]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/ext4/extents.c
ext4: New inode/block allocation algorithms for flex_bg filesystems
[mirror_ubuntu-zesty-kernel.git] / fs / ext4 / extents.c
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>
35 #include <linux/jbd2.h>
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>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
46
47
48 /*
49 * ext_pblock:
50 * combine low and high parts of physical block number into ext4_fsblk_t
51 */
52 static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
53 {
54 ext4_fsblk_t block;
55
56 block = le32_to_cpu(ex->ee_start_lo);
57 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
58 return block;
59 }
60
61 /*
62 * idx_pblock:
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 */
65 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
66 {
67 ext4_fsblk_t block;
68
69 block = le32_to_cpu(ix->ei_leaf_lo);
70 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
71 return block;
72 }
73
74 /*
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
78 */
79 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
80 {
81 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
82 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
83 }
84
85 /*
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
89 */
90 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
91 {
92 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
93 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
94 }
95
96 static int ext4_ext_journal_restart(handle_t *handle, int needed)
97 {
98 int err;
99
100 if (!ext4_handle_valid(handle))
101 return 0;
102 if (handle->h_buffer_credits > needed)
103 return 0;
104 err = ext4_journal_extend(handle, needed);
105 if (err <= 0)
106 return err;
107 return ext4_journal_restart(handle, needed);
108 }
109
110 /*
111 * could return:
112 * - EROFS
113 * - ENOMEM
114 */
115 static 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 */
133 static 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 */
139 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
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
147 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
148 struct ext4_ext_path *path,
149 ext4_lblk_t block)
150 {
151 struct ext4_inode_info *ei = EXT4_I(inode);
152 ext4_fsblk_t bg_start;
153 ext4_fsblk_t last_block;
154 ext4_grpblk_t colour;
155 ext4_group_t block_group;
156 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
157 int depth;
158
159 if (path) {
160 struct ext4_extent *ex;
161 depth = path->p_depth;
162
163 /* try to predict block placement */
164 ex = path[depth].p_ext;
165 if (ex)
166 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
167
168 /* it looks like index is empty;
169 * try to find starting block from index itself */
170 if (path[depth].p_bh)
171 return path[depth].p_bh->b_blocknr;
172 }
173
174 /* OK. use inode's group */
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)) +
190 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
191 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
192
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
200 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
201 colour = (current->pid % 16) *
202 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
203 else
204 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
205 return bg_start + colour + block;
206 }
207
208 /*
209 * Allocation for a meta data block
210 */
211 static ext4_fsblk_t
212 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
213 struct ext4_ext_path *path,
214 struct ext4_extent *ex, int *err)
215 {
216 ext4_fsblk_t goal, newblock;
217
218 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
219 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
220 return newblock;
221 }
222
223 static int ext4_ext_space_block(struct inode *inode)
224 {
225 int size;
226
227 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
228 / sizeof(struct ext4_extent);
229 #ifdef AGGRESSIVE_TEST
230 if (size > 6)
231 size = 6;
232 #endif
233 return size;
234 }
235
236 static int ext4_ext_space_block_idx(struct inode *inode)
237 {
238 int size;
239
240 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
241 / sizeof(struct ext4_extent_idx);
242 #ifdef AGGRESSIVE_TEST
243 if (size > 5)
244 size = 5;
245 #endif
246 return size;
247 }
248
249 static int ext4_ext_space_root(struct inode *inode)
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);
256 #ifdef AGGRESSIVE_TEST
257 if (size > 3)
258 size = 3;
259 #endif
260 return size;
261 }
262
263 static int ext4_ext_space_root_idx(struct inode *inode)
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);
270 #ifdef AGGRESSIVE_TEST
271 if (size > 4)
272 size = 4;
273 #endif
274 return size;
275 }
276
277 /*
278 * Calculate the number of metadata blocks needed
279 * to allocate @blocks
280 * Worse case is one block per extent
281 */
282 int 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
307 static int
308 ext4_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
327 static int __ext4_ext_check_header(const char *function, struct inode *inode,
328 struct ext4_extent_header *eh,
329 int depth)
330 {
331 const char *error_msg;
332 int max = 0;
333
334 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
335 error_msg = "invalid magic";
336 goto corrupted;
337 }
338 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
339 error_msg = "unexpected eh_depth";
340 goto corrupted;
341 }
342 if (unlikely(eh->eh_max == 0)) {
343 error_msg = "invalid eh_max";
344 goto corrupted;
345 }
346 max = ext4_ext_max_entries(inode, depth);
347 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
348 error_msg = "too large eh_max";
349 goto corrupted;
350 }
351 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
352 error_msg = "invalid eh_entries";
353 goto corrupted;
354 }
355 return 0;
356
357 corrupted:
358 ext4_error(inode->i_sb, function,
359 "bad header in inode #%lu: %s - magic %x, "
360 "entries %u, max %u(%u), depth %u(%u)",
361 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
362 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
363 max, le16_to_cpu(eh->eh_depth), depth);
364
365 return -EIO;
366 }
367
368 #define ext4_ext_check_header(inode, eh, depth) \
369 __ext4_ext_check_header(__func__, inode, eh, depth)
370
371 #ifdef EXT_DEBUG
372 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
373 {
374 int k, l = path->p_depth;
375
376 ext_debug("path:");
377 for (k = 0; k <= l; k++, path++) {
378 if (path->p_idx) {
379 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
380 idx_pblock(path->p_idx));
381 } else if (path->p_ext) {
382 ext_debug(" %d:%d:%llu ",
383 le32_to_cpu(path->p_ext->ee_block),
384 ext4_ext_get_actual_len(path->p_ext),
385 ext_pblock(path->p_ext));
386 } else
387 ext_debug(" []");
388 }
389 ext_debug("\n");
390 }
391
392 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
393 {
394 int depth = ext_depth(inode);
395 struct ext4_extent_header *eh;
396 struct ext4_extent *ex;
397 int i;
398
399 if (!path)
400 return;
401
402 eh = path[depth].p_hdr;
403 ex = EXT_FIRST_EXTENT(eh);
404
405 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
406 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
407 ext4_ext_get_actual_len(ex), ext_pblock(ex));
408 }
409 ext_debug("\n");
410 }
411 #else
412 #define ext4_ext_show_path(inode, path)
413 #define ext4_ext_show_leaf(inode, path)
414 #endif
415
416 void ext4_ext_drop_refs(struct ext4_ext_path *path)
417 {
418 int depth = path->p_depth;
419 int i;
420
421 for (i = 0; i <= depth; i++, path++)
422 if (path->p_bh) {
423 brelse(path->p_bh);
424 path->p_bh = NULL;
425 }
426 }
427
428 /*
429 * ext4_ext_binsearch_idx:
430 * binary search for the closest index of the given block
431 * the header must be checked before calling this
432 */
433 static void
434 ext4_ext_binsearch_idx(struct inode *inode,
435 struct ext4_ext_path *path, ext4_lblk_t block)
436 {
437 struct ext4_extent_header *eh = path->p_hdr;
438 struct ext4_extent_idx *r, *l, *m;
439
440
441 ext_debug("binsearch for %u(idx): ", block);
442
443 l = EXT_FIRST_INDEX(eh) + 1;
444 r = EXT_LAST_INDEX(eh);
445 while (l <= r) {
446 m = l + (r - l) / 2;
447 if (block < le32_to_cpu(m->ei_block))
448 r = m - 1;
449 else
450 l = m + 1;
451 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
452 m, le32_to_cpu(m->ei_block),
453 r, le32_to_cpu(r->ei_block));
454 }
455
456 path->p_idx = l - 1;
457 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
458 idx_pblock(path->p_idx));
459
460 #ifdef CHECK_BINSEARCH
461 {
462 struct ext4_extent_idx *chix, *ix;
463 int k;
464
465 chix = ix = EXT_FIRST_INDEX(eh);
466 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
467 if (k != 0 &&
468 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
469 printk(KERN_DEBUG "k=%d, ix=0x%p, "
470 "first=0x%p\n", k,
471 ix, EXT_FIRST_INDEX(eh));
472 printk(KERN_DEBUG "%u <= %u\n",
473 le32_to_cpu(ix->ei_block),
474 le32_to_cpu(ix[-1].ei_block));
475 }
476 BUG_ON(k && le32_to_cpu(ix->ei_block)
477 <= le32_to_cpu(ix[-1].ei_block));
478 if (block < le32_to_cpu(ix->ei_block))
479 break;
480 chix = ix;
481 }
482 BUG_ON(chix != path->p_idx);
483 }
484 #endif
485
486 }
487
488 /*
489 * ext4_ext_binsearch:
490 * binary search for closest extent of the given block
491 * the header must be checked before calling this
492 */
493 static void
494 ext4_ext_binsearch(struct inode *inode,
495 struct ext4_ext_path *path, ext4_lblk_t block)
496 {
497 struct ext4_extent_header *eh = path->p_hdr;
498 struct ext4_extent *r, *l, *m;
499
500 if (eh->eh_entries == 0) {
501 /*
502 * this leaf is empty:
503 * we get such a leaf in split/add case
504 */
505 return;
506 }
507
508 ext_debug("binsearch for %u: ", block);
509
510 l = EXT_FIRST_EXTENT(eh) + 1;
511 r = EXT_LAST_EXTENT(eh);
512
513 while (l <= r) {
514 m = l + (r - l) / 2;
515 if (block < le32_to_cpu(m->ee_block))
516 r = m - 1;
517 else
518 l = m + 1;
519 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
520 m, le32_to_cpu(m->ee_block),
521 r, le32_to_cpu(r->ee_block));
522 }
523
524 path->p_ext = l - 1;
525 ext_debug(" -> %d:%llu:%d ",
526 le32_to_cpu(path->p_ext->ee_block),
527 ext_pblock(path->p_ext),
528 ext4_ext_get_actual_len(path->p_ext));
529
530 #ifdef CHECK_BINSEARCH
531 {
532 struct ext4_extent *chex, *ex;
533 int k;
534
535 chex = ex = EXT_FIRST_EXTENT(eh);
536 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
537 BUG_ON(k && le32_to_cpu(ex->ee_block)
538 <= le32_to_cpu(ex[-1].ee_block));
539 if (block < le32_to_cpu(ex->ee_block))
540 break;
541 chex = ex;
542 }
543 BUG_ON(chex != path->p_ext);
544 }
545 #endif
546
547 }
548
549 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
550 {
551 struct ext4_extent_header *eh;
552
553 eh = ext_inode_hdr(inode);
554 eh->eh_depth = 0;
555 eh->eh_entries = 0;
556 eh->eh_magic = EXT4_EXT_MAGIC;
557 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
558 ext4_mark_inode_dirty(handle, inode);
559 ext4_ext_invalidate_cache(inode);
560 return 0;
561 }
562
563 struct ext4_ext_path *
564 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
565 struct ext4_ext_path *path)
566 {
567 struct ext4_extent_header *eh;
568 struct buffer_head *bh;
569 short int depth, i, ppos = 0, alloc = 0;
570
571 eh = ext_inode_hdr(inode);
572 depth = ext_depth(inode);
573 if (ext4_ext_check_header(inode, eh, depth))
574 return ERR_PTR(-EIO);
575
576
577 /* account possible depth increase */
578 if (!path) {
579 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
580 GFP_NOFS);
581 if (!path)
582 return ERR_PTR(-ENOMEM);
583 alloc = 1;
584 }
585 path[0].p_hdr = eh;
586 path[0].p_bh = NULL;
587
588 i = depth;
589 /* walk through the tree */
590 while (i) {
591 ext_debug("depth %d: num %d, max %d\n",
592 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
593
594 ext4_ext_binsearch_idx(inode, path + ppos, block);
595 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
596 path[ppos].p_depth = i;
597 path[ppos].p_ext = NULL;
598
599 bh = sb_bread(inode->i_sb, path[ppos].p_block);
600 if (!bh)
601 goto err;
602
603 eh = ext_block_hdr(bh);
604 ppos++;
605 BUG_ON(ppos > depth);
606 path[ppos].p_bh = bh;
607 path[ppos].p_hdr = eh;
608 i--;
609
610 if (ext4_ext_check_header(inode, eh, i))
611 goto err;
612 }
613
614 path[ppos].p_depth = i;
615 path[ppos].p_ext = NULL;
616 path[ppos].p_idx = NULL;
617
618 /* find extent */
619 ext4_ext_binsearch(inode, path + ppos, block);
620 /* if not an empty leaf */
621 if (path[ppos].p_ext)
622 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
623
624 ext4_ext_show_path(inode, path);
625
626 return path;
627
628 err:
629 ext4_ext_drop_refs(path);
630 if (alloc)
631 kfree(path);
632 return ERR_PTR(-EIO);
633 }
634
635 /*
636 * ext4_ext_insert_index:
637 * insert new index [@logical;@ptr] into the block at @curp;
638 * check where to insert: before @curp or after @curp
639 */
640 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
641 struct ext4_ext_path *curp,
642 int logical, ext4_fsblk_t ptr)
643 {
644 struct ext4_extent_idx *ix;
645 int len, err;
646
647 err = ext4_ext_get_access(handle, inode, curp);
648 if (err)
649 return err;
650
651 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
652 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
653 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
654 /* insert after */
655 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
656 len = (len - 1) * sizeof(struct ext4_extent_idx);
657 len = len < 0 ? 0 : len;
658 ext_debug("insert new index %d after: %llu. "
659 "move %d from 0x%p to 0x%p\n",
660 logical, ptr, len,
661 (curp->p_idx + 1), (curp->p_idx + 2));
662 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
663 }
664 ix = curp->p_idx + 1;
665 } else {
666 /* insert before */
667 len = len * sizeof(struct ext4_extent_idx);
668 len = len < 0 ? 0 : len;
669 ext_debug("insert new index %d before: %llu. "
670 "move %d from 0x%p to 0x%p\n",
671 logical, ptr, len,
672 curp->p_idx, (curp->p_idx + 1));
673 memmove(curp->p_idx + 1, curp->p_idx, len);
674 ix = curp->p_idx;
675 }
676
677 ix->ei_block = cpu_to_le32(logical);
678 ext4_idx_store_pblock(ix, ptr);
679 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
680
681 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
682 > le16_to_cpu(curp->p_hdr->eh_max));
683 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
684
685 err = ext4_ext_dirty(handle, inode, curp);
686 ext4_std_error(inode->i_sb, err);
687
688 return err;
689 }
690
691 /*
692 * ext4_ext_split:
693 * inserts new subtree into the path, using free index entry
694 * at depth @at:
695 * - allocates all needed blocks (new leaf and all intermediate index blocks)
696 * - makes decision where to split
697 * - moves remaining extents and index entries (right to the split point)
698 * into the newly allocated blocks
699 * - initializes subtree
700 */
701 static int ext4_ext_split(handle_t *handle, struct inode *inode,
702 struct ext4_ext_path *path,
703 struct ext4_extent *newext, int at)
704 {
705 struct buffer_head *bh = NULL;
706 int depth = ext_depth(inode);
707 struct ext4_extent_header *neh;
708 struct ext4_extent_idx *fidx;
709 struct ext4_extent *ex;
710 int i = at, k, m, a;
711 ext4_fsblk_t newblock, oldblock;
712 __le32 border;
713 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
714 int err = 0;
715
716 /* make decision: where to split? */
717 /* FIXME: now decision is simplest: at current extent */
718
719 /* if current leaf will be split, then we should use
720 * border from split point */
721 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
722 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
723 border = path[depth].p_ext[1].ee_block;
724 ext_debug("leaf will be split."
725 " next leaf starts at %d\n",
726 le32_to_cpu(border));
727 } else {
728 border = newext->ee_block;
729 ext_debug("leaf will be added."
730 " next leaf starts at %d\n",
731 le32_to_cpu(border));
732 }
733
734 /*
735 * If error occurs, then we break processing
736 * and mark filesystem read-only. index won't
737 * be inserted and tree will be in consistent
738 * state. Next mount will repair buffers too.
739 */
740
741 /*
742 * Get array to track all allocated blocks.
743 * We need this to handle errors and free blocks
744 * upon them.
745 */
746 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
747 if (!ablocks)
748 return -ENOMEM;
749
750 /* allocate all needed blocks */
751 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
752 for (a = 0; a < depth - at; a++) {
753 newblock = ext4_ext_new_meta_block(handle, inode, path,
754 newext, &err);
755 if (newblock == 0)
756 goto cleanup;
757 ablocks[a] = newblock;
758 }
759
760 /* initialize new leaf */
761 newblock = ablocks[--a];
762 BUG_ON(newblock == 0);
763 bh = sb_getblk(inode->i_sb, newblock);
764 if (!bh) {
765 err = -EIO;
766 goto cleanup;
767 }
768 lock_buffer(bh);
769
770 err = ext4_journal_get_create_access(handle, bh);
771 if (err)
772 goto cleanup;
773
774 neh = ext_block_hdr(bh);
775 neh->eh_entries = 0;
776 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
777 neh->eh_magic = EXT4_EXT_MAGIC;
778 neh->eh_depth = 0;
779 ex = EXT_FIRST_EXTENT(neh);
780
781 /* move remainder of path[depth] to the new leaf */
782 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
783 /* start copy from next extent */
784 /* TODO: we could do it by single memmove */
785 m = 0;
786 path[depth].p_ext++;
787 while (path[depth].p_ext <=
788 EXT_MAX_EXTENT(path[depth].p_hdr)) {
789 ext_debug("move %d:%llu:%d in new leaf %llu\n",
790 le32_to_cpu(path[depth].p_ext->ee_block),
791 ext_pblock(path[depth].p_ext),
792 ext4_ext_get_actual_len(path[depth].p_ext),
793 newblock);
794 /*memmove(ex++, path[depth].p_ext++,
795 sizeof(struct ext4_extent));
796 neh->eh_entries++;*/
797 path[depth].p_ext++;
798 m++;
799 }
800 if (m) {
801 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
802 le16_add_cpu(&neh->eh_entries, m);
803 }
804
805 set_buffer_uptodate(bh);
806 unlock_buffer(bh);
807
808 err = ext4_handle_dirty_metadata(handle, inode, bh);
809 if (err)
810 goto cleanup;
811 brelse(bh);
812 bh = NULL;
813
814 /* correct old leaf */
815 if (m) {
816 err = ext4_ext_get_access(handle, inode, path + depth);
817 if (err)
818 goto cleanup;
819 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
820 err = ext4_ext_dirty(handle, inode, path + depth);
821 if (err)
822 goto cleanup;
823
824 }
825
826 /* create intermediate indexes */
827 k = depth - at - 1;
828 BUG_ON(k < 0);
829 if (k)
830 ext_debug("create %d intermediate indices\n", k);
831 /* insert new index into current index block */
832 /* current depth stored in i var */
833 i = depth - 1;
834 while (k--) {
835 oldblock = newblock;
836 newblock = ablocks[--a];
837 bh = sb_getblk(inode->i_sb, newblock);
838 if (!bh) {
839 err = -EIO;
840 goto cleanup;
841 }
842 lock_buffer(bh);
843
844 err = ext4_journal_get_create_access(handle, bh);
845 if (err)
846 goto cleanup;
847
848 neh = ext_block_hdr(bh);
849 neh->eh_entries = cpu_to_le16(1);
850 neh->eh_magic = EXT4_EXT_MAGIC;
851 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
852 neh->eh_depth = cpu_to_le16(depth - i);
853 fidx = EXT_FIRST_INDEX(neh);
854 fidx->ei_block = border;
855 ext4_idx_store_pblock(fidx, oldblock);
856
857 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
858 i, newblock, le32_to_cpu(border), oldblock);
859 /* copy indexes */
860 m = 0;
861 path[i].p_idx++;
862
863 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
864 EXT_MAX_INDEX(path[i].p_hdr));
865 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
866 EXT_LAST_INDEX(path[i].p_hdr));
867 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
868 ext_debug("%d: move %d:%llu in new index %llu\n", i,
869 le32_to_cpu(path[i].p_idx->ei_block),
870 idx_pblock(path[i].p_idx),
871 newblock);
872 /*memmove(++fidx, path[i].p_idx++,
873 sizeof(struct ext4_extent_idx));
874 neh->eh_entries++;
875 BUG_ON(neh->eh_entries > neh->eh_max);*/
876 path[i].p_idx++;
877 m++;
878 }
879 if (m) {
880 memmove(++fidx, path[i].p_idx - m,
881 sizeof(struct ext4_extent_idx) * m);
882 le16_add_cpu(&neh->eh_entries, m);
883 }
884 set_buffer_uptodate(bh);
885 unlock_buffer(bh);
886
887 err = ext4_handle_dirty_metadata(handle, inode, bh);
888 if (err)
889 goto cleanup;
890 brelse(bh);
891 bh = NULL;
892
893 /* correct old index */
894 if (m) {
895 err = ext4_ext_get_access(handle, inode, path + i);
896 if (err)
897 goto cleanup;
898 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
899 err = ext4_ext_dirty(handle, inode, path + i);
900 if (err)
901 goto cleanup;
902 }
903
904 i--;
905 }
906
907 /* insert new index */
908 err = ext4_ext_insert_index(handle, inode, path + at,
909 le32_to_cpu(border), newblock);
910
911 cleanup:
912 if (bh) {
913 if (buffer_locked(bh))
914 unlock_buffer(bh);
915 brelse(bh);
916 }
917
918 if (err) {
919 /* free all allocated blocks in error case */
920 for (i = 0; i < depth; i++) {
921 if (!ablocks[i])
922 continue;
923 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
924 }
925 }
926 kfree(ablocks);
927
928 return err;
929 }
930
931 /*
932 * ext4_ext_grow_indepth:
933 * implements tree growing procedure:
934 * - allocates new block
935 * - moves top-level data (index block or leaf) into the new block
936 * - initializes new top-level, creating index that points to the
937 * just created block
938 */
939 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
940 struct ext4_ext_path *path,
941 struct ext4_extent *newext)
942 {
943 struct ext4_ext_path *curp = path;
944 struct ext4_extent_header *neh;
945 struct ext4_extent_idx *fidx;
946 struct buffer_head *bh;
947 ext4_fsblk_t newblock;
948 int err = 0;
949
950 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
951 if (newblock == 0)
952 return err;
953
954 bh = sb_getblk(inode->i_sb, newblock);
955 if (!bh) {
956 err = -EIO;
957 ext4_std_error(inode->i_sb, err);
958 return err;
959 }
960 lock_buffer(bh);
961
962 err = ext4_journal_get_create_access(handle, bh);
963 if (err) {
964 unlock_buffer(bh);
965 goto out;
966 }
967
968 /* move top-level index/leaf into new block */
969 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
970
971 /* set size of new block */
972 neh = ext_block_hdr(bh);
973 /* old root could have indexes or leaves
974 * so calculate e_max right way */
975 if (ext_depth(inode))
976 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
977 else
978 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
979 neh->eh_magic = EXT4_EXT_MAGIC;
980 set_buffer_uptodate(bh);
981 unlock_buffer(bh);
982
983 err = ext4_handle_dirty_metadata(handle, inode, bh);
984 if (err)
985 goto out;
986
987 /* create index in new top-level index: num,max,pointer */
988 err = ext4_ext_get_access(handle, inode, curp);
989 if (err)
990 goto out;
991
992 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
993 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
994 curp->p_hdr->eh_entries = cpu_to_le16(1);
995 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
996
997 if (path[0].p_hdr->eh_depth)
998 curp->p_idx->ei_block =
999 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1000 else
1001 curp->p_idx->ei_block =
1002 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1003 ext4_idx_store_pblock(curp->p_idx, newblock);
1004
1005 neh = ext_inode_hdr(inode);
1006 fidx = EXT_FIRST_INDEX(neh);
1007 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1008 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1009 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1010
1011 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1012 err = ext4_ext_dirty(handle, inode, curp);
1013 out:
1014 brelse(bh);
1015
1016 return err;
1017 }
1018
1019 /*
1020 * ext4_ext_create_new_leaf:
1021 * finds empty index and adds new leaf.
1022 * if no free index is found, then it requests in-depth growing.
1023 */
1024 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1025 struct ext4_ext_path *path,
1026 struct ext4_extent *newext)
1027 {
1028 struct ext4_ext_path *curp;
1029 int depth, i, err = 0;
1030
1031 repeat:
1032 i = depth = ext_depth(inode);
1033
1034 /* walk up to the tree and look for free index entry */
1035 curp = path + depth;
1036 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1037 i--;
1038 curp--;
1039 }
1040
1041 /* we use already allocated block for index block,
1042 * so subsequent data blocks should be contiguous */
1043 if (EXT_HAS_FREE_INDEX(curp)) {
1044 /* if we found index with free entry, then use that
1045 * entry: create all needed subtree and add new leaf */
1046 err = ext4_ext_split(handle, inode, path, newext, i);
1047 if (err)
1048 goto out;
1049
1050 /* refill path */
1051 ext4_ext_drop_refs(path);
1052 path = ext4_ext_find_extent(inode,
1053 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1054 path);
1055 if (IS_ERR(path))
1056 err = PTR_ERR(path);
1057 } else {
1058 /* tree is full, time to grow in depth */
1059 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1060 if (err)
1061 goto out;
1062
1063 /* refill path */
1064 ext4_ext_drop_refs(path);
1065 path = ext4_ext_find_extent(inode,
1066 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1067 path);
1068 if (IS_ERR(path)) {
1069 err = PTR_ERR(path);
1070 goto out;
1071 }
1072
1073 /*
1074 * only first (depth 0 -> 1) produces free space;
1075 * in all other cases we have to split the grown tree
1076 */
1077 depth = ext_depth(inode);
1078 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1079 /* now we need to split */
1080 goto repeat;
1081 }
1082 }
1083
1084 out:
1085 return err;
1086 }
1087
1088 /*
1089 * search the closest allocated block to the left for *logical
1090 * and returns it at @logical + it's physical address at @phys
1091 * if *logical is the smallest allocated block, the function
1092 * returns 0 at @phys
1093 * return value contains 0 (success) or error code
1094 */
1095 int
1096 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1097 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1098 {
1099 struct ext4_extent_idx *ix;
1100 struct ext4_extent *ex;
1101 int depth, ee_len;
1102
1103 BUG_ON(path == NULL);
1104 depth = path->p_depth;
1105 *phys = 0;
1106
1107 if (depth == 0 && path->p_ext == NULL)
1108 return 0;
1109
1110 /* usually extent in the path covers blocks smaller
1111 * then *logical, but it can be that extent is the
1112 * first one in the file */
1113
1114 ex = path[depth].p_ext;
1115 ee_len = ext4_ext_get_actual_len(ex);
1116 if (*logical < le32_to_cpu(ex->ee_block)) {
1117 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1118 while (--depth >= 0) {
1119 ix = path[depth].p_idx;
1120 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1121 }
1122 return 0;
1123 }
1124
1125 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1126
1127 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1128 *phys = ext_pblock(ex) + ee_len - 1;
1129 return 0;
1130 }
1131
1132 /*
1133 * search the closest allocated block to the right for *logical
1134 * and returns it at @logical + it's physical address at @phys
1135 * if *logical is the smallest allocated block, the function
1136 * returns 0 at @phys
1137 * return value contains 0 (success) or error code
1138 */
1139 int
1140 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1141 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1142 {
1143 struct buffer_head *bh = NULL;
1144 struct ext4_extent_header *eh;
1145 struct ext4_extent_idx *ix;
1146 struct ext4_extent *ex;
1147 ext4_fsblk_t block;
1148 int depth; /* Note, NOT eh_depth; depth from top of tree */
1149 int ee_len;
1150
1151 BUG_ON(path == NULL);
1152 depth = path->p_depth;
1153 *phys = 0;
1154
1155 if (depth == 0 && path->p_ext == NULL)
1156 return 0;
1157
1158 /* usually extent in the path covers blocks smaller
1159 * then *logical, but it can be that extent is the
1160 * first one in the file */
1161
1162 ex = path[depth].p_ext;
1163 ee_len = ext4_ext_get_actual_len(ex);
1164 if (*logical < le32_to_cpu(ex->ee_block)) {
1165 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1166 while (--depth >= 0) {
1167 ix = path[depth].p_idx;
1168 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1169 }
1170 *logical = le32_to_cpu(ex->ee_block);
1171 *phys = ext_pblock(ex);
1172 return 0;
1173 }
1174
1175 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1176
1177 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1178 /* next allocated block in this leaf */
1179 ex++;
1180 *logical = le32_to_cpu(ex->ee_block);
1181 *phys = ext_pblock(ex);
1182 return 0;
1183 }
1184
1185 /* go up and search for index to the right */
1186 while (--depth >= 0) {
1187 ix = path[depth].p_idx;
1188 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1189 goto got_index;
1190 }
1191
1192 /* we've gone up to the root and found no index to the right */
1193 return 0;
1194
1195 got_index:
1196 /* we've found index to the right, let's
1197 * follow it and find the closest allocated
1198 * block to the right */
1199 ix++;
1200 block = idx_pblock(ix);
1201 while (++depth < path->p_depth) {
1202 bh = sb_bread(inode->i_sb, block);
1203 if (bh == NULL)
1204 return -EIO;
1205 eh = ext_block_hdr(bh);
1206 /* subtract from p_depth to get proper eh_depth */
1207 if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1208 put_bh(bh);
1209 return -EIO;
1210 }
1211 ix = EXT_FIRST_INDEX(eh);
1212 block = idx_pblock(ix);
1213 put_bh(bh);
1214 }
1215
1216 bh = sb_bread(inode->i_sb, block);
1217 if (bh == NULL)
1218 return -EIO;
1219 eh = ext_block_hdr(bh);
1220 if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1221 put_bh(bh);
1222 return -EIO;
1223 }
1224 ex = EXT_FIRST_EXTENT(eh);
1225 *logical = le32_to_cpu(ex->ee_block);
1226 *phys = ext_pblock(ex);
1227 put_bh(bh);
1228 return 0;
1229 }
1230
1231 /*
1232 * ext4_ext_next_allocated_block:
1233 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1234 * NOTE: it considers block number from index entry as
1235 * allocated block. Thus, index entries have to be consistent
1236 * with leaves.
1237 */
1238 static ext4_lblk_t
1239 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1240 {
1241 int depth;
1242
1243 BUG_ON(path == NULL);
1244 depth = path->p_depth;
1245
1246 if (depth == 0 && path->p_ext == NULL)
1247 return EXT_MAX_BLOCK;
1248
1249 while (depth >= 0) {
1250 if (depth == path->p_depth) {
1251 /* leaf */
1252 if (path[depth].p_ext !=
1253 EXT_LAST_EXTENT(path[depth].p_hdr))
1254 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1255 } else {
1256 /* index */
1257 if (path[depth].p_idx !=
1258 EXT_LAST_INDEX(path[depth].p_hdr))
1259 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1260 }
1261 depth--;
1262 }
1263
1264 return EXT_MAX_BLOCK;
1265 }
1266
1267 /*
1268 * ext4_ext_next_leaf_block:
1269 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1270 */
1271 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1272 struct ext4_ext_path *path)
1273 {
1274 int depth;
1275
1276 BUG_ON(path == NULL);
1277 depth = path->p_depth;
1278
1279 /* zero-tree has no leaf blocks at all */
1280 if (depth == 0)
1281 return EXT_MAX_BLOCK;
1282
1283 /* go to index block */
1284 depth--;
1285
1286 while (depth >= 0) {
1287 if (path[depth].p_idx !=
1288 EXT_LAST_INDEX(path[depth].p_hdr))
1289 return (ext4_lblk_t)
1290 le32_to_cpu(path[depth].p_idx[1].ei_block);
1291 depth--;
1292 }
1293
1294 return EXT_MAX_BLOCK;
1295 }
1296
1297 /*
1298 * ext4_ext_correct_indexes:
1299 * if leaf gets modified and modified extent is first in the leaf,
1300 * then we have to correct all indexes above.
1301 * TODO: do we need to correct tree in all cases?
1302 */
1303 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1304 struct ext4_ext_path *path)
1305 {
1306 struct ext4_extent_header *eh;
1307 int depth = ext_depth(inode);
1308 struct ext4_extent *ex;
1309 __le32 border;
1310 int k, err = 0;
1311
1312 eh = path[depth].p_hdr;
1313 ex = path[depth].p_ext;
1314 BUG_ON(ex == NULL);
1315 BUG_ON(eh == NULL);
1316
1317 if (depth == 0) {
1318 /* there is no tree at all */
1319 return 0;
1320 }
1321
1322 if (ex != EXT_FIRST_EXTENT(eh)) {
1323 /* we correct tree if first leaf got modified only */
1324 return 0;
1325 }
1326
1327 /*
1328 * TODO: we need correction if border is smaller than current one
1329 */
1330 k = depth - 1;
1331 border = path[depth].p_ext->ee_block;
1332 err = ext4_ext_get_access(handle, inode, path + k);
1333 if (err)
1334 return err;
1335 path[k].p_idx->ei_block = border;
1336 err = ext4_ext_dirty(handle, inode, path + k);
1337 if (err)
1338 return err;
1339
1340 while (k--) {
1341 /* change all left-side indexes */
1342 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1343 break;
1344 err = ext4_ext_get_access(handle, inode, path + k);
1345 if (err)
1346 break;
1347 path[k].p_idx->ei_block = border;
1348 err = ext4_ext_dirty(handle, inode, path + k);
1349 if (err)
1350 break;
1351 }
1352
1353 return err;
1354 }
1355
1356 static int
1357 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1358 struct ext4_extent *ex2)
1359 {
1360 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1361
1362 /*
1363 * Make sure that either both extents are uninitialized, or
1364 * both are _not_.
1365 */
1366 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1367 return 0;
1368
1369 if (ext4_ext_is_uninitialized(ex1))
1370 max_len = EXT_UNINIT_MAX_LEN;
1371 else
1372 max_len = EXT_INIT_MAX_LEN;
1373
1374 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1375 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1376
1377 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1378 le32_to_cpu(ex2->ee_block))
1379 return 0;
1380
1381 /*
1382 * To allow future support for preallocated extents to be added
1383 * as an RO_COMPAT feature, refuse to merge to extents if
1384 * this can result in the top bit of ee_len being set.
1385 */
1386 if (ext1_ee_len + ext2_ee_len > max_len)
1387 return 0;
1388 #ifdef AGGRESSIVE_TEST
1389 if (ext1_ee_len >= 4)
1390 return 0;
1391 #endif
1392
1393 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1394 return 1;
1395 return 0;
1396 }
1397
1398 /*
1399 * This function tries to merge the "ex" extent to the next extent in the tree.
1400 * It always tries to merge towards right. If you want to merge towards
1401 * left, pass "ex - 1" as argument instead of "ex".
1402 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1403 * 1 if they got merged.
1404 */
1405 int ext4_ext_try_to_merge(struct inode *inode,
1406 struct ext4_ext_path *path,
1407 struct ext4_extent *ex)
1408 {
1409 struct ext4_extent_header *eh;
1410 unsigned int depth, len;
1411 int merge_done = 0;
1412 int uninitialized = 0;
1413
1414 depth = ext_depth(inode);
1415 BUG_ON(path[depth].p_hdr == NULL);
1416 eh = path[depth].p_hdr;
1417
1418 while (ex < EXT_LAST_EXTENT(eh)) {
1419 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1420 break;
1421 /* merge with next extent! */
1422 if (ext4_ext_is_uninitialized(ex))
1423 uninitialized = 1;
1424 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1425 + ext4_ext_get_actual_len(ex + 1));
1426 if (uninitialized)
1427 ext4_ext_mark_uninitialized(ex);
1428
1429 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1430 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1431 * sizeof(struct ext4_extent);
1432 memmove(ex + 1, ex + 2, len);
1433 }
1434 le16_add_cpu(&eh->eh_entries, -1);
1435 merge_done = 1;
1436 WARN_ON(eh->eh_entries == 0);
1437 if (!eh->eh_entries)
1438 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1439 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1440 }
1441
1442 return merge_done;
1443 }
1444
1445 /*
1446 * check if a portion of the "newext" extent overlaps with an
1447 * existing extent.
1448 *
1449 * If there is an overlap discovered, it updates the length of the newext
1450 * such that there will be no overlap, and then returns 1.
1451 * If there is no overlap found, it returns 0.
1452 */
1453 unsigned int ext4_ext_check_overlap(struct inode *inode,
1454 struct ext4_extent *newext,
1455 struct ext4_ext_path *path)
1456 {
1457 ext4_lblk_t b1, b2;
1458 unsigned int depth, len1;
1459 unsigned int ret = 0;
1460
1461 b1 = le32_to_cpu(newext->ee_block);
1462 len1 = ext4_ext_get_actual_len(newext);
1463 depth = ext_depth(inode);
1464 if (!path[depth].p_ext)
1465 goto out;
1466 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1467
1468 /*
1469 * get the next allocated block if the extent in the path
1470 * is before the requested block(s)
1471 */
1472 if (b2 < b1) {
1473 b2 = ext4_ext_next_allocated_block(path);
1474 if (b2 == EXT_MAX_BLOCK)
1475 goto out;
1476 }
1477
1478 /* check for wrap through zero on extent logical start block*/
1479 if (b1 + len1 < b1) {
1480 len1 = EXT_MAX_BLOCK - b1;
1481 newext->ee_len = cpu_to_le16(len1);
1482 ret = 1;
1483 }
1484
1485 /* check for overlap */
1486 if (b1 + len1 > b2) {
1487 newext->ee_len = cpu_to_le16(b2 - b1);
1488 ret = 1;
1489 }
1490 out:
1491 return ret;
1492 }
1493
1494 /*
1495 * ext4_ext_insert_extent:
1496 * tries to merge requsted extent into the existing extent or
1497 * inserts requested extent as new one into the tree,
1498 * creating new leaf in the no-space case.
1499 */
1500 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1501 struct ext4_ext_path *path,
1502 struct ext4_extent *newext)
1503 {
1504 struct ext4_extent_header *eh;
1505 struct ext4_extent *ex, *fex;
1506 struct ext4_extent *nearex; /* nearest extent */
1507 struct ext4_ext_path *npath = NULL;
1508 int depth, len, err;
1509 ext4_lblk_t next;
1510 unsigned uninitialized = 0;
1511
1512 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1513 depth = ext_depth(inode);
1514 ex = path[depth].p_ext;
1515 BUG_ON(path[depth].p_hdr == NULL);
1516
1517 /* try to insert block into found extent and return */
1518 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1519 ext_debug("append %d block to %d:%d (from %llu)\n",
1520 ext4_ext_get_actual_len(newext),
1521 le32_to_cpu(ex->ee_block),
1522 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1523 err = ext4_ext_get_access(handle, inode, path + depth);
1524 if (err)
1525 return err;
1526
1527 /*
1528 * ext4_can_extents_be_merged should have checked that either
1529 * both extents are uninitialized, or both aren't. Thus we
1530 * need to check only one of them here.
1531 */
1532 if (ext4_ext_is_uninitialized(ex))
1533 uninitialized = 1;
1534 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1535 + ext4_ext_get_actual_len(newext));
1536 if (uninitialized)
1537 ext4_ext_mark_uninitialized(ex);
1538 eh = path[depth].p_hdr;
1539 nearex = ex;
1540 goto merge;
1541 }
1542
1543 repeat:
1544 depth = ext_depth(inode);
1545 eh = path[depth].p_hdr;
1546 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1547 goto has_space;
1548
1549 /* probably next leaf has space for us? */
1550 fex = EXT_LAST_EXTENT(eh);
1551 next = ext4_ext_next_leaf_block(inode, path);
1552 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1553 && next != EXT_MAX_BLOCK) {
1554 ext_debug("next leaf block - %d\n", next);
1555 BUG_ON(npath != NULL);
1556 npath = ext4_ext_find_extent(inode, next, NULL);
1557 if (IS_ERR(npath))
1558 return PTR_ERR(npath);
1559 BUG_ON(npath->p_depth != path->p_depth);
1560 eh = npath[depth].p_hdr;
1561 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1562 ext_debug("next leaf isnt full(%d)\n",
1563 le16_to_cpu(eh->eh_entries));
1564 path = npath;
1565 goto repeat;
1566 }
1567 ext_debug("next leaf has no free space(%d,%d)\n",
1568 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1569 }
1570
1571 /*
1572 * There is no free space in the found leaf.
1573 * We're gonna add a new leaf in the tree.
1574 */
1575 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1576 if (err)
1577 goto cleanup;
1578 depth = ext_depth(inode);
1579 eh = path[depth].p_hdr;
1580
1581 has_space:
1582 nearex = path[depth].p_ext;
1583
1584 err = ext4_ext_get_access(handle, inode, path + depth);
1585 if (err)
1586 goto cleanup;
1587
1588 if (!nearex) {
1589 /* there is no extent in this leaf, create first one */
1590 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1591 le32_to_cpu(newext->ee_block),
1592 ext_pblock(newext),
1593 ext4_ext_get_actual_len(newext));
1594 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1595 } else if (le32_to_cpu(newext->ee_block)
1596 > le32_to_cpu(nearex->ee_block)) {
1597 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1598 if (nearex != EXT_LAST_EXTENT(eh)) {
1599 len = EXT_MAX_EXTENT(eh) - nearex;
1600 len = (len - 1) * sizeof(struct ext4_extent);
1601 len = len < 0 ? 0 : len;
1602 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1603 "move %d from 0x%p to 0x%p\n",
1604 le32_to_cpu(newext->ee_block),
1605 ext_pblock(newext),
1606 ext4_ext_get_actual_len(newext),
1607 nearex, len, nearex + 1, nearex + 2);
1608 memmove(nearex + 2, nearex + 1, len);
1609 }
1610 path[depth].p_ext = nearex + 1;
1611 } else {
1612 BUG_ON(newext->ee_block == nearex->ee_block);
1613 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1614 len = len < 0 ? 0 : len;
1615 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1616 "move %d from 0x%p to 0x%p\n",
1617 le32_to_cpu(newext->ee_block),
1618 ext_pblock(newext),
1619 ext4_ext_get_actual_len(newext),
1620 nearex, len, nearex + 1, nearex + 2);
1621 memmove(nearex + 1, nearex, len);
1622 path[depth].p_ext = nearex;
1623 }
1624
1625 le16_add_cpu(&eh->eh_entries, 1);
1626 nearex = path[depth].p_ext;
1627 nearex->ee_block = newext->ee_block;
1628 ext4_ext_store_pblock(nearex, ext_pblock(newext));
1629 nearex->ee_len = newext->ee_len;
1630
1631 merge:
1632 /* try to merge extents to the right */
1633 ext4_ext_try_to_merge(inode, path, nearex);
1634
1635 /* try to merge extents to the left */
1636
1637 /* time to correct all indexes above */
1638 err = ext4_ext_correct_indexes(handle, inode, path);
1639 if (err)
1640 goto cleanup;
1641
1642 err = ext4_ext_dirty(handle, inode, path + depth);
1643
1644 cleanup:
1645 if (npath) {
1646 ext4_ext_drop_refs(npath);
1647 kfree(npath);
1648 }
1649 ext4_ext_invalidate_cache(inode);
1650 return err;
1651 }
1652
1653 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1654 ext4_lblk_t num, ext_prepare_callback func,
1655 void *cbdata)
1656 {
1657 struct ext4_ext_path *path = NULL;
1658 struct ext4_ext_cache cbex;
1659 struct ext4_extent *ex;
1660 ext4_lblk_t next, start = 0, end = 0;
1661 ext4_lblk_t last = block + num;
1662 int depth, exists, err = 0;
1663
1664 BUG_ON(func == NULL);
1665 BUG_ON(inode == NULL);
1666
1667 while (block < last && block != EXT_MAX_BLOCK) {
1668 num = last - block;
1669 /* find extent for this block */
1670 path = ext4_ext_find_extent(inode, block, path);
1671 if (IS_ERR(path)) {
1672 err = PTR_ERR(path);
1673 path = NULL;
1674 break;
1675 }
1676
1677 depth = ext_depth(inode);
1678 BUG_ON(path[depth].p_hdr == NULL);
1679 ex = path[depth].p_ext;
1680 next = ext4_ext_next_allocated_block(path);
1681
1682 exists = 0;
1683 if (!ex) {
1684 /* there is no extent yet, so try to allocate
1685 * all requested space */
1686 start = block;
1687 end = block + num;
1688 } else if (le32_to_cpu(ex->ee_block) > block) {
1689 /* need to allocate space before found extent */
1690 start = block;
1691 end = le32_to_cpu(ex->ee_block);
1692 if (block + num < end)
1693 end = block + num;
1694 } else if (block >= le32_to_cpu(ex->ee_block)
1695 + ext4_ext_get_actual_len(ex)) {
1696 /* need to allocate space after found extent */
1697 start = block;
1698 end = block + num;
1699 if (end >= next)
1700 end = next;
1701 } else if (block >= le32_to_cpu(ex->ee_block)) {
1702 /*
1703 * some part of requested space is covered
1704 * by found extent
1705 */
1706 start = block;
1707 end = le32_to_cpu(ex->ee_block)
1708 + ext4_ext_get_actual_len(ex);
1709 if (block + num < end)
1710 end = block + num;
1711 exists = 1;
1712 } else {
1713 BUG();
1714 }
1715 BUG_ON(end <= start);
1716
1717 if (!exists) {
1718 cbex.ec_block = start;
1719 cbex.ec_len = end - start;
1720 cbex.ec_start = 0;
1721 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1722 } else {
1723 cbex.ec_block = le32_to_cpu(ex->ee_block);
1724 cbex.ec_len = ext4_ext_get_actual_len(ex);
1725 cbex.ec_start = ext_pblock(ex);
1726 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1727 }
1728
1729 BUG_ON(cbex.ec_len == 0);
1730 err = func(inode, path, &cbex, ex, cbdata);
1731 ext4_ext_drop_refs(path);
1732
1733 if (err < 0)
1734 break;
1735
1736 if (err == EXT_REPEAT)
1737 continue;
1738 else if (err == EXT_BREAK) {
1739 err = 0;
1740 break;
1741 }
1742
1743 if (ext_depth(inode) != depth) {
1744 /* depth was changed. we have to realloc path */
1745 kfree(path);
1746 path = NULL;
1747 }
1748
1749 block = cbex.ec_block + cbex.ec_len;
1750 }
1751
1752 if (path) {
1753 ext4_ext_drop_refs(path);
1754 kfree(path);
1755 }
1756
1757 return err;
1758 }
1759
1760 static void
1761 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1762 __u32 len, ext4_fsblk_t start, int type)
1763 {
1764 struct ext4_ext_cache *cex;
1765 BUG_ON(len == 0);
1766 cex = &EXT4_I(inode)->i_cached_extent;
1767 cex->ec_type = type;
1768 cex->ec_block = block;
1769 cex->ec_len = len;
1770 cex->ec_start = start;
1771 }
1772
1773 /*
1774 * ext4_ext_put_gap_in_cache:
1775 * calculate boundaries of the gap that the requested block fits into
1776 * and cache this gap
1777 */
1778 static void
1779 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1780 ext4_lblk_t block)
1781 {
1782 int depth = ext_depth(inode);
1783 unsigned long len;
1784 ext4_lblk_t lblock;
1785 struct ext4_extent *ex;
1786
1787 ex = path[depth].p_ext;
1788 if (ex == NULL) {
1789 /* there is no extent yet, so gap is [0;-] */
1790 lblock = 0;
1791 len = EXT_MAX_BLOCK;
1792 ext_debug("cache gap(whole file):");
1793 } else if (block < le32_to_cpu(ex->ee_block)) {
1794 lblock = block;
1795 len = le32_to_cpu(ex->ee_block) - block;
1796 ext_debug("cache gap(before): %u [%u:%u]",
1797 block,
1798 le32_to_cpu(ex->ee_block),
1799 ext4_ext_get_actual_len(ex));
1800 } else if (block >= le32_to_cpu(ex->ee_block)
1801 + ext4_ext_get_actual_len(ex)) {
1802 ext4_lblk_t next;
1803 lblock = le32_to_cpu(ex->ee_block)
1804 + ext4_ext_get_actual_len(ex);
1805
1806 next = ext4_ext_next_allocated_block(path);
1807 ext_debug("cache gap(after): [%u:%u] %u",
1808 le32_to_cpu(ex->ee_block),
1809 ext4_ext_get_actual_len(ex),
1810 block);
1811 BUG_ON(next == lblock);
1812 len = next - lblock;
1813 } else {
1814 lblock = len = 0;
1815 BUG();
1816 }
1817
1818 ext_debug(" -> %u:%lu\n", lblock, len);
1819 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1820 }
1821
1822 static int
1823 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1824 struct ext4_extent *ex)
1825 {
1826 struct ext4_ext_cache *cex;
1827
1828 cex = &EXT4_I(inode)->i_cached_extent;
1829
1830 /* has cache valid data? */
1831 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1832 return EXT4_EXT_CACHE_NO;
1833
1834 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1835 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1836 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1837 ex->ee_block = cpu_to_le32(cex->ec_block);
1838 ext4_ext_store_pblock(ex, cex->ec_start);
1839 ex->ee_len = cpu_to_le16(cex->ec_len);
1840 ext_debug("%u cached by %u:%u:%llu\n",
1841 block,
1842 cex->ec_block, cex->ec_len, cex->ec_start);
1843 return cex->ec_type;
1844 }
1845
1846 /* not in cache */
1847 return EXT4_EXT_CACHE_NO;
1848 }
1849
1850 /*
1851 * ext4_ext_rm_idx:
1852 * removes index from the index block.
1853 * It's used in truncate case only, thus all requests are for
1854 * last index in the block only.
1855 */
1856 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1857 struct ext4_ext_path *path)
1858 {
1859 struct buffer_head *bh;
1860 int err;
1861 ext4_fsblk_t leaf;
1862
1863 /* free index block */
1864 path--;
1865 leaf = idx_pblock(path->p_idx);
1866 BUG_ON(path->p_hdr->eh_entries == 0);
1867 err = ext4_ext_get_access(handle, inode, path);
1868 if (err)
1869 return err;
1870 le16_add_cpu(&path->p_hdr->eh_entries, -1);
1871 err = ext4_ext_dirty(handle, inode, path);
1872 if (err)
1873 return err;
1874 ext_debug("index is empty, remove it, free block %llu\n", leaf);
1875 bh = sb_find_get_block(inode->i_sb, leaf);
1876 ext4_forget(handle, 1, inode, bh, leaf);
1877 ext4_free_blocks(handle, inode, leaf, 1, 1);
1878 return err;
1879 }
1880
1881 /*
1882 * ext4_ext_calc_credits_for_single_extent:
1883 * This routine returns max. credits that needed to insert an extent
1884 * to the extent tree.
1885 * When pass the actual path, the caller should calculate credits
1886 * under i_data_sem.
1887 */
1888 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1889 struct ext4_ext_path *path)
1890 {
1891 if (path) {
1892 int depth = ext_depth(inode);
1893 int ret = 0;
1894
1895 /* probably there is space in leaf? */
1896 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1897 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
1898
1899 /*
1900 * There are some space in the leaf tree, no
1901 * need to account for leaf block credit
1902 *
1903 * bitmaps and block group descriptor blocks
1904 * and other metadat blocks still need to be
1905 * accounted.
1906 */
1907 /* 1 bitmap, 1 block group descriptor */
1908 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1909 }
1910 }
1911
1912 return ext4_chunk_trans_blocks(inode, nrblocks);
1913 }
1914
1915 /*
1916 * How many index/leaf blocks need to change/allocate to modify nrblocks?
1917 *
1918 * if nrblocks are fit in a single extent (chunk flag is 1), then
1919 * in the worse case, each tree level index/leaf need to be changed
1920 * if the tree split due to insert a new extent, then the old tree
1921 * index/leaf need to be updated too
1922 *
1923 * If the nrblocks are discontiguous, they could cause
1924 * the whole tree split more than once, but this is really rare.
1925 */
1926 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
1927 {
1928 int index;
1929 int depth = ext_depth(inode);
1930
1931 if (chunk)
1932 index = depth * 2;
1933 else
1934 index = depth * 3;
1935
1936 return index;
1937 }
1938
1939 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1940 struct ext4_extent *ex,
1941 ext4_lblk_t from, ext4_lblk_t to)
1942 {
1943 struct buffer_head *bh;
1944 unsigned short ee_len = ext4_ext_get_actual_len(ex);
1945 int i, metadata = 0;
1946
1947 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1948 metadata = 1;
1949 #ifdef EXTENTS_STATS
1950 {
1951 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1952 spin_lock(&sbi->s_ext_stats_lock);
1953 sbi->s_ext_blocks += ee_len;
1954 sbi->s_ext_extents++;
1955 if (ee_len < sbi->s_ext_min)
1956 sbi->s_ext_min = ee_len;
1957 if (ee_len > sbi->s_ext_max)
1958 sbi->s_ext_max = ee_len;
1959 if (ext_depth(inode) > sbi->s_depth_max)
1960 sbi->s_depth_max = ext_depth(inode);
1961 spin_unlock(&sbi->s_ext_stats_lock);
1962 }
1963 #endif
1964 if (from >= le32_to_cpu(ex->ee_block)
1965 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
1966 /* tail removal */
1967 ext4_lblk_t num;
1968 ext4_fsblk_t start;
1969
1970 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1971 start = ext_pblock(ex) + ee_len - num;
1972 ext_debug("free last %u blocks starting %llu\n", num, start);
1973 for (i = 0; i < num; i++) {
1974 bh = sb_find_get_block(inode->i_sb, start + i);
1975 ext4_forget(handle, 0, inode, bh, start + i);
1976 }
1977 ext4_free_blocks(handle, inode, start, num, metadata);
1978 } else if (from == le32_to_cpu(ex->ee_block)
1979 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
1980 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
1981 from, to, le32_to_cpu(ex->ee_block), ee_len);
1982 } else {
1983 printk(KERN_INFO "strange request: removal(2) "
1984 "%u-%u from %u:%u\n",
1985 from, to, le32_to_cpu(ex->ee_block), ee_len);
1986 }
1987 return 0;
1988 }
1989
1990 static int
1991 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1992 struct ext4_ext_path *path, ext4_lblk_t start)
1993 {
1994 int err = 0, correct_index = 0;
1995 int depth = ext_depth(inode), credits;
1996 struct ext4_extent_header *eh;
1997 ext4_lblk_t a, b, block;
1998 unsigned num;
1999 ext4_lblk_t ex_ee_block;
2000 unsigned short ex_ee_len;
2001 unsigned uninitialized = 0;
2002 struct ext4_extent *ex;
2003
2004 /* the header must be checked already in ext4_ext_remove_space() */
2005 ext_debug("truncate since %u in leaf\n", start);
2006 if (!path[depth].p_hdr)
2007 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2008 eh = path[depth].p_hdr;
2009 BUG_ON(eh == NULL);
2010
2011 /* find where to start removing */
2012 ex = EXT_LAST_EXTENT(eh);
2013
2014 ex_ee_block = le32_to_cpu(ex->ee_block);
2015 if (ext4_ext_is_uninitialized(ex))
2016 uninitialized = 1;
2017 ex_ee_len = ext4_ext_get_actual_len(ex);
2018
2019 while (ex >= EXT_FIRST_EXTENT(eh) &&
2020 ex_ee_block + ex_ee_len > start) {
2021 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
2022 path[depth].p_ext = ex;
2023
2024 a = ex_ee_block > start ? ex_ee_block : start;
2025 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2026 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2027
2028 ext_debug(" border %u:%u\n", a, b);
2029
2030 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2031 block = 0;
2032 num = 0;
2033 BUG();
2034 } else if (a != ex_ee_block) {
2035 /* remove tail of the extent */
2036 block = ex_ee_block;
2037 num = a - block;
2038 } else if (b != ex_ee_block + ex_ee_len - 1) {
2039 /* remove head of the extent */
2040 block = a;
2041 num = b - a;
2042 /* there is no "make a hole" API yet */
2043 BUG();
2044 } else {
2045 /* remove whole extent: excellent! */
2046 block = ex_ee_block;
2047 num = 0;
2048 BUG_ON(a != ex_ee_block);
2049 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2050 }
2051
2052 /*
2053 * 3 for leaf, sb, and inode plus 2 (bmap and group
2054 * descriptor) for each block group; assume two block
2055 * groups plus ex_ee_len/blocks_per_block_group for
2056 * the worst case
2057 */
2058 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2059 if (ex == EXT_FIRST_EXTENT(eh)) {
2060 correct_index = 1;
2061 credits += (ext_depth(inode)) + 1;
2062 }
2063 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2064
2065 err = ext4_ext_journal_restart(handle, credits);
2066 if (err)
2067 goto out;
2068
2069 err = ext4_ext_get_access(handle, inode, path + depth);
2070 if (err)
2071 goto out;
2072
2073 err = ext4_remove_blocks(handle, inode, ex, a, b);
2074 if (err)
2075 goto out;
2076
2077 if (num == 0) {
2078 /* this extent is removed; mark slot entirely unused */
2079 ext4_ext_store_pblock(ex, 0);
2080 le16_add_cpu(&eh->eh_entries, -1);
2081 }
2082
2083 ex->ee_block = cpu_to_le32(block);
2084 ex->ee_len = cpu_to_le16(num);
2085 /*
2086 * Do not mark uninitialized if all the blocks in the
2087 * extent have been removed.
2088 */
2089 if (uninitialized && num)
2090 ext4_ext_mark_uninitialized(ex);
2091
2092 err = ext4_ext_dirty(handle, inode, path + depth);
2093 if (err)
2094 goto out;
2095
2096 ext_debug("new extent: %u:%u:%llu\n", block, num,
2097 ext_pblock(ex));
2098 ex--;
2099 ex_ee_block = le32_to_cpu(ex->ee_block);
2100 ex_ee_len = ext4_ext_get_actual_len(ex);
2101 }
2102
2103 if (correct_index && eh->eh_entries)
2104 err = ext4_ext_correct_indexes(handle, inode, path);
2105
2106 /* if this leaf is free, then we should
2107 * remove it from index block above */
2108 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2109 err = ext4_ext_rm_idx(handle, inode, path + depth);
2110
2111 out:
2112 return err;
2113 }
2114
2115 /*
2116 * ext4_ext_more_to_rm:
2117 * returns 1 if current index has to be freed (even partial)
2118 */
2119 static int
2120 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2121 {
2122 BUG_ON(path->p_idx == NULL);
2123
2124 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2125 return 0;
2126
2127 /*
2128 * if truncate on deeper level happened, it wasn't partial,
2129 * so we have to consider current index for truncation
2130 */
2131 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2132 return 0;
2133 return 1;
2134 }
2135
2136 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2137 {
2138 struct super_block *sb = inode->i_sb;
2139 int depth = ext_depth(inode);
2140 struct ext4_ext_path *path;
2141 handle_t *handle;
2142 int i = 0, err = 0;
2143
2144 ext_debug("truncate since %u\n", start);
2145
2146 /* probably first extent we're gonna free will be last in block */
2147 handle = ext4_journal_start(inode, depth + 1);
2148 if (IS_ERR(handle))
2149 return PTR_ERR(handle);
2150
2151 ext4_ext_invalidate_cache(inode);
2152
2153 /*
2154 * We start scanning from right side, freeing all the blocks
2155 * after i_size and walking into the tree depth-wise.
2156 */
2157 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2158 if (path == NULL) {
2159 ext4_journal_stop(handle);
2160 return -ENOMEM;
2161 }
2162 path[0].p_hdr = ext_inode_hdr(inode);
2163 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
2164 err = -EIO;
2165 goto out;
2166 }
2167 path[0].p_depth = depth;
2168
2169 while (i >= 0 && err == 0) {
2170 if (i == depth) {
2171 /* this is leaf block */
2172 err = ext4_ext_rm_leaf(handle, inode, path, start);
2173 /* root level has p_bh == NULL, brelse() eats this */
2174 brelse(path[i].p_bh);
2175 path[i].p_bh = NULL;
2176 i--;
2177 continue;
2178 }
2179
2180 /* this is index block */
2181 if (!path[i].p_hdr) {
2182 ext_debug("initialize header\n");
2183 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2184 }
2185
2186 if (!path[i].p_idx) {
2187 /* this level hasn't been touched yet */
2188 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2189 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2190 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2191 path[i].p_hdr,
2192 le16_to_cpu(path[i].p_hdr->eh_entries));
2193 } else {
2194 /* we were already here, see at next index */
2195 path[i].p_idx--;
2196 }
2197
2198 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2199 i, EXT_FIRST_INDEX(path[i].p_hdr),
2200 path[i].p_idx);
2201 if (ext4_ext_more_to_rm(path + i)) {
2202 struct buffer_head *bh;
2203 /* go to the next level */
2204 ext_debug("move to level %d (block %llu)\n",
2205 i + 1, idx_pblock(path[i].p_idx));
2206 memset(path + i + 1, 0, sizeof(*path));
2207 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2208 if (!bh) {
2209 /* should we reset i_size? */
2210 err = -EIO;
2211 break;
2212 }
2213 if (WARN_ON(i + 1 > depth)) {
2214 err = -EIO;
2215 break;
2216 }
2217 if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2218 depth - i - 1)) {
2219 err = -EIO;
2220 break;
2221 }
2222 path[i + 1].p_bh = bh;
2223
2224 /* save actual number of indexes since this
2225 * number is changed at the next iteration */
2226 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2227 i++;
2228 } else {
2229 /* we finished processing this index, go up */
2230 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2231 /* index is empty, remove it;
2232 * handle must be already prepared by the
2233 * truncatei_leaf() */
2234 err = ext4_ext_rm_idx(handle, inode, path + i);
2235 }
2236 /* root level has p_bh == NULL, brelse() eats this */
2237 brelse(path[i].p_bh);
2238 path[i].p_bh = NULL;
2239 i--;
2240 ext_debug("return to level %d\n", i);
2241 }
2242 }
2243
2244 /* TODO: flexible tree reduction should be here */
2245 if (path->p_hdr->eh_entries == 0) {
2246 /*
2247 * truncate to zero freed all the tree,
2248 * so we need to correct eh_depth
2249 */
2250 err = ext4_ext_get_access(handle, inode, path);
2251 if (err == 0) {
2252 ext_inode_hdr(inode)->eh_depth = 0;
2253 ext_inode_hdr(inode)->eh_max =
2254 cpu_to_le16(ext4_ext_space_root(inode));
2255 err = ext4_ext_dirty(handle, inode, path);
2256 }
2257 }
2258 out:
2259 ext4_ext_drop_refs(path);
2260 kfree(path);
2261 ext4_journal_stop(handle);
2262
2263 return err;
2264 }
2265
2266 /*
2267 * called at mount time
2268 */
2269 void ext4_ext_init(struct super_block *sb)
2270 {
2271 /*
2272 * possible initialization would be here
2273 */
2274
2275 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2276 printk(KERN_INFO "EXT4-fs: file extents enabled");
2277 #ifdef AGGRESSIVE_TEST
2278 printk(", aggressive tests");
2279 #endif
2280 #ifdef CHECK_BINSEARCH
2281 printk(", check binsearch");
2282 #endif
2283 #ifdef EXTENTS_STATS
2284 printk(", stats");
2285 #endif
2286 printk("\n");
2287 #ifdef EXTENTS_STATS
2288 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2289 EXT4_SB(sb)->s_ext_min = 1 << 30;
2290 EXT4_SB(sb)->s_ext_max = 0;
2291 #endif
2292 }
2293 }
2294
2295 /*
2296 * called at umount time
2297 */
2298 void ext4_ext_release(struct super_block *sb)
2299 {
2300 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2301 return;
2302
2303 #ifdef EXTENTS_STATS
2304 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2305 struct ext4_sb_info *sbi = EXT4_SB(sb);
2306 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2307 sbi->s_ext_blocks, sbi->s_ext_extents,
2308 sbi->s_ext_blocks / sbi->s_ext_extents);
2309 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2310 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2311 }
2312 #endif
2313 }
2314
2315 static void bi_complete(struct bio *bio, int error)
2316 {
2317 complete((struct completion *)bio->bi_private);
2318 }
2319
2320 /* FIXME!! we need to try to merge to left or right after zero-out */
2321 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2322 {
2323 int ret = -EIO;
2324 struct bio *bio;
2325 int blkbits, blocksize;
2326 sector_t ee_pblock;
2327 struct completion event;
2328 unsigned int ee_len, len, done, offset;
2329
2330
2331 blkbits = inode->i_blkbits;
2332 blocksize = inode->i_sb->s_blocksize;
2333 ee_len = ext4_ext_get_actual_len(ex);
2334 ee_pblock = ext_pblock(ex);
2335
2336 /* convert ee_pblock to 512 byte sectors */
2337 ee_pblock = ee_pblock << (blkbits - 9);
2338
2339 while (ee_len > 0) {
2340
2341 if (ee_len > BIO_MAX_PAGES)
2342 len = BIO_MAX_PAGES;
2343 else
2344 len = ee_len;
2345
2346 bio = bio_alloc(GFP_NOIO, len);
2347 if (!bio)
2348 return -ENOMEM;
2349 bio->bi_sector = ee_pblock;
2350 bio->bi_bdev = inode->i_sb->s_bdev;
2351
2352 done = 0;
2353 offset = 0;
2354 while (done < len) {
2355 ret = bio_add_page(bio, ZERO_PAGE(0),
2356 blocksize, offset);
2357 if (ret != blocksize) {
2358 /*
2359 * We can't add any more pages because of
2360 * hardware limitations. Start a new bio.
2361 */
2362 break;
2363 }
2364 done++;
2365 offset += blocksize;
2366 if (offset >= PAGE_CACHE_SIZE)
2367 offset = 0;
2368 }
2369
2370 init_completion(&event);
2371 bio->bi_private = &event;
2372 bio->bi_end_io = bi_complete;
2373 submit_bio(WRITE, bio);
2374 wait_for_completion(&event);
2375
2376 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2377 ret = 0;
2378 else {
2379 ret = -EIO;
2380 break;
2381 }
2382 bio_put(bio);
2383 ee_len -= done;
2384 ee_pblock += done << (blkbits - 9);
2385 }
2386 return ret;
2387 }
2388
2389 #define EXT4_EXT_ZERO_LEN 7
2390
2391 /*
2392 * This function is called by ext4_ext_get_blocks() if someone tries to write
2393 * to an uninitialized extent. It may result in splitting the uninitialized
2394 * extent into multiple extents (upto three - one initialized and two
2395 * uninitialized).
2396 * There are three possibilities:
2397 * a> There is no split required: Entire extent should be initialized
2398 * b> Splits in two extents: Write is happening at either end of the extent
2399 * c> Splits in three extents: Somone is writing in middle of the extent
2400 */
2401 static int ext4_ext_convert_to_initialized(handle_t *handle,
2402 struct inode *inode,
2403 struct ext4_ext_path *path,
2404 ext4_lblk_t iblock,
2405 unsigned int max_blocks)
2406 {
2407 struct ext4_extent *ex, newex, orig_ex;
2408 struct ext4_extent *ex1 = NULL;
2409 struct ext4_extent *ex2 = NULL;
2410 struct ext4_extent *ex3 = NULL;
2411 struct ext4_extent_header *eh;
2412 ext4_lblk_t ee_block;
2413 unsigned int allocated, ee_len, depth;
2414 ext4_fsblk_t newblock;
2415 int err = 0;
2416 int ret = 0;
2417
2418 depth = ext_depth(inode);
2419 eh = path[depth].p_hdr;
2420 ex = path[depth].p_ext;
2421 ee_block = le32_to_cpu(ex->ee_block);
2422 ee_len = ext4_ext_get_actual_len(ex);
2423 allocated = ee_len - (iblock - ee_block);
2424 newblock = iblock - ee_block + ext_pblock(ex);
2425 ex2 = ex;
2426 orig_ex.ee_block = ex->ee_block;
2427 orig_ex.ee_len = cpu_to_le16(ee_len);
2428 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2429
2430 err = ext4_ext_get_access(handle, inode, path + depth);
2431 if (err)
2432 goto out;
2433 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2434 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2435 err = ext4_ext_zeroout(inode, &orig_ex);
2436 if (err)
2437 goto fix_extent_len;
2438 /* update the extent length and mark as initialized */
2439 ex->ee_block = orig_ex.ee_block;
2440 ex->ee_len = orig_ex.ee_len;
2441 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2442 ext4_ext_dirty(handle, inode, path + depth);
2443 /* zeroed the full extent */
2444 return allocated;
2445 }
2446
2447 /* ex1: ee_block to iblock - 1 : uninitialized */
2448 if (iblock > ee_block) {
2449 ex1 = ex;
2450 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2451 ext4_ext_mark_uninitialized(ex1);
2452 ex2 = &newex;
2453 }
2454 /*
2455 * for sanity, update the length of the ex2 extent before
2456 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2457 * overlap of blocks.
2458 */
2459 if (!ex1 && allocated > max_blocks)
2460 ex2->ee_len = cpu_to_le16(max_blocks);
2461 /* ex3: to ee_block + ee_len : uninitialised */
2462 if (allocated > max_blocks) {
2463 unsigned int newdepth;
2464 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2465 if (allocated <= EXT4_EXT_ZERO_LEN) {
2466 /*
2467 * iblock == ee_block is handled by the zerouout
2468 * at the beginning.
2469 * Mark first half uninitialized.
2470 * Mark second half initialized and zero out the
2471 * initialized extent
2472 */
2473 ex->ee_block = orig_ex.ee_block;
2474 ex->ee_len = cpu_to_le16(ee_len - allocated);
2475 ext4_ext_mark_uninitialized(ex);
2476 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2477 ext4_ext_dirty(handle, inode, path + depth);
2478
2479 ex3 = &newex;
2480 ex3->ee_block = cpu_to_le32(iblock);
2481 ext4_ext_store_pblock(ex3, newblock);
2482 ex3->ee_len = cpu_to_le16(allocated);
2483 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2484 if (err == -ENOSPC) {
2485 err = ext4_ext_zeroout(inode, &orig_ex);
2486 if (err)
2487 goto fix_extent_len;
2488 ex->ee_block = orig_ex.ee_block;
2489 ex->ee_len = orig_ex.ee_len;
2490 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2491 ext4_ext_dirty(handle, inode, path + depth);
2492 /* blocks available from iblock */
2493 return allocated;
2494
2495 } else if (err)
2496 goto fix_extent_len;
2497
2498 /*
2499 * We need to zero out the second half because
2500 * an fallocate request can update file size and
2501 * converting the second half to initialized extent
2502 * implies that we can leak some junk data to user
2503 * space.
2504 */
2505 err = ext4_ext_zeroout(inode, ex3);
2506 if (err) {
2507 /*
2508 * We should actually mark the
2509 * second half as uninit and return error
2510 * Insert would have changed the extent
2511 */
2512 depth = ext_depth(inode);
2513 ext4_ext_drop_refs(path);
2514 path = ext4_ext_find_extent(inode,
2515 iblock, path);
2516 if (IS_ERR(path)) {
2517 err = PTR_ERR(path);
2518 return err;
2519 }
2520 /* get the second half extent details */
2521 ex = path[depth].p_ext;
2522 err = ext4_ext_get_access(handle, inode,
2523 path + depth);
2524 if (err)
2525 return err;
2526 ext4_ext_mark_uninitialized(ex);
2527 ext4_ext_dirty(handle, inode, path + depth);
2528 return err;
2529 }
2530
2531 /* zeroed the second half */
2532 return allocated;
2533 }
2534 ex3 = &newex;
2535 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2536 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2537 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2538 ext4_ext_mark_uninitialized(ex3);
2539 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2540 if (err == -ENOSPC) {
2541 err = ext4_ext_zeroout(inode, &orig_ex);
2542 if (err)
2543 goto fix_extent_len;
2544 /* update the extent length and mark as initialized */
2545 ex->ee_block = orig_ex.ee_block;
2546 ex->ee_len = orig_ex.ee_len;
2547 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2548 ext4_ext_dirty(handle, inode, path + depth);
2549 /* zeroed the full extent */
2550 /* blocks available from iblock */
2551 return allocated;
2552
2553 } else if (err)
2554 goto fix_extent_len;
2555 /*
2556 * The depth, and hence eh & ex might change
2557 * as part of the insert above.
2558 */
2559 newdepth = ext_depth(inode);
2560 /*
2561 * update the extent length after successful insert of the
2562 * split extent
2563 */
2564 orig_ex.ee_len = cpu_to_le16(ee_len -
2565 ext4_ext_get_actual_len(ex3));
2566 depth = newdepth;
2567 ext4_ext_drop_refs(path);
2568 path = ext4_ext_find_extent(inode, iblock, path);
2569 if (IS_ERR(path)) {
2570 err = PTR_ERR(path);
2571 goto out;
2572 }
2573 eh = path[depth].p_hdr;
2574 ex = path[depth].p_ext;
2575 if (ex2 != &newex)
2576 ex2 = ex;
2577
2578 err = ext4_ext_get_access(handle, inode, path + depth);
2579 if (err)
2580 goto out;
2581
2582 allocated = max_blocks;
2583
2584 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2585 * to insert a extent in the middle zerout directly
2586 * otherwise give the extent a chance to merge to left
2587 */
2588 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2589 iblock != ee_block) {
2590 err = ext4_ext_zeroout(inode, &orig_ex);
2591 if (err)
2592 goto fix_extent_len;
2593 /* update the extent length and mark as initialized */
2594 ex->ee_block = orig_ex.ee_block;
2595 ex->ee_len = orig_ex.ee_len;
2596 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2597 ext4_ext_dirty(handle, inode, path + depth);
2598 /* zero out the first half */
2599 /* blocks available from iblock */
2600 return allocated;
2601 }
2602 }
2603 /*
2604 * If there was a change of depth as part of the
2605 * insertion of ex3 above, we need to update the length
2606 * of the ex1 extent again here
2607 */
2608 if (ex1 && ex1 != ex) {
2609 ex1 = ex;
2610 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2611 ext4_ext_mark_uninitialized(ex1);
2612 ex2 = &newex;
2613 }
2614 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2615 ex2->ee_block = cpu_to_le32(iblock);
2616 ext4_ext_store_pblock(ex2, newblock);
2617 ex2->ee_len = cpu_to_le16(allocated);
2618 if (ex2 != ex)
2619 goto insert;
2620 /*
2621 * New (initialized) extent starts from the first block
2622 * in the current extent. i.e., ex2 == ex
2623 * We have to see if it can be merged with the extent
2624 * on the left.
2625 */
2626 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2627 /*
2628 * To merge left, pass "ex2 - 1" to try_to_merge(),
2629 * since it merges towards right _only_.
2630 */
2631 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2632 if (ret) {
2633 err = ext4_ext_correct_indexes(handle, inode, path);
2634 if (err)
2635 goto out;
2636 depth = ext_depth(inode);
2637 ex2--;
2638 }
2639 }
2640 /*
2641 * Try to Merge towards right. This might be required
2642 * only when the whole extent is being written to.
2643 * i.e. ex2 == ex and ex3 == NULL.
2644 */
2645 if (!ex3) {
2646 ret = ext4_ext_try_to_merge(inode, path, ex2);
2647 if (ret) {
2648 err = ext4_ext_correct_indexes(handle, inode, path);
2649 if (err)
2650 goto out;
2651 }
2652 }
2653 /* Mark modified extent as dirty */
2654 err = ext4_ext_dirty(handle, inode, path + depth);
2655 goto out;
2656 insert:
2657 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2658 if (err == -ENOSPC) {
2659 err = ext4_ext_zeroout(inode, &orig_ex);
2660 if (err)
2661 goto fix_extent_len;
2662 /* update the extent length and mark as initialized */
2663 ex->ee_block = orig_ex.ee_block;
2664 ex->ee_len = orig_ex.ee_len;
2665 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2666 ext4_ext_dirty(handle, inode, path + depth);
2667 /* zero out the first half */
2668 return allocated;
2669 } else if (err)
2670 goto fix_extent_len;
2671 out:
2672 return err ? err : allocated;
2673
2674 fix_extent_len:
2675 ex->ee_block = orig_ex.ee_block;
2676 ex->ee_len = orig_ex.ee_len;
2677 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2678 ext4_ext_mark_uninitialized(ex);
2679 ext4_ext_dirty(handle, inode, path + depth);
2680 return err;
2681 }
2682
2683 /*
2684 * Block allocation/map/preallocation routine for extents based files
2685 *
2686 *
2687 * Need to be called with
2688 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2689 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2690 *
2691 * return > 0, number of of blocks already mapped/allocated
2692 * if create == 0 and these are pre-allocated blocks
2693 * buffer head is unmapped
2694 * otherwise blocks are mapped
2695 *
2696 * return = 0, if plain look up failed (blocks have not been allocated)
2697 * buffer head is unmapped
2698 *
2699 * return < 0, error case.
2700 */
2701 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2702 ext4_lblk_t iblock,
2703 unsigned int max_blocks, struct buffer_head *bh_result,
2704 int create, int extend_disksize)
2705 {
2706 struct ext4_ext_path *path = NULL;
2707 struct ext4_extent_header *eh;
2708 struct ext4_extent newex, *ex;
2709 ext4_fsblk_t newblock;
2710 int err = 0, depth, ret, cache_type;
2711 unsigned int allocated = 0;
2712 struct ext4_allocation_request ar;
2713 loff_t disksize;
2714
2715 __clear_bit(BH_New, &bh_result->b_state);
2716 ext_debug("blocks %u/%u requested for inode %u\n",
2717 iblock, max_blocks, inode->i_ino);
2718
2719 /* check in cache */
2720 cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2721 if (cache_type) {
2722 if (cache_type == EXT4_EXT_CACHE_GAP) {
2723 if (!create) {
2724 /*
2725 * block isn't allocated yet and
2726 * user doesn't want to allocate it
2727 */
2728 goto out2;
2729 }
2730 /* we should allocate requested block */
2731 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
2732 /* block is already allocated */
2733 newblock = iblock
2734 - le32_to_cpu(newex.ee_block)
2735 + ext_pblock(&newex);
2736 /* number of remaining blocks in the extent */
2737 allocated = ext4_ext_get_actual_len(&newex) -
2738 (iblock - le32_to_cpu(newex.ee_block));
2739 goto out;
2740 } else {
2741 BUG();
2742 }
2743 }
2744
2745 /* find extent for this block */
2746 path = ext4_ext_find_extent(inode, iblock, NULL);
2747 if (IS_ERR(path)) {
2748 err = PTR_ERR(path);
2749 path = NULL;
2750 goto out2;
2751 }
2752
2753 depth = ext_depth(inode);
2754
2755 /*
2756 * consistent leaf must not be empty;
2757 * this situation is possible, though, _during_ tree modification;
2758 * this is why assert can't be put in ext4_ext_find_extent()
2759 */
2760 BUG_ON(path[depth].p_ext == NULL && depth != 0);
2761 eh = path[depth].p_hdr;
2762
2763 ex = path[depth].p_ext;
2764 if (ex) {
2765 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2766 ext4_fsblk_t ee_start = ext_pblock(ex);
2767 unsigned short ee_len;
2768
2769 /*
2770 * Uninitialized extents are treated as holes, except that
2771 * we split out initialized portions during a write.
2772 */
2773 ee_len = ext4_ext_get_actual_len(ex);
2774 /* if found extent covers block, simply return it */
2775 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2776 newblock = iblock - ee_block + ee_start;
2777 /* number of remaining blocks in the extent */
2778 allocated = ee_len - (iblock - ee_block);
2779 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2780 ee_block, ee_len, newblock);
2781
2782 /* Do not put uninitialized extent in the cache */
2783 if (!ext4_ext_is_uninitialized(ex)) {
2784 ext4_ext_put_in_cache(inode, ee_block,
2785 ee_len, ee_start,
2786 EXT4_EXT_CACHE_EXTENT);
2787 goto out;
2788 }
2789 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2790 goto out;
2791 if (!create) {
2792 /*
2793 * We have blocks reserved already. We
2794 * return allocated blocks so that delalloc
2795 * won't do block reservation for us. But
2796 * the buffer head will be unmapped so that
2797 * a read from the block returns 0s.
2798 */
2799 if (allocated > max_blocks)
2800 allocated = max_blocks;
2801 set_buffer_unwritten(bh_result);
2802 goto out2;
2803 }
2804
2805 ret = ext4_ext_convert_to_initialized(handle, inode,
2806 path, iblock,
2807 max_blocks);
2808 if (ret <= 0) {
2809 err = ret;
2810 goto out2;
2811 } else
2812 allocated = ret;
2813 goto outnew;
2814 }
2815 }
2816
2817 /*
2818 * requested block isn't allocated yet;
2819 * we couldn't try to create block if create flag is zero
2820 */
2821 if (!create) {
2822 /*
2823 * put just found gap into cache to speed up
2824 * subsequent requests
2825 */
2826 ext4_ext_put_gap_in_cache(inode, path, iblock);
2827 goto out2;
2828 }
2829 /*
2830 * Okay, we need to do block allocation.
2831 */
2832
2833 /* find neighbour allocated blocks */
2834 ar.lleft = iblock;
2835 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2836 if (err)
2837 goto out2;
2838 ar.lright = iblock;
2839 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2840 if (err)
2841 goto out2;
2842
2843 /*
2844 * See if request is beyond maximum number of blocks we can have in
2845 * a single extent. For an initialized extent this limit is
2846 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2847 * EXT_UNINIT_MAX_LEN.
2848 */
2849 if (max_blocks > EXT_INIT_MAX_LEN &&
2850 create != EXT4_CREATE_UNINITIALIZED_EXT)
2851 max_blocks = EXT_INIT_MAX_LEN;
2852 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2853 create == EXT4_CREATE_UNINITIALIZED_EXT)
2854 max_blocks = EXT_UNINIT_MAX_LEN;
2855
2856 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2857 newex.ee_block = cpu_to_le32(iblock);
2858 newex.ee_len = cpu_to_le16(max_blocks);
2859 err = ext4_ext_check_overlap(inode, &newex, path);
2860 if (err)
2861 allocated = ext4_ext_get_actual_len(&newex);
2862 else
2863 allocated = max_blocks;
2864
2865 /* allocate new block */
2866 ar.inode = inode;
2867 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2868 ar.logical = iblock;
2869 ar.len = allocated;
2870 if (S_ISREG(inode->i_mode))
2871 ar.flags = EXT4_MB_HINT_DATA;
2872 else
2873 /* disable in-core preallocation for non-regular files */
2874 ar.flags = 0;
2875 newblock = ext4_mb_new_blocks(handle, &ar, &err);
2876 if (!newblock)
2877 goto out2;
2878 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2879 ar.goal, newblock, allocated);
2880
2881 /* try to insert new extent into found leaf and return */
2882 ext4_ext_store_pblock(&newex, newblock);
2883 newex.ee_len = cpu_to_le16(ar.len);
2884 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2885 ext4_ext_mark_uninitialized(&newex);
2886 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2887 if (err) {
2888 /* free data blocks we just allocated */
2889 /* not a good idea to call discard here directly,
2890 * but otherwise we'd need to call it every free() */
2891 ext4_discard_preallocations(inode);
2892 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2893 ext4_ext_get_actual_len(&newex), 0);
2894 goto out2;
2895 }
2896
2897 /* previous routine could use block we allocated */
2898 newblock = ext_pblock(&newex);
2899 allocated = ext4_ext_get_actual_len(&newex);
2900 outnew:
2901 if (extend_disksize) {
2902 disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits;
2903 if (disksize > i_size_read(inode))
2904 disksize = i_size_read(inode);
2905 if (disksize > EXT4_I(inode)->i_disksize)
2906 EXT4_I(inode)->i_disksize = disksize;
2907 }
2908
2909 set_buffer_new(bh_result);
2910
2911 /* Cache only when it is _not_ an uninitialized extent */
2912 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2913 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2914 EXT4_EXT_CACHE_EXTENT);
2915 out:
2916 if (allocated > max_blocks)
2917 allocated = max_blocks;
2918 ext4_ext_show_leaf(inode, path);
2919 set_buffer_mapped(bh_result);
2920 bh_result->b_bdev = inode->i_sb->s_bdev;
2921 bh_result->b_blocknr = newblock;
2922 out2:
2923 if (path) {
2924 ext4_ext_drop_refs(path);
2925 kfree(path);
2926 }
2927 return err ? err : allocated;
2928 }
2929
2930 void ext4_ext_truncate(struct inode *inode)
2931 {
2932 struct address_space *mapping = inode->i_mapping;
2933 struct super_block *sb = inode->i_sb;
2934 ext4_lblk_t last_block;
2935 handle_t *handle;
2936 int err = 0;
2937
2938 /*
2939 * probably first extent we're gonna free will be last in block
2940 */
2941 err = ext4_writepage_trans_blocks(inode);
2942 handle = ext4_journal_start(inode, err);
2943 if (IS_ERR(handle))
2944 return;
2945
2946 if (inode->i_size & (sb->s_blocksize - 1))
2947 ext4_block_truncate_page(handle, mapping, inode->i_size);
2948
2949 if (ext4_orphan_add(handle, inode))
2950 goto out_stop;
2951
2952 down_write(&EXT4_I(inode)->i_data_sem);
2953 ext4_ext_invalidate_cache(inode);
2954
2955 ext4_discard_preallocations(inode);
2956
2957 /*
2958 * TODO: optimization is possible here.
2959 * Probably we need not scan at all,
2960 * because page truncation is enough.
2961 */
2962
2963 /* we have to know where to truncate from in crash case */
2964 EXT4_I(inode)->i_disksize = inode->i_size;
2965 ext4_mark_inode_dirty(handle, inode);
2966
2967 last_block = (inode->i_size + sb->s_blocksize - 1)
2968 >> EXT4_BLOCK_SIZE_BITS(sb);
2969 err = ext4_ext_remove_space(inode, last_block);
2970
2971 /* In a multi-transaction truncate, we only make the final
2972 * transaction synchronous.
2973 */
2974 if (IS_SYNC(inode))
2975 ext4_handle_sync(handle);
2976
2977 out_stop:
2978 up_write(&EXT4_I(inode)->i_data_sem);
2979 /*
2980 * If this was a simple ftruncate() and the file will remain alive,
2981 * then we need to clear up the orphan record which we created above.
2982 * However, if this was a real unlink then we were called by
2983 * ext4_delete_inode(), and we allow that function to clean up the
2984 * orphan info for us.
2985 */
2986 if (inode->i_nlink)
2987 ext4_orphan_del(handle, inode);
2988
2989 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2990 ext4_mark_inode_dirty(handle, inode);
2991 ext4_journal_stop(handle);
2992 }
2993
2994 static void ext4_falloc_update_inode(struct inode *inode,
2995 int mode, loff_t new_size, int update_ctime)
2996 {
2997 struct timespec now;
2998
2999 if (update_ctime) {
3000 now = current_fs_time(inode->i_sb);
3001 if (!timespec_equal(&inode->i_ctime, &now))
3002 inode->i_ctime = now;
3003 }
3004 /*
3005 * Update only when preallocation was requested beyond
3006 * the file size.
3007 */
3008 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3009 if (new_size > i_size_read(inode))
3010 i_size_write(inode, new_size);
3011 if (new_size > EXT4_I(inode)->i_disksize)
3012 ext4_update_i_disksize(inode, new_size);
3013 }
3014
3015 }
3016
3017 /*
3018 * preallocate space for a file. This implements ext4's fallocate inode
3019 * operation, which gets called from sys_fallocate system call.
3020 * For block-mapped files, posix_fallocate should fall back to the method
3021 * of writing zeroes to the required new blocks (the same behavior which is
3022 * expected for file systems which do not support fallocate() system call).
3023 */
3024 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3025 {
3026 handle_t *handle;
3027 ext4_lblk_t block;
3028 loff_t new_size;
3029 unsigned int max_blocks;
3030 int ret = 0;
3031 int ret2 = 0;
3032 int retries = 0;
3033 struct buffer_head map_bh;
3034 unsigned int credits, blkbits = inode->i_blkbits;
3035
3036 /*
3037 * currently supporting (pre)allocate mode for extent-based
3038 * files _only_
3039 */
3040 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3041 return -EOPNOTSUPP;
3042
3043 /* preallocation to directories is currently not supported */
3044 if (S_ISDIR(inode->i_mode))
3045 return -ENODEV;
3046
3047 block = offset >> blkbits;
3048 /*
3049 * We can't just convert len to max_blocks because
3050 * If blocksize = 4096 offset = 3072 and len = 2048
3051 */
3052 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3053 - block;
3054 /*
3055 * credits to insert 1 extent into extent tree
3056 */
3057 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3058 mutex_lock(&inode->i_mutex);
3059 retry:
3060 while (ret >= 0 && ret < max_blocks) {
3061 block = block + ret;
3062 max_blocks = max_blocks - ret;
3063 handle = ext4_journal_start(inode, credits);
3064 if (IS_ERR(handle)) {
3065 ret = PTR_ERR(handle);
3066 break;
3067 }
3068 ret = ext4_get_blocks_wrap(handle, inode, block,
3069 max_blocks, &map_bh,
3070 EXT4_CREATE_UNINITIALIZED_EXT, 0, 0);
3071 if (ret <= 0) {
3072 #ifdef EXT4FS_DEBUG
3073 WARN_ON(ret <= 0);
3074 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3075 "returned error inode#%lu, block=%u, "
3076 "max_blocks=%u", __func__,
3077 inode->i_ino, block, max_blocks);
3078 #endif
3079 ext4_mark_inode_dirty(handle, inode);
3080 ret2 = ext4_journal_stop(handle);
3081 break;
3082 }
3083 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3084 blkbits) >> blkbits))
3085 new_size = offset + len;
3086 else
3087 new_size = (block + ret) << blkbits;
3088
3089 ext4_falloc_update_inode(inode, mode, new_size,
3090 buffer_new(&map_bh));
3091 ext4_mark_inode_dirty(handle, inode);
3092 ret2 = ext4_journal_stop(handle);
3093 if (ret2)
3094 break;
3095 }
3096 if (ret == -ENOSPC &&
3097 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3098 ret = 0;
3099 goto retry;
3100 }
3101 mutex_unlock(&inode->i_mutex);
3102 return ret > 0 ? ret2 : ret;
3103 }
3104
3105 /*
3106 * Callback function called for each extent to gather FIEMAP information.
3107 */
3108 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3109 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3110 void *data)
3111 {
3112 struct fiemap_extent_info *fieinfo = data;
3113 unsigned long blksize_bits = inode->i_sb->s_blocksize_bits;
3114 __u64 logical;
3115 __u64 physical;
3116 __u64 length;
3117 __u32 flags = 0;
3118 int error;
3119
3120 logical = (__u64)newex->ec_block << blksize_bits;
3121
3122 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3123 pgoff_t offset;
3124 struct page *page;
3125 struct buffer_head *bh = NULL;
3126
3127 offset = logical >> PAGE_SHIFT;
3128 page = find_get_page(inode->i_mapping, offset);
3129 if (!page || !page_has_buffers(page))
3130 return EXT_CONTINUE;
3131
3132 bh = page_buffers(page);
3133
3134 if (!bh)
3135 return EXT_CONTINUE;
3136
3137 if (buffer_delay(bh)) {
3138 flags |= FIEMAP_EXTENT_DELALLOC;
3139 page_cache_release(page);
3140 } else {
3141 page_cache_release(page);
3142 return EXT_CONTINUE;
3143 }
3144 }
3145
3146 physical = (__u64)newex->ec_start << blksize_bits;
3147 length = (__u64)newex->ec_len << blksize_bits;
3148
3149 if (ex && ext4_ext_is_uninitialized(ex))
3150 flags |= FIEMAP_EXTENT_UNWRITTEN;
3151
3152 /*
3153 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3154 *
3155 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3156 * this also indicates no more allocated blocks.
3157 *
3158 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3159 */
3160 if (logical + length - 1 == EXT_MAX_BLOCK ||
3161 ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK)
3162 flags |= FIEMAP_EXTENT_LAST;
3163
3164 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3165 length, flags);
3166 if (error < 0)
3167 return error;
3168 if (error == 1)
3169 return EXT_BREAK;
3170
3171 return EXT_CONTINUE;
3172 }
3173
3174 /* fiemap flags we can handle specified here */
3175 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3176
3177 static int ext4_xattr_fiemap(struct inode *inode,
3178 struct fiemap_extent_info *fieinfo)
3179 {
3180 __u64 physical = 0;
3181 __u64 length;
3182 __u32 flags = FIEMAP_EXTENT_LAST;
3183 int blockbits = inode->i_sb->s_blocksize_bits;
3184 int error = 0;
3185
3186 /* in-inode? */
3187 if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3188 struct ext4_iloc iloc;
3189 int offset; /* offset of xattr in inode */
3190
3191 error = ext4_get_inode_loc(inode, &iloc);
3192 if (error)
3193 return error;
3194 physical = iloc.bh->b_blocknr << blockbits;
3195 offset = EXT4_GOOD_OLD_INODE_SIZE +
3196 EXT4_I(inode)->i_extra_isize;
3197 physical += offset;
3198 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3199 flags |= FIEMAP_EXTENT_DATA_INLINE;
3200 } else { /* external block */
3201 physical = EXT4_I(inode)->i_file_acl << blockbits;
3202 length = inode->i_sb->s_blocksize;
3203 }
3204
3205 if (physical)
3206 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3207 length, flags);
3208 return (error < 0 ? error : 0);
3209 }
3210
3211 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3212 __u64 start, __u64 len)
3213 {
3214 ext4_lblk_t start_blk;
3215 ext4_lblk_t len_blks;
3216 int error = 0;
3217
3218 /* fallback to generic here if not in extents fmt */
3219 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3220 return generic_block_fiemap(inode, fieinfo, start, len,
3221 ext4_get_block);
3222
3223 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3224 return -EBADR;
3225
3226 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3227 error = ext4_xattr_fiemap(inode, fieinfo);
3228 } else {
3229 start_blk = start >> inode->i_sb->s_blocksize_bits;
3230 len_blks = len >> inode->i_sb->s_blocksize_bits;
3231
3232 /*
3233 * Walk the extent tree gathering extent information.
3234 * ext4_ext_fiemap_cb will push extents back to user.
3235 */
3236 down_write(&EXT4_I(inode)->i_data_sem);
3237 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3238 ext4_ext_fiemap_cb, fieinfo);
3239 up_write(&EXT4_I(inode)->i_data_sem);
3240 }
3241
3242 return error;
3243 }
3244