Loading...
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#include <trace/events/ext4.h>
48
49static int ext4_split_extent(handle_t *handle,
50 struct inode *inode,
51 struct ext4_ext_path *path,
52 struct ext4_map_blocks *map,
53 int split_flag,
54 int flags);
55
56static int ext4_ext_truncate_extend_restart(handle_t *handle,
57 struct inode *inode,
58 int needed)
59{
60 int err;
61
62 if (!ext4_handle_valid(handle))
63 return 0;
64 if (handle->h_buffer_credits > needed)
65 return 0;
66 err = ext4_journal_extend(handle, needed);
67 if (err <= 0)
68 return err;
69 err = ext4_truncate_restart_trans(handle, inode, needed);
70 if (err == 0)
71 err = -EAGAIN;
72
73 return err;
74}
75
76/*
77 * could return:
78 * - EROFS
79 * - ENOMEM
80 */
81static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
82 struct ext4_ext_path *path)
83{
84 if (path->p_bh) {
85 /* path points to block */
86 return ext4_journal_get_write_access(handle, path->p_bh);
87 }
88 /* path points to leaf/index in inode body */
89 /* we use in-core data, no need to protect them */
90 return 0;
91}
92
93/*
94 * could return:
95 * - EROFS
96 * - ENOMEM
97 * - EIO
98 */
99static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
100 struct ext4_ext_path *path)
101{
102 int err;
103 if (path->p_bh) {
104 /* path points to block */
105 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
106 } else {
107 /* path points to leaf/index in inode body */
108 err = ext4_mark_inode_dirty(handle, inode);
109 }
110 return err;
111}
112
113static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
114 struct ext4_ext_path *path,
115 ext4_lblk_t block)
116{
117 int depth;
118
119 if (path) {
120 struct ext4_extent *ex;
121 depth = path->p_depth;
122
123 /*
124 * Try to predict block placement assuming that we are
125 * filling in a file which will eventually be
126 * non-sparse --- i.e., in the case of libbfd writing
127 * an ELF object sections out-of-order but in a way
128 * the eventually results in a contiguous object or
129 * executable file, or some database extending a table
130 * space file. However, this is actually somewhat
131 * non-ideal if we are writing a sparse file such as
132 * qemu or KVM writing a raw image file that is going
133 * to stay fairly sparse, since it will end up
134 * fragmenting the file system's free space. Maybe we
135 * should have some hueristics or some way to allow
136 * userspace to pass a hint to file system,
137 * especially if the latter case turns out to be
138 * common.
139 */
140 ex = path[depth].p_ext;
141 if (ex) {
142 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
143 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
144
145 if (block > ext_block)
146 return ext_pblk + (block - ext_block);
147 else
148 return ext_pblk - (ext_block - block);
149 }
150
151 /* it looks like index is empty;
152 * try to find starting block from index itself */
153 if (path[depth].p_bh)
154 return path[depth].p_bh->b_blocknr;
155 }
156
157 /* OK. use inode's group */
158 return ext4_inode_to_goal_block(inode);
159}
160
161/*
162 * Allocation for a meta data block
163 */
164static ext4_fsblk_t
165ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
166 struct ext4_ext_path *path,
167 struct ext4_extent *ex, int *err, unsigned int flags)
168{
169 ext4_fsblk_t goal, newblock;
170
171 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
172 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
173 NULL, err);
174 return newblock;
175}
176
177static inline int ext4_ext_space_block(struct inode *inode, int check)
178{
179 int size;
180
181 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
182 / sizeof(struct ext4_extent);
183 if (!check) {
184#ifdef AGGRESSIVE_TEST
185 if (size > 6)
186 size = 6;
187#endif
188 }
189 return size;
190}
191
192static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
193{
194 int size;
195
196 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
197 / sizeof(struct ext4_extent_idx);
198 if (!check) {
199#ifdef AGGRESSIVE_TEST
200 if (size > 5)
201 size = 5;
202#endif
203 }
204 return size;
205}
206
207static inline int ext4_ext_space_root(struct inode *inode, int check)
208{
209 int size;
210
211 size = sizeof(EXT4_I(inode)->i_data);
212 size -= sizeof(struct ext4_extent_header);
213 size /= sizeof(struct ext4_extent);
214 if (!check) {
215#ifdef AGGRESSIVE_TEST
216 if (size > 3)
217 size = 3;
218#endif
219 }
220 return size;
221}
222
223static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
224{
225 int size;
226
227 size = sizeof(EXT4_I(inode)->i_data);
228 size -= sizeof(struct ext4_extent_header);
229 size /= sizeof(struct ext4_extent_idx);
230 if (!check) {
231#ifdef AGGRESSIVE_TEST
232 if (size > 4)
233 size = 4;
234#endif
235 }
236 return size;
237}
238
239/*
240 * Calculate the number of metadata blocks needed
241 * to allocate @blocks
242 * Worse case is one block per extent
243 */
244int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
245{
246 struct ext4_inode_info *ei = EXT4_I(inode);
247 int idxs, num = 0;
248
249 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
250 / sizeof(struct ext4_extent_idx));
251
252 /*
253 * If the new delayed allocation block is contiguous with the
254 * previous da block, it can share index blocks with the
255 * previous block, so we only need to allocate a new index
256 * block every idxs leaf blocks. At ldxs**2 blocks, we need
257 * an additional index block, and at ldxs**3 blocks, yet
258 * another index blocks.
259 */
260 if (ei->i_da_metadata_calc_len &&
261 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
262 if ((ei->i_da_metadata_calc_len % idxs) == 0)
263 num++;
264 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
265 num++;
266 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
267 num++;
268 ei->i_da_metadata_calc_len = 0;
269 } else
270 ei->i_da_metadata_calc_len++;
271 ei->i_da_metadata_calc_last_lblock++;
272 return num;
273 }
274
275 /*
276 * In the worst case we need a new set of index blocks at
277 * every level of the inode's extent tree.
278 */
279 ei->i_da_metadata_calc_len = 1;
280 ei->i_da_metadata_calc_last_lblock = lblock;
281 return ext_depth(inode) + 1;
282}
283
284static int
285ext4_ext_max_entries(struct inode *inode, int depth)
286{
287 int max;
288
289 if (depth == ext_depth(inode)) {
290 if (depth == 0)
291 max = ext4_ext_space_root(inode, 1);
292 else
293 max = ext4_ext_space_root_idx(inode, 1);
294 } else {
295 if (depth == 0)
296 max = ext4_ext_space_block(inode, 1);
297 else
298 max = ext4_ext_space_block_idx(inode, 1);
299 }
300
301 return max;
302}
303
304static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
305{
306 ext4_fsblk_t block = ext4_ext_pblock(ext);
307 int len = ext4_ext_get_actual_len(ext);
308
309 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
310}
311
312static int ext4_valid_extent_idx(struct inode *inode,
313 struct ext4_extent_idx *ext_idx)
314{
315 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
316
317 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
318}
319
320static int ext4_valid_extent_entries(struct inode *inode,
321 struct ext4_extent_header *eh,
322 int depth)
323{
324 struct ext4_extent *ext;
325 struct ext4_extent_idx *ext_idx;
326 unsigned short entries;
327 if (eh->eh_entries == 0)
328 return 1;
329
330 entries = le16_to_cpu(eh->eh_entries);
331
332 if (depth == 0) {
333 /* leaf entries */
334 ext = EXT_FIRST_EXTENT(eh);
335 while (entries) {
336 if (!ext4_valid_extent(inode, ext))
337 return 0;
338 ext++;
339 entries--;
340 }
341 } else {
342 ext_idx = EXT_FIRST_INDEX(eh);
343 while (entries) {
344 if (!ext4_valid_extent_idx(inode, ext_idx))
345 return 0;
346 ext_idx++;
347 entries--;
348 }
349 }
350 return 1;
351}
352
353static int __ext4_ext_check(const char *function, unsigned int line,
354 struct inode *inode, struct ext4_extent_header *eh,
355 int depth)
356{
357 const char *error_msg;
358 int max = 0;
359
360 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
361 error_msg = "invalid magic";
362 goto corrupted;
363 }
364 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
365 error_msg = "unexpected eh_depth";
366 goto corrupted;
367 }
368 if (unlikely(eh->eh_max == 0)) {
369 error_msg = "invalid eh_max";
370 goto corrupted;
371 }
372 max = ext4_ext_max_entries(inode, depth);
373 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
374 error_msg = "too large eh_max";
375 goto corrupted;
376 }
377 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
378 error_msg = "invalid eh_entries";
379 goto corrupted;
380 }
381 if (!ext4_valid_extent_entries(inode, eh, depth)) {
382 error_msg = "invalid extent entries";
383 goto corrupted;
384 }
385 return 0;
386
387corrupted:
388 ext4_error_inode(inode, function, line, 0,
389 "bad header/extent: %s - magic %x, "
390 "entries %u, max %u(%u), depth %u(%u)",
391 error_msg, le16_to_cpu(eh->eh_magic),
392 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
393 max, le16_to_cpu(eh->eh_depth), depth);
394
395 return -EIO;
396}
397
398#define ext4_ext_check(inode, eh, depth) \
399 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
400
401int ext4_ext_check_inode(struct inode *inode)
402{
403 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
404}
405
406#ifdef EXT_DEBUG
407static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
408{
409 int k, l = path->p_depth;
410
411 ext_debug("path:");
412 for (k = 0; k <= l; k++, path++) {
413 if (path->p_idx) {
414 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
415 ext4_idx_pblock(path->p_idx));
416 } else if (path->p_ext) {
417 ext_debug(" %d:[%d]%d:%llu ",
418 le32_to_cpu(path->p_ext->ee_block),
419 ext4_ext_is_uninitialized(path->p_ext),
420 ext4_ext_get_actual_len(path->p_ext),
421 ext4_ext_pblock(path->p_ext));
422 } else
423 ext_debug(" []");
424 }
425 ext_debug("\n");
426}
427
428static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
429{
430 int depth = ext_depth(inode);
431 struct ext4_extent_header *eh;
432 struct ext4_extent *ex;
433 int i;
434
435 if (!path)
436 return;
437
438 eh = path[depth].p_hdr;
439 ex = EXT_FIRST_EXTENT(eh);
440
441 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
442
443 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
444 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
445 ext4_ext_is_uninitialized(ex),
446 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
447 }
448 ext_debug("\n");
449}
450
451static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
452 ext4_fsblk_t newblock, int level)
453{
454 int depth = ext_depth(inode);
455 struct ext4_extent *ex;
456
457 if (depth != level) {
458 struct ext4_extent_idx *idx;
459 idx = path[level].p_idx;
460 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
461 ext_debug("%d: move %d:%llu in new index %llu\n", level,
462 le32_to_cpu(idx->ei_block),
463 ext4_idx_pblock(idx),
464 newblock);
465 idx++;
466 }
467
468 return;
469 }
470
471 ex = path[depth].p_ext;
472 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
473 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
474 le32_to_cpu(ex->ee_block),
475 ext4_ext_pblock(ex),
476 ext4_ext_is_uninitialized(ex),
477 ext4_ext_get_actual_len(ex),
478 newblock);
479 ex++;
480 }
481}
482
483#else
484#define ext4_ext_show_path(inode, path)
485#define ext4_ext_show_leaf(inode, path)
486#define ext4_ext_show_move(inode, path, newblock, level)
487#endif
488
489void ext4_ext_drop_refs(struct ext4_ext_path *path)
490{
491 int depth = path->p_depth;
492 int i;
493
494 for (i = 0; i <= depth; i++, path++)
495 if (path->p_bh) {
496 brelse(path->p_bh);
497 path->p_bh = NULL;
498 }
499}
500
501/*
502 * ext4_ext_binsearch_idx:
503 * binary search for the closest index of the given block
504 * the header must be checked before calling this
505 */
506static void
507ext4_ext_binsearch_idx(struct inode *inode,
508 struct ext4_ext_path *path, ext4_lblk_t block)
509{
510 struct ext4_extent_header *eh = path->p_hdr;
511 struct ext4_extent_idx *r, *l, *m;
512
513
514 ext_debug("binsearch for %u(idx): ", block);
515
516 l = EXT_FIRST_INDEX(eh) + 1;
517 r = EXT_LAST_INDEX(eh);
518 while (l <= r) {
519 m = l + (r - l) / 2;
520 if (block < le32_to_cpu(m->ei_block))
521 r = m - 1;
522 else
523 l = m + 1;
524 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
525 m, le32_to_cpu(m->ei_block),
526 r, le32_to_cpu(r->ei_block));
527 }
528
529 path->p_idx = l - 1;
530 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
531 ext4_idx_pblock(path->p_idx));
532
533#ifdef CHECK_BINSEARCH
534 {
535 struct ext4_extent_idx *chix, *ix;
536 int k;
537
538 chix = ix = EXT_FIRST_INDEX(eh);
539 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
540 if (k != 0 &&
541 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
542 printk(KERN_DEBUG "k=%d, ix=0x%p, "
543 "first=0x%p\n", k,
544 ix, EXT_FIRST_INDEX(eh));
545 printk(KERN_DEBUG "%u <= %u\n",
546 le32_to_cpu(ix->ei_block),
547 le32_to_cpu(ix[-1].ei_block));
548 }
549 BUG_ON(k && le32_to_cpu(ix->ei_block)
550 <= le32_to_cpu(ix[-1].ei_block));
551 if (block < le32_to_cpu(ix->ei_block))
552 break;
553 chix = ix;
554 }
555 BUG_ON(chix != path->p_idx);
556 }
557#endif
558
559}
560
561/*
562 * ext4_ext_binsearch:
563 * binary search for closest extent of the given block
564 * the header must be checked before calling this
565 */
566static void
567ext4_ext_binsearch(struct inode *inode,
568 struct ext4_ext_path *path, ext4_lblk_t block)
569{
570 struct ext4_extent_header *eh = path->p_hdr;
571 struct ext4_extent *r, *l, *m;
572
573 if (eh->eh_entries == 0) {
574 /*
575 * this leaf is empty:
576 * we get such a leaf in split/add case
577 */
578 return;
579 }
580
581 ext_debug("binsearch for %u: ", block);
582
583 l = EXT_FIRST_EXTENT(eh) + 1;
584 r = EXT_LAST_EXTENT(eh);
585
586 while (l <= r) {
587 m = l + (r - l) / 2;
588 if (block < le32_to_cpu(m->ee_block))
589 r = m - 1;
590 else
591 l = m + 1;
592 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
593 m, le32_to_cpu(m->ee_block),
594 r, le32_to_cpu(r->ee_block));
595 }
596
597 path->p_ext = l - 1;
598 ext_debug(" -> %d:%llu:[%d]%d ",
599 le32_to_cpu(path->p_ext->ee_block),
600 ext4_ext_pblock(path->p_ext),
601 ext4_ext_is_uninitialized(path->p_ext),
602 ext4_ext_get_actual_len(path->p_ext));
603
604#ifdef CHECK_BINSEARCH
605 {
606 struct ext4_extent *chex, *ex;
607 int k;
608
609 chex = ex = EXT_FIRST_EXTENT(eh);
610 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
611 BUG_ON(k && le32_to_cpu(ex->ee_block)
612 <= le32_to_cpu(ex[-1].ee_block));
613 if (block < le32_to_cpu(ex->ee_block))
614 break;
615 chex = ex;
616 }
617 BUG_ON(chex != path->p_ext);
618 }
619#endif
620
621}
622
623int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
624{
625 struct ext4_extent_header *eh;
626
627 eh = ext_inode_hdr(inode);
628 eh->eh_depth = 0;
629 eh->eh_entries = 0;
630 eh->eh_magic = EXT4_EXT_MAGIC;
631 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
632 ext4_mark_inode_dirty(handle, inode);
633 ext4_ext_invalidate_cache(inode);
634 return 0;
635}
636
637struct ext4_ext_path *
638ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
639 struct ext4_ext_path *path)
640{
641 struct ext4_extent_header *eh;
642 struct buffer_head *bh;
643 short int depth, i, ppos = 0, alloc = 0;
644
645 eh = ext_inode_hdr(inode);
646 depth = ext_depth(inode);
647
648 /* account possible depth increase */
649 if (!path) {
650 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
651 GFP_NOFS);
652 if (!path)
653 return ERR_PTR(-ENOMEM);
654 alloc = 1;
655 }
656 path[0].p_hdr = eh;
657 path[0].p_bh = NULL;
658
659 i = depth;
660 /* walk through the tree */
661 while (i) {
662 int need_to_validate = 0;
663
664 ext_debug("depth %d: num %d, max %d\n",
665 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
666
667 ext4_ext_binsearch_idx(inode, path + ppos, block);
668 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
669 path[ppos].p_depth = i;
670 path[ppos].p_ext = NULL;
671
672 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
673 if (unlikely(!bh))
674 goto err;
675 if (!bh_uptodate_or_lock(bh)) {
676 trace_ext4_ext_load_extent(inode, block,
677 path[ppos].p_block);
678 if (bh_submit_read(bh) < 0) {
679 put_bh(bh);
680 goto err;
681 }
682 /* validate the extent entries */
683 need_to_validate = 1;
684 }
685 eh = ext_block_hdr(bh);
686 ppos++;
687 if (unlikely(ppos > depth)) {
688 put_bh(bh);
689 EXT4_ERROR_INODE(inode,
690 "ppos %d > depth %d", ppos, depth);
691 goto err;
692 }
693 path[ppos].p_bh = bh;
694 path[ppos].p_hdr = eh;
695 i--;
696
697 if (need_to_validate && ext4_ext_check(inode, eh, i))
698 goto err;
699 }
700
701 path[ppos].p_depth = i;
702 path[ppos].p_ext = NULL;
703 path[ppos].p_idx = NULL;
704
705 /* find extent */
706 ext4_ext_binsearch(inode, path + ppos, block);
707 /* if not an empty leaf */
708 if (path[ppos].p_ext)
709 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
710
711 ext4_ext_show_path(inode, path);
712
713 return path;
714
715err:
716 ext4_ext_drop_refs(path);
717 if (alloc)
718 kfree(path);
719 return ERR_PTR(-EIO);
720}
721
722/*
723 * ext4_ext_insert_index:
724 * insert new index [@logical;@ptr] into the block at @curp;
725 * check where to insert: before @curp or after @curp
726 */
727static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
728 struct ext4_ext_path *curp,
729 int logical, ext4_fsblk_t ptr)
730{
731 struct ext4_extent_idx *ix;
732 int len, err;
733
734 err = ext4_ext_get_access(handle, inode, curp);
735 if (err)
736 return err;
737
738 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
739 EXT4_ERROR_INODE(inode,
740 "logical %d == ei_block %d!",
741 logical, le32_to_cpu(curp->p_idx->ei_block));
742 return -EIO;
743 }
744
745 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
746 >= le16_to_cpu(curp->p_hdr->eh_max))) {
747 EXT4_ERROR_INODE(inode,
748 "eh_entries %d >= eh_max %d!",
749 le16_to_cpu(curp->p_hdr->eh_entries),
750 le16_to_cpu(curp->p_hdr->eh_max));
751 return -EIO;
752 }
753
754 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
755 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
756 /* insert after */
757 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
758 len = (len - 1) * sizeof(struct ext4_extent_idx);
759 len = len < 0 ? 0 : len;
760 ext_debug("insert new index %d after: %llu. "
761 "move %d from 0x%p to 0x%p\n",
762 logical, ptr, len,
763 (curp->p_idx + 1), (curp->p_idx + 2));
764 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
765 }
766 ix = curp->p_idx + 1;
767 } else {
768 /* insert before */
769 len = len * sizeof(struct ext4_extent_idx);
770 len = len < 0 ? 0 : len;
771 ext_debug("insert new index %d before: %llu. "
772 "move %d from 0x%p to 0x%p\n",
773 logical, ptr, len,
774 curp->p_idx, (curp->p_idx + 1));
775 memmove(curp->p_idx + 1, curp->p_idx, len);
776 ix = curp->p_idx;
777 }
778
779 ix->ei_block = cpu_to_le32(logical);
780 ext4_idx_store_pblock(ix, ptr);
781 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
782
783 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
784 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
785 return -EIO;
786 }
787
788 err = ext4_ext_dirty(handle, inode, curp);
789 ext4_std_error(inode->i_sb, err);
790
791 return err;
792}
793
794/*
795 * ext4_ext_split:
796 * inserts new subtree into the path, using free index entry
797 * at depth @at:
798 * - allocates all needed blocks (new leaf and all intermediate index blocks)
799 * - makes decision where to split
800 * - moves remaining extents and index entries (right to the split point)
801 * into the newly allocated blocks
802 * - initializes subtree
803 */
804static int ext4_ext_split(handle_t *handle, struct inode *inode,
805 unsigned int flags,
806 struct ext4_ext_path *path,
807 struct ext4_extent *newext, int at)
808{
809 struct buffer_head *bh = NULL;
810 int depth = ext_depth(inode);
811 struct ext4_extent_header *neh;
812 struct ext4_extent_idx *fidx;
813 int i = at, k, m, a;
814 ext4_fsblk_t newblock, oldblock;
815 __le32 border;
816 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
817 int err = 0;
818
819 /* make decision: where to split? */
820 /* FIXME: now decision is simplest: at current extent */
821
822 /* if current leaf will be split, then we should use
823 * border from split point */
824 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
825 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
826 return -EIO;
827 }
828 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
829 border = path[depth].p_ext[1].ee_block;
830 ext_debug("leaf will be split."
831 " next leaf starts at %d\n",
832 le32_to_cpu(border));
833 } else {
834 border = newext->ee_block;
835 ext_debug("leaf will be added."
836 " next leaf starts at %d\n",
837 le32_to_cpu(border));
838 }
839
840 /*
841 * If error occurs, then we break processing
842 * and mark filesystem read-only. index won't
843 * be inserted and tree will be in consistent
844 * state. Next mount will repair buffers too.
845 */
846
847 /*
848 * Get array to track all allocated blocks.
849 * We need this to handle errors and free blocks
850 * upon them.
851 */
852 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
853 if (!ablocks)
854 return -ENOMEM;
855
856 /* allocate all needed blocks */
857 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
858 for (a = 0; a < depth - at; a++) {
859 newblock = ext4_ext_new_meta_block(handle, inode, path,
860 newext, &err, flags);
861 if (newblock == 0)
862 goto cleanup;
863 ablocks[a] = newblock;
864 }
865
866 /* initialize new leaf */
867 newblock = ablocks[--a];
868 if (unlikely(newblock == 0)) {
869 EXT4_ERROR_INODE(inode, "newblock == 0!");
870 err = -EIO;
871 goto cleanup;
872 }
873 bh = sb_getblk(inode->i_sb, newblock);
874 if (!bh) {
875 err = -EIO;
876 goto cleanup;
877 }
878 lock_buffer(bh);
879
880 err = ext4_journal_get_create_access(handle, bh);
881 if (err)
882 goto cleanup;
883
884 neh = ext_block_hdr(bh);
885 neh->eh_entries = 0;
886 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
887 neh->eh_magic = EXT4_EXT_MAGIC;
888 neh->eh_depth = 0;
889
890 /* move remainder of path[depth] to the new leaf */
891 if (unlikely(path[depth].p_hdr->eh_entries !=
892 path[depth].p_hdr->eh_max)) {
893 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
894 path[depth].p_hdr->eh_entries,
895 path[depth].p_hdr->eh_max);
896 err = -EIO;
897 goto cleanup;
898 }
899 /* start copy from next extent */
900 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
901 ext4_ext_show_move(inode, path, newblock, depth);
902 if (m) {
903 struct ext4_extent *ex;
904 ex = EXT_FIRST_EXTENT(neh);
905 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
906 le16_add_cpu(&neh->eh_entries, m);
907 }
908
909 set_buffer_uptodate(bh);
910 unlock_buffer(bh);
911
912 err = ext4_handle_dirty_metadata(handle, inode, bh);
913 if (err)
914 goto cleanup;
915 brelse(bh);
916 bh = NULL;
917
918 /* correct old leaf */
919 if (m) {
920 err = ext4_ext_get_access(handle, inode, path + depth);
921 if (err)
922 goto cleanup;
923 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
924 err = ext4_ext_dirty(handle, inode, path + depth);
925 if (err)
926 goto cleanup;
927
928 }
929
930 /* create intermediate indexes */
931 k = depth - at - 1;
932 if (unlikely(k < 0)) {
933 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
934 err = -EIO;
935 goto cleanup;
936 }
937 if (k)
938 ext_debug("create %d intermediate indices\n", k);
939 /* insert new index into current index block */
940 /* current depth stored in i var */
941 i = depth - 1;
942 while (k--) {
943 oldblock = newblock;
944 newblock = ablocks[--a];
945 bh = sb_getblk(inode->i_sb, newblock);
946 if (!bh) {
947 err = -EIO;
948 goto cleanup;
949 }
950 lock_buffer(bh);
951
952 err = ext4_journal_get_create_access(handle, bh);
953 if (err)
954 goto cleanup;
955
956 neh = ext_block_hdr(bh);
957 neh->eh_entries = cpu_to_le16(1);
958 neh->eh_magic = EXT4_EXT_MAGIC;
959 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
960 neh->eh_depth = cpu_to_le16(depth - i);
961 fidx = EXT_FIRST_INDEX(neh);
962 fidx->ei_block = border;
963 ext4_idx_store_pblock(fidx, oldblock);
964
965 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
966 i, newblock, le32_to_cpu(border), oldblock);
967
968 /* move remainder of path[i] to the new index block */
969 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
970 EXT_LAST_INDEX(path[i].p_hdr))) {
971 EXT4_ERROR_INODE(inode,
972 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
973 le32_to_cpu(path[i].p_ext->ee_block));
974 err = -EIO;
975 goto cleanup;
976 }
977 /* start copy indexes */
978 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
979 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
980 EXT_MAX_INDEX(path[i].p_hdr));
981 ext4_ext_show_move(inode, path, newblock, i);
982 if (m) {
983 memmove(++fidx, path[i].p_idx,
984 sizeof(struct ext4_extent_idx) * m);
985 le16_add_cpu(&neh->eh_entries, m);
986 }
987 set_buffer_uptodate(bh);
988 unlock_buffer(bh);
989
990 err = ext4_handle_dirty_metadata(handle, inode, bh);
991 if (err)
992 goto cleanup;
993 brelse(bh);
994 bh = NULL;
995
996 /* correct old index */
997 if (m) {
998 err = ext4_ext_get_access(handle, inode, path + i);
999 if (err)
1000 goto cleanup;
1001 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1002 err = ext4_ext_dirty(handle, inode, path + i);
1003 if (err)
1004 goto cleanup;
1005 }
1006
1007 i--;
1008 }
1009
1010 /* insert new index */
1011 err = ext4_ext_insert_index(handle, inode, path + at,
1012 le32_to_cpu(border), newblock);
1013
1014cleanup:
1015 if (bh) {
1016 if (buffer_locked(bh))
1017 unlock_buffer(bh);
1018 brelse(bh);
1019 }
1020
1021 if (err) {
1022 /* free all allocated blocks in error case */
1023 for (i = 0; i < depth; i++) {
1024 if (!ablocks[i])
1025 continue;
1026 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1027 EXT4_FREE_BLOCKS_METADATA);
1028 }
1029 }
1030 kfree(ablocks);
1031
1032 return err;
1033}
1034
1035/*
1036 * ext4_ext_grow_indepth:
1037 * implements tree growing procedure:
1038 * - allocates new block
1039 * - moves top-level data (index block or leaf) into the new block
1040 * - initializes new top-level, creating index that points to the
1041 * just created block
1042 */
1043static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1044 unsigned int flags,
1045 struct ext4_ext_path *path,
1046 struct ext4_extent *newext)
1047{
1048 struct ext4_ext_path *curp = path;
1049 struct ext4_extent_header *neh;
1050 struct buffer_head *bh;
1051 ext4_fsblk_t newblock;
1052 int err = 0;
1053
1054 newblock = ext4_ext_new_meta_block(handle, inode, path,
1055 newext, &err, flags);
1056 if (newblock == 0)
1057 return err;
1058
1059 bh = sb_getblk(inode->i_sb, newblock);
1060 if (!bh) {
1061 err = -EIO;
1062 ext4_std_error(inode->i_sb, err);
1063 return err;
1064 }
1065 lock_buffer(bh);
1066
1067 err = ext4_journal_get_create_access(handle, bh);
1068 if (err) {
1069 unlock_buffer(bh);
1070 goto out;
1071 }
1072
1073 /* move top-level index/leaf into new block */
1074 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1075
1076 /* set size of new block */
1077 neh = ext_block_hdr(bh);
1078 /* old root could have indexes or leaves
1079 * so calculate e_max right way */
1080 if (ext_depth(inode))
1081 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1082 else
1083 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1084 neh->eh_magic = EXT4_EXT_MAGIC;
1085 set_buffer_uptodate(bh);
1086 unlock_buffer(bh);
1087
1088 err = ext4_handle_dirty_metadata(handle, inode, bh);
1089 if (err)
1090 goto out;
1091
1092 /* create index in new top-level index: num,max,pointer */
1093 err = ext4_ext_get_access(handle, inode, curp);
1094 if (err)
1095 goto out;
1096
1097 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1098 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1099 curp->p_hdr->eh_entries = cpu_to_le16(1);
1100 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1101
1102 if (path[0].p_hdr->eh_depth)
1103 curp->p_idx->ei_block =
1104 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1105 else
1106 curp->p_idx->ei_block =
1107 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1108 ext4_idx_store_pblock(curp->p_idx, newblock);
1109
1110 neh = ext_inode_hdr(inode);
1111 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1112 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1113 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1114 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1115
1116 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1117 err = ext4_ext_dirty(handle, inode, curp);
1118out:
1119 brelse(bh);
1120
1121 return err;
1122}
1123
1124/*
1125 * ext4_ext_create_new_leaf:
1126 * finds empty index and adds new leaf.
1127 * if no free index is found, then it requests in-depth growing.
1128 */
1129static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1130 unsigned int flags,
1131 struct ext4_ext_path *path,
1132 struct ext4_extent *newext)
1133{
1134 struct ext4_ext_path *curp;
1135 int depth, i, err = 0;
1136
1137repeat:
1138 i = depth = ext_depth(inode);
1139
1140 /* walk up to the tree and look for free index entry */
1141 curp = path + depth;
1142 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1143 i--;
1144 curp--;
1145 }
1146
1147 /* we use already allocated block for index block,
1148 * so subsequent data blocks should be contiguous */
1149 if (EXT_HAS_FREE_INDEX(curp)) {
1150 /* if we found index with free entry, then use that
1151 * entry: create all needed subtree and add new leaf */
1152 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1153 if (err)
1154 goto out;
1155
1156 /* refill path */
1157 ext4_ext_drop_refs(path);
1158 path = ext4_ext_find_extent(inode,
1159 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1160 path);
1161 if (IS_ERR(path))
1162 err = PTR_ERR(path);
1163 } else {
1164 /* tree is full, time to grow in depth */
1165 err = ext4_ext_grow_indepth(handle, inode, flags,
1166 path, newext);
1167 if (err)
1168 goto out;
1169
1170 /* refill path */
1171 ext4_ext_drop_refs(path);
1172 path = ext4_ext_find_extent(inode,
1173 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1174 path);
1175 if (IS_ERR(path)) {
1176 err = PTR_ERR(path);
1177 goto out;
1178 }
1179
1180 /*
1181 * only first (depth 0 -> 1) produces free space;
1182 * in all other cases we have to split the grown tree
1183 */
1184 depth = ext_depth(inode);
1185 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1186 /* now we need to split */
1187 goto repeat;
1188 }
1189 }
1190
1191out:
1192 return err;
1193}
1194
1195/*
1196 * search the closest allocated block to the left for *logical
1197 * and returns it at @logical + it's physical address at @phys
1198 * if *logical is the smallest allocated block, the function
1199 * returns 0 at @phys
1200 * return value contains 0 (success) or error code
1201 */
1202static int ext4_ext_search_left(struct inode *inode,
1203 struct ext4_ext_path *path,
1204 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1205{
1206 struct ext4_extent_idx *ix;
1207 struct ext4_extent *ex;
1208 int depth, ee_len;
1209
1210 if (unlikely(path == NULL)) {
1211 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1212 return -EIO;
1213 }
1214 depth = path->p_depth;
1215 *phys = 0;
1216
1217 if (depth == 0 && path->p_ext == NULL)
1218 return 0;
1219
1220 /* usually extent in the path covers blocks smaller
1221 * then *logical, but it can be that extent is the
1222 * first one in the file */
1223
1224 ex = path[depth].p_ext;
1225 ee_len = ext4_ext_get_actual_len(ex);
1226 if (*logical < le32_to_cpu(ex->ee_block)) {
1227 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1228 EXT4_ERROR_INODE(inode,
1229 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1230 *logical, le32_to_cpu(ex->ee_block));
1231 return -EIO;
1232 }
1233 while (--depth >= 0) {
1234 ix = path[depth].p_idx;
1235 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1236 EXT4_ERROR_INODE(inode,
1237 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1238 ix != NULL ? ix->ei_block : 0,
1239 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1240 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1241 depth);
1242 return -EIO;
1243 }
1244 }
1245 return 0;
1246 }
1247
1248 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1249 EXT4_ERROR_INODE(inode,
1250 "logical %d < ee_block %d + ee_len %d!",
1251 *logical, le32_to_cpu(ex->ee_block), ee_len);
1252 return -EIO;
1253 }
1254
1255 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1256 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1257 return 0;
1258}
1259
1260/*
1261 * search the closest allocated block to the right for *logical
1262 * and returns it at @logical + it's physical address at @phys
1263 * if *logical is the smallest allocated block, the function
1264 * returns 0 at @phys
1265 * return value contains 0 (success) or error code
1266 */
1267static int ext4_ext_search_right(struct inode *inode,
1268 struct ext4_ext_path *path,
1269 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1270{
1271 struct buffer_head *bh = NULL;
1272 struct ext4_extent_header *eh;
1273 struct ext4_extent_idx *ix;
1274 struct ext4_extent *ex;
1275 ext4_fsblk_t block;
1276 int depth; /* Note, NOT eh_depth; depth from top of tree */
1277 int ee_len;
1278
1279 if (unlikely(path == NULL)) {
1280 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1281 return -EIO;
1282 }
1283 depth = path->p_depth;
1284 *phys = 0;
1285
1286 if (depth == 0 && path->p_ext == NULL)
1287 return 0;
1288
1289 /* usually extent in the path covers blocks smaller
1290 * then *logical, but it can be that extent is the
1291 * first one in the file */
1292
1293 ex = path[depth].p_ext;
1294 ee_len = ext4_ext_get_actual_len(ex);
1295 if (*logical < le32_to_cpu(ex->ee_block)) {
1296 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1297 EXT4_ERROR_INODE(inode,
1298 "first_extent(path[%d].p_hdr) != ex",
1299 depth);
1300 return -EIO;
1301 }
1302 while (--depth >= 0) {
1303 ix = path[depth].p_idx;
1304 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1305 EXT4_ERROR_INODE(inode,
1306 "ix != EXT_FIRST_INDEX *logical %d!",
1307 *logical);
1308 return -EIO;
1309 }
1310 }
1311 *logical = le32_to_cpu(ex->ee_block);
1312 *phys = ext4_ext_pblock(ex);
1313 return 0;
1314 }
1315
1316 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1317 EXT4_ERROR_INODE(inode,
1318 "logical %d < ee_block %d + ee_len %d!",
1319 *logical, le32_to_cpu(ex->ee_block), ee_len);
1320 return -EIO;
1321 }
1322
1323 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1324 /* next allocated block in this leaf */
1325 ex++;
1326 *logical = le32_to_cpu(ex->ee_block);
1327 *phys = ext4_ext_pblock(ex);
1328 return 0;
1329 }
1330
1331 /* go up and search for index to the right */
1332 while (--depth >= 0) {
1333 ix = path[depth].p_idx;
1334 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1335 goto got_index;
1336 }
1337
1338 /* we've gone up to the root and found no index to the right */
1339 return 0;
1340
1341got_index:
1342 /* we've found index to the right, let's
1343 * follow it and find the closest allocated
1344 * block to the right */
1345 ix++;
1346 block = ext4_idx_pblock(ix);
1347 while (++depth < path->p_depth) {
1348 bh = sb_bread(inode->i_sb, block);
1349 if (bh == NULL)
1350 return -EIO;
1351 eh = ext_block_hdr(bh);
1352 /* subtract from p_depth to get proper eh_depth */
1353 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1354 put_bh(bh);
1355 return -EIO;
1356 }
1357 ix = EXT_FIRST_INDEX(eh);
1358 block = ext4_idx_pblock(ix);
1359 put_bh(bh);
1360 }
1361
1362 bh = sb_bread(inode->i_sb, block);
1363 if (bh == NULL)
1364 return -EIO;
1365 eh = ext_block_hdr(bh);
1366 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1367 put_bh(bh);
1368 return -EIO;
1369 }
1370 ex = EXT_FIRST_EXTENT(eh);
1371 *logical = le32_to_cpu(ex->ee_block);
1372 *phys = ext4_ext_pblock(ex);
1373 put_bh(bh);
1374 return 0;
1375}
1376
1377/*
1378 * ext4_ext_next_allocated_block:
1379 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1380 * NOTE: it considers block number from index entry as
1381 * allocated block. Thus, index entries have to be consistent
1382 * with leaves.
1383 */
1384static ext4_lblk_t
1385ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1386{
1387 int depth;
1388
1389 BUG_ON(path == NULL);
1390 depth = path->p_depth;
1391
1392 if (depth == 0 && path->p_ext == NULL)
1393 return EXT_MAX_BLOCKS;
1394
1395 while (depth >= 0) {
1396 if (depth == path->p_depth) {
1397 /* leaf */
1398 if (path[depth].p_ext !=
1399 EXT_LAST_EXTENT(path[depth].p_hdr))
1400 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1401 } else {
1402 /* index */
1403 if (path[depth].p_idx !=
1404 EXT_LAST_INDEX(path[depth].p_hdr))
1405 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1406 }
1407 depth--;
1408 }
1409
1410 return EXT_MAX_BLOCKS;
1411}
1412
1413/*
1414 * ext4_ext_next_leaf_block:
1415 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1416 */
1417static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1418{
1419 int depth;
1420
1421 BUG_ON(path == NULL);
1422 depth = path->p_depth;
1423
1424 /* zero-tree has no leaf blocks at all */
1425 if (depth == 0)
1426 return EXT_MAX_BLOCKS;
1427
1428 /* go to index block */
1429 depth--;
1430
1431 while (depth >= 0) {
1432 if (path[depth].p_idx !=
1433 EXT_LAST_INDEX(path[depth].p_hdr))
1434 return (ext4_lblk_t)
1435 le32_to_cpu(path[depth].p_idx[1].ei_block);
1436 depth--;
1437 }
1438
1439 return EXT_MAX_BLOCKS;
1440}
1441
1442/*
1443 * ext4_ext_correct_indexes:
1444 * if leaf gets modified and modified extent is first in the leaf,
1445 * then we have to correct all indexes above.
1446 * TODO: do we need to correct tree in all cases?
1447 */
1448static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1449 struct ext4_ext_path *path)
1450{
1451 struct ext4_extent_header *eh;
1452 int depth = ext_depth(inode);
1453 struct ext4_extent *ex;
1454 __le32 border;
1455 int k, err = 0;
1456
1457 eh = path[depth].p_hdr;
1458 ex = path[depth].p_ext;
1459
1460 if (unlikely(ex == NULL || eh == NULL)) {
1461 EXT4_ERROR_INODE(inode,
1462 "ex %p == NULL or eh %p == NULL", ex, eh);
1463 return -EIO;
1464 }
1465
1466 if (depth == 0) {
1467 /* there is no tree at all */
1468 return 0;
1469 }
1470
1471 if (ex != EXT_FIRST_EXTENT(eh)) {
1472 /* we correct tree if first leaf got modified only */
1473 return 0;
1474 }
1475
1476 /*
1477 * TODO: we need correction if border is smaller than current one
1478 */
1479 k = depth - 1;
1480 border = path[depth].p_ext->ee_block;
1481 err = ext4_ext_get_access(handle, inode, path + k);
1482 if (err)
1483 return err;
1484 path[k].p_idx->ei_block = border;
1485 err = ext4_ext_dirty(handle, inode, path + k);
1486 if (err)
1487 return err;
1488
1489 while (k--) {
1490 /* change all left-side indexes */
1491 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1492 break;
1493 err = ext4_ext_get_access(handle, inode, path + k);
1494 if (err)
1495 break;
1496 path[k].p_idx->ei_block = border;
1497 err = ext4_ext_dirty(handle, inode, path + k);
1498 if (err)
1499 break;
1500 }
1501
1502 return err;
1503}
1504
1505int
1506ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1507 struct ext4_extent *ex2)
1508{
1509 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1510
1511 /*
1512 * Make sure that either both extents are uninitialized, or
1513 * both are _not_.
1514 */
1515 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1516 return 0;
1517
1518 if (ext4_ext_is_uninitialized(ex1))
1519 max_len = EXT_UNINIT_MAX_LEN;
1520 else
1521 max_len = EXT_INIT_MAX_LEN;
1522
1523 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1524 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1525
1526 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1527 le32_to_cpu(ex2->ee_block))
1528 return 0;
1529
1530 /*
1531 * To allow future support for preallocated extents to be added
1532 * as an RO_COMPAT feature, refuse to merge to extents if
1533 * this can result in the top bit of ee_len being set.
1534 */
1535 if (ext1_ee_len + ext2_ee_len > max_len)
1536 return 0;
1537#ifdef AGGRESSIVE_TEST
1538 if (ext1_ee_len >= 4)
1539 return 0;
1540#endif
1541
1542 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1543 return 1;
1544 return 0;
1545}
1546
1547/*
1548 * This function tries to merge the "ex" extent to the next extent in the tree.
1549 * It always tries to merge towards right. If you want to merge towards
1550 * left, pass "ex - 1" as argument instead of "ex".
1551 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1552 * 1 if they got merged.
1553 */
1554static int ext4_ext_try_to_merge_right(struct inode *inode,
1555 struct ext4_ext_path *path,
1556 struct ext4_extent *ex)
1557{
1558 struct ext4_extent_header *eh;
1559 unsigned int depth, len;
1560 int merge_done = 0;
1561 int uninitialized = 0;
1562
1563 depth = ext_depth(inode);
1564 BUG_ON(path[depth].p_hdr == NULL);
1565 eh = path[depth].p_hdr;
1566
1567 while (ex < EXT_LAST_EXTENT(eh)) {
1568 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1569 break;
1570 /* merge with next extent! */
1571 if (ext4_ext_is_uninitialized(ex))
1572 uninitialized = 1;
1573 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1574 + ext4_ext_get_actual_len(ex + 1));
1575 if (uninitialized)
1576 ext4_ext_mark_uninitialized(ex);
1577
1578 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1579 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1580 * sizeof(struct ext4_extent);
1581 memmove(ex + 1, ex + 2, len);
1582 }
1583 le16_add_cpu(&eh->eh_entries, -1);
1584 merge_done = 1;
1585 WARN_ON(eh->eh_entries == 0);
1586 if (!eh->eh_entries)
1587 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1588 }
1589
1590 return merge_done;
1591}
1592
1593/*
1594 * This function tries to merge the @ex extent to neighbours in the tree.
1595 * return 1 if merge left else 0.
1596 */
1597static int ext4_ext_try_to_merge(struct inode *inode,
1598 struct ext4_ext_path *path,
1599 struct ext4_extent *ex) {
1600 struct ext4_extent_header *eh;
1601 unsigned int depth;
1602 int merge_done = 0;
1603 int ret = 0;
1604
1605 depth = ext_depth(inode);
1606 BUG_ON(path[depth].p_hdr == NULL);
1607 eh = path[depth].p_hdr;
1608
1609 if (ex > EXT_FIRST_EXTENT(eh))
1610 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1611
1612 if (!merge_done)
1613 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1614
1615 return ret;
1616}
1617
1618/*
1619 * check if a portion of the "newext" extent overlaps with an
1620 * existing extent.
1621 *
1622 * If there is an overlap discovered, it updates the length of the newext
1623 * such that there will be no overlap, and then returns 1.
1624 * If there is no overlap found, it returns 0.
1625 */
1626static unsigned int ext4_ext_check_overlap(struct inode *inode,
1627 struct ext4_extent *newext,
1628 struct ext4_ext_path *path)
1629{
1630 ext4_lblk_t b1, b2;
1631 unsigned int depth, len1;
1632 unsigned int ret = 0;
1633
1634 b1 = le32_to_cpu(newext->ee_block);
1635 len1 = ext4_ext_get_actual_len(newext);
1636 depth = ext_depth(inode);
1637 if (!path[depth].p_ext)
1638 goto out;
1639 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1640
1641 /*
1642 * get the next allocated block if the extent in the path
1643 * is before the requested block(s)
1644 */
1645 if (b2 < b1) {
1646 b2 = ext4_ext_next_allocated_block(path);
1647 if (b2 == EXT_MAX_BLOCKS)
1648 goto out;
1649 }
1650
1651 /* check for wrap through zero on extent logical start block*/
1652 if (b1 + len1 < b1) {
1653 len1 = EXT_MAX_BLOCKS - b1;
1654 newext->ee_len = cpu_to_le16(len1);
1655 ret = 1;
1656 }
1657
1658 /* check for overlap */
1659 if (b1 + len1 > b2) {
1660 newext->ee_len = cpu_to_le16(b2 - b1);
1661 ret = 1;
1662 }
1663out:
1664 return ret;
1665}
1666
1667/*
1668 * ext4_ext_insert_extent:
1669 * tries to merge requsted extent into the existing extent or
1670 * inserts requested extent as new one into the tree,
1671 * creating new leaf in the no-space case.
1672 */
1673int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1674 struct ext4_ext_path *path,
1675 struct ext4_extent *newext, int flag)
1676{
1677 struct ext4_extent_header *eh;
1678 struct ext4_extent *ex, *fex;
1679 struct ext4_extent *nearex; /* nearest extent */
1680 struct ext4_ext_path *npath = NULL;
1681 int depth, len, err;
1682 ext4_lblk_t next;
1683 unsigned uninitialized = 0;
1684 int flags = 0;
1685
1686 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1687 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1688 return -EIO;
1689 }
1690 depth = ext_depth(inode);
1691 ex = path[depth].p_ext;
1692 if (unlikely(path[depth].p_hdr == NULL)) {
1693 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1694 return -EIO;
1695 }
1696
1697 /* try to insert block into found extent and return */
1698 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1699 && ext4_can_extents_be_merged(inode, ex, newext)) {
1700 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1701 ext4_ext_is_uninitialized(newext),
1702 ext4_ext_get_actual_len(newext),
1703 le32_to_cpu(ex->ee_block),
1704 ext4_ext_is_uninitialized(ex),
1705 ext4_ext_get_actual_len(ex),
1706 ext4_ext_pblock(ex));
1707 err = ext4_ext_get_access(handle, inode, path + depth);
1708 if (err)
1709 return err;
1710
1711 /*
1712 * ext4_can_extents_be_merged should have checked that either
1713 * both extents are uninitialized, or both aren't. Thus we
1714 * need to check only one of them here.
1715 */
1716 if (ext4_ext_is_uninitialized(ex))
1717 uninitialized = 1;
1718 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1719 + ext4_ext_get_actual_len(newext));
1720 if (uninitialized)
1721 ext4_ext_mark_uninitialized(ex);
1722 eh = path[depth].p_hdr;
1723 nearex = ex;
1724 goto merge;
1725 }
1726
1727 depth = ext_depth(inode);
1728 eh = path[depth].p_hdr;
1729 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1730 goto has_space;
1731
1732 /* probably next leaf has space for us? */
1733 fex = EXT_LAST_EXTENT(eh);
1734 next = EXT_MAX_BLOCKS;
1735 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1736 next = ext4_ext_next_leaf_block(path);
1737 if (next != EXT_MAX_BLOCKS) {
1738 ext_debug("next leaf block - %d\n", next);
1739 BUG_ON(npath != NULL);
1740 npath = ext4_ext_find_extent(inode, next, NULL);
1741 if (IS_ERR(npath))
1742 return PTR_ERR(npath);
1743 BUG_ON(npath->p_depth != path->p_depth);
1744 eh = npath[depth].p_hdr;
1745 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1746 ext_debug("next leaf isn't full(%d)\n",
1747 le16_to_cpu(eh->eh_entries));
1748 path = npath;
1749 goto has_space;
1750 }
1751 ext_debug("next leaf has no free space(%d,%d)\n",
1752 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1753 }
1754
1755 /*
1756 * There is no free space in the found leaf.
1757 * We're gonna add a new leaf in the tree.
1758 */
1759 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1760 flags = EXT4_MB_USE_ROOT_BLOCKS;
1761 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1762 if (err)
1763 goto cleanup;
1764 depth = ext_depth(inode);
1765 eh = path[depth].p_hdr;
1766
1767has_space:
1768 nearex = path[depth].p_ext;
1769
1770 err = ext4_ext_get_access(handle, inode, path + depth);
1771 if (err)
1772 goto cleanup;
1773
1774 if (!nearex) {
1775 /* there is no extent in this leaf, create first one */
1776 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1777 le32_to_cpu(newext->ee_block),
1778 ext4_ext_pblock(newext),
1779 ext4_ext_is_uninitialized(newext),
1780 ext4_ext_get_actual_len(newext));
1781 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1782 } else if (le32_to_cpu(newext->ee_block)
1783 > le32_to_cpu(nearex->ee_block)) {
1784/* BUG_ON(newext->ee_block == nearex->ee_block); */
1785 if (nearex != EXT_LAST_EXTENT(eh)) {
1786 len = EXT_MAX_EXTENT(eh) - nearex;
1787 len = (len - 1) * sizeof(struct ext4_extent);
1788 len = len < 0 ? 0 : len;
1789 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1790 "move %d from 0x%p to 0x%p\n",
1791 le32_to_cpu(newext->ee_block),
1792 ext4_ext_pblock(newext),
1793 ext4_ext_is_uninitialized(newext),
1794 ext4_ext_get_actual_len(newext),
1795 nearex, len, nearex + 1, nearex + 2);
1796 memmove(nearex + 2, nearex + 1, len);
1797 }
1798 path[depth].p_ext = nearex + 1;
1799 } else {
1800 BUG_ON(newext->ee_block == nearex->ee_block);
1801 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1802 len = len < 0 ? 0 : len;
1803 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1804 "move %d from 0x%p to 0x%p\n",
1805 le32_to_cpu(newext->ee_block),
1806 ext4_ext_pblock(newext),
1807 ext4_ext_is_uninitialized(newext),
1808 ext4_ext_get_actual_len(newext),
1809 nearex, len, nearex, nearex + 1);
1810 memmove(nearex + 1, nearex, len);
1811 path[depth].p_ext = nearex;
1812 }
1813
1814 le16_add_cpu(&eh->eh_entries, 1);
1815 nearex = path[depth].p_ext;
1816 nearex->ee_block = newext->ee_block;
1817 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1818 nearex->ee_len = newext->ee_len;
1819
1820merge:
1821 /* try to merge extents to the right */
1822 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1823 ext4_ext_try_to_merge(inode, path, nearex);
1824
1825 /* try to merge extents to the left */
1826
1827 /* time to correct all indexes above */
1828 err = ext4_ext_correct_indexes(handle, inode, path);
1829 if (err)
1830 goto cleanup;
1831
1832 err = ext4_ext_dirty(handle, inode, path + depth);
1833
1834cleanup:
1835 if (npath) {
1836 ext4_ext_drop_refs(npath);
1837 kfree(npath);
1838 }
1839 ext4_ext_invalidate_cache(inode);
1840 return err;
1841}
1842
1843static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1844 ext4_lblk_t num, ext_prepare_callback func,
1845 void *cbdata)
1846{
1847 struct ext4_ext_path *path = NULL;
1848 struct ext4_ext_cache cbex;
1849 struct ext4_extent *ex;
1850 ext4_lblk_t next, start = 0, end = 0;
1851 ext4_lblk_t last = block + num;
1852 int depth, exists, err = 0;
1853
1854 BUG_ON(func == NULL);
1855 BUG_ON(inode == NULL);
1856
1857 while (block < last && block != EXT_MAX_BLOCKS) {
1858 num = last - block;
1859 /* find extent for this block */
1860 down_read(&EXT4_I(inode)->i_data_sem);
1861 path = ext4_ext_find_extent(inode, block, path);
1862 up_read(&EXT4_I(inode)->i_data_sem);
1863 if (IS_ERR(path)) {
1864 err = PTR_ERR(path);
1865 path = NULL;
1866 break;
1867 }
1868
1869 depth = ext_depth(inode);
1870 if (unlikely(path[depth].p_hdr == NULL)) {
1871 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1872 err = -EIO;
1873 break;
1874 }
1875 ex = path[depth].p_ext;
1876 next = ext4_ext_next_allocated_block(path);
1877
1878 exists = 0;
1879 if (!ex) {
1880 /* there is no extent yet, so try to allocate
1881 * all requested space */
1882 start = block;
1883 end = block + num;
1884 } else if (le32_to_cpu(ex->ee_block) > block) {
1885 /* need to allocate space before found extent */
1886 start = block;
1887 end = le32_to_cpu(ex->ee_block);
1888 if (block + num < end)
1889 end = block + num;
1890 } else if (block >= le32_to_cpu(ex->ee_block)
1891 + ext4_ext_get_actual_len(ex)) {
1892 /* need to allocate space after found extent */
1893 start = block;
1894 end = block + num;
1895 if (end >= next)
1896 end = next;
1897 } else if (block >= le32_to_cpu(ex->ee_block)) {
1898 /*
1899 * some part of requested space is covered
1900 * by found extent
1901 */
1902 start = block;
1903 end = le32_to_cpu(ex->ee_block)
1904 + ext4_ext_get_actual_len(ex);
1905 if (block + num < end)
1906 end = block + num;
1907 exists = 1;
1908 } else {
1909 BUG();
1910 }
1911 BUG_ON(end <= start);
1912
1913 if (!exists) {
1914 cbex.ec_block = start;
1915 cbex.ec_len = end - start;
1916 cbex.ec_start = 0;
1917 } else {
1918 cbex.ec_block = le32_to_cpu(ex->ee_block);
1919 cbex.ec_len = ext4_ext_get_actual_len(ex);
1920 cbex.ec_start = ext4_ext_pblock(ex);
1921 }
1922
1923 if (unlikely(cbex.ec_len == 0)) {
1924 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1925 err = -EIO;
1926 break;
1927 }
1928 err = func(inode, next, &cbex, ex, cbdata);
1929 ext4_ext_drop_refs(path);
1930
1931 if (err < 0)
1932 break;
1933
1934 if (err == EXT_REPEAT)
1935 continue;
1936 else if (err == EXT_BREAK) {
1937 err = 0;
1938 break;
1939 }
1940
1941 if (ext_depth(inode) != depth) {
1942 /* depth was changed. we have to realloc path */
1943 kfree(path);
1944 path = NULL;
1945 }
1946
1947 block = cbex.ec_block + cbex.ec_len;
1948 }
1949
1950 if (path) {
1951 ext4_ext_drop_refs(path);
1952 kfree(path);
1953 }
1954
1955 return err;
1956}
1957
1958static void
1959ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1960 __u32 len, ext4_fsblk_t start)
1961{
1962 struct ext4_ext_cache *cex;
1963 BUG_ON(len == 0);
1964 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1965 cex = &EXT4_I(inode)->i_cached_extent;
1966 cex->ec_block = block;
1967 cex->ec_len = len;
1968 cex->ec_start = start;
1969 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1970}
1971
1972/*
1973 * ext4_ext_put_gap_in_cache:
1974 * calculate boundaries of the gap that the requested block fits into
1975 * and cache this gap
1976 */
1977static void
1978ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1979 ext4_lblk_t block)
1980{
1981 int depth = ext_depth(inode);
1982 unsigned long len;
1983 ext4_lblk_t lblock;
1984 struct ext4_extent *ex;
1985
1986 ex = path[depth].p_ext;
1987 if (ex == NULL) {
1988 /* there is no extent yet, so gap is [0;-] */
1989 lblock = 0;
1990 len = EXT_MAX_BLOCKS;
1991 ext_debug("cache gap(whole file):");
1992 } else if (block < le32_to_cpu(ex->ee_block)) {
1993 lblock = block;
1994 len = le32_to_cpu(ex->ee_block) - block;
1995 ext_debug("cache gap(before): %u [%u:%u]",
1996 block,
1997 le32_to_cpu(ex->ee_block),
1998 ext4_ext_get_actual_len(ex));
1999 } else if (block >= le32_to_cpu(ex->ee_block)
2000 + ext4_ext_get_actual_len(ex)) {
2001 ext4_lblk_t next;
2002 lblock = le32_to_cpu(ex->ee_block)
2003 + ext4_ext_get_actual_len(ex);
2004
2005 next = ext4_ext_next_allocated_block(path);
2006 ext_debug("cache gap(after): [%u:%u] %u",
2007 le32_to_cpu(ex->ee_block),
2008 ext4_ext_get_actual_len(ex),
2009 block);
2010 BUG_ON(next == lblock);
2011 len = next - lblock;
2012 } else {
2013 lblock = len = 0;
2014 BUG();
2015 }
2016
2017 ext_debug(" -> %u:%lu\n", lblock, len);
2018 ext4_ext_put_in_cache(inode, lblock, len, 0);
2019}
2020
2021/*
2022 * ext4_ext_check_cache()
2023 * Checks to see if the given block is in the cache.
2024 * If it is, the cached extent is stored in the given
2025 * cache extent pointer. If the cached extent is a hole,
2026 * this routine should be used instead of
2027 * ext4_ext_in_cache if the calling function needs to
2028 * know the size of the hole.
2029 *
2030 * @inode: The files inode
2031 * @block: The block to look for in the cache
2032 * @ex: Pointer where the cached extent will be stored
2033 * if it contains block
2034 *
2035 * Return 0 if cache is invalid; 1 if the cache is valid
2036 */
2037static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2038 struct ext4_ext_cache *ex){
2039 struct ext4_ext_cache *cex;
2040 struct ext4_sb_info *sbi;
2041 int ret = 0;
2042
2043 /*
2044 * We borrow i_block_reservation_lock to protect i_cached_extent
2045 */
2046 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2047 cex = &EXT4_I(inode)->i_cached_extent;
2048 sbi = EXT4_SB(inode->i_sb);
2049
2050 /* has cache valid data? */
2051 if (cex->ec_len == 0)
2052 goto errout;
2053
2054 if (in_range(block, cex->ec_block, cex->ec_len)) {
2055 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2056 ext_debug("%u cached by %u:%u:%llu\n",
2057 block,
2058 cex->ec_block, cex->ec_len, cex->ec_start);
2059 ret = 1;
2060 }
2061errout:
2062 if (!ret)
2063 sbi->extent_cache_misses++;
2064 else
2065 sbi->extent_cache_hits++;
2066 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2067 return ret;
2068}
2069
2070/*
2071 * ext4_ext_in_cache()
2072 * Checks to see if the given block is in the cache.
2073 * If it is, the cached extent is stored in the given
2074 * extent pointer.
2075 *
2076 * @inode: The files inode
2077 * @block: The block to look for in the cache
2078 * @ex: Pointer where the cached extent will be stored
2079 * if it contains block
2080 *
2081 * Return 0 if cache is invalid; 1 if the cache is valid
2082 */
2083static int
2084ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2085 struct ext4_extent *ex)
2086{
2087 struct ext4_ext_cache cex;
2088 int ret = 0;
2089
2090 if (ext4_ext_check_cache(inode, block, &cex)) {
2091 ex->ee_block = cpu_to_le32(cex.ec_block);
2092 ext4_ext_store_pblock(ex, cex.ec_start);
2093 ex->ee_len = cpu_to_le16(cex.ec_len);
2094 ret = 1;
2095 }
2096
2097 return ret;
2098}
2099
2100
2101/*
2102 * ext4_ext_rm_idx:
2103 * removes index from the index block.
2104 */
2105static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2106 struct ext4_ext_path *path)
2107{
2108 int err;
2109 ext4_fsblk_t leaf;
2110
2111 /* free index block */
2112 path--;
2113 leaf = ext4_idx_pblock(path->p_idx);
2114 if (unlikely(path->p_hdr->eh_entries == 0)) {
2115 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2116 return -EIO;
2117 }
2118 err = ext4_ext_get_access(handle, inode, path);
2119 if (err)
2120 return err;
2121
2122 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2123 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2124 len *= sizeof(struct ext4_extent_idx);
2125 memmove(path->p_idx, path->p_idx + 1, len);
2126 }
2127
2128 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2129 err = ext4_ext_dirty(handle, inode, path);
2130 if (err)
2131 return err;
2132 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2133 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2134 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2135 return err;
2136}
2137
2138/*
2139 * ext4_ext_calc_credits_for_single_extent:
2140 * This routine returns max. credits that needed to insert an extent
2141 * to the extent tree.
2142 * When pass the actual path, the caller should calculate credits
2143 * under i_data_sem.
2144 */
2145int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2146 struct ext4_ext_path *path)
2147{
2148 if (path) {
2149 int depth = ext_depth(inode);
2150 int ret = 0;
2151
2152 /* probably there is space in leaf? */
2153 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2154 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2155
2156 /*
2157 * There are some space in the leaf tree, no
2158 * need to account for leaf block credit
2159 *
2160 * bitmaps and block group descriptor blocks
2161 * and other metadat blocks still need to be
2162 * accounted.
2163 */
2164 /* 1 bitmap, 1 block group descriptor */
2165 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2166 return ret;
2167 }
2168 }
2169
2170 return ext4_chunk_trans_blocks(inode, nrblocks);
2171}
2172
2173/*
2174 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2175 *
2176 * if nrblocks are fit in a single extent (chunk flag is 1), then
2177 * in the worse case, each tree level index/leaf need to be changed
2178 * if the tree split due to insert a new extent, then the old tree
2179 * index/leaf need to be updated too
2180 *
2181 * If the nrblocks are discontiguous, they could cause
2182 * the whole tree split more than once, but this is really rare.
2183 */
2184int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2185{
2186 int index;
2187 int depth = ext_depth(inode);
2188
2189 if (chunk)
2190 index = depth * 2;
2191 else
2192 index = depth * 3;
2193
2194 return index;
2195}
2196
2197static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2198 struct ext4_extent *ex,
2199 ext4_lblk_t from, ext4_lblk_t to)
2200{
2201 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2202 int flags = EXT4_FREE_BLOCKS_FORGET;
2203
2204 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2205 flags |= EXT4_FREE_BLOCKS_METADATA;
2206#ifdef EXTENTS_STATS
2207 {
2208 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2209 spin_lock(&sbi->s_ext_stats_lock);
2210 sbi->s_ext_blocks += ee_len;
2211 sbi->s_ext_extents++;
2212 if (ee_len < sbi->s_ext_min)
2213 sbi->s_ext_min = ee_len;
2214 if (ee_len > sbi->s_ext_max)
2215 sbi->s_ext_max = ee_len;
2216 if (ext_depth(inode) > sbi->s_depth_max)
2217 sbi->s_depth_max = ext_depth(inode);
2218 spin_unlock(&sbi->s_ext_stats_lock);
2219 }
2220#endif
2221 if (from >= le32_to_cpu(ex->ee_block)
2222 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2223 /* tail removal */
2224 ext4_lblk_t num;
2225 ext4_fsblk_t start;
2226
2227 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2228 start = ext4_ext_pblock(ex) + ee_len - num;
2229 ext_debug("free last %u blocks starting %llu\n", num, start);
2230 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2231 } else if (from == le32_to_cpu(ex->ee_block)
2232 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2233 /* head removal */
2234 ext4_lblk_t num;
2235 ext4_fsblk_t start;
2236
2237 num = to - from;
2238 start = ext4_ext_pblock(ex);
2239
2240 ext_debug("free first %u blocks starting %llu\n", num, start);
2241 ext4_free_blocks(handle, inode, 0, start, num, flags);
2242
2243 } else {
2244 printk(KERN_INFO "strange request: removal(2) "
2245 "%u-%u from %u:%u\n",
2246 from, to, le32_to_cpu(ex->ee_block), ee_len);
2247 }
2248 return 0;
2249}
2250
2251
2252/*
2253 * ext4_ext_rm_leaf() Removes the extents associated with the
2254 * blocks appearing between "start" and "end", and splits the extents
2255 * if "start" and "end" appear in the same extent
2256 *
2257 * @handle: The journal handle
2258 * @inode: The files inode
2259 * @path: The path to the leaf
2260 * @start: The first block to remove
2261 * @end: The last block to remove
2262 */
2263static int
2264ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2265 struct ext4_ext_path *path, ext4_lblk_t start,
2266 ext4_lblk_t end)
2267{
2268 int err = 0, correct_index = 0;
2269 int depth = ext_depth(inode), credits;
2270 struct ext4_extent_header *eh;
2271 ext4_lblk_t a, b, block;
2272 unsigned num;
2273 ext4_lblk_t ex_ee_block;
2274 unsigned short ex_ee_len;
2275 unsigned uninitialized = 0;
2276 struct ext4_extent *ex;
2277 struct ext4_map_blocks map;
2278
2279 /* the header must be checked already in ext4_ext_remove_space() */
2280 ext_debug("truncate since %u in leaf\n", start);
2281 if (!path[depth].p_hdr)
2282 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2283 eh = path[depth].p_hdr;
2284 if (unlikely(path[depth].p_hdr == NULL)) {
2285 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2286 return -EIO;
2287 }
2288 /* find where to start removing */
2289 ex = EXT_LAST_EXTENT(eh);
2290
2291 ex_ee_block = le32_to_cpu(ex->ee_block);
2292 ex_ee_len = ext4_ext_get_actual_len(ex);
2293
2294 while (ex >= EXT_FIRST_EXTENT(eh) &&
2295 ex_ee_block + ex_ee_len > start) {
2296
2297 if (ext4_ext_is_uninitialized(ex))
2298 uninitialized = 1;
2299 else
2300 uninitialized = 0;
2301
2302 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2303 uninitialized, ex_ee_len);
2304 path[depth].p_ext = ex;
2305
2306 a = ex_ee_block > start ? ex_ee_block : start;
2307 b = ex_ee_block+ex_ee_len - 1 < end ?
2308 ex_ee_block+ex_ee_len - 1 : end;
2309
2310 ext_debug(" border %u:%u\n", a, b);
2311
2312 /* If this extent is beyond the end of the hole, skip it */
2313 if (end <= ex_ee_block) {
2314 ex--;
2315 ex_ee_block = le32_to_cpu(ex->ee_block);
2316 ex_ee_len = ext4_ext_get_actual_len(ex);
2317 continue;
2318 } else if (a != ex_ee_block &&
2319 b != ex_ee_block + ex_ee_len - 1) {
2320 /*
2321 * If this is a truncate, then this condition should
2322 * never happen because at least one of the end points
2323 * needs to be on the edge of the extent.
2324 */
2325 if (end == EXT_MAX_BLOCKS - 1) {
2326 ext_debug(" bad truncate %u:%u\n",
2327 start, end);
2328 block = 0;
2329 num = 0;
2330 err = -EIO;
2331 goto out;
2332 }
2333 /*
2334 * else this is a hole punch, so the extent needs to
2335 * be split since neither edge of the hole is on the
2336 * extent edge
2337 */
2338 else{
2339 map.m_pblk = ext4_ext_pblock(ex);
2340 map.m_lblk = ex_ee_block;
2341 map.m_len = b - ex_ee_block;
2342
2343 err = ext4_split_extent(handle,
2344 inode, path, &map, 0,
2345 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2346 EXT4_GET_BLOCKS_PRE_IO);
2347
2348 if (err < 0)
2349 goto out;
2350
2351 ex_ee_len = ext4_ext_get_actual_len(ex);
2352
2353 b = ex_ee_block+ex_ee_len - 1 < end ?
2354 ex_ee_block+ex_ee_len - 1 : end;
2355
2356 /* Then remove tail of this extent */
2357 block = ex_ee_block;
2358 num = a - block;
2359 }
2360 } else if (a != ex_ee_block) {
2361 /* remove tail of the extent */
2362 block = ex_ee_block;
2363 num = a - block;
2364 } else if (b != ex_ee_block + ex_ee_len - 1) {
2365 /* remove head of the extent */
2366 block = b;
2367 num = ex_ee_block + ex_ee_len - b;
2368
2369 /*
2370 * If this is a truncate, this condition
2371 * should never happen
2372 */
2373 if (end == EXT_MAX_BLOCKS - 1) {
2374 ext_debug(" bad truncate %u:%u\n",
2375 start, end);
2376 err = -EIO;
2377 goto out;
2378 }
2379 } else {
2380 /* remove whole extent: excellent! */
2381 block = ex_ee_block;
2382 num = 0;
2383 if (a != ex_ee_block) {
2384 ext_debug(" bad truncate %u:%u\n",
2385 start, end);
2386 err = -EIO;
2387 goto out;
2388 }
2389
2390 if (b != ex_ee_block + ex_ee_len - 1) {
2391 ext_debug(" bad truncate %u:%u\n",
2392 start, end);
2393 err = -EIO;
2394 goto out;
2395 }
2396 }
2397
2398 /*
2399 * 3 for leaf, sb, and inode plus 2 (bmap and group
2400 * descriptor) for each block group; assume two block
2401 * groups plus ex_ee_len/blocks_per_block_group for
2402 * the worst case
2403 */
2404 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2405 if (ex == EXT_FIRST_EXTENT(eh)) {
2406 correct_index = 1;
2407 credits += (ext_depth(inode)) + 1;
2408 }
2409 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2410
2411 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2412 if (err)
2413 goto out;
2414
2415 err = ext4_ext_get_access(handle, inode, path + depth);
2416 if (err)
2417 goto out;
2418
2419 err = ext4_remove_blocks(handle, inode, ex, a, b);
2420 if (err)
2421 goto out;
2422
2423 if (num == 0) {
2424 /* this extent is removed; mark slot entirely unused */
2425 ext4_ext_store_pblock(ex, 0);
2426 } else if (block != ex_ee_block) {
2427 /*
2428 * If this was a head removal, then we need to update
2429 * the physical block since it is now at a different
2430 * location
2431 */
2432 ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2433 }
2434
2435 ex->ee_block = cpu_to_le32(block);
2436 ex->ee_len = cpu_to_le16(num);
2437 /*
2438 * Do not mark uninitialized if all the blocks in the
2439 * extent have been removed.
2440 */
2441 if (uninitialized && num)
2442 ext4_ext_mark_uninitialized(ex);
2443
2444 err = ext4_ext_dirty(handle, inode, path + depth);
2445 if (err)
2446 goto out;
2447
2448 /*
2449 * If the extent was completely released,
2450 * we need to remove it from the leaf
2451 */
2452 if (num == 0) {
2453 if (end != EXT_MAX_BLOCKS - 1) {
2454 /*
2455 * For hole punching, we need to scoot all the
2456 * extents up when an extent is removed so that
2457 * we dont have blank extents in the middle
2458 */
2459 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2460 sizeof(struct ext4_extent));
2461
2462 /* Now get rid of the one at the end */
2463 memset(EXT_LAST_EXTENT(eh), 0,
2464 sizeof(struct ext4_extent));
2465 }
2466 le16_add_cpu(&eh->eh_entries, -1);
2467 }
2468
2469 ext_debug("new extent: %u:%u:%llu\n", block, num,
2470 ext4_ext_pblock(ex));
2471 ex--;
2472 ex_ee_block = le32_to_cpu(ex->ee_block);
2473 ex_ee_len = ext4_ext_get_actual_len(ex);
2474 }
2475
2476 if (correct_index && eh->eh_entries)
2477 err = ext4_ext_correct_indexes(handle, inode, path);
2478
2479 /* if this leaf is free, then we should
2480 * remove it from index block above */
2481 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2482 err = ext4_ext_rm_idx(handle, inode, path + depth);
2483
2484out:
2485 return err;
2486}
2487
2488/*
2489 * ext4_ext_more_to_rm:
2490 * returns 1 if current index has to be freed (even partial)
2491 */
2492static int
2493ext4_ext_more_to_rm(struct ext4_ext_path *path)
2494{
2495 BUG_ON(path->p_idx == NULL);
2496
2497 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2498 return 0;
2499
2500 /*
2501 * if truncate on deeper level happened, it wasn't partial,
2502 * so we have to consider current index for truncation
2503 */
2504 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2505 return 0;
2506 return 1;
2507}
2508
2509static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2510{
2511 struct super_block *sb = inode->i_sb;
2512 int depth = ext_depth(inode);
2513 struct ext4_ext_path *path;
2514 handle_t *handle;
2515 int i, err;
2516
2517 ext_debug("truncate since %u\n", start);
2518
2519 /* probably first extent we're gonna free will be last in block */
2520 handle = ext4_journal_start(inode, depth + 1);
2521 if (IS_ERR(handle))
2522 return PTR_ERR(handle);
2523
2524again:
2525 ext4_ext_invalidate_cache(inode);
2526
2527 /*
2528 * We start scanning from right side, freeing all the blocks
2529 * after i_size and walking into the tree depth-wise.
2530 */
2531 depth = ext_depth(inode);
2532 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2533 if (path == NULL) {
2534 ext4_journal_stop(handle);
2535 return -ENOMEM;
2536 }
2537 path[0].p_depth = depth;
2538 path[0].p_hdr = ext_inode_hdr(inode);
2539 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2540 err = -EIO;
2541 goto out;
2542 }
2543 i = err = 0;
2544
2545 while (i >= 0 && err == 0) {
2546 if (i == depth) {
2547 /* this is leaf block */
2548 err = ext4_ext_rm_leaf(handle, inode, path,
2549 start, EXT_MAX_BLOCKS - 1);
2550 /* root level has p_bh == NULL, brelse() eats this */
2551 brelse(path[i].p_bh);
2552 path[i].p_bh = NULL;
2553 i--;
2554 continue;
2555 }
2556
2557 /* this is index block */
2558 if (!path[i].p_hdr) {
2559 ext_debug("initialize header\n");
2560 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2561 }
2562
2563 if (!path[i].p_idx) {
2564 /* this level hasn't been touched yet */
2565 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2566 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2567 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2568 path[i].p_hdr,
2569 le16_to_cpu(path[i].p_hdr->eh_entries));
2570 } else {
2571 /* we were already here, see at next index */
2572 path[i].p_idx--;
2573 }
2574
2575 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2576 i, EXT_FIRST_INDEX(path[i].p_hdr),
2577 path[i].p_idx);
2578 if (ext4_ext_more_to_rm(path + i)) {
2579 struct buffer_head *bh;
2580 /* go to the next level */
2581 ext_debug("move to level %d (block %llu)\n",
2582 i + 1, ext4_idx_pblock(path[i].p_idx));
2583 memset(path + i + 1, 0, sizeof(*path));
2584 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2585 if (!bh) {
2586 /* should we reset i_size? */
2587 err = -EIO;
2588 break;
2589 }
2590 if (WARN_ON(i + 1 > depth)) {
2591 err = -EIO;
2592 break;
2593 }
2594 if (ext4_ext_check(inode, ext_block_hdr(bh),
2595 depth - i - 1)) {
2596 err = -EIO;
2597 break;
2598 }
2599 path[i + 1].p_bh = bh;
2600
2601 /* save actual number of indexes since this
2602 * number is changed at the next iteration */
2603 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2604 i++;
2605 } else {
2606 /* we finished processing this index, go up */
2607 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2608 /* index is empty, remove it;
2609 * handle must be already prepared by the
2610 * truncatei_leaf() */
2611 err = ext4_ext_rm_idx(handle, inode, path + i);
2612 }
2613 /* root level has p_bh == NULL, brelse() eats this */
2614 brelse(path[i].p_bh);
2615 path[i].p_bh = NULL;
2616 i--;
2617 ext_debug("return to level %d\n", i);
2618 }
2619 }
2620
2621 /* TODO: flexible tree reduction should be here */
2622 if (path->p_hdr->eh_entries == 0) {
2623 /*
2624 * truncate to zero freed all the tree,
2625 * so we need to correct eh_depth
2626 */
2627 err = ext4_ext_get_access(handle, inode, path);
2628 if (err == 0) {
2629 ext_inode_hdr(inode)->eh_depth = 0;
2630 ext_inode_hdr(inode)->eh_max =
2631 cpu_to_le16(ext4_ext_space_root(inode, 0));
2632 err = ext4_ext_dirty(handle, inode, path);
2633 }
2634 }
2635out:
2636 ext4_ext_drop_refs(path);
2637 kfree(path);
2638 if (err == -EAGAIN)
2639 goto again;
2640 ext4_journal_stop(handle);
2641
2642 return err;
2643}
2644
2645/*
2646 * called at mount time
2647 */
2648void ext4_ext_init(struct super_block *sb)
2649{
2650 /*
2651 * possible initialization would be here
2652 */
2653
2654 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2655#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2656 printk(KERN_INFO "EXT4-fs: file extents enabled");
2657#ifdef AGGRESSIVE_TEST
2658 printk(", aggressive tests");
2659#endif
2660#ifdef CHECK_BINSEARCH
2661 printk(", check binsearch");
2662#endif
2663#ifdef EXTENTS_STATS
2664 printk(", stats");
2665#endif
2666 printk("\n");
2667#endif
2668#ifdef EXTENTS_STATS
2669 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2670 EXT4_SB(sb)->s_ext_min = 1 << 30;
2671 EXT4_SB(sb)->s_ext_max = 0;
2672#endif
2673 }
2674}
2675
2676/*
2677 * called at umount time
2678 */
2679void ext4_ext_release(struct super_block *sb)
2680{
2681 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2682 return;
2683
2684#ifdef EXTENTS_STATS
2685 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2686 struct ext4_sb_info *sbi = EXT4_SB(sb);
2687 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2688 sbi->s_ext_blocks, sbi->s_ext_extents,
2689 sbi->s_ext_blocks / sbi->s_ext_extents);
2690 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2691 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2692 }
2693#endif
2694}
2695
2696/* FIXME!! we need to try to merge to left or right after zero-out */
2697static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2698{
2699 ext4_fsblk_t ee_pblock;
2700 unsigned int ee_len;
2701 int ret;
2702
2703 ee_len = ext4_ext_get_actual_len(ex);
2704 ee_pblock = ext4_ext_pblock(ex);
2705
2706 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2707 if (ret > 0)
2708 ret = 0;
2709
2710 return ret;
2711}
2712
2713/*
2714 * used by extent splitting.
2715 */
2716#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2717 due to ENOSPC */
2718#define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2719#define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2720
2721/*
2722 * ext4_split_extent_at() splits an extent at given block.
2723 *
2724 * @handle: the journal handle
2725 * @inode: the file inode
2726 * @path: the path to the extent
2727 * @split: the logical block where the extent is splitted.
2728 * @split_flags: indicates if the extent could be zeroout if split fails, and
2729 * the states(init or uninit) of new extents.
2730 * @flags: flags used to insert new extent to extent tree.
2731 *
2732 *
2733 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2734 * of which are deterimined by split_flag.
2735 *
2736 * There are two cases:
2737 * a> the extent are splitted into two extent.
2738 * b> split is not needed, and just mark the extent.
2739 *
2740 * return 0 on success.
2741 */
2742static int ext4_split_extent_at(handle_t *handle,
2743 struct inode *inode,
2744 struct ext4_ext_path *path,
2745 ext4_lblk_t split,
2746 int split_flag,
2747 int flags)
2748{
2749 ext4_fsblk_t newblock;
2750 ext4_lblk_t ee_block;
2751 struct ext4_extent *ex, newex, orig_ex;
2752 struct ext4_extent *ex2 = NULL;
2753 unsigned int ee_len, depth;
2754 int err = 0;
2755
2756 ext_debug("ext4_split_extents_at: inode %lu, logical"
2757 "block %llu\n", inode->i_ino, (unsigned long long)split);
2758
2759 ext4_ext_show_leaf(inode, path);
2760
2761 depth = ext_depth(inode);
2762 ex = path[depth].p_ext;
2763 ee_block = le32_to_cpu(ex->ee_block);
2764 ee_len = ext4_ext_get_actual_len(ex);
2765 newblock = split - ee_block + ext4_ext_pblock(ex);
2766
2767 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2768
2769 err = ext4_ext_get_access(handle, inode, path + depth);
2770 if (err)
2771 goto out;
2772
2773 if (split == ee_block) {
2774 /*
2775 * case b: block @split is the block that the extent begins with
2776 * then we just change the state of the extent, and splitting
2777 * is not needed.
2778 */
2779 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2780 ext4_ext_mark_uninitialized(ex);
2781 else
2782 ext4_ext_mark_initialized(ex);
2783
2784 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2785 ext4_ext_try_to_merge(inode, path, ex);
2786
2787 err = ext4_ext_dirty(handle, inode, path + depth);
2788 goto out;
2789 }
2790
2791 /* case a */
2792 memcpy(&orig_ex, ex, sizeof(orig_ex));
2793 ex->ee_len = cpu_to_le16(split - ee_block);
2794 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2795 ext4_ext_mark_uninitialized(ex);
2796
2797 /*
2798 * path may lead to new leaf, not to original leaf any more
2799 * after ext4_ext_insert_extent() returns,
2800 */
2801 err = ext4_ext_dirty(handle, inode, path + depth);
2802 if (err)
2803 goto fix_extent_len;
2804
2805 ex2 = &newex;
2806 ex2->ee_block = cpu_to_le32(split);
2807 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2808 ext4_ext_store_pblock(ex2, newblock);
2809 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2810 ext4_ext_mark_uninitialized(ex2);
2811
2812 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2813 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2814 err = ext4_ext_zeroout(inode, &orig_ex);
2815 if (err)
2816 goto fix_extent_len;
2817 /* update the extent length and mark as initialized */
2818 ex->ee_len = cpu_to_le32(ee_len);
2819 ext4_ext_try_to_merge(inode, path, ex);
2820 err = ext4_ext_dirty(handle, inode, path + depth);
2821 goto out;
2822 } else if (err)
2823 goto fix_extent_len;
2824
2825out:
2826 ext4_ext_show_leaf(inode, path);
2827 return err;
2828
2829fix_extent_len:
2830 ex->ee_len = orig_ex.ee_len;
2831 ext4_ext_dirty(handle, inode, path + depth);
2832 return err;
2833}
2834
2835/*
2836 * ext4_split_extents() splits an extent and mark extent which is covered
2837 * by @map as split_flags indicates
2838 *
2839 * It may result in splitting the extent into multiple extents (upto three)
2840 * There are three possibilities:
2841 * a> There is no split required
2842 * b> Splits in two extents: Split is happening at either end of the extent
2843 * c> Splits in three extents: Somone is splitting in middle of the extent
2844 *
2845 */
2846static int ext4_split_extent(handle_t *handle,
2847 struct inode *inode,
2848 struct ext4_ext_path *path,
2849 struct ext4_map_blocks *map,
2850 int split_flag,
2851 int flags)
2852{
2853 ext4_lblk_t ee_block;
2854 struct ext4_extent *ex;
2855 unsigned int ee_len, depth;
2856 int err = 0;
2857 int uninitialized;
2858 int split_flag1, flags1;
2859
2860 depth = ext_depth(inode);
2861 ex = path[depth].p_ext;
2862 ee_block = le32_to_cpu(ex->ee_block);
2863 ee_len = ext4_ext_get_actual_len(ex);
2864 uninitialized = ext4_ext_is_uninitialized(ex);
2865
2866 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2867 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2868 EXT4_EXT_MAY_ZEROOUT : 0;
2869 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2870 if (uninitialized)
2871 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2872 EXT4_EXT_MARK_UNINIT2;
2873 err = ext4_split_extent_at(handle, inode, path,
2874 map->m_lblk + map->m_len, split_flag1, flags1);
2875 if (err)
2876 goto out;
2877 }
2878
2879 ext4_ext_drop_refs(path);
2880 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2881 if (IS_ERR(path))
2882 return PTR_ERR(path);
2883
2884 if (map->m_lblk >= ee_block) {
2885 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2886 EXT4_EXT_MAY_ZEROOUT : 0;
2887 if (uninitialized)
2888 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2889 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2890 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2891 err = ext4_split_extent_at(handle, inode, path,
2892 map->m_lblk, split_flag1, flags);
2893 if (err)
2894 goto out;
2895 }
2896
2897 ext4_ext_show_leaf(inode, path);
2898out:
2899 return err ? err : map->m_len;
2900}
2901
2902#define EXT4_EXT_ZERO_LEN 7
2903/*
2904 * This function is called by ext4_ext_map_blocks() if someone tries to write
2905 * to an uninitialized extent. It may result in splitting the uninitialized
2906 * extent into multiple extents (up to three - one initialized and two
2907 * uninitialized).
2908 * There are three possibilities:
2909 * a> There is no split required: Entire extent should be initialized
2910 * b> Splits in two extents: Write is happening at either end of the extent
2911 * c> Splits in three extents: Somone is writing in middle of the extent
2912 */
2913static int ext4_ext_convert_to_initialized(handle_t *handle,
2914 struct inode *inode,
2915 struct ext4_map_blocks *map,
2916 struct ext4_ext_path *path)
2917{
2918 struct ext4_map_blocks split_map;
2919 struct ext4_extent zero_ex;
2920 struct ext4_extent *ex;
2921 ext4_lblk_t ee_block, eof_block;
2922 unsigned int allocated, ee_len, depth;
2923 int err = 0;
2924 int split_flag = 0;
2925
2926 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2927 "block %llu, max_blocks %u\n", inode->i_ino,
2928 (unsigned long long)map->m_lblk, map->m_len);
2929
2930 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2931 inode->i_sb->s_blocksize_bits;
2932 if (eof_block < map->m_lblk + map->m_len)
2933 eof_block = map->m_lblk + map->m_len;
2934
2935 depth = ext_depth(inode);
2936 ex = path[depth].p_ext;
2937 ee_block = le32_to_cpu(ex->ee_block);
2938 ee_len = ext4_ext_get_actual_len(ex);
2939 allocated = ee_len - (map->m_lblk - ee_block);
2940
2941 WARN_ON(map->m_lblk < ee_block);
2942 /*
2943 * It is safe to convert extent to initialized via explicit
2944 * zeroout only if extent is fully insde i_size or new_size.
2945 */
2946 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
2947
2948 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2949 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
2950 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2951 err = ext4_ext_zeroout(inode, ex);
2952 if (err)
2953 goto out;
2954
2955 err = ext4_ext_get_access(handle, inode, path + depth);
2956 if (err)
2957 goto out;
2958 ext4_ext_mark_initialized(ex);
2959 ext4_ext_try_to_merge(inode, path, ex);
2960 err = ext4_ext_dirty(handle, inode, path + depth);
2961 goto out;
2962 }
2963
2964 /*
2965 * four cases:
2966 * 1. split the extent into three extents.
2967 * 2. split the extent into two extents, zeroout the first half.
2968 * 3. split the extent into two extents, zeroout the second half.
2969 * 4. split the extent into two extents with out zeroout.
2970 */
2971 split_map.m_lblk = map->m_lblk;
2972 split_map.m_len = map->m_len;
2973
2974 if (allocated > map->m_len) {
2975 if (allocated <= EXT4_EXT_ZERO_LEN &&
2976 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2977 /* case 3 */
2978 zero_ex.ee_block =
2979 cpu_to_le32(map->m_lblk);
2980 zero_ex.ee_len = cpu_to_le16(allocated);
2981 ext4_ext_store_pblock(&zero_ex,
2982 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
2983 err = ext4_ext_zeroout(inode, &zero_ex);
2984 if (err)
2985 goto out;
2986 split_map.m_lblk = map->m_lblk;
2987 split_map.m_len = allocated;
2988 } else if ((map->m_lblk - ee_block + map->m_len <
2989 EXT4_EXT_ZERO_LEN) &&
2990 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2991 /* case 2 */
2992 if (map->m_lblk != ee_block) {
2993 zero_ex.ee_block = ex->ee_block;
2994 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
2995 ee_block);
2996 ext4_ext_store_pblock(&zero_ex,
2997 ext4_ext_pblock(ex));
2998 err = ext4_ext_zeroout(inode, &zero_ex);
2999 if (err)
3000 goto out;
3001 }
3002
3003 split_map.m_lblk = ee_block;
3004 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3005 allocated = map->m_len;
3006 }
3007 }
3008
3009 allocated = ext4_split_extent(handle, inode, path,
3010 &split_map, split_flag, 0);
3011 if (allocated < 0)
3012 err = allocated;
3013
3014out:
3015 return err ? err : allocated;
3016}
3017
3018/*
3019 * This function is called by ext4_ext_map_blocks() from
3020 * ext4_get_blocks_dio_write() when DIO to write
3021 * to an uninitialized extent.
3022 *
3023 * Writing to an uninitialized extent may result in splitting the uninitialized
3024 * extent into multiple /initialized uninitialized extents (up to three)
3025 * There are three possibilities:
3026 * a> There is no split required: Entire extent should be uninitialized
3027 * b> Splits in two extents: Write is happening at either end of the extent
3028 * c> Splits in three extents: Somone is writing in middle of the extent
3029 *
3030 * One of more index blocks maybe needed if the extent tree grow after
3031 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3032 * complete, we need to split the uninitialized extent before DIO submit
3033 * the IO. The uninitialized extent called at this time will be split
3034 * into three uninitialized extent(at most). After IO complete, the part
3035 * being filled will be convert to initialized by the end_io callback function
3036 * via ext4_convert_unwritten_extents().
3037 *
3038 * Returns the size of uninitialized extent to be written on success.
3039 */
3040static int ext4_split_unwritten_extents(handle_t *handle,
3041 struct inode *inode,
3042 struct ext4_map_blocks *map,
3043 struct ext4_ext_path *path,
3044 int flags)
3045{
3046 ext4_lblk_t eof_block;
3047 ext4_lblk_t ee_block;
3048 struct ext4_extent *ex;
3049 unsigned int ee_len;
3050 int split_flag = 0, depth;
3051
3052 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3053 "block %llu, max_blocks %u\n", inode->i_ino,
3054 (unsigned long long)map->m_lblk, map->m_len);
3055
3056 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3057 inode->i_sb->s_blocksize_bits;
3058 if (eof_block < map->m_lblk + map->m_len)
3059 eof_block = map->m_lblk + map->m_len;
3060 /*
3061 * It is safe to convert extent to initialized via explicit
3062 * zeroout only if extent is fully insde i_size or new_size.
3063 */
3064 depth = ext_depth(inode);
3065 ex = path[depth].p_ext;
3066 ee_block = le32_to_cpu(ex->ee_block);
3067 ee_len = ext4_ext_get_actual_len(ex);
3068
3069 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3070 split_flag |= EXT4_EXT_MARK_UNINIT2;
3071
3072 flags |= EXT4_GET_BLOCKS_PRE_IO;
3073 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3074}
3075
3076static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3077 struct inode *inode,
3078 struct ext4_ext_path *path)
3079{
3080 struct ext4_extent *ex;
3081 int depth;
3082 int err = 0;
3083
3084 depth = ext_depth(inode);
3085 ex = path[depth].p_ext;
3086
3087 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3088 "block %llu, max_blocks %u\n", inode->i_ino,
3089 (unsigned long long)le32_to_cpu(ex->ee_block),
3090 ext4_ext_get_actual_len(ex));
3091
3092 err = ext4_ext_get_access(handle, inode, path + depth);
3093 if (err)
3094 goto out;
3095 /* first mark the extent as initialized */
3096 ext4_ext_mark_initialized(ex);
3097
3098 /* note: ext4_ext_correct_indexes() isn't needed here because
3099 * borders are not changed
3100 */
3101 ext4_ext_try_to_merge(inode, path, ex);
3102
3103 /* Mark modified extent as dirty */
3104 err = ext4_ext_dirty(handle, inode, path + depth);
3105out:
3106 ext4_ext_show_leaf(inode, path);
3107 return err;
3108}
3109
3110static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3111 sector_t block, int count)
3112{
3113 int i;
3114 for (i = 0; i < count; i++)
3115 unmap_underlying_metadata(bdev, block + i);
3116}
3117
3118/*
3119 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3120 */
3121static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3122 ext4_lblk_t lblk,
3123 struct ext4_ext_path *path,
3124 unsigned int len)
3125{
3126 int i, depth;
3127 struct ext4_extent_header *eh;
3128 struct ext4_extent *last_ex;
3129
3130 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3131 return 0;
3132
3133 depth = ext_depth(inode);
3134 eh = path[depth].p_hdr;
3135
3136 if (unlikely(!eh->eh_entries)) {
3137 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3138 "EOFBLOCKS_FL set");
3139 return -EIO;
3140 }
3141 last_ex = EXT_LAST_EXTENT(eh);
3142 /*
3143 * We should clear the EOFBLOCKS_FL flag if we are writing the
3144 * last block in the last extent in the file. We test this by
3145 * first checking to see if the caller to
3146 * ext4_ext_get_blocks() was interested in the last block (or
3147 * a block beyond the last block) in the current extent. If
3148 * this turns out to be false, we can bail out from this
3149 * function immediately.
3150 */
3151 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3152 ext4_ext_get_actual_len(last_ex))
3153 return 0;
3154 /*
3155 * If the caller does appear to be planning to write at or
3156 * beyond the end of the current extent, we then test to see
3157 * if the current extent is the last extent in the file, by
3158 * checking to make sure it was reached via the rightmost node
3159 * at each level of the tree.
3160 */
3161 for (i = depth-1; i >= 0; i--)
3162 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3163 return 0;
3164 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3165 return ext4_mark_inode_dirty(handle, inode);
3166}
3167
3168static int
3169ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3170 struct ext4_map_blocks *map,
3171 struct ext4_ext_path *path, int flags,
3172 unsigned int allocated, ext4_fsblk_t newblock)
3173{
3174 int ret = 0;
3175 int err = 0;
3176 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3177
3178 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3179 "block %llu, max_blocks %u, flags %d, allocated %u",
3180 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3181 flags, allocated);
3182 ext4_ext_show_leaf(inode, path);
3183
3184 /* get_block() before submit the IO, split the extent */
3185 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3186 ret = ext4_split_unwritten_extents(handle, inode, map,
3187 path, flags);
3188 /*
3189 * Flag the inode(non aio case) or end_io struct (aio case)
3190 * that this IO needs to conversion to written when IO is
3191 * completed
3192 */
3193 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3194 io->flag = EXT4_IO_END_UNWRITTEN;
3195 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3196 } else
3197 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3198 if (ext4_should_dioread_nolock(inode))
3199 map->m_flags |= EXT4_MAP_UNINIT;
3200 goto out;
3201 }
3202 /* IO end_io complete, convert the filled extent to written */
3203 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3204 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3205 path);
3206 if (ret >= 0) {
3207 ext4_update_inode_fsync_trans(handle, inode, 1);
3208 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3209 path, map->m_len);
3210 } else
3211 err = ret;
3212 goto out2;
3213 }
3214 /* buffered IO case */
3215 /*
3216 * repeat fallocate creation request
3217 * we already have an unwritten extent
3218 */
3219 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3220 goto map_out;
3221
3222 /* buffered READ or buffered write_begin() lookup */
3223 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3224 /*
3225 * We have blocks reserved already. We
3226 * return allocated blocks so that delalloc
3227 * won't do block reservation for us. But
3228 * the buffer head will be unmapped so that
3229 * a read from the block returns 0s.
3230 */
3231 map->m_flags |= EXT4_MAP_UNWRITTEN;
3232 goto out1;
3233 }
3234
3235 /* buffered write, writepage time, convert*/
3236 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3237 if (ret >= 0) {
3238 ext4_update_inode_fsync_trans(handle, inode, 1);
3239 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3240 map->m_len);
3241 if (err < 0)
3242 goto out2;
3243 }
3244
3245out:
3246 if (ret <= 0) {
3247 err = ret;
3248 goto out2;
3249 } else
3250 allocated = ret;
3251 map->m_flags |= EXT4_MAP_NEW;
3252 /*
3253 * if we allocated more blocks than requested
3254 * we need to make sure we unmap the extra block
3255 * allocated. The actual needed block will get
3256 * unmapped later when we find the buffer_head marked
3257 * new.
3258 */
3259 if (allocated > map->m_len) {
3260 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3261 newblock + map->m_len,
3262 allocated - map->m_len);
3263 allocated = map->m_len;
3264 }
3265
3266 /*
3267 * If we have done fallocate with the offset that is already
3268 * delayed allocated, we would have block reservation
3269 * and quota reservation done in the delayed write path.
3270 * But fallocate would have already updated quota and block
3271 * count for this offset. So cancel these reservation
3272 */
3273 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3274 ext4_da_update_reserve_space(inode, allocated, 0);
3275
3276map_out:
3277 map->m_flags |= EXT4_MAP_MAPPED;
3278out1:
3279 if (allocated > map->m_len)
3280 allocated = map->m_len;
3281 ext4_ext_show_leaf(inode, path);
3282 map->m_pblk = newblock;
3283 map->m_len = allocated;
3284out2:
3285 if (path) {
3286 ext4_ext_drop_refs(path);
3287 kfree(path);
3288 }
3289 return err ? err : allocated;
3290}
3291
3292/*
3293 * Block allocation/map/preallocation routine for extents based files
3294 *
3295 *
3296 * Need to be called with
3297 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3298 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3299 *
3300 * return > 0, number of of blocks already mapped/allocated
3301 * if create == 0 and these are pre-allocated blocks
3302 * buffer head is unmapped
3303 * otherwise blocks are mapped
3304 *
3305 * return = 0, if plain look up failed (blocks have not been allocated)
3306 * buffer head is unmapped
3307 *
3308 * return < 0, error case.
3309 */
3310int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3311 struct ext4_map_blocks *map, int flags)
3312{
3313 struct ext4_ext_path *path = NULL;
3314 struct ext4_extent newex, *ex;
3315 ext4_fsblk_t newblock = 0;
3316 int err = 0, depth, ret;
3317 unsigned int allocated = 0;
3318 unsigned int punched_out = 0;
3319 unsigned int result = 0;
3320 struct ext4_allocation_request ar;
3321 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3322 struct ext4_map_blocks punch_map;
3323
3324 ext_debug("blocks %u/%u requested for inode %lu\n",
3325 map->m_lblk, map->m_len, inode->i_ino);
3326 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3327
3328 /* check in cache */
3329 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3330 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3331 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3332 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3333 /*
3334 * block isn't allocated yet and
3335 * user doesn't want to allocate it
3336 */
3337 goto out2;
3338 }
3339 /* we should allocate requested block */
3340 } else {
3341 /* block is already allocated */
3342 newblock = map->m_lblk
3343 - le32_to_cpu(newex.ee_block)
3344 + ext4_ext_pblock(&newex);
3345 /* number of remaining blocks in the extent */
3346 allocated = ext4_ext_get_actual_len(&newex) -
3347 (map->m_lblk - le32_to_cpu(newex.ee_block));
3348 goto out;
3349 }
3350 }
3351
3352 /* find extent for this block */
3353 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3354 if (IS_ERR(path)) {
3355 err = PTR_ERR(path);
3356 path = NULL;
3357 goto out2;
3358 }
3359
3360 depth = ext_depth(inode);
3361
3362 /*
3363 * consistent leaf must not be empty;
3364 * this situation is possible, though, _during_ tree modification;
3365 * this is why assert can't be put in ext4_ext_find_extent()
3366 */
3367 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3368 EXT4_ERROR_INODE(inode, "bad extent address "
3369 "lblock: %lu, depth: %d pblock %lld",
3370 (unsigned long) map->m_lblk, depth,
3371 path[depth].p_block);
3372 err = -EIO;
3373 goto out2;
3374 }
3375
3376 ex = path[depth].p_ext;
3377 if (ex) {
3378 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3379 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3380 unsigned short ee_len;
3381
3382 /*
3383 * Uninitialized extents are treated as holes, except that
3384 * we split out initialized portions during a write.
3385 */
3386 ee_len = ext4_ext_get_actual_len(ex);
3387 /* if found extent covers block, simply return it */
3388 if (in_range(map->m_lblk, ee_block, ee_len)) {
3389 newblock = map->m_lblk - ee_block + ee_start;
3390 /* number of remaining blocks in the extent */
3391 allocated = ee_len - (map->m_lblk - ee_block);
3392 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3393 ee_block, ee_len, newblock);
3394
3395 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3396 /*
3397 * Do not put uninitialized extent
3398 * in the cache
3399 */
3400 if (!ext4_ext_is_uninitialized(ex)) {
3401 ext4_ext_put_in_cache(inode, ee_block,
3402 ee_len, ee_start);
3403 goto out;
3404 }
3405 ret = ext4_ext_handle_uninitialized_extents(
3406 handle, inode, map, path, flags,
3407 allocated, newblock);
3408 return ret;
3409 }
3410
3411 /*
3412 * Punch out the map length, but only to the
3413 * end of the extent
3414 */
3415 punched_out = allocated < map->m_len ?
3416 allocated : map->m_len;
3417
3418 /*
3419 * Sense extents need to be converted to
3420 * uninitialized, they must fit in an
3421 * uninitialized extent
3422 */
3423 if (punched_out > EXT_UNINIT_MAX_LEN)
3424 punched_out = EXT_UNINIT_MAX_LEN;
3425
3426 punch_map.m_lblk = map->m_lblk;
3427 punch_map.m_pblk = newblock;
3428 punch_map.m_len = punched_out;
3429 punch_map.m_flags = 0;
3430
3431 /* Check to see if the extent needs to be split */
3432 if (punch_map.m_len != ee_len ||
3433 punch_map.m_lblk != ee_block) {
3434
3435 ret = ext4_split_extent(handle, inode,
3436 path, &punch_map, 0,
3437 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3438 EXT4_GET_BLOCKS_PRE_IO);
3439
3440 if (ret < 0) {
3441 err = ret;
3442 goto out2;
3443 }
3444 /*
3445 * find extent for the block at
3446 * the start of the hole
3447 */
3448 ext4_ext_drop_refs(path);
3449 kfree(path);
3450
3451 path = ext4_ext_find_extent(inode,
3452 map->m_lblk, NULL);
3453 if (IS_ERR(path)) {
3454 err = PTR_ERR(path);
3455 path = NULL;
3456 goto out2;
3457 }
3458
3459 depth = ext_depth(inode);
3460 ex = path[depth].p_ext;
3461 ee_len = ext4_ext_get_actual_len(ex);
3462 ee_block = le32_to_cpu(ex->ee_block);
3463 ee_start = ext4_ext_pblock(ex);
3464
3465 }
3466
3467 ext4_ext_mark_uninitialized(ex);
3468
3469 ext4_ext_invalidate_cache(inode);
3470
3471 err = ext4_ext_rm_leaf(handle, inode, path,
3472 map->m_lblk, map->m_lblk + punched_out);
3473
3474 if (!err && path->p_hdr->eh_entries == 0) {
3475 /*
3476 * Punch hole freed all of this sub tree,
3477 * so we need to correct eh_depth
3478 */
3479 err = ext4_ext_get_access(handle, inode, path);
3480 if (err == 0) {
3481 ext_inode_hdr(inode)->eh_depth = 0;
3482 ext_inode_hdr(inode)->eh_max =
3483 cpu_to_le16(ext4_ext_space_root(
3484 inode, 0));
3485
3486 err = ext4_ext_dirty(
3487 handle, inode, path);
3488 }
3489 }
3490
3491 goto out2;
3492 }
3493 }
3494
3495 /*
3496 * requested block isn't allocated yet;
3497 * we couldn't try to create block if create flag is zero
3498 */
3499 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3500 /*
3501 * put just found gap into cache to speed up
3502 * subsequent requests
3503 */
3504 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3505 goto out2;
3506 }
3507 /*
3508 * Okay, we need to do block allocation.
3509 */
3510
3511 /* find neighbour allocated blocks */
3512 ar.lleft = map->m_lblk;
3513 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3514 if (err)
3515 goto out2;
3516 ar.lright = map->m_lblk;
3517 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3518 if (err)
3519 goto out2;
3520
3521 /*
3522 * See if request is beyond maximum number of blocks we can have in
3523 * a single extent. For an initialized extent this limit is
3524 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3525 * EXT_UNINIT_MAX_LEN.
3526 */
3527 if (map->m_len > EXT_INIT_MAX_LEN &&
3528 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3529 map->m_len = EXT_INIT_MAX_LEN;
3530 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3531 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3532 map->m_len = EXT_UNINIT_MAX_LEN;
3533
3534 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3535 newex.ee_block = cpu_to_le32(map->m_lblk);
3536 newex.ee_len = cpu_to_le16(map->m_len);
3537 err = ext4_ext_check_overlap(inode, &newex, path);
3538 if (err)
3539 allocated = ext4_ext_get_actual_len(&newex);
3540 else
3541 allocated = map->m_len;
3542
3543 /* allocate new block */
3544 ar.inode = inode;
3545 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3546 ar.logical = map->m_lblk;
3547 ar.len = allocated;
3548 if (S_ISREG(inode->i_mode))
3549 ar.flags = EXT4_MB_HINT_DATA;
3550 else
3551 /* disable in-core preallocation for non-regular files */
3552 ar.flags = 0;
3553 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
3554 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
3555 newblock = ext4_mb_new_blocks(handle, &ar, &err);
3556 if (!newblock)
3557 goto out2;
3558 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3559 ar.goal, newblock, allocated);
3560
3561 /* try to insert new extent into found leaf and return */
3562 ext4_ext_store_pblock(&newex, newblock);
3563 newex.ee_len = cpu_to_le16(ar.len);
3564 /* Mark uninitialized */
3565 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3566 ext4_ext_mark_uninitialized(&newex);
3567 /*
3568 * io_end structure was created for every IO write to an
3569 * uninitialized extent. To avoid unnecessary conversion,
3570 * here we flag the IO that really needs the conversion.
3571 * For non asycn direct IO case, flag the inode state
3572 * that we need to perform conversion when IO is done.
3573 */
3574 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3575 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3576 io->flag = EXT4_IO_END_UNWRITTEN;
3577 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3578 } else
3579 ext4_set_inode_state(inode,
3580 EXT4_STATE_DIO_UNWRITTEN);
3581 }
3582 if (ext4_should_dioread_nolock(inode))
3583 map->m_flags |= EXT4_MAP_UNINIT;
3584 }
3585
3586 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3587 if (!err)
3588 err = ext4_ext_insert_extent(handle, inode, path,
3589 &newex, flags);
3590 if (err) {
3591 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3592 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
3593 /* free data blocks we just allocated */
3594 /* not a good idea to call discard here directly,
3595 * but otherwise we'd need to call it every free() */
3596 ext4_discard_preallocations(inode);
3597 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3598 ext4_ext_get_actual_len(&newex), fb_flags);
3599 goto out2;
3600 }
3601
3602 /* previous routine could use block we allocated */
3603 newblock = ext4_ext_pblock(&newex);
3604 allocated = ext4_ext_get_actual_len(&newex);
3605 if (allocated > map->m_len)
3606 allocated = map->m_len;
3607 map->m_flags |= EXT4_MAP_NEW;
3608
3609 /*
3610 * Update reserved blocks/metadata blocks after successful
3611 * block allocation which had been deferred till now.
3612 */
3613 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3614 ext4_da_update_reserve_space(inode, allocated, 1);
3615
3616 /*
3617 * Cache the extent and update transaction to commit on fdatasync only
3618 * when it is _not_ an uninitialized extent.
3619 */
3620 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3621 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3622 ext4_update_inode_fsync_trans(handle, inode, 1);
3623 } else
3624 ext4_update_inode_fsync_trans(handle, inode, 0);
3625out:
3626 if (allocated > map->m_len)
3627 allocated = map->m_len;
3628 ext4_ext_show_leaf(inode, path);
3629 map->m_flags |= EXT4_MAP_MAPPED;
3630 map->m_pblk = newblock;
3631 map->m_len = allocated;
3632out2:
3633 if (path) {
3634 ext4_ext_drop_refs(path);
3635 kfree(path);
3636 }
3637 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3638 newblock, map->m_len, err ? err : allocated);
3639
3640 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3641 punched_out : allocated;
3642
3643 return err ? err : result;
3644}
3645
3646void ext4_ext_truncate(struct inode *inode)
3647{
3648 struct address_space *mapping = inode->i_mapping;
3649 struct super_block *sb = inode->i_sb;
3650 ext4_lblk_t last_block;
3651 handle_t *handle;
3652 int err = 0;
3653
3654 /*
3655 * finish any pending end_io work so we won't run the risk of
3656 * converting any truncated blocks to initialized later
3657 */
3658 ext4_flush_completed_IO(inode);
3659
3660 /*
3661 * probably first extent we're gonna free will be last in block
3662 */
3663 err = ext4_writepage_trans_blocks(inode);
3664 handle = ext4_journal_start(inode, err);
3665 if (IS_ERR(handle))
3666 return;
3667
3668 if (inode->i_size & (sb->s_blocksize - 1))
3669 ext4_block_truncate_page(handle, mapping, inode->i_size);
3670
3671 if (ext4_orphan_add(handle, inode))
3672 goto out_stop;
3673
3674 down_write(&EXT4_I(inode)->i_data_sem);
3675 ext4_ext_invalidate_cache(inode);
3676
3677 ext4_discard_preallocations(inode);
3678
3679 /*
3680 * TODO: optimization is possible here.
3681 * Probably we need not scan at all,
3682 * because page truncation is enough.
3683 */
3684
3685 /* we have to know where to truncate from in crash case */
3686 EXT4_I(inode)->i_disksize = inode->i_size;
3687 ext4_mark_inode_dirty(handle, inode);
3688
3689 last_block = (inode->i_size + sb->s_blocksize - 1)
3690 >> EXT4_BLOCK_SIZE_BITS(sb);
3691 err = ext4_ext_remove_space(inode, last_block);
3692
3693 /* In a multi-transaction truncate, we only make the final
3694 * transaction synchronous.
3695 */
3696 if (IS_SYNC(inode))
3697 ext4_handle_sync(handle);
3698
3699 up_write(&EXT4_I(inode)->i_data_sem);
3700
3701out_stop:
3702 /*
3703 * If this was a simple ftruncate() and the file will remain alive,
3704 * then we need to clear up the orphan record which we created above.
3705 * However, if this was a real unlink then we were called by
3706 * ext4_delete_inode(), and we allow that function to clean up the
3707 * orphan info for us.
3708 */
3709 if (inode->i_nlink)
3710 ext4_orphan_del(handle, inode);
3711
3712 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3713 ext4_mark_inode_dirty(handle, inode);
3714 ext4_journal_stop(handle);
3715}
3716
3717static void ext4_falloc_update_inode(struct inode *inode,
3718 int mode, loff_t new_size, int update_ctime)
3719{
3720 struct timespec now;
3721
3722 if (update_ctime) {
3723 now = current_fs_time(inode->i_sb);
3724 if (!timespec_equal(&inode->i_ctime, &now))
3725 inode->i_ctime = now;
3726 }
3727 /*
3728 * Update only when preallocation was requested beyond
3729 * the file size.
3730 */
3731 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3732 if (new_size > i_size_read(inode))
3733 i_size_write(inode, new_size);
3734 if (new_size > EXT4_I(inode)->i_disksize)
3735 ext4_update_i_disksize(inode, new_size);
3736 } else {
3737 /*
3738 * Mark that we allocate beyond EOF so the subsequent truncate
3739 * can proceed even if the new size is the same as i_size.
3740 */
3741 if (new_size > i_size_read(inode))
3742 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3743 }
3744
3745}
3746
3747/*
3748 * preallocate space for a file. This implements ext4's fallocate file
3749 * operation, which gets called from sys_fallocate system call.
3750 * For block-mapped files, posix_fallocate should fall back to the method
3751 * of writing zeroes to the required new blocks (the same behavior which is
3752 * expected for file systems which do not support fallocate() system call).
3753 */
3754long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3755{
3756 struct inode *inode = file->f_path.dentry->d_inode;
3757 handle_t *handle;
3758 loff_t new_size;
3759 unsigned int max_blocks;
3760 int ret = 0;
3761 int ret2 = 0;
3762 int retries = 0;
3763 struct ext4_map_blocks map;
3764 unsigned int credits, blkbits = inode->i_blkbits;
3765
3766 /*
3767 * currently supporting (pre)allocate mode for extent-based
3768 * files _only_
3769 */
3770 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3771 return -EOPNOTSUPP;
3772
3773 /* Return error if mode is not supported */
3774 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3775 return -EOPNOTSUPP;
3776
3777 if (mode & FALLOC_FL_PUNCH_HOLE)
3778 return ext4_punch_hole(file, offset, len);
3779
3780 trace_ext4_fallocate_enter(inode, offset, len, mode);
3781 map.m_lblk = offset >> blkbits;
3782 /*
3783 * We can't just convert len to max_blocks because
3784 * If blocksize = 4096 offset = 3072 and len = 2048
3785 */
3786 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3787 - map.m_lblk;
3788 /*
3789 * credits to insert 1 extent into extent tree
3790 */
3791 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3792 mutex_lock(&inode->i_mutex);
3793 ret = inode_newsize_ok(inode, (len + offset));
3794 if (ret) {
3795 mutex_unlock(&inode->i_mutex);
3796 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
3797 return ret;
3798 }
3799retry:
3800 while (ret >= 0 && ret < max_blocks) {
3801 map.m_lblk = map.m_lblk + ret;
3802 map.m_len = max_blocks = max_blocks - ret;
3803 handle = ext4_journal_start(inode, credits);
3804 if (IS_ERR(handle)) {
3805 ret = PTR_ERR(handle);
3806 break;
3807 }
3808 ret = ext4_map_blocks(handle, inode, &map,
3809 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
3810 EXT4_GET_BLOCKS_NO_NORMALIZE);
3811 if (ret <= 0) {
3812#ifdef EXT4FS_DEBUG
3813 WARN_ON(ret <= 0);
3814 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3815 "returned error inode#%lu, block=%u, "
3816 "max_blocks=%u", __func__,
3817 inode->i_ino, map.m_lblk, max_blocks);
3818#endif
3819 ext4_mark_inode_dirty(handle, inode);
3820 ret2 = ext4_journal_stop(handle);
3821 break;
3822 }
3823 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3824 blkbits) >> blkbits))
3825 new_size = offset + len;
3826 else
3827 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
3828
3829 ext4_falloc_update_inode(inode, mode, new_size,
3830 (map.m_flags & EXT4_MAP_NEW));
3831 ext4_mark_inode_dirty(handle, inode);
3832 ret2 = ext4_journal_stop(handle);
3833 if (ret2)
3834 break;
3835 }
3836 if (ret == -ENOSPC &&
3837 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3838 ret = 0;
3839 goto retry;
3840 }
3841 mutex_unlock(&inode->i_mutex);
3842 trace_ext4_fallocate_exit(inode, offset, max_blocks,
3843 ret > 0 ? ret2 : ret);
3844 return ret > 0 ? ret2 : ret;
3845}
3846
3847/*
3848 * This function convert a range of blocks to written extents
3849 * The caller of this function will pass the start offset and the size.
3850 * all unwritten extents within this range will be converted to
3851 * written extents.
3852 *
3853 * This function is called from the direct IO end io call back
3854 * function, to convert the fallocated extents after IO is completed.
3855 * Returns 0 on success.
3856 */
3857int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3858 ssize_t len)
3859{
3860 handle_t *handle;
3861 unsigned int max_blocks;
3862 int ret = 0;
3863 int ret2 = 0;
3864 struct ext4_map_blocks map;
3865 unsigned int credits, blkbits = inode->i_blkbits;
3866
3867 map.m_lblk = offset >> blkbits;
3868 /*
3869 * We can't just convert len to max_blocks because
3870 * If blocksize = 4096 offset = 3072 and len = 2048
3871 */
3872 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3873 map.m_lblk);
3874 /*
3875 * credits to insert 1 extent into extent tree
3876 */
3877 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3878 while (ret >= 0 && ret < max_blocks) {
3879 map.m_lblk += ret;
3880 map.m_len = (max_blocks -= ret);
3881 handle = ext4_journal_start(inode, credits);
3882 if (IS_ERR(handle)) {
3883 ret = PTR_ERR(handle);
3884 break;
3885 }
3886 ret = ext4_map_blocks(handle, inode, &map,
3887 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3888 if (ret <= 0) {
3889 WARN_ON(ret <= 0);
3890 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3891 "returned error inode#%lu, block=%u, "
3892 "max_blocks=%u", __func__,
3893 inode->i_ino, map.m_lblk, map.m_len);
3894 }
3895 ext4_mark_inode_dirty(handle, inode);
3896 ret2 = ext4_journal_stop(handle);
3897 if (ret <= 0 || ret2 )
3898 break;
3899 }
3900 return ret > 0 ? ret2 : ret;
3901}
3902
3903/*
3904 * Callback function called for each extent to gather FIEMAP information.
3905 */
3906static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
3907 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3908 void *data)
3909{
3910 __u64 logical;
3911 __u64 physical;
3912 __u64 length;
3913 __u32 flags = 0;
3914 int ret = 0;
3915 struct fiemap_extent_info *fieinfo = data;
3916 unsigned char blksize_bits;
3917
3918 blksize_bits = inode->i_sb->s_blocksize_bits;
3919 logical = (__u64)newex->ec_block << blksize_bits;
3920
3921 if (newex->ec_start == 0) {
3922 /*
3923 * No extent in extent-tree contains block @newex->ec_start,
3924 * then the block may stay in 1)a hole or 2)delayed-extent.
3925 *
3926 * Holes or delayed-extents are processed as follows.
3927 * 1. lookup dirty pages with specified range in pagecache.
3928 * If no page is got, then there is no delayed-extent and
3929 * return with EXT_CONTINUE.
3930 * 2. find the 1st mapped buffer,
3931 * 3. check if the mapped buffer is both in the request range
3932 * and a delayed buffer. If not, there is no delayed-extent,
3933 * then return.
3934 * 4. a delayed-extent is found, the extent will be collected.
3935 */
3936 ext4_lblk_t end = 0;
3937 pgoff_t last_offset;
3938 pgoff_t offset;
3939 pgoff_t index;
3940 pgoff_t start_index = 0;
3941 struct page **pages = NULL;
3942 struct buffer_head *bh = NULL;
3943 struct buffer_head *head = NULL;
3944 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
3945
3946 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
3947 if (pages == NULL)
3948 return -ENOMEM;
3949
3950 offset = logical >> PAGE_SHIFT;
3951repeat:
3952 last_offset = offset;
3953 head = NULL;
3954 ret = find_get_pages_tag(inode->i_mapping, &offset,
3955 PAGECACHE_TAG_DIRTY, nr_pages, pages);
3956
3957 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
3958 /* First time, try to find a mapped buffer. */
3959 if (ret == 0) {
3960out:
3961 for (index = 0; index < ret; index++)
3962 page_cache_release(pages[index]);
3963 /* just a hole. */
3964 kfree(pages);
3965 return EXT_CONTINUE;
3966 }
3967 index = 0;
3968
3969next_page:
3970 /* Try to find the 1st mapped buffer. */
3971 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
3972 blksize_bits;
3973 if (!page_has_buffers(pages[index]))
3974 goto out;
3975 head = page_buffers(pages[index]);
3976 if (!head)
3977 goto out;
3978
3979 index++;
3980 bh = head;
3981 do {
3982 if (end >= newex->ec_block +
3983 newex->ec_len)
3984 /* The buffer is out of
3985 * the request range.
3986 */
3987 goto out;
3988
3989 if (buffer_mapped(bh) &&
3990 end >= newex->ec_block) {
3991 start_index = index - 1;
3992 /* get the 1st mapped buffer. */
3993 goto found_mapped_buffer;
3994 }
3995
3996 bh = bh->b_this_page;
3997 end++;
3998 } while (bh != head);
3999
4000 /* No mapped buffer in the range found in this page,
4001 * We need to look up next page.
4002 */
4003 if (index >= ret) {
4004 /* There is no page left, but we need to limit
4005 * newex->ec_len.
4006 */
4007 newex->ec_len = end - newex->ec_block;
4008 goto out;
4009 }
4010 goto next_page;
4011 } else {
4012 /*Find contiguous delayed buffers. */
4013 if (ret > 0 && pages[0]->index == last_offset)
4014 head = page_buffers(pages[0]);
4015 bh = head;
4016 index = 1;
4017 start_index = 0;
4018 }
4019
4020found_mapped_buffer:
4021 if (bh != NULL && buffer_delay(bh)) {
4022 /* 1st or contiguous delayed buffer found. */
4023 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4024 /*
4025 * 1st delayed buffer found, record
4026 * the start of extent.
4027 */
4028 flags |= FIEMAP_EXTENT_DELALLOC;
4029 newex->ec_block = end;
4030 logical = (__u64)end << blksize_bits;
4031 }
4032 /* Find contiguous delayed buffers. */
4033 do {
4034 if (!buffer_delay(bh))
4035 goto found_delayed_extent;
4036 bh = bh->b_this_page;
4037 end++;
4038 } while (bh != head);
4039
4040 for (; index < ret; index++) {
4041 if (!page_has_buffers(pages[index])) {
4042 bh = NULL;
4043 break;
4044 }
4045 head = page_buffers(pages[index]);
4046 if (!head) {
4047 bh = NULL;
4048 break;
4049 }
4050
4051 if (pages[index]->index !=
4052 pages[start_index]->index + index
4053 - start_index) {
4054 /* Blocks are not contiguous. */
4055 bh = NULL;
4056 break;
4057 }
4058 bh = head;
4059 do {
4060 if (!buffer_delay(bh))
4061 /* Delayed-extent ends. */
4062 goto found_delayed_extent;
4063 bh = bh->b_this_page;
4064 end++;
4065 } while (bh != head);
4066 }
4067 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4068 /* a hole found. */
4069 goto out;
4070
4071found_delayed_extent:
4072 newex->ec_len = min(end - newex->ec_block,
4073 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4074 if (ret == nr_pages && bh != NULL &&
4075 newex->ec_len < EXT_INIT_MAX_LEN &&
4076 buffer_delay(bh)) {
4077 /* Have not collected an extent and continue. */
4078 for (index = 0; index < ret; index++)
4079 page_cache_release(pages[index]);
4080 goto repeat;
4081 }
4082
4083 for (index = 0; index < ret; index++)
4084 page_cache_release(pages[index]);
4085 kfree(pages);
4086 }
4087
4088 physical = (__u64)newex->ec_start << blksize_bits;
4089 length = (__u64)newex->ec_len << blksize_bits;
4090
4091 if (ex && ext4_ext_is_uninitialized(ex))
4092 flags |= FIEMAP_EXTENT_UNWRITTEN;
4093
4094 if (next == EXT_MAX_BLOCKS)
4095 flags |= FIEMAP_EXTENT_LAST;
4096
4097 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4098 length, flags);
4099 if (ret < 0)
4100 return ret;
4101 if (ret == 1)
4102 return EXT_BREAK;
4103 return EXT_CONTINUE;
4104}
4105
4106/* fiemap flags we can handle specified here */
4107#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4108
4109static int ext4_xattr_fiemap(struct inode *inode,
4110 struct fiemap_extent_info *fieinfo)
4111{
4112 __u64 physical = 0;
4113 __u64 length;
4114 __u32 flags = FIEMAP_EXTENT_LAST;
4115 int blockbits = inode->i_sb->s_blocksize_bits;
4116 int error = 0;
4117
4118 /* in-inode? */
4119 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4120 struct ext4_iloc iloc;
4121 int offset; /* offset of xattr in inode */
4122
4123 error = ext4_get_inode_loc(inode, &iloc);
4124 if (error)
4125 return error;
4126 physical = iloc.bh->b_blocknr << blockbits;
4127 offset = EXT4_GOOD_OLD_INODE_SIZE +
4128 EXT4_I(inode)->i_extra_isize;
4129 physical += offset;
4130 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4131 flags |= FIEMAP_EXTENT_DATA_INLINE;
4132 brelse(iloc.bh);
4133 } else { /* external block */
4134 physical = EXT4_I(inode)->i_file_acl << blockbits;
4135 length = inode->i_sb->s_blocksize;
4136 }
4137
4138 if (physical)
4139 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4140 length, flags);
4141 return (error < 0 ? error : 0);
4142}
4143
4144/*
4145 * ext4_ext_punch_hole
4146 *
4147 * Punches a hole of "length" bytes in a file starting
4148 * at byte "offset"
4149 *
4150 * @inode: The inode of the file to punch a hole in
4151 * @offset: The starting byte offset of the hole
4152 * @length: The length of the hole
4153 *
4154 * Returns the number of blocks removed or negative on err
4155 */
4156int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4157{
4158 struct inode *inode = file->f_path.dentry->d_inode;
4159 struct super_block *sb = inode->i_sb;
4160 struct ext4_ext_cache cache_ex;
4161 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4162 struct address_space *mapping = inode->i_mapping;
4163 struct ext4_map_blocks map;
4164 handle_t *handle;
4165 loff_t first_block_offset, last_block_offset, block_len;
4166 loff_t first_page, last_page, first_page_offset, last_page_offset;
4167 int ret, credits, blocks_released, err = 0;
4168
4169 first_block = (offset + sb->s_blocksize - 1) >>
4170 EXT4_BLOCK_SIZE_BITS(sb);
4171 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4172
4173 first_block_offset = first_block << EXT4_BLOCK_SIZE_BITS(sb);
4174 last_block_offset = last_block << EXT4_BLOCK_SIZE_BITS(sb);
4175
4176 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4177 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4178
4179 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4180 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4181
4182 /*
4183 * Write out all dirty pages to avoid race conditions
4184 * Then release them.
4185 */
4186 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4187 err = filemap_write_and_wait_range(mapping,
4188 first_page_offset == 0 ? 0 : first_page_offset-1,
4189 last_page_offset);
4190
4191 if (err)
4192 return err;
4193 }
4194
4195 /* Now release the pages */
4196 if (last_page_offset > first_page_offset) {
4197 truncate_inode_pages_range(mapping, first_page_offset,
4198 last_page_offset-1);
4199 }
4200
4201 /* finish any pending end_io work */
4202 ext4_flush_completed_IO(inode);
4203
4204 credits = ext4_writepage_trans_blocks(inode);
4205 handle = ext4_journal_start(inode, credits);
4206 if (IS_ERR(handle))
4207 return PTR_ERR(handle);
4208
4209 err = ext4_orphan_add(handle, inode);
4210 if (err)
4211 goto out;
4212
4213 /*
4214 * Now we need to zero out the un block aligned data.
4215 * If the file is smaller than a block, just
4216 * zero out the middle
4217 */
4218 if (first_block > last_block)
4219 ext4_block_zero_page_range(handle, mapping, offset, length);
4220 else {
4221 /* zero out the head of the hole before the first block */
4222 block_len = first_block_offset - offset;
4223 if (block_len > 0)
4224 ext4_block_zero_page_range(handle, mapping,
4225 offset, block_len);
4226
4227 /* zero out the tail of the hole after the last block */
4228 block_len = offset + length - last_block_offset;
4229 if (block_len > 0) {
4230 ext4_block_zero_page_range(handle, mapping,
4231 last_block_offset, block_len);
4232 }
4233 }
4234
4235 /* If there are no blocks to remove, return now */
4236 if (first_block >= last_block)
4237 goto out;
4238
4239 down_write(&EXT4_I(inode)->i_data_sem);
4240 ext4_ext_invalidate_cache(inode);
4241 ext4_discard_preallocations(inode);
4242
4243 /*
4244 * Loop over all the blocks and identify blocks
4245 * that need to be punched out
4246 */
4247 iblock = first_block;
4248 blocks_released = 0;
4249 while (iblock < last_block) {
4250 max_blocks = last_block - iblock;
4251 num_blocks = 1;
4252 memset(&map, 0, sizeof(map));
4253 map.m_lblk = iblock;
4254 map.m_len = max_blocks;
4255 ret = ext4_ext_map_blocks(handle, inode, &map,
4256 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4257
4258 if (ret > 0) {
4259 blocks_released += ret;
4260 num_blocks = ret;
4261 } else if (ret == 0) {
4262 /*
4263 * If map blocks could not find the block,
4264 * then it is in a hole. If the hole was
4265 * not already cached, then map blocks should
4266 * put it in the cache. So we can get the hole
4267 * out of the cache
4268 */
4269 memset(&cache_ex, 0, sizeof(cache_ex));
4270 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4271 !cache_ex.ec_start) {
4272
4273 /* The hole is cached */
4274 num_blocks = cache_ex.ec_block +
4275 cache_ex.ec_len - iblock;
4276
4277 } else {
4278 /* The block could not be identified */
4279 err = -EIO;
4280 break;
4281 }
4282 } else {
4283 /* Map blocks error */
4284 err = ret;
4285 break;
4286 }
4287
4288 if (num_blocks == 0) {
4289 /* This condition should never happen */
4290 ext_debug("Block lookup failed");
4291 err = -EIO;
4292 break;
4293 }
4294
4295 iblock += num_blocks;
4296 }
4297
4298 if (blocks_released > 0) {
4299 ext4_ext_invalidate_cache(inode);
4300 ext4_discard_preallocations(inode);
4301 }
4302
4303 if (IS_SYNC(inode))
4304 ext4_handle_sync(handle);
4305
4306 up_write(&EXT4_I(inode)->i_data_sem);
4307
4308out:
4309 ext4_orphan_del(handle, inode);
4310 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4311 ext4_mark_inode_dirty(handle, inode);
4312 ext4_journal_stop(handle);
4313 return err;
4314}
4315int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4316 __u64 start, __u64 len)
4317{
4318 ext4_lblk_t start_blk;
4319 int error = 0;
4320
4321 /* fallback to generic here if not in extents fmt */
4322 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4323 return generic_block_fiemap(inode, fieinfo, start, len,
4324 ext4_get_block);
4325
4326 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4327 return -EBADR;
4328
4329 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4330 error = ext4_xattr_fiemap(inode, fieinfo);
4331 } else {
4332 ext4_lblk_t len_blks;
4333 __u64 last_blk;
4334
4335 start_blk = start >> inode->i_sb->s_blocksize_bits;
4336 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4337 if (last_blk >= EXT_MAX_BLOCKS)
4338 last_blk = EXT_MAX_BLOCKS-1;
4339 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4340
4341 /*
4342 * Walk the extent tree gathering extent information.
4343 * ext4_ext_fiemap_cb will push extents back to user.
4344 */
4345 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4346 ext4_ext_fiemap_cb, fieinfo);
4347 }
4348
4349 return error;
4350}
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/fs.h>
33#include <linux/time.h>
34#include <linux/jbd2.h>
35#include <linux/highuid.h>
36#include <linux/pagemap.h>
37#include <linux/quotaops.h>
38#include <linux/string.h>
39#include <linux/slab.h>
40#include <asm/uaccess.h>
41#include <linux/fiemap.h>
42#include "ext4_jbd2.h"
43#include "ext4_extents.h"
44#include "xattr.h"
45
46#include <trace/events/ext4.h>
47
48/*
49 * used by extent splitting.
50 */
51#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
52 due to ENOSPC */
53#define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
54#define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
55
56#define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57#define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
58
59static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
61{
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
64 __u32 csum;
65
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
69}
70
71static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
73{
74 struct ext4_extent_tail *et;
75
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
78 return 1;
79
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
82 return 0;
83 return 1;
84}
85
86static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
88{
89 struct ext4_extent_tail *et;
90
91 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
93 return;
94
95 et = find_ext4_extent_tail(eh);
96 et->et_checksum = ext4_extent_block_csum(inode, eh);
97}
98
99static int ext4_split_extent(handle_t *handle,
100 struct inode *inode,
101 struct ext4_ext_path *path,
102 struct ext4_map_blocks *map,
103 int split_flag,
104 int flags);
105
106static int ext4_split_extent_at(handle_t *handle,
107 struct inode *inode,
108 struct ext4_ext_path *path,
109 ext4_lblk_t split,
110 int split_flag,
111 int flags);
112
113static int ext4_find_delayed_extent(struct inode *inode,
114 struct extent_status *newes);
115
116static int ext4_ext_truncate_extend_restart(handle_t *handle,
117 struct inode *inode,
118 int needed)
119{
120 int err;
121
122 if (!ext4_handle_valid(handle))
123 return 0;
124 if (handle->h_buffer_credits > needed)
125 return 0;
126 err = ext4_journal_extend(handle, needed);
127 if (err <= 0)
128 return err;
129 err = ext4_truncate_restart_trans(handle, inode, needed);
130 if (err == 0)
131 err = -EAGAIN;
132
133 return err;
134}
135
136/*
137 * could return:
138 * - EROFS
139 * - ENOMEM
140 */
141static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
142 struct ext4_ext_path *path)
143{
144 if (path->p_bh) {
145 /* path points to block */
146 return ext4_journal_get_write_access(handle, path->p_bh);
147 }
148 /* path points to leaf/index in inode body */
149 /* we use in-core data, no need to protect them */
150 return 0;
151}
152
153/*
154 * could return:
155 * - EROFS
156 * - ENOMEM
157 * - EIO
158 */
159int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
160 struct inode *inode, struct ext4_ext_path *path)
161{
162 int err;
163 if (path->p_bh) {
164 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
165 /* path points to block */
166 err = __ext4_handle_dirty_metadata(where, line, handle,
167 inode, path->p_bh);
168 } else {
169 /* path points to leaf/index in inode body */
170 err = ext4_mark_inode_dirty(handle, inode);
171 }
172 return err;
173}
174
175static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
176 struct ext4_ext_path *path,
177 ext4_lblk_t block)
178{
179 if (path) {
180 int depth = path->p_depth;
181 struct ext4_extent *ex;
182
183 /*
184 * Try to predict block placement assuming that we are
185 * filling in a file which will eventually be
186 * non-sparse --- i.e., in the case of libbfd writing
187 * an ELF object sections out-of-order but in a way
188 * the eventually results in a contiguous object or
189 * executable file, or some database extending a table
190 * space file. However, this is actually somewhat
191 * non-ideal if we are writing a sparse file such as
192 * qemu or KVM writing a raw image file that is going
193 * to stay fairly sparse, since it will end up
194 * fragmenting the file system's free space. Maybe we
195 * should have some hueristics or some way to allow
196 * userspace to pass a hint to file system,
197 * especially if the latter case turns out to be
198 * common.
199 */
200 ex = path[depth].p_ext;
201 if (ex) {
202 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
203 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
204
205 if (block > ext_block)
206 return ext_pblk + (block - ext_block);
207 else
208 return ext_pblk - (ext_block - block);
209 }
210
211 /* it looks like index is empty;
212 * try to find starting block from index itself */
213 if (path[depth].p_bh)
214 return path[depth].p_bh->b_blocknr;
215 }
216
217 /* OK. use inode's group */
218 return ext4_inode_to_goal_block(inode);
219}
220
221/*
222 * Allocation for a meta data block
223 */
224static ext4_fsblk_t
225ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
226 struct ext4_ext_path *path,
227 struct ext4_extent *ex, int *err, unsigned int flags)
228{
229 ext4_fsblk_t goal, newblock;
230
231 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
232 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
233 NULL, err);
234 return newblock;
235}
236
237static inline int ext4_ext_space_block(struct inode *inode, int check)
238{
239 int size;
240
241 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
242 / sizeof(struct ext4_extent);
243#ifdef AGGRESSIVE_TEST
244 if (!check && size > 6)
245 size = 6;
246#endif
247 return size;
248}
249
250static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
251{
252 int size;
253
254 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
255 / sizeof(struct ext4_extent_idx);
256#ifdef AGGRESSIVE_TEST
257 if (!check && size > 5)
258 size = 5;
259#endif
260 return size;
261}
262
263static inline int ext4_ext_space_root(struct inode *inode, int check)
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);
270#ifdef AGGRESSIVE_TEST
271 if (!check && size > 3)
272 size = 3;
273#endif
274 return size;
275}
276
277static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
278{
279 int size;
280
281 size = sizeof(EXT4_I(inode)->i_data);
282 size -= sizeof(struct ext4_extent_header);
283 size /= sizeof(struct ext4_extent_idx);
284#ifdef AGGRESSIVE_TEST
285 if (!check && size > 4)
286 size = 4;
287#endif
288 return size;
289}
290
291/*
292 * Calculate the number of metadata blocks needed
293 * to allocate @blocks
294 * Worse case is one block per extent
295 */
296int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
297{
298 struct ext4_inode_info *ei = EXT4_I(inode);
299 int idxs;
300
301 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
302 / sizeof(struct ext4_extent_idx));
303
304 /*
305 * If the new delayed allocation block is contiguous with the
306 * previous da block, it can share index blocks with the
307 * previous block, so we only need to allocate a new index
308 * block every idxs leaf blocks. At ldxs**2 blocks, we need
309 * an additional index block, and at ldxs**3 blocks, yet
310 * another index blocks.
311 */
312 if (ei->i_da_metadata_calc_len &&
313 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
314 int num = 0;
315
316 if ((ei->i_da_metadata_calc_len % idxs) == 0)
317 num++;
318 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
319 num++;
320 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
321 num++;
322 ei->i_da_metadata_calc_len = 0;
323 } else
324 ei->i_da_metadata_calc_len++;
325 ei->i_da_metadata_calc_last_lblock++;
326 return num;
327 }
328
329 /*
330 * In the worst case we need a new set of index blocks at
331 * every level of the inode's extent tree.
332 */
333 ei->i_da_metadata_calc_len = 1;
334 ei->i_da_metadata_calc_last_lblock = lblock;
335 return ext_depth(inode) + 1;
336}
337
338static int
339ext4_ext_max_entries(struct inode *inode, int depth)
340{
341 int max;
342
343 if (depth == ext_depth(inode)) {
344 if (depth == 0)
345 max = ext4_ext_space_root(inode, 1);
346 else
347 max = ext4_ext_space_root_idx(inode, 1);
348 } else {
349 if (depth == 0)
350 max = ext4_ext_space_block(inode, 1);
351 else
352 max = ext4_ext_space_block_idx(inode, 1);
353 }
354
355 return max;
356}
357
358static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
359{
360 ext4_fsblk_t block = ext4_ext_pblock(ext);
361 int len = ext4_ext_get_actual_len(ext);
362 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
363 ext4_lblk_t last = lblock + len - 1;
364
365 if (lblock > last)
366 return 0;
367 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
368}
369
370static int ext4_valid_extent_idx(struct inode *inode,
371 struct ext4_extent_idx *ext_idx)
372{
373 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
374
375 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
376}
377
378static int ext4_valid_extent_entries(struct inode *inode,
379 struct ext4_extent_header *eh,
380 int depth)
381{
382 unsigned short entries;
383 if (eh->eh_entries == 0)
384 return 1;
385
386 entries = le16_to_cpu(eh->eh_entries);
387
388 if (depth == 0) {
389 /* leaf entries */
390 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
391 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
392 ext4_fsblk_t pblock = 0;
393 ext4_lblk_t lblock = 0;
394 ext4_lblk_t prev = 0;
395 int len = 0;
396 while (entries) {
397 if (!ext4_valid_extent(inode, ext))
398 return 0;
399
400 /* Check for overlapping extents */
401 lblock = le32_to_cpu(ext->ee_block);
402 len = ext4_ext_get_actual_len(ext);
403 if ((lblock <= prev) && prev) {
404 pblock = ext4_ext_pblock(ext);
405 es->s_last_error_block = cpu_to_le64(pblock);
406 return 0;
407 }
408 ext++;
409 entries--;
410 prev = lblock + len - 1;
411 }
412 } else {
413 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
414 while (entries) {
415 if (!ext4_valid_extent_idx(inode, ext_idx))
416 return 0;
417 ext_idx++;
418 entries--;
419 }
420 }
421 return 1;
422}
423
424static int __ext4_ext_check(const char *function, unsigned int line,
425 struct inode *inode, struct ext4_extent_header *eh,
426 int depth, ext4_fsblk_t pblk)
427{
428 const char *error_msg;
429 int max = 0;
430
431 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
432 error_msg = "invalid magic";
433 goto corrupted;
434 }
435 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
436 error_msg = "unexpected eh_depth";
437 goto corrupted;
438 }
439 if (unlikely(eh->eh_max == 0)) {
440 error_msg = "invalid eh_max";
441 goto corrupted;
442 }
443 max = ext4_ext_max_entries(inode, depth);
444 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
445 error_msg = "too large eh_max";
446 goto corrupted;
447 }
448 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
449 error_msg = "invalid eh_entries";
450 goto corrupted;
451 }
452 if (!ext4_valid_extent_entries(inode, eh, depth)) {
453 error_msg = "invalid extent entries";
454 goto corrupted;
455 }
456 /* Verify checksum on non-root extent tree nodes */
457 if (ext_depth(inode) != depth &&
458 !ext4_extent_block_csum_verify(inode, eh)) {
459 error_msg = "extent tree corrupted";
460 goto corrupted;
461 }
462 return 0;
463
464corrupted:
465 ext4_error_inode(inode, function, line, 0,
466 "pblk %llu bad header/extent: %s - magic %x, "
467 "entries %u, max %u(%u), depth %u(%u)",
468 (unsigned long long) pblk, error_msg,
469 le16_to_cpu(eh->eh_magic),
470 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
471 max, le16_to_cpu(eh->eh_depth), depth);
472 return -EIO;
473}
474
475#define ext4_ext_check(inode, eh, depth, pblk) \
476 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
477
478int ext4_ext_check_inode(struct inode *inode)
479{
480 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
481}
482
483static struct buffer_head *
484__read_extent_tree_block(const char *function, unsigned int line,
485 struct inode *inode, ext4_fsblk_t pblk, int depth,
486 int flags)
487{
488 struct buffer_head *bh;
489 int err;
490
491 bh = sb_getblk(inode->i_sb, pblk);
492 if (unlikely(!bh))
493 return ERR_PTR(-ENOMEM);
494
495 if (!bh_uptodate_or_lock(bh)) {
496 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
497 err = bh_submit_read(bh);
498 if (err < 0)
499 goto errout;
500 }
501 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
502 return bh;
503 err = __ext4_ext_check(function, line, inode,
504 ext_block_hdr(bh), depth, pblk);
505 if (err)
506 goto errout;
507 set_buffer_verified(bh);
508 /*
509 * If this is a leaf block, cache all of its entries
510 */
511 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
512 struct ext4_extent_header *eh = ext_block_hdr(bh);
513 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
514 ext4_lblk_t prev = 0;
515 int i;
516
517 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
518 unsigned int status = EXTENT_STATUS_WRITTEN;
519 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
520 int len = ext4_ext_get_actual_len(ex);
521
522 if (prev && (prev != lblk))
523 ext4_es_cache_extent(inode, prev,
524 lblk - prev, ~0,
525 EXTENT_STATUS_HOLE);
526
527 if (ext4_ext_is_uninitialized(ex))
528 status = EXTENT_STATUS_UNWRITTEN;
529 ext4_es_cache_extent(inode, lblk, len,
530 ext4_ext_pblock(ex), status);
531 prev = lblk + len;
532 }
533 }
534 return bh;
535errout:
536 put_bh(bh);
537 return ERR_PTR(err);
538
539}
540
541#define read_extent_tree_block(inode, pblk, depth, flags) \
542 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
543 (depth), (flags))
544
545/*
546 * This function is called to cache a file's extent information in the
547 * extent status tree
548 */
549int ext4_ext_precache(struct inode *inode)
550{
551 struct ext4_inode_info *ei = EXT4_I(inode);
552 struct ext4_ext_path *path = NULL;
553 struct buffer_head *bh;
554 int i = 0, depth, ret = 0;
555
556 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
557 return 0; /* not an extent-mapped inode */
558
559 down_read(&ei->i_data_sem);
560 depth = ext_depth(inode);
561
562 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
563 GFP_NOFS);
564 if (path == NULL) {
565 up_read(&ei->i_data_sem);
566 return -ENOMEM;
567 }
568
569 /* Don't cache anything if there are no external extent blocks */
570 if (depth == 0)
571 goto out;
572 path[0].p_hdr = ext_inode_hdr(inode);
573 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
574 if (ret)
575 goto out;
576 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
577 while (i >= 0) {
578 /*
579 * If this is a leaf block or we've reached the end of
580 * the index block, go up
581 */
582 if ((i == depth) ||
583 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
584 brelse(path[i].p_bh);
585 path[i].p_bh = NULL;
586 i--;
587 continue;
588 }
589 bh = read_extent_tree_block(inode,
590 ext4_idx_pblock(path[i].p_idx++),
591 depth - i - 1,
592 EXT4_EX_FORCE_CACHE);
593 if (IS_ERR(bh)) {
594 ret = PTR_ERR(bh);
595 break;
596 }
597 i++;
598 path[i].p_bh = bh;
599 path[i].p_hdr = ext_block_hdr(bh);
600 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
601 }
602 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
603out:
604 up_read(&ei->i_data_sem);
605 ext4_ext_drop_refs(path);
606 kfree(path);
607 return ret;
608}
609
610#ifdef EXT_DEBUG
611static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
612{
613 int k, l = path->p_depth;
614
615 ext_debug("path:");
616 for (k = 0; k <= l; k++, path++) {
617 if (path->p_idx) {
618 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
619 ext4_idx_pblock(path->p_idx));
620 } else if (path->p_ext) {
621 ext_debug(" %d:[%d]%d:%llu ",
622 le32_to_cpu(path->p_ext->ee_block),
623 ext4_ext_is_uninitialized(path->p_ext),
624 ext4_ext_get_actual_len(path->p_ext),
625 ext4_ext_pblock(path->p_ext));
626 } else
627 ext_debug(" []");
628 }
629 ext_debug("\n");
630}
631
632static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
633{
634 int depth = ext_depth(inode);
635 struct ext4_extent_header *eh;
636 struct ext4_extent *ex;
637 int i;
638
639 if (!path)
640 return;
641
642 eh = path[depth].p_hdr;
643 ex = EXT_FIRST_EXTENT(eh);
644
645 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
646
647 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
648 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
649 ext4_ext_is_uninitialized(ex),
650 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
651 }
652 ext_debug("\n");
653}
654
655static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
656 ext4_fsblk_t newblock, int level)
657{
658 int depth = ext_depth(inode);
659 struct ext4_extent *ex;
660
661 if (depth != level) {
662 struct ext4_extent_idx *idx;
663 idx = path[level].p_idx;
664 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
665 ext_debug("%d: move %d:%llu in new index %llu\n", level,
666 le32_to_cpu(idx->ei_block),
667 ext4_idx_pblock(idx),
668 newblock);
669 idx++;
670 }
671
672 return;
673 }
674
675 ex = path[depth].p_ext;
676 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
677 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
678 le32_to_cpu(ex->ee_block),
679 ext4_ext_pblock(ex),
680 ext4_ext_is_uninitialized(ex),
681 ext4_ext_get_actual_len(ex),
682 newblock);
683 ex++;
684 }
685}
686
687#else
688#define ext4_ext_show_path(inode, path)
689#define ext4_ext_show_leaf(inode, path)
690#define ext4_ext_show_move(inode, path, newblock, level)
691#endif
692
693void ext4_ext_drop_refs(struct ext4_ext_path *path)
694{
695 int depth = path->p_depth;
696 int i;
697
698 for (i = 0; i <= depth; i++, path++)
699 if (path->p_bh) {
700 brelse(path->p_bh);
701 path->p_bh = NULL;
702 }
703}
704
705/*
706 * ext4_ext_binsearch_idx:
707 * binary search for the closest index of the given block
708 * the header must be checked before calling this
709 */
710static void
711ext4_ext_binsearch_idx(struct inode *inode,
712 struct ext4_ext_path *path, ext4_lblk_t block)
713{
714 struct ext4_extent_header *eh = path->p_hdr;
715 struct ext4_extent_idx *r, *l, *m;
716
717
718 ext_debug("binsearch for %u(idx): ", block);
719
720 l = EXT_FIRST_INDEX(eh) + 1;
721 r = EXT_LAST_INDEX(eh);
722 while (l <= r) {
723 m = l + (r - l) / 2;
724 if (block < le32_to_cpu(m->ei_block))
725 r = m - 1;
726 else
727 l = m + 1;
728 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
729 m, le32_to_cpu(m->ei_block),
730 r, le32_to_cpu(r->ei_block));
731 }
732
733 path->p_idx = l - 1;
734 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
735 ext4_idx_pblock(path->p_idx));
736
737#ifdef CHECK_BINSEARCH
738 {
739 struct ext4_extent_idx *chix, *ix;
740 int k;
741
742 chix = ix = EXT_FIRST_INDEX(eh);
743 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
744 if (k != 0 &&
745 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
746 printk(KERN_DEBUG "k=%d, ix=0x%p, "
747 "first=0x%p\n", k,
748 ix, EXT_FIRST_INDEX(eh));
749 printk(KERN_DEBUG "%u <= %u\n",
750 le32_to_cpu(ix->ei_block),
751 le32_to_cpu(ix[-1].ei_block));
752 }
753 BUG_ON(k && le32_to_cpu(ix->ei_block)
754 <= le32_to_cpu(ix[-1].ei_block));
755 if (block < le32_to_cpu(ix->ei_block))
756 break;
757 chix = ix;
758 }
759 BUG_ON(chix != path->p_idx);
760 }
761#endif
762
763}
764
765/*
766 * ext4_ext_binsearch:
767 * binary search for closest extent of the given block
768 * the header must be checked before calling this
769 */
770static void
771ext4_ext_binsearch(struct inode *inode,
772 struct ext4_ext_path *path, ext4_lblk_t block)
773{
774 struct ext4_extent_header *eh = path->p_hdr;
775 struct ext4_extent *r, *l, *m;
776
777 if (eh->eh_entries == 0) {
778 /*
779 * this leaf is empty:
780 * we get such a leaf in split/add case
781 */
782 return;
783 }
784
785 ext_debug("binsearch for %u: ", block);
786
787 l = EXT_FIRST_EXTENT(eh) + 1;
788 r = EXT_LAST_EXTENT(eh);
789
790 while (l <= r) {
791 m = l + (r - l) / 2;
792 if (block < le32_to_cpu(m->ee_block))
793 r = m - 1;
794 else
795 l = m + 1;
796 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
797 m, le32_to_cpu(m->ee_block),
798 r, le32_to_cpu(r->ee_block));
799 }
800
801 path->p_ext = l - 1;
802 ext_debug(" -> %d:%llu:[%d]%d ",
803 le32_to_cpu(path->p_ext->ee_block),
804 ext4_ext_pblock(path->p_ext),
805 ext4_ext_is_uninitialized(path->p_ext),
806 ext4_ext_get_actual_len(path->p_ext));
807
808#ifdef CHECK_BINSEARCH
809 {
810 struct ext4_extent *chex, *ex;
811 int k;
812
813 chex = ex = EXT_FIRST_EXTENT(eh);
814 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
815 BUG_ON(k && le32_to_cpu(ex->ee_block)
816 <= le32_to_cpu(ex[-1].ee_block));
817 if (block < le32_to_cpu(ex->ee_block))
818 break;
819 chex = ex;
820 }
821 BUG_ON(chex != path->p_ext);
822 }
823#endif
824
825}
826
827int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
828{
829 struct ext4_extent_header *eh;
830
831 eh = ext_inode_hdr(inode);
832 eh->eh_depth = 0;
833 eh->eh_entries = 0;
834 eh->eh_magic = EXT4_EXT_MAGIC;
835 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
836 ext4_mark_inode_dirty(handle, inode);
837 return 0;
838}
839
840struct ext4_ext_path *
841ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
842 struct ext4_ext_path *path, int flags)
843{
844 struct ext4_extent_header *eh;
845 struct buffer_head *bh;
846 short int depth, i, ppos = 0, alloc = 0;
847 int ret;
848
849 eh = ext_inode_hdr(inode);
850 depth = ext_depth(inode);
851
852 /* account possible depth increase */
853 if (!path) {
854 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
855 GFP_NOFS);
856 if (!path)
857 return ERR_PTR(-ENOMEM);
858 alloc = 1;
859 }
860 path[0].p_hdr = eh;
861 path[0].p_bh = NULL;
862
863 i = depth;
864 /* walk through the tree */
865 while (i) {
866 ext_debug("depth %d: num %d, max %d\n",
867 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
868
869 ext4_ext_binsearch_idx(inode, path + ppos, block);
870 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
871 path[ppos].p_depth = i;
872 path[ppos].p_ext = NULL;
873
874 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
875 flags);
876 if (IS_ERR(bh)) {
877 ret = PTR_ERR(bh);
878 goto err;
879 }
880
881 eh = ext_block_hdr(bh);
882 ppos++;
883 if (unlikely(ppos > depth)) {
884 put_bh(bh);
885 EXT4_ERROR_INODE(inode,
886 "ppos %d > depth %d", ppos, depth);
887 ret = -EIO;
888 goto err;
889 }
890 path[ppos].p_bh = bh;
891 path[ppos].p_hdr = eh;
892 }
893
894 path[ppos].p_depth = i;
895 path[ppos].p_ext = NULL;
896 path[ppos].p_idx = NULL;
897
898 /* find extent */
899 ext4_ext_binsearch(inode, path + ppos, block);
900 /* if not an empty leaf */
901 if (path[ppos].p_ext)
902 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
903
904 ext4_ext_show_path(inode, path);
905
906 return path;
907
908err:
909 ext4_ext_drop_refs(path);
910 if (alloc)
911 kfree(path);
912 return ERR_PTR(ret);
913}
914
915/*
916 * ext4_ext_insert_index:
917 * insert new index [@logical;@ptr] into the block at @curp;
918 * check where to insert: before @curp or after @curp
919 */
920static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
921 struct ext4_ext_path *curp,
922 int logical, ext4_fsblk_t ptr)
923{
924 struct ext4_extent_idx *ix;
925 int len, err;
926
927 err = ext4_ext_get_access(handle, inode, curp);
928 if (err)
929 return err;
930
931 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
932 EXT4_ERROR_INODE(inode,
933 "logical %d == ei_block %d!",
934 logical, le32_to_cpu(curp->p_idx->ei_block));
935 return -EIO;
936 }
937
938 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
939 >= le16_to_cpu(curp->p_hdr->eh_max))) {
940 EXT4_ERROR_INODE(inode,
941 "eh_entries %d >= eh_max %d!",
942 le16_to_cpu(curp->p_hdr->eh_entries),
943 le16_to_cpu(curp->p_hdr->eh_max));
944 return -EIO;
945 }
946
947 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
948 /* insert after */
949 ext_debug("insert new index %d after: %llu\n", logical, ptr);
950 ix = curp->p_idx + 1;
951 } else {
952 /* insert before */
953 ext_debug("insert new index %d before: %llu\n", logical, ptr);
954 ix = curp->p_idx;
955 }
956
957 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
958 BUG_ON(len < 0);
959 if (len > 0) {
960 ext_debug("insert new index %d: "
961 "move %d indices from 0x%p to 0x%p\n",
962 logical, len, ix, ix + 1);
963 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
964 }
965
966 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
967 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
968 return -EIO;
969 }
970
971 ix->ei_block = cpu_to_le32(logical);
972 ext4_idx_store_pblock(ix, ptr);
973 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
974
975 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
976 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
977 return -EIO;
978 }
979
980 err = ext4_ext_dirty(handle, inode, curp);
981 ext4_std_error(inode->i_sb, err);
982
983 return err;
984}
985
986/*
987 * ext4_ext_split:
988 * inserts new subtree into the path, using free index entry
989 * at depth @at:
990 * - allocates all needed blocks (new leaf and all intermediate index blocks)
991 * - makes decision where to split
992 * - moves remaining extents and index entries (right to the split point)
993 * into the newly allocated blocks
994 * - initializes subtree
995 */
996static int ext4_ext_split(handle_t *handle, struct inode *inode,
997 unsigned int flags,
998 struct ext4_ext_path *path,
999 struct ext4_extent *newext, int at)
1000{
1001 struct buffer_head *bh = NULL;
1002 int depth = ext_depth(inode);
1003 struct ext4_extent_header *neh;
1004 struct ext4_extent_idx *fidx;
1005 int i = at, k, m, a;
1006 ext4_fsblk_t newblock, oldblock;
1007 __le32 border;
1008 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1009 int err = 0;
1010
1011 /* make decision: where to split? */
1012 /* FIXME: now decision is simplest: at current extent */
1013
1014 /* if current leaf will be split, then we should use
1015 * border from split point */
1016 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1017 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1018 return -EIO;
1019 }
1020 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1021 border = path[depth].p_ext[1].ee_block;
1022 ext_debug("leaf will be split."
1023 " next leaf starts at %d\n",
1024 le32_to_cpu(border));
1025 } else {
1026 border = newext->ee_block;
1027 ext_debug("leaf will be added."
1028 " next leaf starts at %d\n",
1029 le32_to_cpu(border));
1030 }
1031
1032 /*
1033 * If error occurs, then we break processing
1034 * and mark filesystem read-only. index won't
1035 * be inserted and tree will be in consistent
1036 * state. Next mount will repair buffers too.
1037 */
1038
1039 /*
1040 * Get array to track all allocated blocks.
1041 * We need this to handle errors and free blocks
1042 * upon them.
1043 */
1044 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1045 if (!ablocks)
1046 return -ENOMEM;
1047
1048 /* allocate all needed blocks */
1049 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1050 for (a = 0; a < depth - at; a++) {
1051 newblock = ext4_ext_new_meta_block(handle, inode, path,
1052 newext, &err, flags);
1053 if (newblock == 0)
1054 goto cleanup;
1055 ablocks[a] = newblock;
1056 }
1057
1058 /* initialize new leaf */
1059 newblock = ablocks[--a];
1060 if (unlikely(newblock == 0)) {
1061 EXT4_ERROR_INODE(inode, "newblock == 0!");
1062 err = -EIO;
1063 goto cleanup;
1064 }
1065 bh = sb_getblk(inode->i_sb, newblock);
1066 if (unlikely(!bh)) {
1067 err = -ENOMEM;
1068 goto cleanup;
1069 }
1070 lock_buffer(bh);
1071
1072 err = ext4_journal_get_create_access(handle, bh);
1073 if (err)
1074 goto cleanup;
1075
1076 neh = ext_block_hdr(bh);
1077 neh->eh_entries = 0;
1078 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1079 neh->eh_magic = EXT4_EXT_MAGIC;
1080 neh->eh_depth = 0;
1081
1082 /* move remainder of path[depth] to the new leaf */
1083 if (unlikely(path[depth].p_hdr->eh_entries !=
1084 path[depth].p_hdr->eh_max)) {
1085 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1086 path[depth].p_hdr->eh_entries,
1087 path[depth].p_hdr->eh_max);
1088 err = -EIO;
1089 goto cleanup;
1090 }
1091 /* start copy from next extent */
1092 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1093 ext4_ext_show_move(inode, path, newblock, depth);
1094 if (m) {
1095 struct ext4_extent *ex;
1096 ex = EXT_FIRST_EXTENT(neh);
1097 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1098 le16_add_cpu(&neh->eh_entries, m);
1099 }
1100
1101 ext4_extent_block_csum_set(inode, neh);
1102 set_buffer_uptodate(bh);
1103 unlock_buffer(bh);
1104
1105 err = ext4_handle_dirty_metadata(handle, inode, bh);
1106 if (err)
1107 goto cleanup;
1108 brelse(bh);
1109 bh = NULL;
1110
1111 /* correct old leaf */
1112 if (m) {
1113 err = ext4_ext_get_access(handle, inode, path + depth);
1114 if (err)
1115 goto cleanup;
1116 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1117 err = ext4_ext_dirty(handle, inode, path + depth);
1118 if (err)
1119 goto cleanup;
1120
1121 }
1122
1123 /* create intermediate indexes */
1124 k = depth - at - 1;
1125 if (unlikely(k < 0)) {
1126 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1127 err = -EIO;
1128 goto cleanup;
1129 }
1130 if (k)
1131 ext_debug("create %d intermediate indices\n", k);
1132 /* insert new index into current index block */
1133 /* current depth stored in i var */
1134 i = depth - 1;
1135 while (k--) {
1136 oldblock = newblock;
1137 newblock = ablocks[--a];
1138 bh = sb_getblk(inode->i_sb, newblock);
1139 if (unlikely(!bh)) {
1140 err = -ENOMEM;
1141 goto cleanup;
1142 }
1143 lock_buffer(bh);
1144
1145 err = ext4_journal_get_create_access(handle, bh);
1146 if (err)
1147 goto cleanup;
1148
1149 neh = ext_block_hdr(bh);
1150 neh->eh_entries = cpu_to_le16(1);
1151 neh->eh_magic = EXT4_EXT_MAGIC;
1152 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1153 neh->eh_depth = cpu_to_le16(depth - i);
1154 fidx = EXT_FIRST_INDEX(neh);
1155 fidx->ei_block = border;
1156 ext4_idx_store_pblock(fidx, oldblock);
1157
1158 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1159 i, newblock, le32_to_cpu(border), oldblock);
1160
1161 /* move remainder of path[i] to the new index block */
1162 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1163 EXT_LAST_INDEX(path[i].p_hdr))) {
1164 EXT4_ERROR_INODE(inode,
1165 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1166 le32_to_cpu(path[i].p_ext->ee_block));
1167 err = -EIO;
1168 goto cleanup;
1169 }
1170 /* start copy indexes */
1171 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1172 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1173 EXT_MAX_INDEX(path[i].p_hdr));
1174 ext4_ext_show_move(inode, path, newblock, i);
1175 if (m) {
1176 memmove(++fidx, path[i].p_idx,
1177 sizeof(struct ext4_extent_idx) * m);
1178 le16_add_cpu(&neh->eh_entries, m);
1179 }
1180 ext4_extent_block_csum_set(inode, neh);
1181 set_buffer_uptodate(bh);
1182 unlock_buffer(bh);
1183
1184 err = ext4_handle_dirty_metadata(handle, inode, bh);
1185 if (err)
1186 goto cleanup;
1187 brelse(bh);
1188 bh = NULL;
1189
1190 /* correct old index */
1191 if (m) {
1192 err = ext4_ext_get_access(handle, inode, path + i);
1193 if (err)
1194 goto cleanup;
1195 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1196 err = ext4_ext_dirty(handle, inode, path + i);
1197 if (err)
1198 goto cleanup;
1199 }
1200
1201 i--;
1202 }
1203
1204 /* insert new index */
1205 err = ext4_ext_insert_index(handle, inode, path + at,
1206 le32_to_cpu(border), newblock);
1207
1208cleanup:
1209 if (bh) {
1210 if (buffer_locked(bh))
1211 unlock_buffer(bh);
1212 brelse(bh);
1213 }
1214
1215 if (err) {
1216 /* free all allocated blocks in error case */
1217 for (i = 0; i < depth; i++) {
1218 if (!ablocks[i])
1219 continue;
1220 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1221 EXT4_FREE_BLOCKS_METADATA);
1222 }
1223 }
1224 kfree(ablocks);
1225
1226 return err;
1227}
1228
1229/*
1230 * ext4_ext_grow_indepth:
1231 * implements tree growing procedure:
1232 * - allocates new block
1233 * - moves top-level data (index block or leaf) into the new block
1234 * - initializes new top-level, creating index that points to the
1235 * just created block
1236 */
1237static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1238 unsigned int flags,
1239 struct ext4_extent *newext)
1240{
1241 struct ext4_extent_header *neh;
1242 struct buffer_head *bh;
1243 ext4_fsblk_t newblock;
1244 int err = 0;
1245
1246 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1247 newext, &err, flags);
1248 if (newblock == 0)
1249 return err;
1250
1251 bh = sb_getblk(inode->i_sb, newblock);
1252 if (unlikely(!bh))
1253 return -ENOMEM;
1254 lock_buffer(bh);
1255
1256 err = ext4_journal_get_create_access(handle, bh);
1257 if (err) {
1258 unlock_buffer(bh);
1259 goto out;
1260 }
1261
1262 /* move top-level index/leaf into new block */
1263 memmove(bh->b_data, EXT4_I(inode)->i_data,
1264 sizeof(EXT4_I(inode)->i_data));
1265
1266 /* set size of new block */
1267 neh = ext_block_hdr(bh);
1268 /* old root could have indexes or leaves
1269 * so calculate e_max right way */
1270 if (ext_depth(inode))
1271 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1272 else
1273 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1274 neh->eh_magic = EXT4_EXT_MAGIC;
1275 ext4_extent_block_csum_set(inode, neh);
1276 set_buffer_uptodate(bh);
1277 unlock_buffer(bh);
1278
1279 err = ext4_handle_dirty_metadata(handle, inode, bh);
1280 if (err)
1281 goto out;
1282
1283 /* Update top-level index: num,max,pointer */
1284 neh = ext_inode_hdr(inode);
1285 neh->eh_entries = cpu_to_le16(1);
1286 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1287 if (neh->eh_depth == 0) {
1288 /* Root extent block becomes index block */
1289 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1290 EXT_FIRST_INDEX(neh)->ei_block =
1291 EXT_FIRST_EXTENT(neh)->ee_block;
1292 }
1293 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1294 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1295 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1296 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1297
1298 le16_add_cpu(&neh->eh_depth, 1);
1299 ext4_mark_inode_dirty(handle, inode);
1300out:
1301 brelse(bh);
1302
1303 return err;
1304}
1305
1306/*
1307 * ext4_ext_create_new_leaf:
1308 * finds empty index and adds new leaf.
1309 * if no free index is found, then it requests in-depth growing.
1310 */
1311static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1312 unsigned int mb_flags,
1313 unsigned int gb_flags,
1314 struct ext4_ext_path *path,
1315 struct ext4_extent *newext)
1316{
1317 struct ext4_ext_path *curp;
1318 int depth, i, err = 0;
1319
1320repeat:
1321 i = depth = ext_depth(inode);
1322
1323 /* walk up to the tree and look for free index entry */
1324 curp = path + depth;
1325 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1326 i--;
1327 curp--;
1328 }
1329
1330 /* we use already allocated block for index block,
1331 * so subsequent data blocks should be contiguous */
1332 if (EXT_HAS_FREE_INDEX(curp)) {
1333 /* if we found index with free entry, then use that
1334 * entry: create all needed subtree and add new leaf */
1335 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1336 if (err)
1337 goto out;
1338
1339 /* refill path */
1340 ext4_ext_drop_refs(path);
1341 path = ext4_ext_find_extent(inode,
1342 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1343 path, gb_flags);
1344 if (IS_ERR(path))
1345 err = PTR_ERR(path);
1346 } else {
1347 /* tree is full, time to grow in depth */
1348 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1349 if (err)
1350 goto out;
1351
1352 /* refill path */
1353 ext4_ext_drop_refs(path);
1354 path = ext4_ext_find_extent(inode,
1355 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1356 path, gb_flags);
1357 if (IS_ERR(path)) {
1358 err = PTR_ERR(path);
1359 goto out;
1360 }
1361
1362 /*
1363 * only first (depth 0 -> 1) produces free space;
1364 * in all other cases we have to split the grown tree
1365 */
1366 depth = ext_depth(inode);
1367 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1368 /* now we need to split */
1369 goto repeat;
1370 }
1371 }
1372
1373out:
1374 return err;
1375}
1376
1377/*
1378 * search the closest allocated block to the left for *logical
1379 * and returns it at @logical + it's physical address at @phys
1380 * if *logical is the smallest allocated block, the function
1381 * returns 0 at @phys
1382 * return value contains 0 (success) or error code
1383 */
1384static int ext4_ext_search_left(struct inode *inode,
1385 struct ext4_ext_path *path,
1386 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1387{
1388 struct ext4_extent_idx *ix;
1389 struct ext4_extent *ex;
1390 int depth, ee_len;
1391
1392 if (unlikely(path == NULL)) {
1393 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1394 return -EIO;
1395 }
1396 depth = path->p_depth;
1397 *phys = 0;
1398
1399 if (depth == 0 && path->p_ext == NULL)
1400 return 0;
1401
1402 /* usually extent in the path covers blocks smaller
1403 * then *logical, but it can be that extent is the
1404 * first one in the file */
1405
1406 ex = path[depth].p_ext;
1407 ee_len = ext4_ext_get_actual_len(ex);
1408 if (*logical < le32_to_cpu(ex->ee_block)) {
1409 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1410 EXT4_ERROR_INODE(inode,
1411 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1412 *logical, le32_to_cpu(ex->ee_block));
1413 return -EIO;
1414 }
1415 while (--depth >= 0) {
1416 ix = path[depth].p_idx;
1417 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1418 EXT4_ERROR_INODE(inode,
1419 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1420 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1421 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1422 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1423 depth);
1424 return -EIO;
1425 }
1426 }
1427 return 0;
1428 }
1429
1430 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1431 EXT4_ERROR_INODE(inode,
1432 "logical %d < ee_block %d + ee_len %d!",
1433 *logical, le32_to_cpu(ex->ee_block), ee_len);
1434 return -EIO;
1435 }
1436
1437 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1438 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1439 return 0;
1440}
1441
1442/*
1443 * search the closest allocated block to the right for *logical
1444 * and returns it at @logical + it's physical address at @phys
1445 * if *logical is the largest allocated block, the function
1446 * returns 0 at @phys
1447 * return value contains 0 (success) or error code
1448 */
1449static int ext4_ext_search_right(struct inode *inode,
1450 struct ext4_ext_path *path,
1451 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1452 struct ext4_extent **ret_ex)
1453{
1454 struct buffer_head *bh = NULL;
1455 struct ext4_extent_header *eh;
1456 struct ext4_extent_idx *ix;
1457 struct ext4_extent *ex;
1458 ext4_fsblk_t block;
1459 int depth; /* Note, NOT eh_depth; depth from top of tree */
1460 int ee_len;
1461
1462 if (unlikely(path == NULL)) {
1463 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1464 return -EIO;
1465 }
1466 depth = path->p_depth;
1467 *phys = 0;
1468
1469 if (depth == 0 && path->p_ext == NULL)
1470 return 0;
1471
1472 /* usually extent in the path covers blocks smaller
1473 * then *logical, but it can be that extent is the
1474 * first one in the file */
1475
1476 ex = path[depth].p_ext;
1477 ee_len = ext4_ext_get_actual_len(ex);
1478 if (*logical < le32_to_cpu(ex->ee_block)) {
1479 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1480 EXT4_ERROR_INODE(inode,
1481 "first_extent(path[%d].p_hdr) != ex",
1482 depth);
1483 return -EIO;
1484 }
1485 while (--depth >= 0) {
1486 ix = path[depth].p_idx;
1487 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1488 EXT4_ERROR_INODE(inode,
1489 "ix != EXT_FIRST_INDEX *logical %d!",
1490 *logical);
1491 return -EIO;
1492 }
1493 }
1494 goto found_extent;
1495 }
1496
1497 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1498 EXT4_ERROR_INODE(inode,
1499 "logical %d < ee_block %d + ee_len %d!",
1500 *logical, le32_to_cpu(ex->ee_block), ee_len);
1501 return -EIO;
1502 }
1503
1504 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1505 /* next allocated block in this leaf */
1506 ex++;
1507 goto found_extent;
1508 }
1509
1510 /* go up and search for index to the right */
1511 while (--depth >= 0) {
1512 ix = path[depth].p_idx;
1513 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1514 goto got_index;
1515 }
1516
1517 /* we've gone up to the root and found no index to the right */
1518 return 0;
1519
1520got_index:
1521 /* we've found index to the right, let's
1522 * follow it and find the closest allocated
1523 * block to the right */
1524 ix++;
1525 block = ext4_idx_pblock(ix);
1526 while (++depth < path->p_depth) {
1527 /* subtract from p_depth to get proper eh_depth */
1528 bh = read_extent_tree_block(inode, block,
1529 path->p_depth - depth, 0);
1530 if (IS_ERR(bh))
1531 return PTR_ERR(bh);
1532 eh = ext_block_hdr(bh);
1533 ix = EXT_FIRST_INDEX(eh);
1534 block = ext4_idx_pblock(ix);
1535 put_bh(bh);
1536 }
1537
1538 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1539 if (IS_ERR(bh))
1540 return PTR_ERR(bh);
1541 eh = ext_block_hdr(bh);
1542 ex = EXT_FIRST_EXTENT(eh);
1543found_extent:
1544 *logical = le32_to_cpu(ex->ee_block);
1545 *phys = ext4_ext_pblock(ex);
1546 *ret_ex = ex;
1547 if (bh)
1548 put_bh(bh);
1549 return 0;
1550}
1551
1552/*
1553 * ext4_ext_next_allocated_block:
1554 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1555 * NOTE: it considers block number from index entry as
1556 * allocated block. Thus, index entries have to be consistent
1557 * with leaves.
1558 */
1559static ext4_lblk_t
1560ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1561{
1562 int depth;
1563
1564 BUG_ON(path == NULL);
1565 depth = path->p_depth;
1566
1567 if (depth == 0 && path->p_ext == NULL)
1568 return EXT_MAX_BLOCKS;
1569
1570 while (depth >= 0) {
1571 if (depth == path->p_depth) {
1572 /* leaf */
1573 if (path[depth].p_ext &&
1574 path[depth].p_ext !=
1575 EXT_LAST_EXTENT(path[depth].p_hdr))
1576 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1577 } else {
1578 /* index */
1579 if (path[depth].p_idx !=
1580 EXT_LAST_INDEX(path[depth].p_hdr))
1581 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1582 }
1583 depth--;
1584 }
1585
1586 return EXT_MAX_BLOCKS;
1587}
1588
1589/*
1590 * ext4_ext_next_leaf_block:
1591 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1592 */
1593static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1594{
1595 int depth;
1596
1597 BUG_ON(path == NULL);
1598 depth = path->p_depth;
1599
1600 /* zero-tree has no leaf blocks at all */
1601 if (depth == 0)
1602 return EXT_MAX_BLOCKS;
1603
1604 /* go to index block */
1605 depth--;
1606
1607 while (depth >= 0) {
1608 if (path[depth].p_idx !=
1609 EXT_LAST_INDEX(path[depth].p_hdr))
1610 return (ext4_lblk_t)
1611 le32_to_cpu(path[depth].p_idx[1].ei_block);
1612 depth--;
1613 }
1614
1615 return EXT_MAX_BLOCKS;
1616}
1617
1618/*
1619 * ext4_ext_correct_indexes:
1620 * if leaf gets modified and modified extent is first in the leaf,
1621 * then we have to correct all indexes above.
1622 * TODO: do we need to correct tree in all cases?
1623 */
1624static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1625 struct ext4_ext_path *path)
1626{
1627 struct ext4_extent_header *eh;
1628 int depth = ext_depth(inode);
1629 struct ext4_extent *ex;
1630 __le32 border;
1631 int k, err = 0;
1632
1633 eh = path[depth].p_hdr;
1634 ex = path[depth].p_ext;
1635
1636 if (unlikely(ex == NULL || eh == NULL)) {
1637 EXT4_ERROR_INODE(inode,
1638 "ex %p == NULL or eh %p == NULL", ex, eh);
1639 return -EIO;
1640 }
1641
1642 if (depth == 0) {
1643 /* there is no tree at all */
1644 return 0;
1645 }
1646
1647 if (ex != EXT_FIRST_EXTENT(eh)) {
1648 /* we correct tree if first leaf got modified only */
1649 return 0;
1650 }
1651
1652 /*
1653 * TODO: we need correction if border is smaller than current one
1654 */
1655 k = depth - 1;
1656 border = path[depth].p_ext->ee_block;
1657 err = ext4_ext_get_access(handle, inode, path + k);
1658 if (err)
1659 return err;
1660 path[k].p_idx->ei_block = border;
1661 err = ext4_ext_dirty(handle, inode, path + k);
1662 if (err)
1663 return err;
1664
1665 while (k--) {
1666 /* change all left-side indexes */
1667 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1668 break;
1669 err = ext4_ext_get_access(handle, inode, path + k);
1670 if (err)
1671 break;
1672 path[k].p_idx->ei_block = border;
1673 err = ext4_ext_dirty(handle, inode, path + k);
1674 if (err)
1675 break;
1676 }
1677
1678 return err;
1679}
1680
1681int
1682ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1683 struct ext4_extent *ex2)
1684{
1685 unsigned short ext1_ee_len, ext2_ee_len;
1686
1687 /*
1688 * Make sure that both extents are initialized. We don't merge
1689 * uninitialized extents so that we can be sure that end_io code has
1690 * the extent that was written properly split out and conversion to
1691 * initialized is trivial.
1692 */
1693 if (ext4_ext_is_uninitialized(ex1) != ext4_ext_is_uninitialized(ex2))
1694 return 0;
1695
1696 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1697 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1698
1699 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1700 le32_to_cpu(ex2->ee_block))
1701 return 0;
1702
1703 /*
1704 * To allow future support for preallocated extents to be added
1705 * as an RO_COMPAT feature, refuse to merge to extents if
1706 * this can result in the top bit of ee_len being set.
1707 */
1708 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1709 return 0;
1710 if (ext4_ext_is_uninitialized(ex1) &&
1711 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1712 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1713 (ext1_ee_len + ext2_ee_len > EXT_UNINIT_MAX_LEN)))
1714 return 0;
1715#ifdef AGGRESSIVE_TEST
1716 if (ext1_ee_len >= 4)
1717 return 0;
1718#endif
1719
1720 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1721 return 1;
1722 return 0;
1723}
1724
1725/*
1726 * This function tries to merge the "ex" extent to the next extent in the tree.
1727 * It always tries to merge towards right. If you want to merge towards
1728 * left, pass "ex - 1" as argument instead of "ex".
1729 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1730 * 1 if they got merged.
1731 */
1732static int ext4_ext_try_to_merge_right(struct inode *inode,
1733 struct ext4_ext_path *path,
1734 struct ext4_extent *ex)
1735{
1736 struct ext4_extent_header *eh;
1737 unsigned int depth, len;
1738 int merge_done = 0, uninit;
1739
1740 depth = ext_depth(inode);
1741 BUG_ON(path[depth].p_hdr == NULL);
1742 eh = path[depth].p_hdr;
1743
1744 while (ex < EXT_LAST_EXTENT(eh)) {
1745 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1746 break;
1747 /* merge with next extent! */
1748 uninit = ext4_ext_is_uninitialized(ex);
1749 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1750 + ext4_ext_get_actual_len(ex + 1));
1751 if (uninit)
1752 ext4_ext_mark_uninitialized(ex);
1753
1754 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1755 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1756 * sizeof(struct ext4_extent);
1757 memmove(ex + 1, ex + 2, len);
1758 }
1759 le16_add_cpu(&eh->eh_entries, -1);
1760 merge_done = 1;
1761 WARN_ON(eh->eh_entries == 0);
1762 if (!eh->eh_entries)
1763 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1764 }
1765
1766 return merge_done;
1767}
1768
1769/*
1770 * This function does a very simple check to see if we can collapse
1771 * an extent tree with a single extent tree leaf block into the inode.
1772 */
1773static void ext4_ext_try_to_merge_up(handle_t *handle,
1774 struct inode *inode,
1775 struct ext4_ext_path *path)
1776{
1777 size_t s;
1778 unsigned max_root = ext4_ext_space_root(inode, 0);
1779 ext4_fsblk_t blk;
1780
1781 if ((path[0].p_depth != 1) ||
1782 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1783 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1784 return;
1785
1786 /*
1787 * We need to modify the block allocation bitmap and the block
1788 * group descriptor to release the extent tree block. If we
1789 * can't get the journal credits, give up.
1790 */
1791 if (ext4_journal_extend(handle, 2))
1792 return;
1793
1794 /*
1795 * Copy the extent data up to the inode
1796 */
1797 blk = ext4_idx_pblock(path[0].p_idx);
1798 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1799 sizeof(struct ext4_extent_idx);
1800 s += sizeof(struct ext4_extent_header);
1801
1802 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1803 path[0].p_depth = 0;
1804 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1805 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1806 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1807
1808 brelse(path[1].p_bh);
1809 ext4_free_blocks(handle, inode, NULL, blk, 1,
1810 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET |
1811 EXT4_FREE_BLOCKS_RESERVE);
1812}
1813
1814/*
1815 * This function tries to merge the @ex extent to neighbours in the tree.
1816 * return 1 if merge left else 0.
1817 */
1818static void ext4_ext_try_to_merge(handle_t *handle,
1819 struct inode *inode,
1820 struct ext4_ext_path *path,
1821 struct ext4_extent *ex) {
1822 struct ext4_extent_header *eh;
1823 unsigned int depth;
1824 int merge_done = 0;
1825
1826 depth = ext_depth(inode);
1827 BUG_ON(path[depth].p_hdr == NULL);
1828 eh = path[depth].p_hdr;
1829
1830 if (ex > EXT_FIRST_EXTENT(eh))
1831 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1832
1833 if (!merge_done)
1834 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1835
1836 ext4_ext_try_to_merge_up(handle, inode, path);
1837}
1838
1839/*
1840 * check if a portion of the "newext" extent overlaps with an
1841 * existing extent.
1842 *
1843 * If there is an overlap discovered, it updates the length of the newext
1844 * such that there will be no overlap, and then returns 1.
1845 * If there is no overlap found, it returns 0.
1846 */
1847static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1848 struct inode *inode,
1849 struct ext4_extent *newext,
1850 struct ext4_ext_path *path)
1851{
1852 ext4_lblk_t b1, b2;
1853 unsigned int depth, len1;
1854 unsigned int ret = 0;
1855
1856 b1 = le32_to_cpu(newext->ee_block);
1857 len1 = ext4_ext_get_actual_len(newext);
1858 depth = ext_depth(inode);
1859 if (!path[depth].p_ext)
1860 goto out;
1861 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1862
1863 /*
1864 * get the next allocated block if the extent in the path
1865 * is before the requested block(s)
1866 */
1867 if (b2 < b1) {
1868 b2 = ext4_ext_next_allocated_block(path);
1869 if (b2 == EXT_MAX_BLOCKS)
1870 goto out;
1871 b2 = EXT4_LBLK_CMASK(sbi, b2);
1872 }
1873
1874 /* check for wrap through zero on extent logical start block*/
1875 if (b1 + len1 < b1) {
1876 len1 = EXT_MAX_BLOCKS - b1;
1877 newext->ee_len = cpu_to_le16(len1);
1878 ret = 1;
1879 }
1880
1881 /* check for overlap */
1882 if (b1 + len1 > b2) {
1883 newext->ee_len = cpu_to_le16(b2 - b1);
1884 ret = 1;
1885 }
1886out:
1887 return ret;
1888}
1889
1890/*
1891 * ext4_ext_insert_extent:
1892 * tries to merge requsted extent into the existing extent or
1893 * inserts requested extent as new one into the tree,
1894 * creating new leaf in the no-space case.
1895 */
1896int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1897 struct ext4_ext_path *path,
1898 struct ext4_extent *newext, int gb_flags)
1899{
1900 struct ext4_extent_header *eh;
1901 struct ext4_extent *ex, *fex;
1902 struct ext4_extent *nearex; /* nearest extent */
1903 struct ext4_ext_path *npath = NULL;
1904 int depth, len, err;
1905 ext4_lblk_t next;
1906 int mb_flags = 0, uninit;
1907
1908 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1909 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1910 return -EIO;
1911 }
1912 depth = ext_depth(inode);
1913 ex = path[depth].p_ext;
1914 eh = path[depth].p_hdr;
1915 if (unlikely(path[depth].p_hdr == NULL)) {
1916 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1917 return -EIO;
1918 }
1919
1920 /* try to insert block into found extent and return */
1921 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1922
1923 /*
1924 * Try to see whether we should rather test the extent on
1925 * right from ex, or from the left of ex. This is because
1926 * ext4_ext_find_extent() can return either extent on the
1927 * left, or on the right from the searched position. This
1928 * will make merging more effective.
1929 */
1930 if (ex < EXT_LAST_EXTENT(eh) &&
1931 (le32_to_cpu(ex->ee_block) +
1932 ext4_ext_get_actual_len(ex) <
1933 le32_to_cpu(newext->ee_block))) {
1934 ex += 1;
1935 goto prepend;
1936 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1937 (le32_to_cpu(newext->ee_block) +
1938 ext4_ext_get_actual_len(newext) <
1939 le32_to_cpu(ex->ee_block)))
1940 ex -= 1;
1941
1942 /* Try to append newex to the ex */
1943 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1944 ext_debug("append [%d]%d block to %u:[%d]%d"
1945 "(from %llu)\n",
1946 ext4_ext_is_uninitialized(newext),
1947 ext4_ext_get_actual_len(newext),
1948 le32_to_cpu(ex->ee_block),
1949 ext4_ext_is_uninitialized(ex),
1950 ext4_ext_get_actual_len(ex),
1951 ext4_ext_pblock(ex));
1952 err = ext4_ext_get_access(handle, inode,
1953 path + depth);
1954 if (err)
1955 return err;
1956 uninit = ext4_ext_is_uninitialized(ex);
1957 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1958 + ext4_ext_get_actual_len(newext));
1959 if (uninit)
1960 ext4_ext_mark_uninitialized(ex);
1961 eh = path[depth].p_hdr;
1962 nearex = ex;
1963 goto merge;
1964 }
1965
1966prepend:
1967 /* Try to prepend newex to the ex */
1968 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1969 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1970 "(from %llu)\n",
1971 le32_to_cpu(newext->ee_block),
1972 ext4_ext_is_uninitialized(newext),
1973 ext4_ext_get_actual_len(newext),
1974 le32_to_cpu(ex->ee_block),
1975 ext4_ext_is_uninitialized(ex),
1976 ext4_ext_get_actual_len(ex),
1977 ext4_ext_pblock(ex));
1978 err = ext4_ext_get_access(handle, inode,
1979 path + depth);
1980 if (err)
1981 return err;
1982
1983 uninit = ext4_ext_is_uninitialized(ex);
1984 ex->ee_block = newext->ee_block;
1985 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1986 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1987 + ext4_ext_get_actual_len(newext));
1988 if (uninit)
1989 ext4_ext_mark_uninitialized(ex);
1990 eh = path[depth].p_hdr;
1991 nearex = ex;
1992 goto merge;
1993 }
1994 }
1995
1996 depth = ext_depth(inode);
1997 eh = path[depth].p_hdr;
1998 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1999 goto has_space;
2000
2001 /* probably next leaf has space for us? */
2002 fex = EXT_LAST_EXTENT(eh);
2003 next = EXT_MAX_BLOCKS;
2004 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2005 next = ext4_ext_next_leaf_block(path);
2006 if (next != EXT_MAX_BLOCKS) {
2007 ext_debug("next leaf block - %u\n", next);
2008 BUG_ON(npath != NULL);
2009 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2010 if (IS_ERR(npath))
2011 return PTR_ERR(npath);
2012 BUG_ON(npath->p_depth != path->p_depth);
2013 eh = npath[depth].p_hdr;
2014 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2015 ext_debug("next leaf isn't full(%d)\n",
2016 le16_to_cpu(eh->eh_entries));
2017 path = npath;
2018 goto has_space;
2019 }
2020 ext_debug("next leaf has no free space(%d,%d)\n",
2021 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2022 }
2023
2024 /*
2025 * There is no free space in the found leaf.
2026 * We're gonna add a new leaf in the tree.
2027 */
2028 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2029 mb_flags = EXT4_MB_USE_RESERVED;
2030 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2031 path, newext);
2032 if (err)
2033 goto cleanup;
2034 depth = ext_depth(inode);
2035 eh = path[depth].p_hdr;
2036
2037has_space:
2038 nearex = path[depth].p_ext;
2039
2040 err = ext4_ext_get_access(handle, inode, path + depth);
2041 if (err)
2042 goto cleanup;
2043
2044 if (!nearex) {
2045 /* there is no extent in this leaf, create first one */
2046 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2047 le32_to_cpu(newext->ee_block),
2048 ext4_ext_pblock(newext),
2049 ext4_ext_is_uninitialized(newext),
2050 ext4_ext_get_actual_len(newext));
2051 nearex = EXT_FIRST_EXTENT(eh);
2052 } else {
2053 if (le32_to_cpu(newext->ee_block)
2054 > le32_to_cpu(nearex->ee_block)) {
2055 /* Insert after */
2056 ext_debug("insert %u:%llu:[%d]%d before: "
2057 "nearest %p\n",
2058 le32_to_cpu(newext->ee_block),
2059 ext4_ext_pblock(newext),
2060 ext4_ext_is_uninitialized(newext),
2061 ext4_ext_get_actual_len(newext),
2062 nearex);
2063 nearex++;
2064 } else {
2065 /* Insert before */
2066 BUG_ON(newext->ee_block == nearex->ee_block);
2067 ext_debug("insert %u:%llu:[%d]%d after: "
2068 "nearest %p\n",
2069 le32_to_cpu(newext->ee_block),
2070 ext4_ext_pblock(newext),
2071 ext4_ext_is_uninitialized(newext),
2072 ext4_ext_get_actual_len(newext),
2073 nearex);
2074 }
2075 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2076 if (len > 0) {
2077 ext_debug("insert %u:%llu:[%d]%d: "
2078 "move %d extents from 0x%p to 0x%p\n",
2079 le32_to_cpu(newext->ee_block),
2080 ext4_ext_pblock(newext),
2081 ext4_ext_is_uninitialized(newext),
2082 ext4_ext_get_actual_len(newext),
2083 len, nearex, nearex + 1);
2084 memmove(nearex + 1, nearex,
2085 len * sizeof(struct ext4_extent));
2086 }
2087 }
2088
2089 le16_add_cpu(&eh->eh_entries, 1);
2090 path[depth].p_ext = nearex;
2091 nearex->ee_block = newext->ee_block;
2092 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2093 nearex->ee_len = newext->ee_len;
2094
2095merge:
2096 /* try to merge extents */
2097 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2098 ext4_ext_try_to_merge(handle, inode, path, nearex);
2099
2100
2101 /* time to correct all indexes above */
2102 err = ext4_ext_correct_indexes(handle, inode, path);
2103 if (err)
2104 goto cleanup;
2105
2106 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2107
2108cleanup:
2109 if (npath) {
2110 ext4_ext_drop_refs(npath);
2111 kfree(npath);
2112 }
2113 return err;
2114}
2115
2116static int ext4_fill_fiemap_extents(struct inode *inode,
2117 ext4_lblk_t block, ext4_lblk_t num,
2118 struct fiemap_extent_info *fieinfo)
2119{
2120 struct ext4_ext_path *path = NULL;
2121 struct ext4_extent *ex;
2122 struct extent_status es;
2123 ext4_lblk_t next, next_del, start = 0, end = 0;
2124 ext4_lblk_t last = block + num;
2125 int exists, depth = 0, err = 0;
2126 unsigned int flags = 0;
2127 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2128
2129 while (block < last && block != EXT_MAX_BLOCKS) {
2130 num = last - block;
2131 /* find extent for this block */
2132 down_read(&EXT4_I(inode)->i_data_sem);
2133
2134 if (path && ext_depth(inode) != depth) {
2135 /* depth was changed. we have to realloc path */
2136 kfree(path);
2137 path = NULL;
2138 }
2139
2140 path = ext4_ext_find_extent(inode, block, path, 0);
2141 if (IS_ERR(path)) {
2142 up_read(&EXT4_I(inode)->i_data_sem);
2143 err = PTR_ERR(path);
2144 path = NULL;
2145 break;
2146 }
2147
2148 depth = ext_depth(inode);
2149 if (unlikely(path[depth].p_hdr == NULL)) {
2150 up_read(&EXT4_I(inode)->i_data_sem);
2151 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2152 err = -EIO;
2153 break;
2154 }
2155 ex = path[depth].p_ext;
2156 next = ext4_ext_next_allocated_block(path);
2157 ext4_ext_drop_refs(path);
2158
2159 flags = 0;
2160 exists = 0;
2161 if (!ex) {
2162 /* there is no extent yet, so try to allocate
2163 * all requested space */
2164 start = block;
2165 end = block + num;
2166 } else if (le32_to_cpu(ex->ee_block) > block) {
2167 /* need to allocate space before found extent */
2168 start = block;
2169 end = le32_to_cpu(ex->ee_block);
2170 if (block + num < end)
2171 end = block + num;
2172 } else if (block >= le32_to_cpu(ex->ee_block)
2173 + ext4_ext_get_actual_len(ex)) {
2174 /* need to allocate space after found extent */
2175 start = block;
2176 end = block + num;
2177 if (end >= next)
2178 end = next;
2179 } else if (block >= le32_to_cpu(ex->ee_block)) {
2180 /*
2181 * some part of requested space is covered
2182 * by found extent
2183 */
2184 start = block;
2185 end = le32_to_cpu(ex->ee_block)
2186 + ext4_ext_get_actual_len(ex);
2187 if (block + num < end)
2188 end = block + num;
2189 exists = 1;
2190 } else {
2191 BUG();
2192 }
2193 BUG_ON(end <= start);
2194
2195 if (!exists) {
2196 es.es_lblk = start;
2197 es.es_len = end - start;
2198 es.es_pblk = 0;
2199 } else {
2200 es.es_lblk = le32_to_cpu(ex->ee_block);
2201 es.es_len = ext4_ext_get_actual_len(ex);
2202 es.es_pblk = ext4_ext_pblock(ex);
2203 if (ext4_ext_is_uninitialized(ex))
2204 flags |= FIEMAP_EXTENT_UNWRITTEN;
2205 }
2206
2207 /*
2208 * Find delayed extent and update es accordingly. We call
2209 * it even in !exists case to find out whether es is the
2210 * last existing extent or not.
2211 */
2212 next_del = ext4_find_delayed_extent(inode, &es);
2213 if (!exists && next_del) {
2214 exists = 1;
2215 flags |= (FIEMAP_EXTENT_DELALLOC |
2216 FIEMAP_EXTENT_UNKNOWN);
2217 }
2218 up_read(&EXT4_I(inode)->i_data_sem);
2219
2220 if (unlikely(es.es_len == 0)) {
2221 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2222 err = -EIO;
2223 break;
2224 }
2225
2226 /*
2227 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2228 * we need to check next == EXT_MAX_BLOCKS because it is
2229 * possible that an extent is with unwritten and delayed
2230 * status due to when an extent is delayed allocated and
2231 * is allocated by fallocate status tree will track both of
2232 * them in a extent.
2233 *
2234 * So we could return a unwritten and delayed extent, and
2235 * its block is equal to 'next'.
2236 */
2237 if (next == next_del && next == EXT_MAX_BLOCKS) {
2238 flags |= FIEMAP_EXTENT_LAST;
2239 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2240 next != EXT_MAX_BLOCKS)) {
2241 EXT4_ERROR_INODE(inode,
2242 "next extent == %u, next "
2243 "delalloc extent = %u",
2244 next, next_del);
2245 err = -EIO;
2246 break;
2247 }
2248 }
2249
2250 if (exists) {
2251 err = fiemap_fill_next_extent(fieinfo,
2252 (__u64)es.es_lblk << blksize_bits,
2253 (__u64)es.es_pblk << blksize_bits,
2254 (__u64)es.es_len << blksize_bits,
2255 flags);
2256 if (err < 0)
2257 break;
2258 if (err == 1) {
2259 err = 0;
2260 break;
2261 }
2262 }
2263
2264 block = es.es_lblk + es.es_len;
2265 }
2266
2267 if (path) {
2268 ext4_ext_drop_refs(path);
2269 kfree(path);
2270 }
2271
2272 return err;
2273}
2274
2275/*
2276 * ext4_ext_put_gap_in_cache:
2277 * calculate boundaries of the gap that the requested block fits into
2278 * and cache this gap
2279 */
2280static void
2281ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2282 ext4_lblk_t block)
2283{
2284 int depth = ext_depth(inode);
2285 unsigned long len = 0;
2286 ext4_lblk_t lblock = 0;
2287 struct ext4_extent *ex;
2288
2289 ex = path[depth].p_ext;
2290 if (ex == NULL) {
2291 /*
2292 * there is no extent yet, so gap is [0;-] and we
2293 * don't cache it
2294 */
2295 ext_debug("cache gap(whole file):");
2296 } else if (block < le32_to_cpu(ex->ee_block)) {
2297 lblock = block;
2298 len = le32_to_cpu(ex->ee_block) - block;
2299 ext_debug("cache gap(before): %u [%u:%u]",
2300 block,
2301 le32_to_cpu(ex->ee_block),
2302 ext4_ext_get_actual_len(ex));
2303 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2304 ext4_es_insert_extent(inode, lblock, len, ~0,
2305 EXTENT_STATUS_HOLE);
2306 } else if (block >= le32_to_cpu(ex->ee_block)
2307 + ext4_ext_get_actual_len(ex)) {
2308 ext4_lblk_t next;
2309 lblock = le32_to_cpu(ex->ee_block)
2310 + ext4_ext_get_actual_len(ex);
2311
2312 next = ext4_ext_next_allocated_block(path);
2313 ext_debug("cache gap(after): [%u:%u] %u",
2314 le32_to_cpu(ex->ee_block),
2315 ext4_ext_get_actual_len(ex),
2316 block);
2317 BUG_ON(next == lblock);
2318 len = next - lblock;
2319 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2320 ext4_es_insert_extent(inode, lblock, len, ~0,
2321 EXTENT_STATUS_HOLE);
2322 } else {
2323 BUG();
2324 }
2325
2326 ext_debug(" -> %u:%lu\n", lblock, len);
2327}
2328
2329/*
2330 * ext4_ext_rm_idx:
2331 * removes index from the index block.
2332 */
2333static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2334 struct ext4_ext_path *path, int depth)
2335{
2336 int err;
2337 ext4_fsblk_t leaf;
2338
2339 /* free index block */
2340 depth--;
2341 path = path + depth;
2342 leaf = ext4_idx_pblock(path->p_idx);
2343 if (unlikely(path->p_hdr->eh_entries == 0)) {
2344 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2345 return -EIO;
2346 }
2347 err = ext4_ext_get_access(handle, inode, path);
2348 if (err)
2349 return err;
2350
2351 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2352 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2353 len *= sizeof(struct ext4_extent_idx);
2354 memmove(path->p_idx, path->p_idx + 1, len);
2355 }
2356
2357 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2358 err = ext4_ext_dirty(handle, inode, path);
2359 if (err)
2360 return err;
2361 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2362 trace_ext4_ext_rm_idx(inode, leaf);
2363
2364 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2365 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2366
2367 while (--depth >= 0) {
2368 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2369 break;
2370 path--;
2371 err = ext4_ext_get_access(handle, inode, path);
2372 if (err)
2373 break;
2374 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2375 err = ext4_ext_dirty(handle, inode, path);
2376 if (err)
2377 break;
2378 }
2379 return err;
2380}
2381
2382/*
2383 * ext4_ext_calc_credits_for_single_extent:
2384 * This routine returns max. credits that needed to insert an extent
2385 * to the extent tree.
2386 * When pass the actual path, the caller should calculate credits
2387 * under i_data_sem.
2388 */
2389int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2390 struct ext4_ext_path *path)
2391{
2392 if (path) {
2393 int depth = ext_depth(inode);
2394 int ret = 0;
2395
2396 /* probably there is space in leaf? */
2397 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2398 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2399
2400 /*
2401 * There are some space in the leaf tree, no
2402 * need to account for leaf block credit
2403 *
2404 * bitmaps and block group descriptor blocks
2405 * and other metadata blocks still need to be
2406 * accounted.
2407 */
2408 /* 1 bitmap, 1 block group descriptor */
2409 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2410 return ret;
2411 }
2412 }
2413
2414 return ext4_chunk_trans_blocks(inode, nrblocks);
2415}
2416
2417/*
2418 * How many index/leaf blocks need to change/allocate to add @extents extents?
2419 *
2420 * If we add a single extent, then in the worse case, each tree level
2421 * index/leaf need to be changed in case of the tree split.
2422 *
2423 * If more extents are inserted, they could cause the whole tree split more
2424 * than once, but this is really rare.
2425 */
2426int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2427{
2428 int index;
2429 int depth;
2430
2431 /* If we are converting the inline data, only one is needed here. */
2432 if (ext4_has_inline_data(inode))
2433 return 1;
2434
2435 depth = ext_depth(inode);
2436
2437 if (extents <= 1)
2438 index = depth * 2;
2439 else
2440 index = depth * 3;
2441
2442 return index;
2443}
2444
2445static inline int get_default_free_blocks_flags(struct inode *inode)
2446{
2447 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2448 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2449 else if (ext4_should_journal_data(inode))
2450 return EXT4_FREE_BLOCKS_FORGET;
2451 return 0;
2452}
2453
2454static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2455 struct ext4_extent *ex,
2456 long long *partial_cluster,
2457 ext4_lblk_t from, ext4_lblk_t to)
2458{
2459 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2460 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2461 ext4_fsblk_t pblk;
2462 int flags = get_default_free_blocks_flags(inode);
2463
2464 /*
2465 * For bigalloc file systems, we never free a partial cluster
2466 * at the beginning of the extent. Instead, we make a note
2467 * that we tried freeing the cluster, and check to see if we
2468 * need to free it on a subsequent call to ext4_remove_blocks,
2469 * or at the end of the ext4_truncate() operation.
2470 */
2471 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2472
2473 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2474 /*
2475 * If we have a partial cluster, and it's different from the
2476 * cluster of the last block, we need to explicitly free the
2477 * partial cluster here.
2478 */
2479 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2480 if ((*partial_cluster > 0) &&
2481 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2482 ext4_free_blocks(handle, inode, NULL,
2483 EXT4_C2B(sbi, *partial_cluster),
2484 sbi->s_cluster_ratio, flags);
2485 *partial_cluster = 0;
2486 }
2487
2488#ifdef EXTENTS_STATS
2489 {
2490 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2491 spin_lock(&sbi->s_ext_stats_lock);
2492 sbi->s_ext_blocks += ee_len;
2493 sbi->s_ext_extents++;
2494 if (ee_len < sbi->s_ext_min)
2495 sbi->s_ext_min = ee_len;
2496 if (ee_len > sbi->s_ext_max)
2497 sbi->s_ext_max = ee_len;
2498 if (ext_depth(inode) > sbi->s_depth_max)
2499 sbi->s_depth_max = ext_depth(inode);
2500 spin_unlock(&sbi->s_ext_stats_lock);
2501 }
2502#endif
2503 if (from >= le32_to_cpu(ex->ee_block)
2504 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2505 /* tail removal */
2506 ext4_lblk_t num;
2507 unsigned int unaligned;
2508
2509 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2510 pblk = ext4_ext_pblock(ex) + ee_len - num;
2511 /*
2512 * Usually we want to free partial cluster at the end of the
2513 * extent, except for the situation when the cluster is still
2514 * used by any other extent (partial_cluster is negative).
2515 */
2516 if (*partial_cluster < 0 &&
2517 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2518 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2519
2520 ext_debug("free last %u blocks starting %llu partial %lld\n",
2521 num, pblk, *partial_cluster);
2522 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2523 /*
2524 * If the block range to be freed didn't start at the
2525 * beginning of a cluster, and we removed the entire
2526 * extent and the cluster is not used by any other extent,
2527 * save the partial cluster here, since we might need to
2528 * delete if we determine that the truncate operation has
2529 * removed all of the blocks in the cluster.
2530 *
2531 * On the other hand, if we did not manage to free the whole
2532 * extent, we have to mark the cluster as used (store negative
2533 * cluster number in partial_cluster).
2534 */
2535 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2536 if (unaligned && (ee_len == num) &&
2537 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2538 *partial_cluster = EXT4_B2C(sbi, pblk);
2539 else if (unaligned)
2540 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2541 else if (*partial_cluster > 0)
2542 *partial_cluster = 0;
2543 } else
2544 ext4_error(sbi->s_sb, "strange request: removal(2) "
2545 "%u-%u from %u:%u\n",
2546 from, to, le32_to_cpu(ex->ee_block), ee_len);
2547 return 0;
2548}
2549
2550
2551/*
2552 * ext4_ext_rm_leaf() Removes the extents associated with the
2553 * blocks appearing between "start" and "end", and splits the extents
2554 * if "start" and "end" appear in the same extent
2555 *
2556 * @handle: The journal handle
2557 * @inode: The files inode
2558 * @path: The path to the leaf
2559 * @partial_cluster: The cluster which we'll have to free if all extents
2560 * has been released from it. It gets negative in case
2561 * that the cluster is still used.
2562 * @start: The first block to remove
2563 * @end: The last block to remove
2564 */
2565static int
2566ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2567 struct ext4_ext_path *path,
2568 long long *partial_cluster,
2569 ext4_lblk_t start, ext4_lblk_t end)
2570{
2571 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2572 int err = 0, correct_index = 0;
2573 int depth = ext_depth(inode), credits;
2574 struct ext4_extent_header *eh;
2575 ext4_lblk_t a, b;
2576 unsigned num;
2577 ext4_lblk_t ex_ee_block;
2578 unsigned short ex_ee_len;
2579 unsigned uninitialized = 0;
2580 struct ext4_extent *ex;
2581 ext4_fsblk_t pblk;
2582
2583 /* the header must be checked already in ext4_ext_remove_space() */
2584 ext_debug("truncate since %u in leaf to %u\n", start, end);
2585 if (!path[depth].p_hdr)
2586 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2587 eh = path[depth].p_hdr;
2588 if (unlikely(path[depth].p_hdr == NULL)) {
2589 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2590 return -EIO;
2591 }
2592 /* find where to start removing */
2593 ex = path[depth].p_ext;
2594 if (!ex)
2595 ex = EXT_LAST_EXTENT(eh);
2596
2597 ex_ee_block = le32_to_cpu(ex->ee_block);
2598 ex_ee_len = ext4_ext_get_actual_len(ex);
2599
2600 /*
2601 * If we're starting with an extent other than the last one in the
2602 * node, we need to see if it shares a cluster with the extent to
2603 * the right (towards the end of the file). If its leftmost cluster
2604 * is this extent's rightmost cluster and it is not cluster aligned,
2605 * we'll mark it as a partial that is not to be deallocated.
2606 */
2607
2608 if (ex != EXT_LAST_EXTENT(eh)) {
2609 ext4_fsblk_t current_pblk, right_pblk;
2610 long long current_cluster, right_cluster;
2611
2612 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2613 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2614 right_pblk = ext4_ext_pblock(ex + 1);
2615 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2616 if (current_cluster == right_cluster &&
2617 EXT4_PBLK_COFF(sbi, right_pblk))
2618 *partial_cluster = -right_cluster;
2619 }
2620
2621 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2622
2623 while (ex >= EXT_FIRST_EXTENT(eh) &&
2624 ex_ee_block + ex_ee_len > start) {
2625
2626 if (ext4_ext_is_uninitialized(ex))
2627 uninitialized = 1;
2628 else
2629 uninitialized = 0;
2630
2631 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2632 uninitialized, ex_ee_len);
2633 path[depth].p_ext = ex;
2634
2635 a = ex_ee_block > start ? ex_ee_block : start;
2636 b = ex_ee_block+ex_ee_len - 1 < end ?
2637 ex_ee_block+ex_ee_len - 1 : end;
2638
2639 ext_debug(" border %u:%u\n", a, b);
2640
2641 /* If this extent is beyond the end of the hole, skip it */
2642 if (end < ex_ee_block) {
2643 /*
2644 * We're going to skip this extent and move to another,
2645 * so if this extent is not cluster aligned we have
2646 * to mark the current cluster as used to avoid
2647 * accidentally freeing it later on
2648 */
2649 pblk = ext4_ext_pblock(ex);
2650 if (EXT4_PBLK_COFF(sbi, pblk))
2651 *partial_cluster =
2652 -((long long)EXT4_B2C(sbi, pblk));
2653 ex--;
2654 ex_ee_block = le32_to_cpu(ex->ee_block);
2655 ex_ee_len = ext4_ext_get_actual_len(ex);
2656 continue;
2657 } else if (b != ex_ee_block + ex_ee_len - 1) {
2658 EXT4_ERROR_INODE(inode,
2659 "can not handle truncate %u:%u "
2660 "on extent %u:%u",
2661 start, end, ex_ee_block,
2662 ex_ee_block + ex_ee_len - 1);
2663 err = -EIO;
2664 goto out;
2665 } else if (a != ex_ee_block) {
2666 /* remove tail of the extent */
2667 num = a - ex_ee_block;
2668 } else {
2669 /* remove whole extent: excellent! */
2670 num = 0;
2671 }
2672 /*
2673 * 3 for leaf, sb, and inode plus 2 (bmap and group
2674 * descriptor) for each block group; assume two block
2675 * groups plus ex_ee_len/blocks_per_block_group for
2676 * the worst case
2677 */
2678 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2679 if (ex == EXT_FIRST_EXTENT(eh)) {
2680 correct_index = 1;
2681 credits += (ext_depth(inode)) + 1;
2682 }
2683 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2684
2685 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2686 if (err)
2687 goto out;
2688
2689 err = ext4_ext_get_access(handle, inode, path + depth);
2690 if (err)
2691 goto out;
2692
2693 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2694 a, b);
2695 if (err)
2696 goto out;
2697
2698 if (num == 0)
2699 /* this extent is removed; mark slot entirely unused */
2700 ext4_ext_store_pblock(ex, 0);
2701
2702 ex->ee_len = cpu_to_le16(num);
2703 /*
2704 * Do not mark uninitialized if all the blocks in the
2705 * extent have been removed.
2706 */
2707 if (uninitialized && num)
2708 ext4_ext_mark_uninitialized(ex);
2709 /*
2710 * If the extent was completely released,
2711 * we need to remove it from the leaf
2712 */
2713 if (num == 0) {
2714 if (end != EXT_MAX_BLOCKS - 1) {
2715 /*
2716 * For hole punching, we need to scoot all the
2717 * extents up when an extent is removed so that
2718 * we dont have blank extents in the middle
2719 */
2720 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2721 sizeof(struct ext4_extent));
2722
2723 /* Now get rid of the one at the end */
2724 memset(EXT_LAST_EXTENT(eh), 0,
2725 sizeof(struct ext4_extent));
2726 }
2727 le16_add_cpu(&eh->eh_entries, -1);
2728 } else if (*partial_cluster > 0)
2729 *partial_cluster = 0;
2730
2731 err = ext4_ext_dirty(handle, inode, path + depth);
2732 if (err)
2733 goto out;
2734
2735 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2736 ext4_ext_pblock(ex));
2737 ex--;
2738 ex_ee_block = le32_to_cpu(ex->ee_block);
2739 ex_ee_len = ext4_ext_get_actual_len(ex);
2740 }
2741
2742 if (correct_index && eh->eh_entries)
2743 err = ext4_ext_correct_indexes(handle, inode, path);
2744
2745 /*
2746 * If there's a partial cluster and at least one extent remains in
2747 * the leaf, free the partial cluster if it isn't shared with the
2748 * current extent. If there's a partial cluster and no extents
2749 * remain in the leaf, it can't be freed here. It can only be
2750 * freed when it's possible to determine if it's not shared with
2751 * any other extent - when the next leaf is processed or when space
2752 * removal is complete.
2753 */
2754 if (*partial_cluster > 0 && eh->eh_entries &&
2755 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2756 *partial_cluster)) {
2757 int flags = get_default_free_blocks_flags(inode);
2758
2759 ext4_free_blocks(handle, inode, NULL,
2760 EXT4_C2B(sbi, *partial_cluster),
2761 sbi->s_cluster_ratio, flags);
2762 *partial_cluster = 0;
2763 }
2764
2765 /* if this leaf is free, then we should
2766 * remove it from index block above */
2767 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2768 err = ext4_ext_rm_idx(handle, inode, path, depth);
2769
2770out:
2771 return err;
2772}
2773
2774/*
2775 * ext4_ext_more_to_rm:
2776 * returns 1 if current index has to be freed (even partial)
2777 */
2778static int
2779ext4_ext_more_to_rm(struct ext4_ext_path *path)
2780{
2781 BUG_ON(path->p_idx == NULL);
2782
2783 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2784 return 0;
2785
2786 /*
2787 * if truncate on deeper level happened, it wasn't partial,
2788 * so we have to consider current index for truncation
2789 */
2790 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2791 return 0;
2792 return 1;
2793}
2794
2795int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2796 ext4_lblk_t end)
2797{
2798 struct super_block *sb = inode->i_sb;
2799 int depth = ext_depth(inode);
2800 struct ext4_ext_path *path = NULL;
2801 long long partial_cluster = 0;
2802 handle_t *handle;
2803 int i = 0, err = 0;
2804
2805 ext_debug("truncate since %u to %u\n", start, end);
2806
2807 /* probably first extent we're gonna free will be last in block */
2808 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2809 if (IS_ERR(handle))
2810 return PTR_ERR(handle);
2811
2812again:
2813 trace_ext4_ext_remove_space(inode, start, end, depth);
2814
2815 /*
2816 * Check if we are removing extents inside the extent tree. If that
2817 * is the case, we are going to punch a hole inside the extent tree
2818 * so we have to check whether we need to split the extent covering
2819 * the last block to remove so we can easily remove the part of it
2820 * in ext4_ext_rm_leaf().
2821 */
2822 if (end < EXT_MAX_BLOCKS - 1) {
2823 struct ext4_extent *ex;
2824 ext4_lblk_t ee_block;
2825
2826 /* find extent for this block */
2827 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2828 if (IS_ERR(path)) {
2829 ext4_journal_stop(handle);
2830 return PTR_ERR(path);
2831 }
2832 depth = ext_depth(inode);
2833 /* Leaf not may not exist only if inode has no blocks at all */
2834 ex = path[depth].p_ext;
2835 if (!ex) {
2836 if (depth) {
2837 EXT4_ERROR_INODE(inode,
2838 "path[%d].p_hdr == NULL",
2839 depth);
2840 err = -EIO;
2841 }
2842 goto out;
2843 }
2844
2845 ee_block = le32_to_cpu(ex->ee_block);
2846
2847 /*
2848 * See if the last block is inside the extent, if so split
2849 * the extent at 'end' block so we can easily remove the
2850 * tail of the first part of the split extent in
2851 * ext4_ext_rm_leaf().
2852 */
2853 if (end >= ee_block &&
2854 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2855 int split_flag = 0;
2856
2857 if (ext4_ext_is_uninitialized(ex))
2858 split_flag = EXT4_EXT_MARK_UNINIT1 |
2859 EXT4_EXT_MARK_UNINIT2;
2860
2861 /*
2862 * Split the extent in two so that 'end' is the last
2863 * block in the first new extent. Also we should not
2864 * fail removing space due to ENOSPC so try to use
2865 * reserved block if that happens.
2866 */
2867 err = ext4_split_extent_at(handle, inode, path,
2868 end + 1, split_flag,
2869 EXT4_EX_NOCACHE |
2870 EXT4_GET_BLOCKS_PRE_IO |
2871 EXT4_GET_BLOCKS_METADATA_NOFAIL);
2872
2873 if (err < 0)
2874 goto out;
2875 }
2876 }
2877 /*
2878 * We start scanning from right side, freeing all the blocks
2879 * after i_size and walking into the tree depth-wise.
2880 */
2881 depth = ext_depth(inode);
2882 if (path) {
2883 int k = i = depth;
2884 while (--k > 0)
2885 path[k].p_block =
2886 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2887 } else {
2888 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2889 GFP_NOFS);
2890 if (path == NULL) {
2891 ext4_journal_stop(handle);
2892 return -ENOMEM;
2893 }
2894 path[0].p_depth = depth;
2895 path[0].p_hdr = ext_inode_hdr(inode);
2896 i = 0;
2897
2898 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2899 err = -EIO;
2900 goto out;
2901 }
2902 }
2903 err = 0;
2904
2905 while (i >= 0 && err == 0) {
2906 if (i == depth) {
2907 /* this is leaf block */
2908 err = ext4_ext_rm_leaf(handle, inode, path,
2909 &partial_cluster, start,
2910 end);
2911 /* root level has p_bh == NULL, brelse() eats this */
2912 brelse(path[i].p_bh);
2913 path[i].p_bh = NULL;
2914 i--;
2915 continue;
2916 }
2917
2918 /* this is index block */
2919 if (!path[i].p_hdr) {
2920 ext_debug("initialize header\n");
2921 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2922 }
2923
2924 if (!path[i].p_idx) {
2925 /* this level hasn't been touched yet */
2926 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2927 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2928 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2929 path[i].p_hdr,
2930 le16_to_cpu(path[i].p_hdr->eh_entries));
2931 } else {
2932 /* we were already here, see at next index */
2933 path[i].p_idx--;
2934 }
2935
2936 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2937 i, EXT_FIRST_INDEX(path[i].p_hdr),
2938 path[i].p_idx);
2939 if (ext4_ext_more_to_rm(path + i)) {
2940 struct buffer_head *bh;
2941 /* go to the next level */
2942 ext_debug("move to level %d (block %llu)\n",
2943 i + 1, ext4_idx_pblock(path[i].p_idx));
2944 memset(path + i + 1, 0, sizeof(*path));
2945 bh = read_extent_tree_block(inode,
2946 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2947 EXT4_EX_NOCACHE);
2948 if (IS_ERR(bh)) {
2949 /* should we reset i_size? */
2950 err = PTR_ERR(bh);
2951 break;
2952 }
2953 /* Yield here to deal with large extent trees.
2954 * Should be a no-op if we did IO above. */
2955 cond_resched();
2956 if (WARN_ON(i + 1 > depth)) {
2957 err = -EIO;
2958 break;
2959 }
2960 path[i + 1].p_bh = bh;
2961
2962 /* save actual number of indexes since this
2963 * number is changed at the next iteration */
2964 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2965 i++;
2966 } else {
2967 /* we finished processing this index, go up */
2968 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2969 /* index is empty, remove it;
2970 * handle must be already prepared by the
2971 * truncatei_leaf() */
2972 err = ext4_ext_rm_idx(handle, inode, path, i);
2973 }
2974 /* root level has p_bh == NULL, brelse() eats this */
2975 brelse(path[i].p_bh);
2976 path[i].p_bh = NULL;
2977 i--;
2978 ext_debug("return to level %d\n", i);
2979 }
2980 }
2981
2982 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2983 partial_cluster, path->p_hdr->eh_entries);
2984
2985 /* If we still have something in the partial cluster and we have removed
2986 * even the first extent, then we should free the blocks in the partial
2987 * cluster as well. */
2988 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2989 int flags = get_default_free_blocks_flags(inode);
2990
2991 ext4_free_blocks(handle, inode, NULL,
2992 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2993 EXT4_SB(sb)->s_cluster_ratio, flags);
2994 partial_cluster = 0;
2995 }
2996
2997 /* TODO: flexible tree reduction should be here */
2998 if (path->p_hdr->eh_entries == 0) {
2999 /*
3000 * truncate to zero freed all the tree,
3001 * so we need to correct eh_depth
3002 */
3003 err = ext4_ext_get_access(handle, inode, path);
3004 if (err == 0) {
3005 ext_inode_hdr(inode)->eh_depth = 0;
3006 ext_inode_hdr(inode)->eh_max =
3007 cpu_to_le16(ext4_ext_space_root(inode, 0));
3008 err = ext4_ext_dirty(handle, inode, path);
3009 }
3010 }
3011out:
3012 ext4_ext_drop_refs(path);
3013 kfree(path);
3014 if (err == -EAGAIN) {
3015 path = NULL;
3016 goto again;
3017 }
3018 ext4_journal_stop(handle);
3019
3020 return err;
3021}
3022
3023/*
3024 * called at mount time
3025 */
3026void ext4_ext_init(struct super_block *sb)
3027{
3028 /*
3029 * possible initialization would be here
3030 */
3031
3032 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3033#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3034 printk(KERN_INFO "EXT4-fs: file extents enabled"
3035#ifdef AGGRESSIVE_TEST
3036 ", aggressive tests"
3037#endif
3038#ifdef CHECK_BINSEARCH
3039 ", check binsearch"
3040#endif
3041#ifdef EXTENTS_STATS
3042 ", stats"
3043#endif
3044 "\n");
3045#endif
3046#ifdef EXTENTS_STATS
3047 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3048 EXT4_SB(sb)->s_ext_min = 1 << 30;
3049 EXT4_SB(sb)->s_ext_max = 0;
3050#endif
3051 }
3052}
3053
3054/*
3055 * called at umount time
3056 */
3057void ext4_ext_release(struct super_block *sb)
3058{
3059 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3060 return;
3061
3062#ifdef EXTENTS_STATS
3063 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3064 struct ext4_sb_info *sbi = EXT4_SB(sb);
3065 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3066 sbi->s_ext_blocks, sbi->s_ext_extents,
3067 sbi->s_ext_blocks / sbi->s_ext_extents);
3068 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3069 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3070 }
3071#endif
3072}
3073
3074static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3075{
3076 ext4_lblk_t ee_block;
3077 ext4_fsblk_t ee_pblock;
3078 unsigned int ee_len;
3079
3080 ee_block = le32_to_cpu(ex->ee_block);
3081 ee_len = ext4_ext_get_actual_len(ex);
3082 ee_pblock = ext4_ext_pblock(ex);
3083
3084 if (ee_len == 0)
3085 return 0;
3086
3087 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3088 EXTENT_STATUS_WRITTEN);
3089}
3090
3091/* FIXME!! we need to try to merge to left or right after zero-out */
3092static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3093{
3094 ext4_fsblk_t ee_pblock;
3095 unsigned int ee_len;
3096 int ret;
3097
3098 ee_len = ext4_ext_get_actual_len(ex);
3099 ee_pblock = ext4_ext_pblock(ex);
3100
3101 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3102 if (ret > 0)
3103 ret = 0;
3104
3105 return ret;
3106}
3107
3108/*
3109 * ext4_split_extent_at() splits an extent at given block.
3110 *
3111 * @handle: the journal handle
3112 * @inode: the file inode
3113 * @path: the path to the extent
3114 * @split: the logical block where the extent is splitted.
3115 * @split_flags: indicates if the extent could be zeroout if split fails, and
3116 * the states(init or uninit) of new extents.
3117 * @flags: flags used to insert new extent to extent tree.
3118 *
3119 *
3120 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3121 * of which are deterimined by split_flag.
3122 *
3123 * There are two cases:
3124 * a> the extent are splitted into two extent.
3125 * b> split is not needed, and just mark the extent.
3126 *
3127 * return 0 on success.
3128 */
3129static int ext4_split_extent_at(handle_t *handle,
3130 struct inode *inode,
3131 struct ext4_ext_path *path,
3132 ext4_lblk_t split,
3133 int split_flag,
3134 int flags)
3135{
3136 ext4_fsblk_t newblock;
3137 ext4_lblk_t ee_block;
3138 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3139 struct ext4_extent *ex2 = NULL;
3140 unsigned int ee_len, depth;
3141 int err = 0;
3142
3143 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3144 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3145
3146 ext_debug("ext4_split_extents_at: inode %lu, logical"
3147 "block %llu\n", inode->i_ino, (unsigned long long)split);
3148
3149 ext4_ext_show_leaf(inode, path);
3150
3151 depth = ext_depth(inode);
3152 ex = path[depth].p_ext;
3153 ee_block = le32_to_cpu(ex->ee_block);
3154 ee_len = ext4_ext_get_actual_len(ex);
3155 newblock = split - ee_block + ext4_ext_pblock(ex);
3156
3157 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3158 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3159 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3160 EXT4_EXT_MARK_UNINIT1 |
3161 EXT4_EXT_MARK_UNINIT2));
3162
3163 err = ext4_ext_get_access(handle, inode, path + depth);
3164 if (err)
3165 goto out;
3166
3167 if (split == ee_block) {
3168 /*
3169 * case b: block @split is the block that the extent begins with
3170 * then we just change the state of the extent, and splitting
3171 * is not needed.
3172 */
3173 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3174 ext4_ext_mark_uninitialized(ex);
3175 else
3176 ext4_ext_mark_initialized(ex);
3177
3178 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3179 ext4_ext_try_to_merge(handle, inode, path, ex);
3180
3181 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3182 goto out;
3183 }
3184
3185 /* case a */
3186 memcpy(&orig_ex, ex, sizeof(orig_ex));
3187 ex->ee_len = cpu_to_le16(split - ee_block);
3188 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3189 ext4_ext_mark_uninitialized(ex);
3190
3191 /*
3192 * path may lead to new leaf, not to original leaf any more
3193 * after ext4_ext_insert_extent() returns,
3194 */
3195 err = ext4_ext_dirty(handle, inode, path + depth);
3196 if (err)
3197 goto fix_extent_len;
3198
3199 ex2 = &newex;
3200 ex2->ee_block = cpu_to_le32(split);
3201 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3202 ext4_ext_store_pblock(ex2, newblock);
3203 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3204 ext4_ext_mark_uninitialized(ex2);
3205
3206 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3207 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3208 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3209 if (split_flag & EXT4_EXT_DATA_VALID1) {
3210 err = ext4_ext_zeroout(inode, ex2);
3211 zero_ex.ee_block = ex2->ee_block;
3212 zero_ex.ee_len = cpu_to_le16(
3213 ext4_ext_get_actual_len(ex2));
3214 ext4_ext_store_pblock(&zero_ex,
3215 ext4_ext_pblock(ex2));
3216 } else {
3217 err = ext4_ext_zeroout(inode, ex);
3218 zero_ex.ee_block = ex->ee_block;
3219 zero_ex.ee_len = cpu_to_le16(
3220 ext4_ext_get_actual_len(ex));
3221 ext4_ext_store_pblock(&zero_ex,
3222 ext4_ext_pblock(ex));
3223 }
3224 } else {
3225 err = ext4_ext_zeroout(inode, &orig_ex);
3226 zero_ex.ee_block = orig_ex.ee_block;
3227 zero_ex.ee_len = cpu_to_le16(
3228 ext4_ext_get_actual_len(&orig_ex));
3229 ext4_ext_store_pblock(&zero_ex,
3230 ext4_ext_pblock(&orig_ex));
3231 }
3232
3233 if (err)
3234 goto fix_extent_len;
3235 /* update the extent length and mark as initialized */
3236 ex->ee_len = cpu_to_le16(ee_len);
3237 ext4_ext_try_to_merge(handle, inode, path, ex);
3238 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3239 if (err)
3240 goto fix_extent_len;
3241
3242 /* update extent status tree */
3243 err = ext4_zeroout_es(inode, &zero_ex);
3244
3245 goto out;
3246 } else if (err)
3247 goto fix_extent_len;
3248
3249out:
3250 ext4_ext_show_leaf(inode, path);
3251 return err;
3252
3253fix_extent_len:
3254 ex->ee_len = orig_ex.ee_len;
3255 ext4_ext_dirty(handle, inode, path + depth);
3256 return err;
3257}
3258
3259/*
3260 * ext4_split_extents() splits an extent and mark extent which is covered
3261 * by @map as split_flags indicates
3262 *
3263 * It may result in splitting the extent into multiple extents (up to three)
3264 * There are three possibilities:
3265 * a> There is no split required
3266 * b> Splits in two extents: Split is happening at either end of the extent
3267 * c> Splits in three extents: Somone is splitting in middle of the extent
3268 *
3269 */
3270static int ext4_split_extent(handle_t *handle,
3271 struct inode *inode,
3272 struct ext4_ext_path *path,
3273 struct ext4_map_blocks *map,
3274 int split_flag,
3275 int flags)
3276{
3277 ext4_lblk_t ee_block;
3278 struct ext4_extent *ex;
3279 unsigned int ee_len, depth;
3280 int err = 0;
3281 int uninitialized;
3282 int split_flag1, flags1;
3283 int allocated = map->m_len;
3284
3285 depth = ext_depth(inode);
3286 ex = path[depth].p_ext;
3287 ee_block = le32_to_cpu(ex->ee_block);
3288 ee_len = ext4_ext_get_actual_len(ex);
3289 uninitialized = ext4_ext_is_uninitialized(ex);
3290
3291 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3292 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3293 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3294 if (uninitialized)
3295 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3296 EXT4_EXT_MARK_UNINIT2;
3297 if (split_flag & EXT4_EXT_DATA_VALID2)
3298 split_flag1 |= EXT4_EXT_DATA_VALID1;
3299 err = ext4_split_extent_at(handle, inode, path,
3300 map->m_lblk + map->m_len, split_flag1, flags1);
3301 if (err)
3302 goto out;
3303 } else {
3304 allocated = ee_len - (map->m_lblk - ee_block);
3305 }
3306 /*
3307 * Update path is required because previous ext4_split_extent_at() may
3308 * result in split of original leaf or extent zeroout.
3309 */
3310 ext4_ext_drop_refs(path);
3311 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3312 if (IS_ERR(path))
3313 return PTR_ERR(path);
3314 depth = ext_depth(inode);
3315 ex = path[depth].p_ext;
3316 if (!ex) {
3317 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3318 (unsigned long) map->m_lblk);
3319 return -EIO;
3320 }
3321 uninitialized = ext4_ext_is_uninitialized(ex);
3322 split_flag1 = 0;
3323
3324 if (map->m_lblk >= ee_block) {
3325 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3326 if (uninitialized) {
3327 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3328 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3329 EXT4_EXT_MARK_UNINIT2);
3330 }
3331 err = ext4_split_extent_at(handle, inode, path,
3332 map->m_lblk, split_flag1, flags);
3333 if (err)
3334 goto out;
3335 }
3336
3337 ext4_ext_show_leaf(inode, path);
3338out:
3339 return err ? err : allocated;
3340}
3341
3342/*
3343 * This function is called by ext4_ext_map_blocks() if someone tries to write
3344 * to an uninitialized extent. It may result in splitting the uninitialized
3345 * extent into multiple extents (up to three - one initialized and two
3346 * uninitialized).
3347 * There are three possibilities:
3348 * a> There is no split required: Entire extent should be initialized
3349 * b> Splits in two extents: Write is happening at either end of the extent
3350 * c> Splits in three extents: Somone is writing in middle of the extent
3351 *
3352 * Pre-conditions:
3353 * - The extent pointed to by 'path' is uninitialized.
3354 * - The extent pointed to by 'path' contains a superset
3355 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3356 *
3357 * Post-conditions on success:
3358 * - the returned value is the number of blocks beyond map->l_lblk
3359 * that are allocated and initialized.
3360 * It is guaranteed to be >= map->m_len.
3361 */
3362static int ext4_ext_convert_to_initialized(handle_t *handle,
3363 struct inode *inode,
3364 struct ext4_map_blocks *map,
3365 struct ext4_ext_path *path,
3366 int flags)
3367{
3368 struct ext4_sb_info *sbi;
3369 struct ext4_extent_header *eh;
3370 struct ext4_map_blocks split_map;
3371 struct ext4_extent zero_ex;
3372 struct ext4_extent *ex, *abut_ex;
3373 ext4_lblk_t ee_block, eof_block;
3374 unsigned int ee_len, depth, map_len = map->m_len;
3375 int allocated = 0, max_zeroout = 0;
3376 int err = 0;
3377 int split_flag = 0;
3378
3379 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3380 "block %llu, max_blocks %u\n", inode->i_ino,
3381 (unsigned long long)map->m_lblk, map_len);
3382
3383 sbi = EXT4_SB(inode->i_sb);
3384 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3385 inode->i_sb->s_blocksize_bits;
3386 if (eof_block < map->m_lblk + map_len)
3387 eof_block = map->m_lblk + map_len;
3388
3389 depth = ext_depth(inode);
3390 eh = path[depth].p_hdr;
3391 ex = path[depth].p_ext;
3392 ee_block = le32_to_cpu(ex->ee_block);
3393 ee_len = ext4_ext_get_actual_len(ex);
3394 zero_ex.ee_len = 0;
3395
3396 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3397
3398 /* Pre-conditions */
3399 BUG_ON(!ext4_ext_is_uninitialized(ex));
3400 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3401
3402 /*
3403 * Attempt to transfer newly initialized blocks from the currently
3404 * uninitialized extent to its neighbor. This is much cheaper
3405 * than an insertion followed by a merge as those involve costly
3406 * memmove() calls. Transferring to the left is the common case in
3407 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3408 * followed by append writes.
3409 *
3410 * Limitations of the current logic:
3411 * - L1: we do not deal with writes covering the whole extent.
3412 * This would require removing the extent if the transfer
3413 * is possible.
3414 * - L2: we only attempt to merge with an extent stored in the
3415 * same extent tree node.
3416 */
3417 if ((map->m_lblk == ee_block) &&
3418 /* See if we can merge left */
3419 (map_len < ee_len) && /*L1*/
3420 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3421 ext4_lblk_t prev_lblk;
3422 ext4_fsblk_t prev_pblk, ee_pblk;
3423 unsigned int prev_len;
3424
3425 abut_ex = ex - 1;
3426 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3427 prev_len = ext4_ext_get_actual_len(abut_ex);
3428 prev_pblk = ext4_ext_pblock(abut_ex);
3429 ee_pblk = ext4_ext_pblock(ex);
3430
3431 /*
3432 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3433 * upon those conditions:
3434 * - C1: abut_ex is initialized,
3435 * - C2: abut_ex is logically abutting ex,
3436 * - C3: abut_ex is physically abutting ex,
3437 * - C4: abut_ex can receive the additional blocks without
3438 * overflowing the (initialized) length limit.
3439 */
3440 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3441 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3442 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3443 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3444 err = ext4_ext_get_access(handle, inode, path + depth);
3445 if (err)
3446 goto out;
3447
3448 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3449 map, ex, abut_ex);
3450
3451 /* Shift the start of ex by 'map_len' blocks */
3452 ex->ee_block = cpu_to_le32(ee_block + map_len);
3453 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3454 ex->ee_len = cpu_to_le16(ee_len - map_len);
3455 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3456
3457 /* Extend abut_ex by 'map_len' blocks */
3458 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3459
3460 /* Result: number of initialized blocks past m_lblk */
3461 allocated = map_len;
3462 }
3463 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3464 (map_len < ee_len) && /*L1*/
3465 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3466 /* See if we can merge right */
3467 ext4_lblk_t next_lblk;
3468 ext4_fsblk_t next_pblk, ee_pblk;
3469 unsigned int next_len;
3470
3471 abut_ex = ex + 1;
3472 next_lblk = le32_to_cpu(abut_ex->ee_block);
3473 next_len = ext4_ext_get_actual_len(abut_ex);
3474 next_pblk = ext4_ext_pblock(abut_ex);
3475 ee_pblk = ext4_ext_pblock(ex);
3476
3477 /*
3478 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3479 * upon those conditions:
3480 * - C1: abut_ex is initialized,
3481 * - C2: abut_ex is logically abutting ex,
3482 * - C3: abut_ex is physically abutting ex,
3483 * - C4: abut_ex can receive the additional blocks without
3484 * overflowing the (initialized) length limit.
3485 */
3486 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3487 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3488 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3489 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3490 err = ext4_ext_get_access(handle, inode, path + depth);
3491 if (err)
3492 goto out;
3493
3494 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3495 map, ex, abut_ex);
3496
3497 /* Shift the start of abut_ex by 'map_len' blocks */
3498 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3499 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3500 ex->ee_len = cpu_to_le16(ee_len - map_len);
3501 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3502
3503 /* Extend abut_ex by 'map_len' blocks */
3504 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3505
3506 /* Result: number of initialized blocks past m_lblk */
3507 allocated = map_len;
3508 }
3509 }
3510 if (allocated) {
3511 /* Mark the block containing both extents as dirty */
3512 ext4_ext_dirty(handle, inode, path + depth);
3513
3514 /* Update path to point to the right extent */
3515 path[depth].p_ext = abut_ex;
3516 goto out;
3517 } else
3518 allocated = ee_len - (map->m_lblk - ee_block);
3519
3520 WARN_ON(map->m_lblk < ee_block);
3521 /*
3522 * It is safe to convert extent to initialized via explicit
3523 * zeroout only if extent is fully inside i_size or new_size.
3524 */
3525 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3526
3527 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3528 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3529 (inode->i_sb->s_blocksize_bits - 10);
3530
3531 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3532 if (max_zeroout && (ee_len <= max_zeroout)) {
3533 err = ext4_ext_zeroout(inode, ex);
3534 if (err)
3535 goto out;
3536 zero_ex.ee_block = ex->ee_block;
3537 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3538 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3539
3540 err = ext4_ext_get_access(handle, inode, path + depth);
3541 if (err)
3542 goto out;
3543 ext4_ext_mark_initialized(ex);
3544 ext4_ext_try_to_merge(handle, inode, path, ex);
3545 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3546 goto out;
3547 }
3548
3549 /*
3550 * four cases:
3551 * 1. split the extent into three extents.
3552 * 2. split the extent into two extents, zeroout the first half.
3553 * 3. split the extent into two extents, zeroout the second half.
3554 * 4. split the extent into two extents with out zeroout.
3555 */
3556 split_map.m_lblk = map->m_lblk;
3557 split_map.m_len = map->m_len;
3558
3559 if (max_zeroout && (allocated > map->m_len)) {
3560 if (allocated <= max_zeroout) {
3561 /* case 3 */
3562 zero_ex.ee_block =
3563 cpu_to_le32(map->m_lblk);
3564 zero_ex.ee_len = cpu_to_le16(allocated);
3565 ext4_ext_store_pblock(&zero_ex,
3566 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3567 err = ext4_ext_zeroout(inode, &zero_ex);
3568 if (err)
3569 goto out;
3570 split_map.m_lblk = map->m_lblk;
3571 split_map.m_len = allocated;
3572 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3573 /* case 2 */
3574 if (map->m_lblk != ee_block) {
3575 zero_ex.ee_block = ex->ee_block;
3576 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3577 ee_block);
3578 ext4_ext_store_pblock(&zero_ex,
3579 ext4_ext_pblock(ex));
3580 err = ext4_ext_zeroout(inode, &zero_ex);
3581 if (err)
3582 goto out;
3583 }
3584
3585 split_map.m_lblk = ee_block;
3586 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3587 allocated = map->m_len;
3588 }
3589 }
3590
3591 allocated = ext4_split_extent(handle, inode, path,
3592 &split_map, split_flag, flags);
3593 if (allocated < 0)
3594 err = allocated;
3595
3596out:
3597 /* If we have gotten a failure, don't zero out status tree */
3598 if (!err)
3599 err = ext4_zeroout_es(inode, &zero_ex);
3600 return err ? err : allocated;
3601}
3602
3603/*
3604 * This function is called by ext4_ext_map_blocks() from
3605 * ext4_get_blocks_dio_write() when DIO to write
3606 * to an uninitialized extent.
3607 *
3608 * Writing to an uninitialized extent may result in splitting the uninitialized
3609 * extent into multiple initialized/uninitialized extents (up to three)
3610 * There are three possibilities:
3611 * a> There is no split required: Entire extent should be uninitialized
3612 * b> Splits in two extents: Write is happening at either end of the extent
3613 * c> Splits in three extents: Somone is writing in middle of the extent
3614 *
3615 * This works the same way in the case of initialized -> unwritten conversion.
3616 *
3617 * One of more index blocks maybe needed if the extent tree grow after
3618 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3619 * complete, we need to split the uninitialized extent before DIO submit
3620 * the IO. The uninitialized extent called at this time will be split
3621 * into three uninitialized extent(at most). After IO complete, the part
3622 * being filled will be convert to initialized by the end_io callback function
3623 * via ext4_convert_unwritten_extents().
3624 *
3625 * Returns the size of uninitialized extent to be written on success.
3626 */
3627static int ext4_split_convert_extents(handle_t *handle,
3628 struct inode *inode,
3629 struct ext4_map_blocks *map,
3630 struct ext4_ext_path *path,
3631 int flags)
3632{
3633 ext4_lblk_t eof_block;
3634 ext4_lblk_t ee_block;
3635 struct ext4_extent *ex;
3636 unsigned int ee_len;
3637 int split_flag = 0, depth;
3638
3639 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3640 __func__, inode->i_ino,
3641 (unsigned long long)map->m_lblk, map->m_len);
3642
3643 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3644 inode->i_sb->s_blocksize_bits;
3645 if (eof_block < map->m_lblk + map->m_len)
3646 eof_block = map->m_lblk + map->m_len;
3647 /*
3648 * It is safe to convert extent to initialized via explicit
3649 * zeroout only if extent is fully insde i_size or new_size.
3650 */
3651 depth = ext_depth(inode);
3652 ex = path[depth].p_ext;
3653 ee_block = le32_to_cpu(ex->ee_block);
3654 ee_len = ext4_ext_get_actual_len(ex);
3655
3656 /* Convert to unwritten */
3657 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3658 split_flag |= EXT4_EXT_DATA_VALID1;
3659 /* Convert to initialized */
3660 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3661 split_flag |= ee_block + ee_len <= eof_block ?
3662 EXT4_EXT_MAY_ZEROOUT : 0;
3663 split_flag |= (EXT4_EXT_MARK_UNINIT2 | EXT4_EXT_DATA_VALID2);
3664 }
3665 flags |= EXT4_GET_BLOCKS_PRE_IO;
3666 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3667}
3668
3669static int ext4_convert_initialized_extents(handle_t *handle,
3670 struct inode *inode,
3671 struct ext4_map_blocks *map,
3672 struct ext4_ext_path *path)
3673{
3674 struct ext4_extent *ex;
3675 ext4_lblk_t ee_block;
3676 unsigned int ee_len;
3677 int depth;
3678 int err = 0;
3679
3680 depth = ext_depth(inode);
3681 ex = path[depth].p_ext;
3682 ee_block = le32_to_cpu(ex->ee_block);
3683 ee_len = ext4_ext_get_actual_len(ex);
3684
3685 ext_debug("%s: inode %lu, logical"
3686 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3687 (unsigned long long)ee_block, ee_len);
3688
3689 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3690 err = ext4_split_convert_extents(handle, inode, map, path,
3691 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3692 if (err < 0)
3693 goto out;
3694 ext4_ext_drop_refs(path);
3695 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3696 if (IS_ERR(path)) {
3697 err = PTR_ERR(path);
3698 goto out;
3699 }
3700 depth = ext_depth(inode);
3701 ex = path[depth].p_ext;
3702 if (!ex) {
3703 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3704 (unsigned long) map->m_lblk);
3705 err = -EIO;
3706 goto out;
3707 }
3708 }
3709
3710 err = ext4_ext_get_access(handle, inode, path + depth);
3711 if (err)
3712 goto out;
3713 /* first mark the extent as uninitialized */
3714 ext4_ext_mark_uninitialized(ex);
3715
3716 /* note: ext4_ext_correct_indexes() isn't needed here because
3717 * borders are not changed
3718 */
3719 ext4_ext_try_to_merge(handle, inode, path, ex);
3720
3721 /* Mark modified extent as dirty */
3722 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3723out:
3724 ext4_ext_show_leaf(inode, path);
3725 return err;
3726}
3727
3728
3729static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3730 struct inode *inode,
3731 struct ext4_map_blocks *map,
3732 struct ext4_ext_path *path)
3733{
3734 struct ext4_extent *ex;
3735 ext4_lblk_t ee_block;
3736 unsigned int ee_len;
3737 int depth;
3738 int err = 0;
3739
3740 depth = ext_depth(inode);
3741 ex = path[depth].p_ext;
3742 ee_block = le32_to_cpu(ex->ee_block);
3743 ee_len = ext4_ext_get_actual_len(ex);
3744
3745 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3746 "block %llu, max_blocks %u\n", inode->i_ino,
3747 (unsigned long long)ee_block, ee_len);
3748
3749 /* If extent is larger than requested it is a clear sign that we still
3750 * have some extent state machine issues left. So extent_split is still
3751 * required.
3752 * TODO: Once all related issues will be fixed this situation should be
3753 * illegal.
3754 */
3755 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3756#ifdef EXT4_DEBUG
3757 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3758 " len %u; IO logical block %llu, len %u\n",
3759 inode->i_ino, (unsigned long long)ee_block, ee_len,
3760 (unsigned long long)map->m_lblk, map->m_len);
3761#endif
3762 err = ext4_split_convert_extents(handle, inode, map, path,
3763 EXT4_GET_BLOCKS_CONVERT);
3764 if (err < 0)
3765 goto out;
3766 ext4_ext_drop_refs(path);
3767 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3768 if (IS_ERR(path)) {
3769 err = PTR_ERR(path);
3770 goto out;
3771 }
3772 depth = ext_depth(inode);
3773 ex = path[depth].p_ext;
3774 }
3775
3776 err = ext4_ext_get_access(handle, inode, path + depth);
3777 if (err)
3778 goto out;
3779 /* first mark the extent as initialized */
3780 ext4_ext_mark_initialized(ex);
3781
3782 /* note: ext4_ext_correct_indexes() isn't needed here because
3783 * borders are not changed
3784 */
3785 ext4_ext_try_to_merge(handle, inode, path, ex);
3786
3787 /* Mark modified extent as dirty */
3788 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3789out:
3790 ext4_ext_show_leaf(inode, path);
3791 return err;
3792}
3793
3794static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3795 sector_t block, int count)
3796{
3797 int i;
3798 for (i = 0; i < count; i++)
3799 unmap_underlying_metadata(bdev, block + i);
3800}
3801
3802/*
3803 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3804 */
3805static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3806 ext4_lblk_t lblk,
3807 struct ext4_ext_path *path,
3808 unsigned int len)
3809{
3810 int i, depth;
3811 struct ext4_extent_header *eh;
3812 struct ext4_extent *last_ex;
3813
3814 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3815 return 0;
3816
3817 depth = ext_depth(inode);
3818 eh = path[depth].p_hdr;
3819
3820 /*
3821 * We're going to remove EOFBLOCKS_FL entirely in future so we
3822 * do not care for this case anymore. Simply remove the flag
3823 * if there are no extents.
3824 */
3825 if (unlikely(!eh->eh_entries))
3826 goto out;
3827 last_ex = EXT_LAST_EXTENT(eh);
3828 /*
3829 * We should clear the EOFBLOCKS_FL flag if we are writing the
3830 * last block in the last extent in the file. We test this by
3831 * first checking to see if the caller to
3832 * ext4_ext_get_blocks() was interested in the last block (or
3833 * a block beyond the last block) in the current extent. If
3834 * this turns out to be false, we can bail out from this
3835 * function immediately.
3836 */
3837 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3838 ext4_ext_get_actual_len(last_ex))
3839 return 0;
3840 /*
3841 * If the caller does appear to be planning to write at or
3842 * beyond the end of the current extent, we then test to see
3843 * if the current extent is the last extent in the file, by
3844 * checking to make sure it was reached via the rightmost node
3845 * at each level of the tree.
3846 */
3847 for (i = depth-1; i >= 0; i--)
3848 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3849 return 0;
3850out:
3851 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3852 return ext4_mark_inode_dirty(handle, inode);
3853}
3854
3855/**
3856 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3857 *
3858 * Return 1 if there is a delalloc block in the range, otherwise 0.
3859 */
3860int ext4_find_delalloc_range(struct inode *inode,
3861 ext4_lblk_t lblk_start,
3862 ext4_lblk_t lblk_end)
3863{
3864 struct extent_status es;
3865
3866 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3867 if (es.es_len == 0)
3868 return 0; /* there is no delay extent in this tree */
3869 else if (es.es_lblk <= lblk_start &&
3870 lblk_start < es.es_lblk + es.es_len)
3871 return 1;
3872 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3873 return 1;
3874 else
3875 return 0;
3876}
3877
3878int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3879{
3880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3881 ext4_lblk_t lblk_start, lblk_end;
3882 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3883 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3884
3885 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3886}
3887
3888/**
3889 * Determines how many complete clusters (out of those specified by the 'map')
3890 * are under delalloc and were reserved quota for.
3891 * This function is called when we are writing out the blocks that were
3892 * originally written with their allocation delayed, but then the space was
3893 * allocated using fallocate() before the delayed allocation could be resolved.
3894 * The cases to look for are:
3895 * ('=' indicated delayed allocated blocks
3896 * '-' indicates non-delayed allocated blocks)
3897 * (a) partial clusters towards beginning and/or end outside of allocated range
3898 * are not delalloc'ed.
3899 * Ex:
3900 * |----c---=|====c====|====c====|===-c----|
3901 * |++++++ allocated ++++++|
3902 * ==> 4 complete clusters in above example
3903 *
3904 * (b) partial cluster (outside of allocated range) towards either end is
3905 * marked for delayed allocation. In this case, we will exclude that
3906 * cluster.
3907 * Ex:
3908 * |----====c========|========c========|
3909 * |++++++ allocated ++++++|
3910 * ==> 1 complete clusters in above example
3911 *
3912 * Ex:
3913 * |================c================|
3914 * |++++++ allocated ++++++|
3915 * ==> 0 complete clusters in above example
3916 *
3917 * The ext4_da_update_reserve_space will be called only if we
3918 * determine here that there were some "entire" clusters that span
3919 * this 'allocated' range.
3920 * In the non-bigalloc case, this function will just end up returning num_blks
3921 * without ever calling ext4_find_delalloc_range.
3922 */
3923static unsigned int
3924get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3925 unsigned int num_blks)
3926{
3927 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3928 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3929 ext4_lblk_t lblk_from, lblk_to, c_offset;
3930 unsigned int allocated_clusters = 0;
3931
3932 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3933 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3934
3935 /* max possible clusters for this allocation */
3936 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3937
3938 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3939
3940 /* Check towards left side */
3941 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3942 if (c_offset) {
3943 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3944 lblk_to = lblk_from + c_offset - 1;
3945
3946 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3947 allocated_clusters--;
3948 }
3949
3950 /* Now check towards right. */
3951 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3952 if (allocated_clusters && c_offset) {
3953 lblk_from = lblk_start + num_blks;
3954 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3955
3956 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3957 allocated_clusters--;
3958 }
3959
3960 return allocated_clusters;
3961}
3962
3963static int
3964ext4_ext_convert_initialized_extent(handle_t *handle, struct inode *inode,
3965 struct ext4_map_blocks *map,
3966 struct ext4_ext_path *path, int flags,
3967 unsigned int allocated, ext4_fsblk_t newblock)
3968{
3969 int ret = 0;
3970 int err = 0;
3971
3972 /*
3973 * Make sure that the extent is no bigger than we support with
3974 * uninitialized extent
3975 */
3976 if (map->m_len > EXT_UNINIT_MAX_LEN)
3977 map->m_len = EXT_UNINIT_MAX_LEN / 2;
3978
3979 ret = ext4_convert_initialized_extents(handle, inode, map,
3980 path);
3981 if (ret >= 0) {
3982 ext4_update_inode_fsync_trans(handle, inode, 1);
3983 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3984 path, map->m_len);
3985 } else
3986 err = ret;
3987 map->m_flags |= EXT4_MAP_UNWRITTEN;
3988 if (allocated > map->m_len)
3989 allocated = map->m_len;
3990 map->m_len = allocated;
3991
3992 return err ? err : allocated;
3993}
3994
3995static int
3996ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3997 struct ext4_map_blocks *map,
3998 struct ext4_ext_path *path, int flags,
3999 unsigned int allocated, ext4_fsblk_t newblock)
4000{
4001 int ret = 0;
4002 int err = 0;
4003 ext4_io_end_t *io = ext4_inode_aio(inode);
4004
4005 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
4006 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4007 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4008 flags, allocated);
4009 ext4_ext_show_leaf(inode, path);
4010
4011 /*
4012 * When writing into uninitialized space, we should not fail to
4013 * allocate metadata blocks for the new extent block if needed.
4014 */
4015 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4016
4017 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
4018 allocated, newblock);
4019
4020 /* get_block() before submit the IO, split the extent */
4021 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4022 ret = ext4_split_convert_extents(handle, inode, map,
4023 path, flags | EXT4_GET_BLOCKS_CONVERT);
4024 if (ret <= 0)
4025 goto out;
4026 /*
4027 * Flag the inode(non aio case) or end_io struct (aio case)
4028 * that this IO needs to conversion to written when IO is
4029 * completed
4030 */
4031 if (io)
4032 ext4_set_io_unwritten_flag(inode, io);
4033 else
4034 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4035 map->m_flags |= EXT4_MAP_UNWRITTEN;
4036 if (ext4_should_dioread_nolock(inode))
4037 map->m_flags |= EXT4_MAP_UNINIT;
4038 goto out;
4039 }
4040 /* IO end_io complete, convert the filled extent to written */
4041 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
4042 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4043 path);
4044 if (ret >= 0) {
4045 ext4_update_inode_fsync_trans(handle, inode, 1);
4046 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4047 path, map->m_len);
4048 } else
4049 err = ret;
4050 map->m_flags |= EXT4_MAP_MAPPED;
4051 map->m_pblk = newblock;
4052 if (allocated > map->m_len)
4053 allocated = map->m_len;
4054 map->m_len = allocated;
4055 goto out2;
4056 }
4057 /* buffered IO case */
4058 /*
4059 * repeat fallocate creation request
4060 * we already have an unwritten extent
4061 */
4062 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
4063 map->m_flags |= EXT4_MAP_UNWRITTEN;
4064 goto map_out;
4065 }
4066
4067 /* buffered READ or buffered write_begin() lookup */
4068 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4069 /*
4070 * We have blocks reserved already. We
4071 * return allocated blocks so that delalloc
4072 * won't do block reservation for us. But
4073 * the buffer head will be unmapped so that
4074 * a read from the block returns 0s.
4075 */
4076 map->m_flags |= EXT4_MAP_UNWRITTEN;
4077 goto out1;
4078 }
4079
4080 /* buffered write, writepage time, convert*/
4081 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
4082 if (ret >= 0)
4083 ext4_update_inode_fsync_trans(handle, inode, 1);
4084out:
4085 if (ret <= 0) {
4086 err = ret;
4087 goto out2;
4088 } else
4089 allocated = ret;
4090 map->m_flags |= EXT4_MAP_NEW;
4091 /*
4092 * if we allocated more blocks than requested
4093 * we need to make sure we unmap the extra block
4094 * allocated. The actual needed block will get
4095 * unmapped later when we find the buffer_head marked
4096 * new.
4097 */
4098 if (allocated > map->m_len) {
4099 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4100 newblock + map->m_len,
4101 allocated - map->m_len);
4102 allocated = map->m_len;
4103 }
4104 map->m_len = allocated;
4105
4106 /*
4107 * If we have done fallocate with the offset that is already
4108 * delayed allocated, we would have block reservation
4109 * and quota reservation done in the delayed write path.
4110 * But fallocate would have already updated quota and block
4111 * count for this offset. So cancel these reservation
4112 */
4113 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4114 unsigned int reserved_clusters;
4115 reserved_clusters = get_reserved_cluster_alloc(inode,
4116 map->m_lblk, map->m_len);
4117 if (reserved_clusters)
4118 ext4_da_update_reserve_space(inode,
4119 reserved_clusters,
4120 0);
4121 }
4122
4123map_out:
4124 map->m_flags |= EXT4_MAP_MAPPED;
4125 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4126 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4127 map->m_len);
4128 if (err < 0)
4129 goto out2;
4130 }
4131out1:
4132 if (allocated > map->m_len)
4133 allocated = map->m_len;
4134 ext4_ext_show_leaf(inode, path);
4135 map->m_pblk = newblock;
4136 map->m_len = allocated;
4137out2:
4138 return err ? err : allocated;
4139}
4140
4141/*
4142 * get_implied_cluster_alloc - check to see if the requested
4143 * allocation (in the map structure) overlaps with a cluster already
4144 * allocated in an extent.
4145 * @sb The filesystem superblock structure
4146 * @map The requested lblk->pblk mapping
4147 * @ex The extent structure which might contain an implied
4148 * cluster allocation
4149 *
4150 * This function is called by ext4_ext_map_blocks() after we failed to
4151 * find blocks that were already in the inode's extent tree. Hence,
4152 * we know that the beginning of the requested region cannot overlap
4153 * the extent from the inode's extent tree. There are three cases we
4154 * want to catch. The first is this case:
4155 *
4156 * |--- cluster # N--|
4157 * |--- extent ---| |---- requested region ---|
4158 * |==========|
4159 *
4160 * The second case that we need to test for is this one:
4161 *
4162 * |--------- cluster # N ----------------|
4163 * |--- requested region --| |------- extent ----|
4164 * |=======================|
4165 *
4166 * The third case is when the requested region lies between two extents
4167 * within the same cluster:
4168 * |------------- cluster # N-------------|
4169 * |----- ex -----| |---- ex_right ----|
4170 * |------ requested region ------|
4171 * |================|
4172 *
4173 * In each of the above cases, we need to set the map->m_pblk and
4174 * map->m_len so it corresponds to the return the extent labelled as
4175 * "|====|" from cluster #N, since it is already in use for data in
4176 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4177 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4178 * as a new "allocated" block region. Otherwise, we will return 0 and
4179 * ext4_ext_map_blocks() will then allocate one or more new clusters
4180 * by calling ext4_mb_new_blocks().
4181 */
4182static int get_implied_cluster_alloc(struct super_block *sb,
4183 struct ext4_map_blocks *map,
4184 struct ext4_extent *ex,
4185 struct ext4_ext_path *path)
4186{
4187 struct ext4_sb_info *sbi = EXT4_SB(sb);
4188 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4189 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4190 ext4_lblk_t rr_cluster_start;
4191 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4192 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4193 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4194
4195 /* The extent passed in that we are trying to match */
4196 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4197 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4198
4199 /* The requested region passed into ext4_map_blocks() */
4200 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4201
4202 if ((rr_cluster_start == ex_cluster_end) ||
4203 (rr_cluster_start == ex_cluster_start)) {
4204 if (rr_cluster_start == ex_cluster_end)
4205 ee_start += ee_len - 1;
4206 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4207 map->m_len = min(map->m_len,
4208 (unsigned) sbi->s_cluster_ratio - c_offset);
4209 /*
4210 * Check for and handle this case:
4211 *
4212 * |--------- cluster # N-------------|
4213 * |------- extent ----|
4214 * |--- requested region ---|
4215 * |===========|
4216 */
4217
4218 if (map->m_lblk < ee_block)
4219 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4220
4221 /*
4222 * Check for the case where there is already another allocated
4223 * block to the right of 'ex' but before the end of the cluster.
4224 *
4225 * |------------- cluster # N-------------|
4226 * |----- ex -----| |---- ex_right ----|
4227 * |------ requested region ------|
4228 * |================|
4229 */
4230 if (map->m_lblk > ee_block) {
4231 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4232 map->m_len = min(map->m_len, next - map->m_lblk);
4233 }
4234
4235 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4236 return 1;
4237 }
4238
4239 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4240 return 0;
4241}
4242
4243
4244/*
4245 * Block allocation/map/preallocation routine for extents based files
4246 *
4247 *
4248 * Need to be called with
4249 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4250 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4251 *
4252 * return > 0, number of of blocks already mapped/allocated
4253 * if create == 0 and these are pre-allocated blocks
4254 * buffer head is unmapped
4255 * otherwise blocks are mapped
4256 *
4257 * return = 0, if plain look up failed (blocks have not been allocated)
4258 * buffer head is unmapped
4259 *
4260 * return < 0, error case.
4261 */
4262int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4263 struct ext4_map_blocks *map, int flags)
4264{
4265 struct ext4_ext_path *path = NULL;
4266 struct ext4_extent newex, *ex, *ex2;
4267 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4268 ext4_fsblk_t newblock = 0;
4269 int free_on_err = 0, err = 0, depth, ret;
4270 unsigned int allocated = 0, offset = 0;
4271 unsigned int allocated_clusters = 0;
4272 struct ext4_allocation_request ar;
4273 ext4_io_end_t *io = ext4_inode_aio(inode);
4274 ext4_lblk_t cluster_offset;
4275 int set_unwritten = 0;
4276
4277 ext_debug("blocks %u/%u requested for inode %lu\n",
4278 map->m_lblk, map->m_len, inode->i_ino);
4279 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4280
4281 /* find extent for this block */
4282 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4283 if (IS_ERR(path)) {
4284 err = PTR_ERR(path);
4285 path = NULL;
4286 goto out2;
4287 }
4288
4289 depth = ext_depth(inode);
4290
4291 /*
4292 * consistent leaf must not be empty;
4293 * this situation is possible, though, _during_ tree modification;
4294 * this is why assert can't be put in ext4_ext_find_extent()
4295 */
4296 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4297 EXT4_ERROR_INODE(inode, "bad extent address "
4298 "lblock: %lu, depth: %d pblock %lld",
4299 (unsigned long) map->m_lblk, depth,
4300 path[depth].p_block);
4301 err = -EIO;
4302 goto out2;
4303 }
4304
4305 ex = path[depth].p_ext;
4306 if (ex) {
4307 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4308 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4309 unsigned short ee_len;
4310
4311
4312 /*
4313 * Uninitialized extents are treated as holes, except that
4314 * we split out initialized portions during a write.
4315 */
4316 ee_len = ext4_ext_get_actual_len(ex);
4317
4318 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4319
4320 /* if found extent covers block, simply return it */
4321 if (in_range(map->m_lblk, ee_block, ee_len)) {
4322 newblock = map->m_lblk - ee_block + ee_start;
4323 /* number of remaining blocks in the extent */
4324 allocated = ee_len - (map->m_lblk - ee_block);
4325 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4326 ee_block, ee_len, newblock);
4327
4328 /*
4329 * If the extent is initialized check whether the
4330 * caller wants to convert it to unwritten.
4331 */
4332 if ((!ext4_ext_is_uninitialized(ex)) &&
4333 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4334 allocated = ext4_ext_convert_initialized_extent(
4335 handle, inode, map, path, flags,
4336 allocated, newblock);
4337 goto out2;
4338 } else if (!ext4_ext_is_uninitialized(ex))
4339 goto out;
4340
4341 ret = ext4_ext_handle_uninitialized_extents(
4342 handle, inode, map, path, flags,
4343 allocated, newblock);
4344 if (ret < 0)
4345 err = ret;
4346 else
4347 allocated = ret;
4348 goto out2;
4349 }
4350 }
4351
4352 if ((sbi->s_cluster_ratio > 1) &&
4353 ext4_find_delalloc_cluster(inode, map->m_lblk))
4354 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4355
4356 /*
4357 * requested block isn't allocated yet;
4358 * we couldn't try to create block if create flag is zero
4359 */
4360 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4361 /*
4362 * put just found gap into cache to speed up
4363 * subsequent requests
4364 */
4365 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4366 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4367 goto out2;
4368 }
4369
4370 /*
4371 * Okay, we need to do block allocation.
4372 */
4373 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4374 newex.ee_block = cpu_to_le32(map->m_lblk);
4375 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4376
4377 /*
4378 * If we are doing bigalloc, check to see if the extent returned
4379 * by ext4_ext_find_extent() implies a cluster we can use.
4380 */
4381 if (cluster_offset && ex &&
4382 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4383 ar.len = allocated = map->m_len;
4384 newblock = map->m_pblk;
4385 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4386 goto got_allocated_blocks;
4387 }
4388
4389 /* find neighbour allocated blocks */
4390 ar.lleft = map->m_lblk;
4391 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4392 if (err)
4393 goto out2;
4394 ar.lright = map->m_lblk;
4395 ex2 = NULL;
4396 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4397 if (err)
4398 goto out2;
4399
4400 /* Check if the extent after searching to the right implies a
4401 * cluster we can use. */
4402 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4403 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4404 ar.len = allocated = map->m_len;
4405 newblock = map->m_pblk;
4406 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4407 goto got_allocated_blocks;
4408 }
4409
4410 /*
4411 * See if request is beyond maximum number of blocks we can have in
4412 * a single extent. For an initialized extent this limit is
4413 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4414 * EXT_UNINIT_MAX_LEN.
4415 */
4416 if (map->m_len > EXT_INIT_MAX_LEN &&
4417 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4418 map->m_len = EXT_INIT_MAX_LEN;
4419 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4420 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4421 map->m_len = EXT_UNINIT_MAX_LEN;
4422
4423 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4424 newex.ee_len = cpu_to_le16(map->m_len);
4425 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4426 if (err)
4427 allocated = ext4_ext_get_actual_len(&newex);
4428 else
4429 allocated = map->m_len;
4430
4431 /* allocate new block */
4432 ar.inode = inode;
4433 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4434 ar.logical = map->m_lblk;
4435 /*
4436 * We calculate the offset from the beginning of the cluster
4437 * for the logical block number, since when we allocate a
4438 * physical cluster, the physical block should start at the
4439 * same offset from the beginning of the cluster. This is
4440 * needed so that future calls to get_implied_cluster_alloc()
4441 * work correctly.
4442 */
4443 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4444 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4445 ar.goal -= offset;
4446 ar.logical -= offset;
4447 if (S_ISREG(inode->i_mode))
4448 ar.flags = EXT4_MB_HINT_DATA;
4449 else
4450 /* disable in-core preallocation for non-regular files */
4451 ar.flags = 0;
4452 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4453 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4454 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4455 if (!newblock)
4456 goto out2;
4457 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4458 ar.goal, newblock, allocated);
4459 free_on_err = 1;
4460 allocated_clusters = ar.len;
4461 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4462 if (ar.len > allocated)
4463 ar.len = allocated;
4464
4465got_allocated_blocks:
4466 /* try to insert new extent into found leaf and return */
4467 ext4_ext_store_pblock(&newex, newblock + offset);
4468 newex.ee_len = cpu_to_le16(ar.len);
4469 /* Mark uninitialized */
4470 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4471 ext4_ext_mark_uninitialized(&newex);
4472 map->m_flags |= EXT4_MAP_UNWRITTEN;
4473 /*
4474 * io_end structure was created for every IO write to an
4475 * uninitialized extent. To avoid unnecessary conversion,
4476 * here we flag the IO that really needs the conversion.
4477 * For non asycn direct IO case, flag the inode state
4478 * that we need to perform conversion when IO is done.
4479 */
4480 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4481 set_unwritten = 1;
4482 if (ext4_should_dioread_nolock(inode))
4483 map->m_flags |= EXT4_MAP_UNINIT;
4484 }
4485
4486 err = 0;
4487 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4488 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4489 path, ar.len);
4490 if (!err)
4491 err = ext4_ext_insert_extent(handle, inode, path,
4492 &newex, flags);
4493
4494 if (!err && set_unwritten) {
4495 if (io)
4496 ext4_set_io_unwritten_flag(inode, io);
4497 else
4498 ext4_set_inode_state(inode,
4499 EXT4_STATE_DIO_UNWRITTEN);
4500 }
4501
4502 if (err && free_on_err) {
4503 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4504 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4505 /* free data blocks we just allocated */
4506 /* not a good idea to call discard here directly,
4507 * but otherwise we'd need to call it every free() */
4508 ext4_discard_preallocations(inode);
4509 ext4_free_blocks(handle, inode, NULL, newblock,
4510 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4511 goto out2;
4512 }
4513
4514 /* previous routine could use block we allocated */
4515 newblock = ext4_ext_pblock(&newex);
4516 allocated = ext4_ext_get_actual_len(&newex);
4517 if (allocated > map->m_len)
4518 allocated = map->m_len;
4519 map->m_flags |= EXT4_MAP_NEW;
4520
4521 /*
4522 * Update reserved blocks/metadata blocks after successful
4523 * block allocation which had been deferred till now.
4524 */
4525 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4526 unsigned int reserved_clusters;
4527 /*
4528 * Check how many clusters we had reserved this allocated range
4529 */
4530 reserved_clusters = get_reserved_cluster_alloc(inode,
4531 map->m_lblk, allocated);
4532 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4533 if (reserved_clusters) {
4534 /*
4535 * We have clusters reserved for this range.
4536 * But since we are not doing actual allocation
4537 * and are simply using blocks from previously
4538 * allocated cluster, we should release the
4539 * reservation and not claim quota.
4540 */
4541 ext4_da_update_reserve_space(inode,
4542 reserved_clusters, 0);
4543 }
4544 } else {
4545 BUG_ON(allocated_clusters < reserved_clusters);
4546 if (reserved_clusters < allocated_clusters) {
4547 struct ext4_inode_info *ei = EXT4_I(inode);
4548 int reservation = allocated_clusters -
4549 reserved_clusters;
4550 /*
4551 * It seems we claimed few clusters outside of
4552 * the range of this allocation. We should give
4553 * it back to the reservation pool. This can
4554 * happen in the following case:
4555 *
4556 * * Suppose s_cluster_ratio is 4 (i.e., each
4557 * cluster has 4 blocks. Thus, the clusters
4558 * are [0-3],[4-7],[8-11]...
4559 * * First comes delayed allocation write for
4560 * logical blocks 10 & 11. Since there were no
4561 * previous delayed allocated blocks in the
4562 * range [8-11], we would reserve 1 cluster
4563 * for this write.
4564 * * Next comes write for logical blocks 3 to 8.
4565 * In this case, we will reserve 2 clusters
4566 * (for [0-3] and [4-7]; and not for [8-11] as
4567 * that range has a delayed allocated blocks.
4568 * Thus total reserved clusters now becomes 3.
4569 * * Now, during the delayed allocation writeout
4570 * time, we will first write blocks [3-8] and
4571 * allocate 3 clusters for writing these
4572 * blocks. Also, we would claim all these
4573 * three clusters above.
4574 * * Now when we come here to writeout the
4575 * blocks [10-11], we would expect to claim
4576 * the reservation of 1 cluster we had made
4577 * (and we would claim it since there are no
4578 * more delayed allocated blocks in the range
4579 * [8-11]. But our reserved cluster count had
4580 * already gone to 0.
4581 *
4582 * Thus, at the step 4 above when we determine
4583 * that there are still some unwritten delayed
4584 * allocated blocks outside of our current
4585 * block range, we should increment the
4586 * reserved clusters count so that when the
4587 * remaining blocks finally gets written, we
4588 * could claim them.
4589 */
4590 dquot_reserve_block(inode,
4591 EXT4_C2B(sbi, reservation));
4592 spin_lock(&ei->i_block_reservation_lock);
4593 ei->i_reserved_data_blocks += reservation;
4594 spin_unlock(&ei->i_block_reservation_lock);
4595 }
4596 /*
4597 * We will claim quota for all newly allocated blocks.
4598 * We're updating the reserved space *after* the
4599 * correction above so we do not accidentally free
4600 * all the metadata reservation because we might
4601 * actually need it later on.
4602 */
4603 ext4_da_update_reserve_space(inode, allocated_clusters,
4604 1);
4605 }
4606 }
4607
4608 /*
4609 * Cache the extent and update transaction to commit on fdatasync only
4610 * when it is _not_ an uninitialized extent.
4611 */
4612 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
4613 ext4_update_inode_fsync_trans(handle, inode, 1);
4614 else
4615 ext4_update_inode_fsync_trans(handle, inode, 0);
4616out:
4617 if (allocated > map->m_len)
4618 allocated = map->m_len;
4619 ext4_ext_show_leaf(inode, path);
4620 map->m_flags |= EXT4_MAP_MAPPED;
4621 map->m_pblk = newblock;
4622 map->m_len = allocated;
4623out2:
4624 if (path) {
4625 ext4_ext_drop_refs(path);
4626 kfree(path);
4627 }
4628
4629 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4630 err ? err : allocated);
4631 ext4_es_lru_add(inode);
4632 return err ? err : allocated;
4633}
4634
4635void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4636{
4637 struct super_block *sb = inode->i_sb;
4638 ext4_lblk_t last_block;
4639 int err = 0;
4640
4641 /*
4642 * TODO: optimization is possible here.
4643 * Probably we need not scan at all,
4644 * because page truncation is enough.
4645 */
4646
4647 /* we have to know where to truncate from in crash case */
4648 EXT4_I(inode)->i_disksize = inode->i_size;
4649 ext4_mark_inode_dirty(handle, inode);
4650
4651 last_block = (inode->i_size + sb->s_blocksize - 1)
4652 >> EXT4_BLOCK_SIZE_BITS(sb);
4653retry:
4654 err = ext4_es_remove_extent(inode, last_block,
4655 EXT_MAX_BLOCKS - last_block);
4656 if (err == -ENOMEM) {
4657 cond_resched();
4658 congestion_wait(BLK_RW_ASYNC, HZ/50);
4659 goto retry;
4660 }
4661 if (err) {
4662 ext4_std_error(inode->i_sb, err);
4663 return;
4664 }
4665 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4666 ext4_std_error(inode->i_sb, err);
4667}
4668
4669static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4670 ext4_lblk_t len, int flags, int mode)
4671{
4672 struct inode *inode = file_inode(file);
4673 handle_t *handle;
4674 int ret = 0;
4675 int ret2 = 0;
4676 int retries = 0;
4677 struct ext4_map_blocks map;
4678 unsigned int credits;
4679
4680 map.m_lblk = offset;
4681 /*
4682 * Don't normalize the request if it can fit in one extent so
4683 * that it doesn't get unnecessarily split into multiple
4684 * extents.
4685 */
4686 if (len <= EXT_UNINIT_MAX_LEN)
4687 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4688
4689 /*
4690 * credits to insert 1 extent into extent tree
4691 */
4692 credits = ext4_chunk_trans_blocks(inode, len);
4693
4694retry:
4695 while (ret >= 0 && ret < len) {
4696 map.m_lblk = map.m_lblk + ret;
4697 map.m_len = len = len - ret;
4698 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4699 credits);
4700 if (IS_ERR(handle)) {
4701 ret = PTR_ERR(handle);
4702 break;
4703 }
4704 ret = ext4_map_blocks(handle, inode, &map, flags);
4705 if (ret <= 0) {
4706 ext4_debug("inode #%lu: block %u: len %u: "
4707 "ext4_ext_map_blocks returned %d",
4708 inode->i_ino, map.m_lblk,
4709 map.m_len, ret);
4710 ext4_mark_inode_dirty(handle, inode);
4711 ret2 = ext4_journal_stop(handle);
4712 break;
4713 }
4714 ret2 = ext4_journal_stop(handle);
4715 if (ret2)
4716 break;
4717 }
4718 if (ret == -ENOSPC &&
4719 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4720 ret = 0;
4721 goto retry;
4722 }
4723
4724 return ret > 0 ? ret2 : ret;
4725}
4726
4727static long ext4_zero_range(struct file *file, loff_t offset,
4728 loff_t len, int mode)
4729{
4730 struct inode *inode = file_inode(file);
4731 handle_t *handle = NULL;
4732 unsigned int max_blocks;
4733 loff_t new_size = 0;
4734 int ret = 0;
4735 int flags;
4736 int partial;
4737 loff_t start, end;
4738 ext4_lblk_t lblk;
4739 struct address_space *mapping = inode->i_mapping;
4740 unsigned int blkbits = inode->i_blkbits;
4741
4742 trace_ext4_zero_range(inode, offset, len, mode);
4743
4744 if (!S_ISREG(inode->i_mode))
4745 return -EINVAL;
4746
4747 /*
4748 * Write out all dirty pages to avoid race conditions
4749 * Then release them.
4750 */
4751 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4752 ret = filemap_write_and_wait_range(mapping, offset,
4753 offset + len - 1);
4754 if (ret)
4755 return ret;
4756 }
4757
4758 /*
4759 * Round up offset. This is not fallocate, we neet to zero out
4760 * blocks, so convert interior block aligned part of the range to
4761 * unwritten and possibly manually zero out unaligned parts of the
4762 * range.
4763 */
4764 start = round_up(offset, 1 << blkbits);
4765 end = round_down((offset + len), 1 << blkbits);
4766
4767 if (start < offset || end > offset + len)
4768 return -EINVAL;
4769 partial = (offset + len) & ((1 << blkbits) - 1);
4770
4771 lblk = start >> blkbits;
4772 max_blocks = (end >> blkbits);
4773 if (max_blocks < lblk)
4774 max_blocks = 0;
4775 else
4776 max_blocks -= lblk;
4777
4778 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
4779 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN;
4780 if (mode & FALLOC_FL_KEEP_SIZE)
4781 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4782
4783 mutex_lock(&inode->i_mutex);
4784
4785 /*
4786 * Indirect files do not support unwritten extnets
4787 */
4788 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4789 ret = -EOPNOTSUPP;
4790 goto out_mutex;
4791 }
4792
4793 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4794 offset + len > i_size_read(inode)) {
4795 new_size = offset + len;
4796 ret = inode_newsize_ok(inode, new_size);
4797 if (ret)
4798 goto out_mutex;
4799 /*
4800 * If we have a partial block after EOF we have to allocate
4801 * the entire block.
4802 */
4803 if (partial)
4804 max_blocks += 1;
4805 }
4806
4807 if (max_blocks > 0) {
4808
4809 /* Now release the pages and zero block aligned part of pages*/
4810 truncate_pagecache_range(inode, start, end - 1);
4811
4812 /* Wait all existing dio workers, newcomers will block on i_mutex */
4813 ext4_inode_block_unlocked_dio(inode);
4814 inode_dio_wait(inode);
4815
4816 /*
4817 * Remove entire range from the extent status tree.
4818 */
4819 ret = ext4_es_remove_extent(inode, lblk, max_blocks);
4820 if (ret)
4821 goto out_dio;
4822
4823 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, flags,
4824 mode);
4825 if (ret)
4826 goto out_dio;
4827 }
4828
4829 handle = ext4_journal_start(inode, EXT4_HT_MISC, 4);
4830 if (IS_ERR(handle)) {
4831 ret = PTR_ERR(handle);
4832 ext4_std_error(inode->i_sb, ret);
4833 goto out_dio;
4834 }
4835
4836 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4837
4838 if (new_size) {
4839 if (new_size > i_size_read(inode))
4840 i_size_write(inode, new_size);
4841 if (new_size > EXT4_I(inode)->i_disksize)
4842 ext4_update_i_disksize(inode, new_size);
4843 } else {
4844 /*
4845 * Mark that we allocate beyond EOF so the subsequent truncate
4846 * can proceed even if the new size is the same as i_size.
4847 */
4848 if ((offset + len) > i_size_read(inode))
4849 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4850 }
4851
4852 ext4_mark_inode_dirty(handle, inode);
4853
4854 /* Zero out partial block at the edges of the range */
4855 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4856
4857 if (file->f_flags & O_SYNC)
4858 ext4_handle_sync(handle);
4859
4860 ext4_journal_stop(handle);
4861out_dio:
4862 ext4_inode_resume_unlocked_dio(inode);
4863out_mutex:
4864 mutex_unlock(&inode->i_mutex);
4865 return ret;
4866}
4867
4868/*
4869 * preallocate space for a file. This implements ext4's fallocate file
4870 * operation, which gets called from sys_fallocate system call.
4871 * For block-mapped files, posix_fallocate should fall back to the method
4872 * of writing zeroes to the required new blocks (the same behavior which is
4873 * expected for file systems which do not support fallocate() system call).
4874 */
4875long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4876{
4877 struct inode *inode = file_inode(file);
4878 handle_t *handle;
4879 loff_t new_size = 0;
4880 unsigned int max_blocks;
4881 int ret = 0;
4882 int flags;
4883 ext4_lblk_t lblk;
4884 struct timespec tv;
4885 unsigned int blkbits = inode->i_blkbits;
4886
4887 /* Return error if mode is not supported */
4888 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4889 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4890 return -EOPNOTSUPP;
4891
4892 if (mode & FALLOC_FL_PUNCH_HOLE)
4893 return ext4_punch_hole(inode, offset, len);
4894
4895 ret = ext4_convert_inline_data(inode);
4896 if (ret)
4897 return ret;
4898
4899 /*
4900 * currently supporting (pre)allocate mode for extent-based
4901 * files _only_
4902 */
4903 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4904 return -EOPNOTSUPP;
4905
4906 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4907 return ext4_collapse_range(inode, offset, len);
4908
4909 if (mode & FALLOC_FL_ZERO_RANGE)
4910 return ext4_zero_range(file, offset, len, mode);
4911
4912 trace_ext4_fallocate_enter(inode, offset, len, mode);
4913 lblk = offset >> blkbits;
4914 /*
4915 * We can't just convert len to max_blocks because
4916 * If blocksize = 4096 offset = 3072 and len = 2048
4917 */
4918 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4919 - lblk;
4920
4921 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4922 if (mode & FALLOC_FL_KEEP_SIZE)
4923 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4924
4925 mutex_lock(&inode->i_mutex);
4926
4927 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4928 offset + len > i_size_read(inode)) {
4929 new_size = offset + len;
4930 ret = inode_newsize_ok(inode, new_size);
4931 if (ret)
4932 goto out;
4933 }
4934
4935 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, flags, mode);
4936 if (ret)
4937 goto out;
4938
4939 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4940 if (IS_ERR(handle))
4941 goto out;
4942
4943 tv = inode->i_ctime = ext4_current_time(inode);
4944
4945 if (new_size) {
4946 if (new_size > i_size_read(inode)) {
4947 i_size_write(inode, new_size);
4948 inode->i_mtime = tv;
4949 }
4950 if (new_size > EXT4_I(inode)->i_disksize)
4951 ext4_update_i_disksize(inode, new_size);
4952 } else {
4953 /*
4954 * Mark that we allocate beyond EOF so the subsequent truncate
4955 * can proceed even if the new size is the same as i_size.
4956 */
4957 if ((offset + len) > i_size_read(inode))
4958 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4959 }
4960 ext4_mark_inode_dirty(handle, inode);
4961 if (file->f_flags & O_SYNC)
4962 ext4_handle_sync(handle);
4963
4964 ext4_journal_stop(handle);
4965out:
4966 mutex_unlock(&inode->i_mutex);
4967 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4968 return ret;
4969}
4970
4971/*
4972 * This function convert a range of blocks to written extents
4973 * The caller of this function will pass the start offset and the size.
4974 * all unwritten extents within this range will be converted to
4975 * written extents.
4976 *
4977 * This function is called from the direct IO end io call back
4978 * function, to convert the fallocated extents after IO is completed.
4979 * Returns 0 on success.
4980 */
4981int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4982 loff_t offset, ssize_t len)
4983{
4984 unsigned int max_blocks;
4985 int ret = 0;
4986 int ret2 = 0;
4987 struct ext4_map_blocks map;
4988 unsigned int credits, blkbits = inode->i_blkbits;
4989
4990 map.m_lblk = offset >> blkbits;
4991 /*
4992 * We can't just convert len to max_blocks because
4993 * If blocksize = 4096 offset = 3072 and len = 2048
4994 */
4995 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4996 map.m_lblk);
4997 /*
4998 * This is somewhat ugly but the idea is clear: When transaction is
4999 * reserved, everything goes into it. Otherwise we rather start several
5000 * smaller transactions for conversion of each extent separately.
5001 */
5002 if (handle) {
5003 handle = ext4_journal_start_reserved(handle,
5004 EXT4_HT_EXT_CONVERT);
5005 if (IS_ERR(handle))
5006 return PTR_ERR(handle);
5007 credits = 0;
5008 } else {
5009 /*
5010 * credits to insert 1 extent into extent tree
5011 */
5012 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5013 }
5014 while (ret >= 0 && ret < max_blocks) {
5015 map.m_lblk += ret;
5016 map.m_len = (max_blocks -= ret);
5017 if (credits) {
5018 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5019 credits);
5020 if (IS_ERR(handle)) {
5021 ret = PTR_ERR(handle);
5022 break;
5023 }
5024 }
5025 ret = ext4_map_blocks(handle, inode, &map,
5026 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5027 if (ret <= 0)
5028 ext4_warning(inode->i_sb,
5029 "inode #%lu: block %u: len %u: "
5030 "ext4_ext_map_blocks returned %d",
5031 inode->i_ino, map.m_lblk,
5032 map.m_len, ret);
5033 ext4_mark_inode_dirty(handle, inode);
5034 if (credits)
5035 ret2 = ext4_journal_stop(handle);
5036 if (ret <= 0 || ret2)
5037 break;
5038 }
5039 if (!credits)
5040 ret2 = ext4_journal_stop(handle);
5041 return ret > 0 ? ret2 : ret;
5042}
5043
5044/*
5045 * If newes is not existing extent (newes->ec_pblk equals zero) find
5046 * delayed extent at start of newes and update newes accordingly and
5047 * return start of the next delayed extent.
5048 *
5049 * If newes is existing extent (newes->ec_pblk is not equal zero)
5050 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5051 * extent found. Leave newes unmodified.
5052 */
5053static int ext4_find_delayed_extent(struct inode *inode,
5054 struct extent_status *newes)
5055{
5056 struct extent_status es;
5057 ext4_lblk_t block, next_del;
5058
5059 if (newes->es_pblk == 0) {
5060 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5061 newes->es_lblk + newes->es_len - 1, &es);
5062
5063 /*
5064 * No extent in extent-tree contains block @newes->es_pblk,
5065 * then the block may stay in 1)a hole or 2)delayed-extent.
5066 */
5067 if (es.es_len == 0)
5068 /* A hole found. */
5069 return 0;
5070
5071 if (es.es_lblk > newes->es_lblk) {
5072 /* A hole found. */
5073 newes->es_len = min(es.es_lblk - newes->es_lblk,
5074 newes->es_len);
5075 return 0;
5076 }
5077
5078 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5079 }
5080
5081 block = newes->es_lblk + newes->es_len;
5082 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5083 if (es.es_len == 0)
5084 next_del = EXT_MAX_BLOCKS;
5085 else
5086 next_del = es.es_lblk;
5087
5088 return next_del;
5089}
5090/* fiemap flags we can handle specified here */
5091#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5092
5093static int ext4_xattr_fiemap(struct inode *inode,
5094 struct fiemap_extent_info *fieinfo)
5095{
5096 __u64 physical = 0;
5097 __u64 length;
5098 __u32 flags = FIEMAP_EXTENT_LAST;
5099 int blockbits = inode->i_sb->s_blocksize_bits;
5100 int error = 0;
5101
5102 /* in-inode? */
5103 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5104 struct ext4_iloc iloc;
5105 int offset; /* offset of xattr in inode */
5106
5107 error = ext4_get_inode_loc(inode, &iloc);
5108 if (error)
5109 return error;
5110 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5111 offset = EXT4_GOOD_OLD_INODE_SIZE +
5112 EXT4_I(inode)->i_extra_isize;
5113 physical += offset;
5114 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5115 flags |= FIEMAP_EXTENT_DATA_INLINE;
5116 brelse(iloc.bh);
5117 } else { /* external block */
5118 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5119 length = inode->i_sb->s_blocksize;
5120 }
5121
5122 if (physical)
5123 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5124 length, flags);
5125 return (error < 0 ? error : 0);
5126}
5127
5128int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5129 __u64 start, __u64 len)
5130{
5131 ext4_lblk_t start_blk;
5132 int error = 0;
5133
5134 if (ext4_has_inline_data(inode)) {
5135 int has_inline = 1;
5136
5137 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5138
5139 if (has_inline)
5140 return error;
5141 }
5142
5143 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5144 error = ext4_ext_precache(inode);
5145 if (error)
5146 return error;
5147 }
5148
5149 /* fallback to generic here if not in extents fmt */
5150 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5151 return generic_block_fiemap(inode, fieinfo, start, len,
5152 ext4_get_block);
5153
5154 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5155 return -EBADR;
5156
5157 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5158 error = ext4_xattr_fiemap(inode, fieinfo);
5159 } else {
5160 ext4_lblk_t len_blks;
5161 __u64 last_blk;
5162
5163 start_blk = start >> inode->i_sb->s_blocksize_bits;
5164 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5165 if (last_blk >= EXT_MAX_BLOCKS)
5166 last_blk = EXT_MAX_BLOCKS-1;
5167 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5168
5169 /*
5170 * Walk the extent tree gathering extent information
5171 * and pushing extents back to the user.
5172 */
5173 error = ext4_fill_fiemap_extents(inode, start_blk,
5174 len_blks, fieinfo);
5175 }
5176 ext4_es_lru_add(inode);
5177 return error;
5178}
5179
5180/*
5181 * ext4_access_path:
5182 * Function to access the path buffer for marking it dirty.
5183 * It also checks if there are sufficient credits left in the journal handle
5184 * to update path.
5185 */
5186static int
5187ext4_access_path(handle_t *handle, struct inode *inode,
5188 struct ext4_ext_path *path)
5189{
5190 int credits, err;
5191
5192 if (!ext4_handle_valid(handle))
5193 return 0;
5194
5195 /*
5196 * Check if need to extend journal credits
5197 * 3 for leaf, sb, and inode plus 2 (bmap and group
5198 * descriptor) for each block group; assume two block
5199 * groups
5200 */
5201 if (handle->h_buffer_credits < 7) {
5202 credits = ext4_writepage_trans_blocks(inode);
5203 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5204 /* EAGAIN is success */
5205 if (err && err != -EAGAIN)
5206 return err;
5207 }
5208
5209 err = ext4_ext_get_access(handle, inode, path);
5210 return err;
5211}
5212
5213/*
5214 * ext4_ext_shift_path_extents:
5215 * Shift the extents of a path structure lying between path[depth].p_ext
5216 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5217 * from starting block for each extent.
5218 */
5219static int
5220ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5221 struct inode *inode, handle_t *handle,
5222 ext4_lblk_t *start)
5223{
5224 int depth, err = 0;
5225 struct ext4_extent *ex_start, *ex_last;
5226 bool update = 0;
5227 depth = path->p_depth;
5228
5229 while (depth >= 0) {
5230 if (depth == path->p_depth) {
5231 ex_start = path[depth].p_ext;
5232 if (!ex_start)
5233 return -EIO;
5234
5235 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5236 if (!ex_last)
5237 return -EIO;
5238
5239 err = ext4_access_path(handle, inode, path + depth);
5240 if (err)
5241 goto out;
5242
5243 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5244 update = 1;
5245
5246 *start = le32_to_cpu(ex_last->ee_block) +
5247 ext4_ext_get_actual_len(ex_last);
5248
5249 while (ex_start <= ex_last) {
5250 le32_add_cpu(&ex_start->ee_block, -shift);
5251 /* Try to merge to the left. */
5252 if ((ex_start >
5253 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5254 ext4_ext_try_to_merge_right(inode,
5255 path, ex_start - 1))
5256 ex_last--;
5257 else
5258 ex_start++;
5259 }
5260 err = ext4_ext_dirty(handle, inode, path + depth);
5261 if (err)
5262 goto out;
5263
5264 if (--depth < 0 || !update)
5265 break;
5266 }
5267
5268 /* Update index too */
5269 err = ext4_access_path(handle, inode, path + depth);
5270 if (err)
5271 goto out;
5272
5273 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5274 err = ext4_ext_dirty(handle, inode, path + depth);
5275 if (err)
5276 goto out;
5277
5278 /* we are done if current index is not a starting index */
5279 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5280 break;
5281
5282 depth--;
5283 }
5284
5285out:
5286 return err;
5287}
5288
5289/*
5290 * ext4_ext_shift_extents:
5291 * All the extents which lies in the range from start to the last allocated
5292 * block for the file are shifted downwards by shift blocks.
5293 * On success, 0 is returned, error otherwise.
5294 */
5295static int
5296ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5297 ext4_lblk_t start, ext4_lblk_t shift)
5298{
5299 struct ext4_ext_path *path;
5300 int ret = 0, depth;
5301 struct ext4_extent *extent;
5302 ext4_lblk_t stop_block, current_block;
5303 ext4_lblk_t ex_start, ex_end;
5304
5305 /* Let path point to the last extent */
5306 path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5307 if (IS_ERR(path))
5308 return PTR_ERR(path);
5309
5310 depth = path->p_depth;
5311 extent = path[depth].p_ext;
5312 if (!extent) {
5313 ext4_ext_drop_refs(path);
5314 kfree(path);
5315 return ret;
5316 }
5317
5318 stop_block = le32_to_cpu(extent->ee_block) +
5319 ext4_ext_get_actual_len(extent);
5320 ext4_ext_drop_refs(path);
5321 kfree(path);
5322
5323 /* Nothing to shift, if hole is at the end of file */
5324 if (start >= stop_block)
5325 return ret;
5326
5327 /*
5328 * Don't start shifting extents until we make sure the hole is big
5329 * enough to accomodate the shift.
5330 */
5331 path = ext4_ext_find_extent(inode, start - 1, NULL, 0);
5332 if (IS_ERR(path))
5333 return PTR_ERR(path);
5334 depth = path->p_depth;
5335 extent = path[depth].p_ext;
5336 if (extent) {
5337 ex_start = le32_to_cpu(extent->ee_block);
5338 ex_end = le32_to_cpu(extent->ee_block) +
5339 ext4_ext_get_actual_len(extent);
5340 } else {
5341 ex_start = 0;
5342 ex_end = 0;
5343 }
5344 ext4_ext_drop_refs(path);
5345 kfree(path);
5346
5347 if ((start == ex_start && shift > ex_start) ||
5348 (shift > start - ex_end))
5349 return -EINVAL;
5350
5351 /* Its safe to start updating extents */
5352 while (start < stop_block) {
5353 path = ext4_ext_find_extent(inode, start, NULL, 0);
5354 if (IS_ERR(path))
5355 return PTR_ERR(path);
5356 depth = path->p_depth;
5357 extent = path[depth].p_ext;
5358 if (!extent) {
5359 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5360 (unsigned long) start);
5361 return -EIO;
5362 }
5363
5364 current_block = le32_to_cpu(extent->ee_block);
5365 if (start > current_block) {
5366 /* Hole, move to the next extent */
5367 ret = mext_next_extent(inode, path, &extent);
5368 if (ret != 0) {
5369 ext4_ext_drop_refs(path);
5370 kfree(path);
5371 if (ret == 1)
5372 ret = 0;
5373 break;
5374 }
5375 }
5376 ret = ext4_ext_shift_path_extents(path, shift, inode,
5377 handle, &start);
5378 ext4_ext_drop_refs(path);
5379 kfree(path);
5380 if (ret)
5381 break;
5382 }
5383
5384 return ret;
5385}
5386
5387/*
5388 * ext4_collapse_range:
5389 * This implements the fallocate's collapse range functionality for ext4
5390 * Returns: 0 and non-zero on error.
5391 */
5392int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5393{
5394 struct super_block *sb = inode->i_sb;
5395 ext4_lblk_t punch_start, punch_stop;
5396 handle_t *handle;
5397 unsigned int credits;
5398 loff_t new_size, ioffset;
5399 int ret;
5400
5401 /* Collapse range works only on fs block size aligned offsets. */
5402 if (offset & (EXT4_BLOCK_SIZE(sb) - 1) ||
5403 len & (EXT4_BLOCK_SIZE(sb) - 1))
5404 return -EINVAL;
5405
5406 if (!S_ISREG(inode->i_mode))
5407 return -EINVAL;
5408
5409 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1)
5410 return -EOPNOTSUPP;
5411
5412 trace_ext4_collapse_range(inode, offset, len);
5413
5414 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5415 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5416
5417 /* Call ext4_force_commit to flush all data in case of data=journal. */
5418 if (ext4_should_journal_data(inode)) {
5419 ret = ext4_force_commit(inode->i_sb);
5420 if (ret)
5421 return ret;
5422 }
5423
5424 /*
5425 * Need to round down offset to be aligned with page size boundary
5426 * for page size > block size.
5427 */
5428 ioffset = round_down(offset, PAGE_SIZE);
5429
5430 /* Write out all dirty pages */
5431 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5432 LLONG_MAX);
5433 if (ret)
5434 return ret;
5435
5436 /* Take mutex lock */
5437 mutex_lock(&inode->i_mutex);
5438
5439 /*
5440 * There is no need to overlap collapse range with EOF, in which case
5441 * it is effectively a truncate operation
5442 */
5443 if (offset + len >= i_size_read(inode)) {
5444 ret = -EINVAL;
5445 goto out_mutex;
5446 }
5447
5448 /* Currently just for extent based files */
5449 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5450 ret = -EOPNOTSUPP;
5451 goto out_mutex;
5452 }
5453
5454 truncate_pagecache(inode, ioffset);
5455
5456 /* Wait for existing dio to complete */
5457 ext4_inode_block_unlocked_dio(inode);
5458 inode_dio_wait(inode);
5459
5460 credits = ext4_writepage_trans_blocks(inode);
5461 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5462 if (IS_ERR(handle)) {
5463 ret = PTR_ERR(handle);
5464 goto out_dio;
5465 }
5466
5467 down_write(&EXT4_I(inode)->i_data_sem);
5468 ext4_discard_preallocations(inode);
5469
5470 ret = ext4_es_remove_extent(inode, punch_start,
5471 EXT_MAX_BLOCKS - punch_start);
5472 if (ret) {
5473 up_write(&EXT4_I(inode)->i_data_sem);
5474 goto out_stop;
5475 }
5476
5477 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5478 if (ret) {
5479 up_write(&EXT4_I(inode)->i_data_sem);
5480 goto out_stop;
5481 }
5482 ext4_discard_preallocations(inode);
5483
5484 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5485 punch_stop - punch_start);
5486 if (ret) {
5487 up_write(&EXT4_I(inode)->i_data_sem);
5488 goto out_stop;
5489 }
5490
5491 new_size = i_size_read(inode) - len;
5492 i_size_write(inode, new_size);
5493 EXT4_I(inode)->i_disksize = new_size;
5494
5495 up_write(&EXT4_I(inode)->i_data_sem);
5496 if (IS_SYNC(inode))
5497 ext4_handle_sync(handle);
5498 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5499 ext4_mark_inode_dirty(handle, inode);
5500
5501out_stop:
5502 ext4_journal_stop(handle);
5503out_dio:
5504 ext4_inode_resume_unlocked_dio(inode);
5505out_mutex:
5506 mutex_unlock(&inode->i_mutex);
5507 return ret;
5508}