6181
AT		 2087 struct ext4_extent *ex;
2088
c29c0ae7 2089 /* the header must be checked already in ext4_ext_remove_space() */
725d26d3 2090 ext_debug("truncate since %u in leaf\n", start);
a86c6181
AT		 2091 if (!path[depth].p_hdr)
2092 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2093 eh = path[depth].p_hdr;
2094 BUG_ON(eh == NULL);
a86c6181
AT		 2095
2096 /* find where to start removing */
2097 ex = EXT_LAST_EXTENT(eh);
2098
2099 ex_ee_block = le32_to_cpu(ex->ee_block);
a2df2a63
AA		 2100 if (ext4_ext_is_uninitialized(ex))
2101 uninitialized = 1;
2102 ex_ee_len = ext4_ext_get_actual_len(ex);
a86c6181
AT		 2103
2104 while (ex >= EXT_FIRST_EXTENT(eh) &&
2105 ex_ee_block + ex_ee_len > start) {
2106 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
2107 path[depth].p_ext = ex;
2108
2109 a = ex_ee_block > start ? ex_ee_block : start;
2110 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2111 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2112
2113 ext_debug(" border %u:%u\n", a, b);
2114
2115 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2116 block = 0;
2117 num = 0;
2118 BUG();
2119 } else if (a != ex_ee_block) {
2120 /* remove tail of the extent */
2121 block = ex_ee_block;
2122 num = a - block;
2123 } else if (b != ex_ee_block + ex_ee_len - 1) {
2124 /* remove head of the extent */
2125 block = a;
2126 num = b - a;
2127 /* there is no "make a hole" API yet */
2128 BUG();
2129 } else {
2130 /* remove whole extent: excellent! */
2131 block = ex_ee_block;
2132 num = 0;
2133 BUG_ON(a != ex_ee_block);
2134 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2135 }
2136
34071da7
TT		 2137 /*
2138 * 3 for leaf, sb, and inode plus 2 (bmap and group
2139 * descriptor) for each block group; assume two block
2140 * groups plus ex_ee_len/blocks_per_block_group for
2141 * the worst case
2142 */
2143 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
a86c6181
AT		 2144 if (ex == EXT_FIRST_EXTENT(eh)) {
2145 correct_index = 1;
2146 credits += (ext_depth(inode)) + 1;
2147 }
a86c6181 2148 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
a86c6181 2149
9102e4fa
SF		 2150 err = ext4_ext_journal_restart(handle, credits);
2151 if (err)
a86c6181 2152 goto out;
a86c6181
AT		 2153
2154 err = ext4_ext_get_access(handle, inode, path + depth);
2155 if (err)
2156 goto out;
2157
2158 err = ext4_remove_blocks(handle, inode, ex, a, b);
2159 if (err)
2160 goto out;
2161
2162 if (num == 0) {
d0d856e8 2163 /* this extent is removed; mark slot entirely unused */
f65e6fba 2164 ext4_ext_store_pblock(ex, 0);
e8546d06 2165 le16_add_cpu(&eh->eh_entries, -1);
a86c6181
AT		 2166 }
2167
2168 ex->ee_block = cpu_to_le32(block);
2169 ex->ee_len = cpu_to_le16(num);
749269fa
AA		 2170 /*
2171 * Do not mark uninitialized if all the blocks in the
2172 * extent have been removed.
2173 */
2174 if (uninitialized && num)
a2df2a63 2175 ext4_ext_mark_uninitialized(ex);
a86c6181
AT		 2176
2177 err = ext4_ext_dirty(handle, inode, path + depth);
2178 if (err)
2179 goto out;
2180
2ae02107 2181 ext_debug("new extent: %u:%u:%llu\n", block, num,
f65e6fba 2182 ext_pblock(ex));
a86c6181
AT		 2183 ex--;
2184 ex_ee_block = le32_to_cpu(ex->ee_block);
a2df2a63 2185 ex_ee_len = ext4_ext_get_actual_len(ex);
a86c6181
AT		 2186 }
2187
2188 if (correct_index && eh->eh_entries)
2189 err = ext4_ext_correct_indexes(handle, inode, path);
2190
2191 /* if this leaf is free, then we should
2192 * remove it from index block above */
2193 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2194 err = ext4_ext_rm_idx(handle, inode, path + depth);
2195
2196out:
2197 return err;
2198}
2199
2200/*
d0d856e8
RD		 2201 * ext4_ext_more_to_rm:
2202 * returns 1 if current index has to be freed (even partial)
a86c6181 2203 */
09b88252 2204static int
a86c6181
AT		 2205ext4_ext_more_to_rm(struct ext4_ext_path *path)
2206{
2207 BUG_ON(path->p_idx == NULL);
2208
2209 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2210 return 0;
2211
2212 /*
d0d856e8 2213 * if truncate on deeper level happened, it wasn't partial,
a86c6181
AT		 2214 * so we have to consider current index for truncation
2215 */
2216 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2217 return 0;
2218 return 1;
2219}
2220
1d03ec98 2221static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
a86c6181
AT		 2222{
2223 struct super_block *sb = inode->i_sb;
2224 int depth = ext_depth(inode);
2225 struct ext4_ext_path *path;
2226 handle_t *handle;
2227 int i = 0, err = 0;
2228
725d26d3 2229 ext_debug("truncate since %u\n", start);
a86c6181
AT		 2230
2231 /* probably first extent we're gonna free will be last in block */
2232 handle = ext4_journal_start(inode, depth + 1);
2233 if (IS_ERR(handle))
2234 return PTR_ERR(handle);
2235
2236 ext4_ext_invalidate_cache(inode);
2237
2238 /*
d0d856e8
RD		 2239 * We start scanning from right side, freeing all the blocks
2240 * after i_size and walking into the tree depth-wise.
a86c6181 2241 */
216553c4 2242 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
a86c6181
AT		 2243 if (path == NULL) {
2244 ext4_journal_stop(handle);
2245 return -ENOMEM;
2246 }
a86c6181 2247 path[0].p_hdr = ext_inode_hdr(inode);
56b19868 2248 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
a86c6181
AT		 2249 err = -EIO;
2250 goto out;
2251 }
2252 path[0].p_depth = depth;
2253
2254 while (i >= 0 && err == 0) {
2255 if (i == depth) {
2256 /* this is leaf block */
2257 err = ext4_ext_rm_leaf(handle, inode, path, start);
d0d856e8 2258 /* root level has p_bh == NULL, brelse() eats this */
a86c6181
AT		 2259 brelse(path[i].p_bh);
2260 path[i].p_bh = NULL;
2261 i--;
2262 continue;
2263 }
2264
2265 /* this is index block */
2266 if (!path[i].p_hdr) {
2267 ext_debug("initialize header\n");
2268 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
a86c6181
AT		 2269 }
2270
a86c6181 2271 if (!path[i].p_idx) {
d0d856e8 2272 /* this level hasn't been touched yet */
a86c6181
AT		 2273 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2274 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2275 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2276 path[i].p_hdr,
2277 le16_to_cpu(path[i].p_hdr->eh_entries));
2278 } else {
d0d856e8 2279 /* we were already here, see at next index */
a86c6181
AT		 2280 path[i].p_idx--;
2281 }
2282
2283 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2284 i, EXT_FIRST_INDEX(path[i].p_hdr),
2285 path[i].p_idx);
2286 if (ext4_ext_more_to_rm(path + i)) {
c29c0ae7 2287 struct buffer_head *bh;
a86c6181 2288 /* go to the next level */
2ae02107 2289 ext_debug("move to level %d (block %llu)\n",
f65e6fba 2290 i + 1, idx_pblock(path[i].p_idx));
a86c6181 2291 memset(path + i + 1, 0, sizeof(*path));
c29c0ae7
AT		 2292 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2293 if (!bh) {
a86c6181
AT		 2294 /* should we reset i_size? */
2295 err = -EIO;
2296 break;
2297 }
c29c0ae7
AT		 2298 if (WARN_ON(i + 1 > depth)) {
2299 err = -EIO;
2300 break;
2301 }
56b19868 2302 if (ext4_ext_check(inode, ext_block_hdr(bh),
c29c0ae7
AT		 2303 depth - i - 1)) {
2304 err = -EIO;
2305 break;
2306 }
2307 path[i + 1].p_bh = bh;
a86c6181 2308
d0d856e8
RD		 2309 /* save actual number of indexes since this
2310 * number is changed at the next iteration */
a86c6181
AT		 2311 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2312 i++;
2313 } else {
d0d856e8 2314 /* we finished processing this index, go up */
a86c6181 2315 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
d0d856e8 2316 /* index is empty, remove it;
a86c6181
AT		 2317 * handle must be already prepared by the
2318 * truncatei_leaf() */
2319 err = ext4_ext_rm_idx(handle, inode, path + i);
2320 }
d0d856e8 2321 /* root level has p_bh == NULL, brelse() eats this */
a86c6181
AT		 2322 brelse(path[i].p_bh);
2323 path[i].p_bh = NULL;
2324 i--;
2325 ext_debug("return to level %d\n", i);
2326 }
2327 }
2328
2329 /* TODO: flexible tree reduction should be here */
2330 if (path->p_hdr->eh_entries == 0) {
2331 /*
d0d856e8
RD		 2332 * truncate to zero freed all the tree,
2333 * so we need to correct eh_depth
a86c6181
AT		 2334 */
2335 err = ext4_ext_get_access(handle, inode, path);
2336 if (err == 0) {
2337 ext_inode_hdr(inode)->eh_depth = 0;
2338 ext_inode_hdr(inode)->eh_max =
2339 cpu_to_le16(ext4_ext_space_root(inode));
2340 err = ext4_ext_dirty(handle, inode, path);
2341 }
2342 }
2343out:
a86c6181
AT		 2344 ext4_ext_drop_refs(path);
2345 kfree(path);
2346 ext4_journal_stop(handle);
2347
2348 return err;
2349}
2350
2351/*
2352 * called at mount time
2353 */
2354void ext4_ext_init(struct super_block *sb)
2355{
2356 /*
2357 * possible initialization would be here
2358 */
2359
83982b6f 2360 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
4776004f 2361 printk(KERN_INFO "EXT4-fs: file extents enabled");
bbf2f9fb
RD		 2362#ifdef AGGRESSIVE_TEST
2363 printk(", aggressive tests");
a86c6181
AT		 2364#endif
2365#ifdef CHECK_BINSEARCH
2366 printk(", check binsearch");
2367#endif
2368#ifdef EXTENTS_STATS
2369 printk(", stats");
2370#endif
2371 printk("\n");
2372#ifdef EXTENTS_STATS
2373 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2374 EXT4_SB(sb)->s_ext_min = 1 << 30;
2375 EXT4_SB(sb)->s_ext_max = 0;
2376#endif
2377 }
2378}
2379
2380/*
2381 * called at umount time
2382 */
2383void ext4_ext_release(struct super_block *sb)
2384{
83982b6f 2385 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
a86c6181
AT		 2386 return;
2387
2388#ifdef EXTENTS_STATS
2389 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2390 struct ext4_sb_info *sbi = EXT4_SB(sb);
2391 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2392 sbi->s_ext_blocks, sbi->s_ext_extents,
2393 sbi->s_ext_blocks / sbi->s_ext_extents);
2394 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2395 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2396 }
2397#endif
2398}
2399
093a088b
AK		 2400static void bi_complete(struct bio *bio, int error)
2401{
2402 complete((struct completion *)bio->bi_private);
2403}
2404
2405/* FIXME!! we need to try to merge to left or right after zero-out */
2406static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2407{
2408 int ret = -EIO;
2409 struct bio *bio;
2410 int blkbits, blocksize;
2411 sector_t ee_pblock;
2412 struct completion event;
2413 unsigned int ee_len, len, done, offset;
2414
2415
2416 blkbits = inode->i_blkbits;
2417 blocksize = inode->i_sb->s_blocksize;
2418 ee_len = ext4_ext_get_actual_len(ex);
2419 ee_pblock = ext_pblock(ex);
2420
2421 /* convert ee_pblock to 512 byte sectors */
2422 ee_pblock = ee_pblock << (blkbits - 9);
2423
2424 while (ee_len > 0) {
2425
2426 if (ee_len > BIO_MAX_PAGES)
2427 len = BIO_MAX_PAGES;
2428 else
2429 len = ee_len;
2430
2431 bio = bio_alloc(GFP_NOIO, len);
093a088b
AK		 2432 bio->bi_sector = ee_pblock;
2433 bio->bi_bdev = inode->i_sb->s_bdev;
2434
2435 done = 0;
2436 offset = 0;
2437 while (done < len) {
2438 ret = bio_add_page(bio, ZERO_PAGE(0),
2439 blocksize, offset);
2440 if (ret != blocksize) {
2441 /*
2442 * We can't add any more pages because of
2443 * hardware limitations. Start a new bio.
2444 */
2445 break;
2446 }
2447 done++;
2448 offset += blocksize;
2449 if (offset >= PAGE_CACHE_SIZE)
2450 offset = 0;
2451 }
2452
2453 init_completion(&event);
2454 bio->bi_private = &event;
2455 bio->bi_end_io = bi_complete;
2456 submit_bio(WRITE, bio);
2457 wait_for_completion(&event);
2458
2459 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2460 ret = 0;
2461 else {
2462 ret = -EIO;
2463 break;
2464 }
2465 bio_put(bio);
2466 ee_len -= done;
2467 ee_pblock += done << (blkbits - 9);
2468 }
2469 return ret;
2470}
2471
3977c965
AK		 2472#define EXT4_EXT_ZERO_LEN 7
2473
56055d3a
AA		 2474/*
2475 * This function is called by ext4_ext_get_blocks() if someone tries to write
2476 * to an uninitialized extent. It may result in splitting the uninitialized
2477 * extent into multiple extents (upto three - one initialized and two
2478 * uninitialized).
2479 * There are three possibilities:
2480 * a> There is no split required: Entire extent should be initialized
2481 * b> Splits in two extents: Write is happening at either end of the extent
2482 * c> Splits in three extents: Somone is writing in middle of the extent
2483 */
725d26d3
AK		 2484static int ext4_ext_convert_to_initialized(handle_t *handle,
2485 struct inode *inode,
2486 struct ext4_ext_path *path,
2487 ext4_lblk_t iblock,
498e5f24 2488 unsigned int max_blocks)
56055d3a 2489{
95c3889c 2490 struct ext4_extent *ex, newex, orig_ex;
56055d3a
AA		 2491 struct ext4_extent *ex1 = NULL;
2492 struct ext4_extent *ex2 = NULL;
2493 struct ext4_extent *ex3 = NULL;
2494 struct ext4_extent_header *eh;
725d26d3
AK		 2495 ext4_lblk_t ee_block;
2496 unsigned int allocated, ee_len, depth;
56055d3a
AA		 2497 ext4_fsblk_t newblock;
2498 int err = 0;
2499 int ret = 0;
2500
2501 depth = ext_depth(inode);
2502 eh = path[depth].p_hdr;
2503 ex = path[depth].p_ext;
2504 ee_block = le32_to_cpu(ex->ee_block);
2505 ee_len = ext4_ext_get_actual_len(ex);
2506 allocated = ee_len - (iblock - ee_block);
2507 newblock = iblock - ee_block + ext_pblock(ex);
2508 ex2 = ex;
95c3889c
AK		 2509 orig_ex.ee_block = ex->ee_block;
2510 orig_ex.ee_len = cpu_to_le16(ee_len);
2511 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
56055d3a 2512
9df5643a
AK		 2513 err = ext4_ext_get_access(handle, inode, path + depth);
2514 if (err)
2515 goto out;
3977c965
AK		 2516 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2517 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2518 err = ext4_ext_zeroout(inode, &orig_ex);
2519 if (err)
2520 goto fix_extent_len;
2521 /* update the extent length and mark as initialized */
2522 ex->ee_block = orig_ex.ee_block;
2523 ex->ee_len = orig_ex.ee_len;
2524 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2525 ext4_ext_dirty(handle, inode, path + depth);
161e7b7c
AK		 2526 /* zeroed the full extent */
2527 return allocated;
3977c965 2528 }
9df5643a 2529
56055d3a
AA		 2530 /* ex1: ee_block to iblock - 1 : uninitialized */
2531 if (iblock > ee_block) {
2532 ex1 = ex;
2533 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2534 ext4_ext_mark_uninitialized(ex1);
2535 ex2 = &newex;
2536 }
2537 /*
2538 * for sanity, update the length of the ex2 extent before
2539 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2540 * overlap of blocks.
2541 */
2542 if (!ex1 && allocated > max_blocks)
2543 ex2->ee_len = cpu_to_le16(max_blocks);
2544 /* ex3: to ee_block + ee_len : uninitialised */
2545 if (allocated > max_blocks) {
2546 unsigned int newdepth;
3977c965
AK		 2547 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2548 if (allocated <= EXT4_EXT_ZERO_LEN) {
d03856bd
AK		 2549 /*
2550 * iblock == ee_block is handled by the zerouout
2551 * at the beginning.
2552 * Mark first half uninitialized.
3977c965
AK		 2553 * Mark second half initialized and zero out the
2554 * initialized extent
2555 */
2556 ex->ee_block = orig_ex.ee_block;
2557 ex->ee_len = cpu_to_le16(ee_len - allocated);
2558 ext4_ext_mark_uninitialized(ex);
2559 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2560 ext4_ext_dirty(handle, inode, path + depth);
2561
2562 ex3 = &newex;
2563 ex3->ee_block = cpu_to_le32(iblock);
2564 ext4_ext_store_pblock(ex3, newblock);
2565 ex3->ee_len = cpu_to_le16(allocated);
2566 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2567 if (err == -ENOSPC) {
2568 err = ext4_ext_zeroout(inode, &orig_ex);
2569 if (err)
2570 goto fix_extent_len;
2571 ex->ee_block = orig_ex.ee_block;
2572 ex->ee_len = orig_ex.ee_len;
2573 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2574 ext4_ext_dirty(handle, inode, path + depth);
d03856bd 2575 /* blocks available from iblock */
161e7b7c 2576 return allocated;
3977c965
AK		 2577
2578 } else if (err)
2579 goto fix_extent_len;
2580
161e7b7c
AK		 2581 /*
2582 * We need to zero out the second half because
2583 * an fallocate request can update file size and
2584 * converting the second half to initialized extent
2585 * implies that we can leak some junk data to user
2586 * space.
2587 */
2588 err = ext4_ext_zeroout(inode, ex3);
2589 if (err) {
2590 /*
2591 * We should actually mark the
2592 * second half as uninit and return error
2593 * Insert would have changed the extent
2594 */
2595 depth = ext_depth(inode);
2596 ext4_ext_drop_refs(path);
2597 path = ext4_ext_find_extent(inode,
2598 iblock, path);
2599 if (IS_ERR(path)) {
2600 err = PTR_ERR(path);
2601 return err;
2602 }
d03856bd 2603 /* get the second half extent details */
161e7b7c
AK		 2604 ex = path[depth].p_ext;
2605 err = ext4_ext_get_access(handle, inode,
2606 path + depth);
2607 if (err)
2608 return err;
2609 ext4_ext_mark_uninitialized(ex);
2610 ext4_ext_dirty(handle, inode, path + depth);
2611 return err;
2612 }
2613
2614 /* zeroed the second half */
3977c965
AK		 2615 return allocated;
2616 }
56055d3a
AA		 2617 ex3 = &newex;
2618 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2619 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2620 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2621 ext4_ext_mark_uninitialized(ex3);
2622 err = ext4_ext_insert_extent(handle, inode, path, ex3);
093a088b
AK		 2623 if (err == -ENOSPC) {
2624 err = ext4_ext_zeroout(inode, &orig_ex);
2625 if (err)
2626 goto fix_extent_len;
2627 /* update the extent length and mark as initialized */
95c3889c
AK		 2628 ex->ee_block = orig_ex.ee_block;
2629 ex->ee_len = orig_ex.ee_len;
2630 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
95c3889c 2631 ext4_ext_dirty(handle, inode, path + depth);
161e7b7c 2632 /* zeroed the full extent */
d03856bd 2633 /* blocks available from iblock */
161e7b7c 2634 return allocated;
093a088b
AK		 2635
2636 } else if (err)
2637 goto fix_extent_len;
56055d3a
AA		 2638 /*
2639 * The depth, and hence eh & ex might change
2640 * as part of the insert above.
2641 */
2642 newdepth = ext_depth(inode);
95c3889c 2643 /*
73ac36ea 2644 * update the extent length after successful insert of the
95c3889c
AK		 2645 * split extent
2646 */
2647 orig_ex.ee_len = cpu_to_le16(ee_len -
2648 ext4_ext_get_actual_len(ex3));
d03856bd
AK		 2649 depth = newdepth;
2650 ext4_ext_drop_refs(path);
2651 path = ext4_ext_find_extent(inode, iblock, path);
2652 if (IS_ERR(path)) {
2653 err = PTR_ERR(path);
2654 goto out;
56055d3a 2655 }
d03856bd
AK		 2656 eh = path[depth].p_hdr;
2657 ex = path[depth].p_ext;
2658 if (ex2 != &newex)
2659 ex2 = ex;
2660
2661 err = ext4_ext_get_access(handle, inode, path + depth);
2662 if (err)
2663 goto out;
2664
56055d3a 2665 allocated = max_blocks;
3977c965
AK		 2666
2667 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2668 * to insert a extent in the middle zerout directly
2669 * otherwise give the extent a chance to merge to left
2670 */
2671 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2672 iblock != ee_block) {
2673 err = ext4_ext_zeroout(inode, &orig_ex);
2674 if (err)
2675 goto fix_extent_len;
2676 /* update the extent length and mark as initialized */
2677 ex->ee_block = orig_ex.ee_block;
2678 ex->ee_len = orig_ex.ee_len;
2679 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2680 ext4_ext_dirty(handle, inode, path + depth);
161e7b7c 2681 /* zero out the first half */
d03856bd 2682 /* blocks available from iblock */
161e7b7c 2683 return allocated;
3977c965 2684 }
56055d3a
AA		 2685 }
2686 /*
2687 * If there was a change of depth as part of the
2688 * insertion of ex3 above, we need to update the length
2689 * of the ex1 extent again here
2690 */
2691 if (ex1 && ex1 != ex) {
2692 ex1 = ex;
2693 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2694 ext4_ext_mark_uninitialized(ex1);
2695 ex2 = &newex;
2696 }
2697 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2698 ex2->ee_block = cpu_to_le32(iblock);
56055d3a
AA		 2699 ext4_ext_store_pblock(ex2, newblock);
2700 ex2->ee_len = cpu_to_le16(allocated);
2701 if (ex2 != ex)
2702 goto insert;
56055d3a
AA		 2703 /*
2704 * New (initialized) extent starts from the first block
2705 * in the current extent. i.e., ex2 == ex
2706 * We have to see if it can be merged with the extent
2707 * on the left.
2708 */
2709 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2710 /*
2711 * To merge left, pass "ex2 - 1" to try_to_merge(),
2712 * since it merges towards right _only_.
2713 */
2714 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2715 if (ret) {
2716 err = ext4_ext_correct_indexes(handle, inode, path);
2717 if (err)
2718 goto out;
2719 depth = ext_depth(inode);
2720 ex2--;
2721 }
2722 }
2723 /*
2724 * Try to Merge towards right. This might be required
2725 * only when the whole extent is being written to.
2726 * i.e. ex2 == ex and ex3 == NULL.
2727 */
2728 if (!ex3) {
2729 ret = ext4_ext_try_to_merge(inode, path, ex2);
2730 if (ret) {
2731 err = ext4_ext_correct_indexes(handle, inode, path);
2732 if (err)
2733 goto out;
2734 }
2735 }
2736 /* Mark modified extent as dirty */
2737 err = ext4_ext_dirty(handle, inode, path + depth);
2738 goto out;
2739insert:
2740 err = ext4_ext_insert_extent(handle, inode, path, &newex);
093a088b
AK		 2741 if (err == -ENOSPC) {
2742 err = ext4_ext_zeroout(inode, &orig_ex);
2743 if (err)
2744 goto fix_extent_len;
2745 /* update the extent length and mark as initialized */
95c3889c
AK		 2746 ex->ee_block = orig_ex.ee_block;
2747 ex->ee_len = orig_ex.ee_len;
2748 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
95c3889c 2749 ext4_ext_dirty(handle, inode, path + depth);
161e7b7c
AK		 2750 /* zero out the first half */
2751 return allocated;
093a088b
AK		 2752 } else if (err)
2753 goto fix_extent_len;
56055d3a
AA		 2754out:
2755 return err ? err : allocated;
093a088b
AK		 2756
2757fix_extent_len:
2758 ex->ee_block = orig_ex.ee_block;
2759 ex->ee_len = orig_ex.ee_len;
2760 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2761 ext4_ext_mark_uninitialized(ex);
2762 ext4_ext_dirty(handle, inode, path + depth);
2763 return err;
56055d3a
AA		 2764}
2765
c278bfec 2766/*
f5ab0d1f
MC		 2767 * Block allocation/map/preallocation routine for extents based files
2768 *
2769 *
c278bfec 2770 * Need to be called with
0e855ac8
AK		 2771 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2772 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
f5ab0d1f
MC		 2773 *
2774 * return > 0, number of of blocks already mapped/allocated
2775 * if create == 0 and these are pre-allocated blocks
2776 * buffer head is unmapped
2777 * otherwise blocks are mapped
2778 *
2779 * return = 0, if plain look up failed (blocks have not been allocated)
2780 * buffer head is unmapped
2781 *
2782 * return < 0, error case.
c278bfec 2783 */
f65e6fba 2784int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
725d26d3 2785 ext4_lblk_t iblock,
498e5f24 2786 unsigned int max_blocks, struct buffer_head *bh_result,
a86c6181
AT		 2787 int create, int extend_disksize)
2788{
2789 struct ext4_ext_path *path = NULL;
56055d3a 2790 struct ext4_extent_header *eh;
a86c6181 2791 struct ext4_extent newex, *ex;
498e5f24
TT		 2792 ext4_fsblk_t newblock;
2793 int err = 0, depth, ret, cache_type;
2794 unsigned int allocated = 0;
c9de560d 2795 struct ext4_allocation_request ar;
61628a3f 2796 loff_t disksize;
a86c6181
AT		 2797
2798 __clear_bit(BH_New, &bh_result->b_state);
498e5f24 2799 ext_debug("blocks %u/%u requested for inode %u\n",
bba90743 2800 iblock, max_blocks, inode->i_ino);
a86c6181
AT		 2801
2802 /* check in cache */
498e5f24
TT		 2803 cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2804 if (cache_type) {
2805 if (cache_type == EXT4_EXT_CACHE_GAP) {
a86c6181 2806 if (!create) {
56055d3a
AA		 2807 /*
2808 * block isn't allocated yet and
2809 * user doesn't want to allocate it
2810 */
a86c6181
AT		 2811 goto out2;
2812 }
2813 /* we should allocate requested block */
498e5f24 2814 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
a86c6181 2815 /* block is already allocated */
8c55e204
DK		 2816 newblock = iblock
2817 - le32_to_cpu(newex.ee_block)
2818 + ext_pblock(&newex);
d0d856e8 2819 /* number of remaining blocks in the extent */
b939e376 2820 allocated = ext4_ext_get_actual_len(&newex) -
a86c6181
AT		 2821 (iblock - le32_to_cpu(newex.ee_block));
2822 goto out;
2823 } else {
2824 BUG();
2825 }
2826 }
2827
2828 /* find extent for this block */
2829 path = ext4_ext_find_extent(inode, iblock, NULL);
2830 if (IS_ERR(path)) {
2831 err = PTR_ERR(path);
2832 path = NULL;
2833 goto out2;
2834 }
2835
2836 depth = ext_depth(inode);
2837
2838 /*
d0d856e8
RD		 2839 * consistent leaf must not be empty;
2840 * this situation is possible, though, _during_ tree modification;
a86c6181
AT		 2841 * this is why assert can't be put in ext4_ext_find_extent()
2842 */
2843 BUG_ON(path[depth].p_ext == NULL && depth != 0);
56055d3a 2844 eh = path[depth].p_hdr;
a86c6181 2845
7e028976
AM		 2846 ex = path[depth].p_ext;
2847 if (ex) {
725d26d3 2848 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
f65e6fba 2849 ext4_fsblk_t ee_start = ext_pblock(ex);
a2df2a63 2850 unsigned short ee_len;
471d4011
SB		 2851
2852 /*
471d4011 2853 * Uninitialized extents are treated as holes, except that
56055d3a 2854 * we split out initialized portions during a write.
471d4011 2855 */
a2df2a63 2856 ee_len = ext4_ext_get_actual_len(ex);
d0d856e8 2857 /* if found extent covers block, simply return it */
8c55e204 2858 if (iblock >= ee_block && iblock < ee_block + ee_len) {
a86c6181 2859 newblock = iblock - ee_block + ee_start;
d0d856e8 2860 /* number of remaining blocks in the extent */
a86c6181 2861 allocated = ee_len - (iblock - ee_block);
bba90743 2862 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
a86c6181 2863 ee_block, ee_len, newblock);
56055d3a 2864
a2df2a63 2865 /* Do not put uninitialized extent in the cache */
56055d3a 2866 if (!ext4_ext_is_uninitialized(ex)) {
a2df2a63
AA		 2867 ext4_ext_put_in_cache(inode, ee_block,
2868 ee_len, ee_start,
2869 EXT4_EXT_CACHE_EXTENT);
56055d3a
AA		 2870 goto out;
2871 }
2872 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2873 goto out;
e067ba00 2874 if (!create) {
29fa89d0
AK		 2875 if (allocated > max_blocks)
2876 allocated = max_blocks;
e067ba00
AK		 2877 /*
2878 * We have blocks reserved already. We
2879 * return allocated blocks so that delalloc
2880 * won't do block reservation for us. But
2881 * the buffer head will be unmapped so that
2882 * a read from the block returns 0s.
2883 */
953e622b 2884 set_buffer_unwritten(bh_result);
9c1ee184
AK		 2885 bh_result->b_bdev = inode->i_sb->s_bdev;
2886 bh_result->b_blocknr = newblock;
56055d3a 2887 goto out2;
e067ba00 2888 }
56055d3a
AA		 2889
2890 ret = ext4_ext_convert_to_initialized(handle, inode,
2891 path, iblock,
2892 max_blocks);
dbf9d7da
DM		 2893 if (ret <= 0) {
2894 err = ret;
56055d3a 2895 goto out2;
dbf9d7da 2896 } else
56055d3a
AA		 2897 allocated = ret;
2898 goto outnew;
a86c6181
AT		 2899 }
2900 }
2901
2902 /*
d0d856e8 2903 * requested block isn't allocated yet;
a86c6181
AT		 2904 * we couldn't try to create block if create flag is zero
2905 */
2906 if (!create) {
56055d3a
AA		 2907 /*
2908 * put just found gap into cache to speed up
2909 * subsequent requests
2910 */
a86c6181
AT		 2911 ext4_ext_put_gap_in_cache(inode, path, iblock);
2912 goto out2;
2913 }
2914 /*
c2ea3fde 2915 * Okay, we need to do block allocation.
63f57933 2916 */
a86c6181 2917
c9de560d
AT		 2918 /* find neighbour allocated blocks */
2919 ar.lleft = iblock;
2920 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2921 if (err)
2922 goto out2;
2923 ar.lright = iblock;
2924 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2925 if (err)
2926 goto out2;
25d14f98 2927
749269fa
AA		 2928 /*
2929 * See if request is beyond maximum number of blocks we can have in
2930 * a single extent. For an initialized extent this limit is
2931 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2932 * EXT_UNINIT_MAX_LEN.
2933 */
2934 if (max_blocks > EXT_INIT_MAX_LEN &&
2935 create != EXT4_CREATE_UNINITIALIZED_EXT)
2936 max_blocks = EXT_INIT_MAX_LEN;
2937 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2938 create == EXT4_CREATE_UNINITIALIZED_EXT)
2939 max_blocks = EXT_UNINIT_MAX_LEN;
2940
25d14f98
AA		 2941 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2942 newex.ee_block = cpu_to_le32(iblock);
2943 newex.ee_len = cpu_to_le16(max_blocks);
2944 err = ext4_ext_check_overlap(inode, &newex, path);
2945 if (err)
b939e376 2946 allocated = ext4_ext_get_actual_len(&newex);
25d14f98
AA		 2947 else
2948 allocated = max_blocks;
c9de560d
AT		 2949
2950 /* allocate new block */
2951 ar.inode = inode;
2952 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2953 ar.logical = iblock;
2954 ar.len = allocated;
2955 if (S_ISREG(inode->i_mode))
2956 ar.flags = EXT4_MB_HINT_DATA;
2957 else
2958 /* disable in-core preallocation for non-regular files */
2959 ar.flags = 0;
2960 newblock = ext4_mb_new_blocks(handle, &ar, &err);
a86c6181
AT		 2961 if (!newblock)
2962 goto out2;
2ae02107 2963 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
498e5f24 2964 ar.goal, newblock, allocated);
a86c6181
AT		 2965
2966 /* try to insert new extent into found leaf and return */
f65e6fba 2967 ext4_ext_store_pblock(&newex, newblock);
c9de560d 2968 newex.ee_len = cpu_to_le16(ar.len);
a2df2a63
AA		 2969 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2970 ext4_ext_mark_uninitialized(&newex);
a86c6181 2971 err = ext4_ext_insert_extent(handle, inode, path, &newex);
315054f0
AT		 2972 if (err) {
2973 /* free data blocks we just allocated */
c9de560d
AT		 2974 /* not a good idea to call discard here directly,
2975 * but otherwise we'd need to call it every free() */
c2ea3fde 2976 ext4_discard_preallocations(inode);
315054f0 2977 ext4_free_blocks(handle, inode, ext_pblock(&newex),
b939e376 2978 ext4_ext_get_actual_len(&newex), 0);
a86c6181 2979 goto out2;
315054f0 2980 }
a86c6181 2981
a86c6181 2982 /* previous routine could use block we allocated */
f65e6fba 2983 newblock = ext_pblock(&newex);
b939e376 2984 allocated = ext4_ext_get_actual_len(&newex);
56055d3a 2985outnew:
61628a3f
MC		 2986 if (extend_disksize) {
2987 disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits;
2988 if (disksize > i_size_read(inode))
2989 disksize = i_size_read(inode);
2990 if (disksize > EXT4_I(inode)->i_disksize)
2991 EXT4_I(inode)->i_disksize = disksize;
2992 }
a379cd1d 2993
953e622b 2994 set_buffer_new(bh_result);
a86c6181 2995
a2df2a63
AA		 2996 /* Cache only when it is _not_ an uninitialized extent */
2997 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2998 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2999 EXT4_EXT_CACHE_EXTENT);
a86c6181
AT		 3000out:
3001 if (allocated > max_blocks)
3002 allocated = max_blocks;
3003 ext4_ext_show_leaf(inode, path);
953e622b 3004 set_buffer_mapped(bh_result);
a86c6181
AT		 3005 bh_result->b_bdev = inode->i_sb->s_bdev;
3006 bh_result->b_blocknr = newblock;
3007out2:
3008 if (path) {
3009 ext4_ext_drop_refs(path);
3010 kfree(path);
3011 }
a86c6181
AT		 3012 return err ? err : allocated;
3013}
3014
cf108bca 3015void ext4_ext_truncate(struct inode *inode)
a86c6181
AT		 3016{
3017 struct address_space *mapping = inode->i_mapping;
3018 struct super_block *sb = inode->i_sb;
725d26d3 3019 ext4_lblk_t last_block;
a86c6181
AT		 3020 handle_t *handle;
3021 int err = 0;
3022
3023 /*
3024 * probably first extent we're gonna free will be last in block
3025 */
f3bd1f3f 3026 err = ext4_writepage_trans_blocks(inode);
a86c6181 3027 handle = ext4_journal_start(inode, err);
cf108bca 3028 if (IS_ERR(handle))
a86c6181 3029 return;
a86c6181 3030
cf108bca
JK		 3031 if (inode->i_size & (sb->s_blocksize - 1))
3032 ext4_block_truncate_page(handle, mapping, inode->i_size);
a86c6181 3033
9ddfc3dc
JK		 3034 if (ext4_orphan_add(handle, inode))
3035 goto out_stop;
3036
0e855ac8 3037 down_write(&EXT4_I(inode)->i_data_sem);
a86c6181
AT		 3038 ext4_ext_invalidate_cache(inode);
3039
c2ea3fde 3040 ext4_discard_preallocations(inode);
c9de560d 3041
a86c6181 3042 /*
d0d856e8
RD		 3043 * TODO: optimization is possible here.
3044 * Probably we need not scan at all,
3045 * because page truncation is enough.
a86c6181 3046 */
a86c6181
AT		 3047
3048 /* we have to know where to truncate from in crash case */
3049 EXT4_I(inode)->i_disksize = inode->i_size;
3050 ext4_mark_inode_dirty(handle, inode);
3051
3052 last_block = (inode->i_size + sb->s_blocksize - 1)
3053 >> EXT4_BLOCK_SIZE_BITS(sb);
3054 err = ext4_ext_remove_space(inode, last_block);
3055
3056 /* In a multi-transaction truncate, we only make the final
56055d3a
AA		 3057 * transaction synchronous.
3058 */
a86c6181 3059 if (IS_SYNC(inode))
0390131b 3060 ext4_handle_sync(handle);
a86c6181
AT		 3061
3062out_stop:
9ddfc3dc 3063 up_write(&EXT4_I(inode)->i_data_sem);
a86c6181 3064 /*
d0d856e8 3065 * If this was a simple ftruncate() and the file will remain alive,
a86c6181
AT		 3066 * then we need to clear up the orphan record which we created above.
3067 * However, if this was a real unlink then we were called by
3068 * ext4_delete_inode(), and we allow that function to clean up the
3069 * orphan info for us.
3070 */
3071 if (inode->i_nlink)
3072 ext4_orphan_del(handle, inode);
3073
ef737728
SR		 3074 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3075 ext4_mark_inode_dirty(handle, inode);
a86c6181
AT		 3076 ext4_journal_stop(handle);
3077}
3078
fd28784a
AK		 3079static void ext4_falloc_update_inode(struct inode *inode,
3080 int mode, loff_t new_size, int update_ctime)
3081{
3082 struct timespec now;
3083
3084 if (update_ctime) {
3085 now = current_fs_time(inode->i_sb);
3086 if (!timespec_equal(&inode->i_ctime, &now))
3087 inode->i_ctime = now;
3088 }
3089 /*
3090 * Update only when preallocation was requested beyond
3091 * the file size.
3092 */
cf17fea6
AK		 3093 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3094 if (new_size > i_size_read(inode))
3095 i_size_write(inode, new_size);
3096 if (new_size > EXT4_I(inode)->i_disksize)
3097 ext4_update_i_disksize(inode, new_size);
fd28784a
AK		 3098 }
3099
3100}
3101
a2df2a63
AA		 3102/*
3103 * preallocate space for a file. This implements ext4's fallocate inode
3104 * operation, which gets called from sys_fallocate system call.
3105 * For block-mapped files, posix_fallocate should fall back to the method
3106 * of writing zeroes to the required new blocks (the same behavior which is
3107 * expected for file systems which do not support fallocate() system call).
3108 */
3109long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3110{
3111 handle_t *handle;
725d26d3 3112 ext4_lblk_t block;
fd28784a 3113 loff_t new_size;
498e5f24 3114 unsigned int max_blocks;
a2df2a63
AA		 3115 int ret = 0;
3116 int ret2 = 0;
3117 int retries = 0;
3118 struct buffer_head map_bh;
3119 unsigned int credits, blkbits = inode->i_blkbits;
3120
3121 /*
3122 * currently supporting (pre)allocate mode for extent-based
3123 * files _only_
3124 */
3125 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3126 return -EOPNOTSUPP;
3127
3128 /* preallocation to directories is currently not supported */
3129 if (S_ISDIR(inode->i_mode))
3130 return -ENODEV;
3131
3132 block = offset >> blkbits;
fd28784a
AK		 3133 /*
3134 * We can't just convert len to max_blocks because
3135 * If blocksize = 4096 offset = 3072 and len = 2048
3136 */
a2df2a63 3137 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
fd28784a 3138 - block;
a2df2a63 3139 /*
f3bd1f3f 3140 * credits to insert 1 extent into extent tree
a2df2a63 3141 */
f3bd1f3f 3142 credits = ext4_chunk_trans_blocks(inode, max_blocks);
55bd725a 3143 mutex_lock(&inode->i_mutex);
a2df2a63
AA		 3144retry:
3145 while (ret >= 0 && ret < max_blocks) {
3146 block = block + ret;
3147 max_blocks = max_blocks - ret;
3148 handle = ext4_journal_start(inode, credits);
3149 if (IS_ERR(handle)) {
3150 ret = PTR_ERR(handle);
3151 break;
3152 }
79ffab34 3153 map_bh.b_state = 0;
55bd725a 3154 ret = ext4_get_blocks_wrap(handle, inode, block,
a2df2a63 3155 max_blocks, &map_bh,
d2a17637 3156 EXT4_CREATE_UNINITIALIZED_EXT, 0, 0);
221879c9 3157 if (ret <= 0) {
2c98615d
AK		 3158#ifdef EXT4FS_DEBUG
3159 WARN_ON(ret <= 0);
3160 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3161 "returned error inode#%lu, block=%u, "
9fd9784c 3162 "max_blocks=%u", __func__,
221879c9 3163 inode->i_ino, block, max_blocks);
2c98615d 3164#endif
a2df2a63
AA		 3165 ext4_mark_inode_dirty(handle, inode);
3166 ret2 = ext4_journal_stop(handle);
3167 break;
3168 }
fd28784a
AK		 3169 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3170 blkbits) >> blkbits))
3171 new_size = offset + len;
3172 else
3173 new_size = (block + ret) << blkbits;
a2df2a63 3174
fd28784a
AK		 3175 ext4_falloc_update_inode(inode, mode, new_size,
3176 buffer_new(&map_bh));
a2df2a63
AA		 3177 ext4_mark_inode_dirty(handle, inode);
3178 ret2 = ext4_journal_stop(handle);
3179 if (ret2)
3180 break;
3181 }
fd28784a
AK		 3182 if (ret == -ENOSPC &&
3183 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3184 ret = 0;
a2df2a63 3185 goto retry;
a2df2a63 3186 }
55bd725a 3187 mutex_unlock(&inode->i_mutex);
a2df2a63
AA		 3188 return ret > 0 ? ret2 : ret;
3189}
6873fa0d
ES		 3190
3191/*
3192 * Callback function called for each extent to gather FIEMAP information.
3193 */
3a06d778 3194static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
6873fa0d
ES		 3195 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3196 void *data)
3197{
3198 struct fiemap_extent_info *fieinfo = data;
c9877b20 3199 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
6873fa0d
ES		 3200 __u64 logical;
3201 __u64 physical;
3202 __u64 length;
3203 __u32 flags = 0;
3204 int error;
3205
3206 logical = (__u64)newex->ec_block << blksize_bits;
3207
3208 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3209 pgoff_t offset;
3210 struct page *page;
3211 struct buffer_head *bh = NULL;
3212
3213 offset = logical >> PAGE_SHIFT;
3214 page = find_get_page(inode->i_mapping, offset);
3215 if (!page || !page_has_buffers(page))
3216 return EXT_CONTINUE;
3217
3218 bh = page_buffers(page);
3219
3220 if (!bh)
3221 return EXT_CONTINUE;
3222
3223 if (buffer_delay(bh)) {
3224 flags |= FIEMAP_EXTENT_DELALLOC;
3225 page_cache_release(page);
3226 } else {
3227 page_cache_release(page);
3228 return EXT_CONTINUE;
3229 }
3230 }
3231
3232 physical = (__u64)newex->ec_start << blksize_bits;
3233 length = (__u64)newex->ec_len << blksize_bits;
3234
3235 if (ex && ext4_ext_is_uninitialized(ex))
3236 flags |= FIEMAP_EXTENT_UNWRITTEN;
3237
3238 /*
3239 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3240 *
3241 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3242 * this also indicates no more allocated blocks.
3243 *
3244 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3245 */
c9877b20
ES		 3246 if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3247 newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK)
6873fa0d
ES		 3248 flags |= FIEMAP_EXTENT_LAST;
3249
3250 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3251 length, flags);
3252 if (error < 0)
3253 return error;
3254 if (error == 1)
3255 return EXT_BREAK;
3256
3257 return EXT_CONTINUE;
3258}
3259
3260/* fiemap flags we can handle specified here */
3261#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3262
3a06d778
AK		 3263static int ext4_xattr_fiemap(struct inode *inode,
3264 struct fiemap_extent_info *fieinfo)
6873fa0d
ES		 3265{
3266 __u64 physical = 0;
3267 __u64 length;
3268 __u32 flags = FIEMAP_EXTENT_LAST;
3269 int blockbits = inode->i_sb->s_blocksize_bits;
3270 int error = 0;
3271
3272 /* in-inode? */
3273 if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3274 struct ext4_iloc iloc;
3275 int offset; /* offset of xattr in inode */
3276
3277 error = ext4_get_inode_loc(inode, &iloc);
3278 if (error)
3279 return error;
3280 physical = iloc.bh->b_blocknr << blockbits;
3281 offset = EXT4_GOOD_OLD_INODE_SIZE +
3282 EXT4_I(inode)->i_extra_isize;
3283 physical += offset;
3284 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3285 flags |= FIEMAP_EXTENT_DATA_INLINE;
3286 } else { /* external block */
3287 physical = EXT4_I(inode)->i_file_acl << blockbits;
3288 length = inode->i_sb->s_blocksize;
3289 }
3290
3291 if (physical)
3292 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3293 length, flags);
3294 return (error < 0 ? error : 0);
3295}
3296
3297int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3298 __u64 start, __u64 len)
3299{
3300 ext4_lblk_t start_blk;
3301 ext4_lblk_t len_blks;
3302 int error = 0;
3303
3304 /* fallback to generic here if not in extents fmt */
3305 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3306 return generic_block_fiemap(inode, fieinfo, start, len,
3307 ext4_get_block);
3308
3309 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3310 return -EBADR;
3311
3312 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3313 error = ext4_xattr_fiemap(inode, fieinfo);
3314 } else {
3315 start_blk = start >> inode->i_sb->s_blocksize_bits;
3316 len_blks = len >> inode->i_sb->s_blocksize_bits;
3317
3318 /*
3319 * Walk the extent tree gathering extent information.
3320 * ext4_ext_fiemap_cb will push extents back to user.
3321 */
3322 down_write(&EXT4_I(inode)->i_data_sem);
3323 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3324 ext4_ext_fiemap_cb, fieinfo);
3325 up_write(&EXT4_I(inode)->i_data_sem);
3326 }
3327
3328 return error;
3329}
3330
Linux kernel - Deliverables
This page took 0.591033 seconds and 5 git commands to generate.