Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
[linux-2.6] / fs / ext4 / extents.c
1 /*
2  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3  * Written by Alex Tomas <alex@clusterfs.com>
4  *
5  * Architecture independence:
6  *   Copyright (c) 2005, Bull S.A.
7  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public Licens
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
21  */
22
23 /*
24  * Extents support for EXT4
25  *
26  * TODO:
27  *   - ext4*_error() should be used in some situations
28  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29  *   - smart tree reduction
30  */
31
32 #include <linux/module.h>
33 #include <linux/fs.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
45
46
47 /*
48  * ext_pblock:
49  * combine low and high parts of physical block number into ext4_fsblk_t
50  */
51 static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
52 {
53         ext4_fsblk_t block;
54
55         block = le32_to_cpu(ex->ee_start_lo);
56         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
57         return block;
58 }
59
60 /*
61  * idx_pblock:
62  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
63  */
64 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
65 {
66         ext4_fsblk_t block;
67
68         block = le32_to_cpu(ix->ei_leaf_lo);
69         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
70         return block;
71 }
72
73 /*
74  * ext4_ext_store_pblock:
75  * stores a large physical block number into an extent struct,
76  * breaking it into parts
77  */
78 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
79 {
80         ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
81         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
82 }
83
84 /*
85  * ext4_idx_store_pblock:
86  * stores a large physical block number into an index struct,
87  * breaking it into parts
88  */
89 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
90 {
91         ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
92         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
93 }
94
95 static int ext4_ext_journal_restart(handle_t *handle, int needed)
96 {
97         int err;
98
99         if (handle->h_buffer_credits > needed)
100                 return 0;
101         err = ext4_journal_extend(handle, needed);
102         if (err <= 0)
103                 return err;
104         return ext4_journal_restart(handle, needed);
105 }
106
107 /*
108  * could return:
109  *  - EROFS
110  *  - ENOMEM
111  */
112 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
113                                 struct ext4_ext_path *path)
114 {
115         if (path->p_bh) {
116                 /* path points to block */
117                 return ext4_journal_get_write_access(handle, path->p_bh);
118         }
119         /* path points to leaf/index in inode body */
120         /* we use in-core data, no need to protect them */
121         return 0;
122 }
123
124 /*
125  * could return:
126  *  - EROFS
127  *  - ENOMEM
128  *  - EIO
129  */
130 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
131                                 struct ext4_ext_path *path)
132 {
133         int err;
134         if (path->p_bh) {
135                 /* path points to block */
136                 err = ext4_journal_dirty_metadata(handle, path->p_bh);
137         } else {
138                 /* path points to leaf/index in inode body */
139                 err = ext4_mark_inode_dirty(handle, inode);
140         }
141         return err;
142 }
143
144 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
145                               struct ext4_ext_path *path,
146                               ext4_lblk_t block)
147 {
148         struct ext4_inode_info *ei = EXT4_I(inode);
149         ext4_fsblk_t bg_start;
150         ext4_fsblk_t last_block;
151         ext4_grpblk_t colour;
152         int depth;
153
154         if (path) {
155                 struct ext4_extent *ex;
156                 depth = path->p_depth;
157
158                 /* try to predict block placement */
159                 ex = path[depth].p_ext;
160                 if (ex)
161                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
162
163                 /* it looks like index is empty;
164                  * try to find starting block from index itself */
165                 if (path[depth].p_bh)
166                         return path[depth].p_bh->b_blocknr;
167         }
168
169         /* OK. use inode's group */
170         bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
171                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
172         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
173
174         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
175                 colour = (current->pid % 16) *
176                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
177         else
178                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
179         return bg_start + colour + block;
180 }
181
182 /*
183  * Allocation for a meta data block
184  */
185 static ext4_fsblk_t
186 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
187                         struct ext4_ext_path *path,
188                         struct ext4_extent *ex, int *err)
189 {
190         ext4_fsblk_t goal, newblock;
191
192         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
193         newblock = ext4_new_meta_block(handle, inode, goal, err);
194         return newblock;
195 }
196
197 static int ext4_ext_space_block(struct inode *inode)
198 {
199         int size;
200
201         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
202                         / sizeof(struct ext4_extent);
203 #ifdef AGGRESSIVE_TEST
204         if (size > 6)
205                 size = 6;
206 #endif
207         return size;
208 }
209
210 static int ext4_ext_space_block_idx(struct inode *inode)
211 {
212         int size;
213
214         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
215                         / sizeof(struct ext4_extent_idx);
216 #ifdef AGGRESSIVE_TEST
217         if (size > 5)
218                 size = 5;
219 #endif
220         return size;
221 }
222
223 static int ext4_ext_space_root(struct inode *inode)
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);
230 #ifdef AGGRESSIVE_TEST
231         if (size > 3)
232                 size = 3;
233 #endif
234         return size;
235 }
236
237 static int ext4_ext_space_root_idx(struct inode *inode)
238 {
239         int size;
240
241         size = sizeof(EXT4_I(inode)->i_data);
242         size -= sizeof(struct ext4_extent_header);
243         size /= sizeof(struct ext4_extent_idx);
244 #ifdef AGGRESSIVE_TEST
245         if (size > 4)
246                 size = 4;
247 #endif
248         return size;
249 }
250
251 /*
252  * Calculate the number of metadata blocks needed
253  * to allocate @blocks
254  * Worse case is one block per extent
255  */
256 int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
257 {
258         int lcap, icap, rcap, leafs, idxs, num;
259         int newextents = blocks;
260
261         rcap = ext4_ext_space_root_idx(inode);
262         lcap = ext4_ext_space_block(inode);
263         icap = ext4_ext_space_block_idx(inode);
264
265         /* number of new leaf blocks needed */
266         num = leafs = (newextents + lcap - 1) / lcap;
267
268         /*
269          * Worse case, we need separate index block(s)
270          * to link all new leaf blocks
271          */
272         idxs = (leafs + icap - 1) / icap;
273         do {
274                 num += idxs;
275                 idxs = (idxs + icap - 1) / icap;
276         } while (idxs > rcap);
277
278         return num;
279 }
280
281 static int
282 ext4_ext_max_entries(struct inode *inode, int depth)
283 {
284         int max;
285
286         if (depth == ext_depth(inode)) {
287                 if (depth == 0)
288                         max = ext4_ext_space_root(inode);
289                 else
290                         max = ext4_ext_space_root_idx(inode);
291         } else {
292                 if (depth == 0)
293                         max = ext4_ext_space_block(inode);
294                 else
295                         max = ext4_ext_space_block_idx(inode);
296         }
297
298         return max;
299 }
300
301 static int __ext4_ext_check_header(const char *function, struct inode *inode,
302                                         struct ext4_extent_header *eh,
303                                         int depth)
304 {
305         const char *error_msg;
306         int max = 0;
307
308         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
309                 error_msg = "invalid magic";
310                 goto corrupted;
311         }
312         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
313                 error_msg = "unexpected eh_depth";
314                 goto corrupted;
315         }
316         if (unlikely(eh->eh_max == 0)) {
317                 error_msg = "invalid eh_max";
318                 goto corrupted;
319         }
320         max = ext4_ext_max_entries(inode, depth);
321         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
322                 error_msg = "too large eh_max";
323                 goto corrupted;
324         }
325         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
326                 error_msg = "invalid eh_entries";
327                 goto corrupted;
328         }
329         return 0;
330
331 corrupted:
332         ext4_error(inode->i_sb, function,
333                         "bad header in inode #%lu: %s - magic %x, "
334                         "entries %u, max %u(%u), depth %u(%u)",
335                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
336                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
337                         max, le16_to_cpu(eh->eh_depth), depth);
338
339         return -EIO;
340 }
341
342 #define ext4_ext_check_header(inode, eh, depth) \
343         __ext4_ext_check_header(__func__, inode, eh, depth)
344
345 #ifdef EXT_DEBUG
346 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
347 {
348         int k, l = path->p_depth;
349
350         ext_debug("path:");
351         for (k = 0; k <= l; k++, path++) {
352                 if (path->p_idx) {
353                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
354                             idx_pblock(path->p_idx));
355                 } else if (path->p_ext) {
356                         ext_debug("  %d:%d:%llu ",
357                                   le32_to_cpu(path->p_ext->ee_block),
358                                   ext4_ext_get_actual_len(path->p_ext),
359                                   ext_pblock(path->p_ext));
360                 } else
361                         ext_debug("  []");
362         }
363         ext_debug("\n");
364 }
365
366 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
367 {
368         int depth = ext_depth(inode);
369         struct ext4_extent_header *eh;
370         struct ext4_extent *ex;
371         int i;
372
373         if (!path)
374                 return;
375
376         eh = path[depth].p_hdr;
377         ex = EXT_FIRST_EXTENT(eh);
378
379         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
380                 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
381                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
382         }
383         ext_debug("\n");
384 }
385 #else
386 #define ext4_ext_show_path(inode,path)
387 #define ext4_ext_show_leaf(inode,path)
388 #endif
389
390 void ext4_ext_drop_refs(struct ext4_ext_path *path)
391 {
392         int depth = path->p_depth;
393         int i;
394
395         for (i = 0; i <= depth; i++, path++)
396                 if (path->p_bh) {
397                         brelse(path->p_bh);
398                         path->p_bh = NULL;
399                 }
400 }
401
402 /*
403  * ext4_ext_binsearch_idx:
404  * binary search for the closest index of the given block
405  * the header must be checked before calling this
406  */
407 static void
408 ext4_ext_binsearch_idx(struct inode *inode,
409                         struct ext4_ext_path *path, ext4_lblk_t block)
410 {
411         struct ext4_extent_header *eh = path->p_hdr;
412         struct ext4_extent_idx *r, *l, *m;
413
414
415         ext_debug("binsearch for %u(idx):  ", block);
416
417         l = EXT_FIRST_INDEX(eh) + 1;
418         r = EXT_LAST_INDEX(eh);
419         while (l <= r) {
420                 m = l + (r - l) / 2;
421                 if (block < le32_to_cpu(m->ei_block))
422                         r = m - 1;
423                 else
424                         l = m + 1;
425                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
426                                 m, le32_to_cpu(m->ei_block),
427                                 r, le32_to_cpu(r->ei_block));
428         }
429
430         path->p_idx = l - 1;
431         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
432                   idx_pblock(path->p_idx));
433
434 #ifdef CHECK_BINSEARCH
435         {
436                 struct ext4_extent_idx *chix, *ix;
437                 int k;
438
439                 chix = ix = EXT_FIRST_INDEX(eh);
440                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
441                   if (k != 0 &&
442                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
443                                 printk("k=%d, ix=0x%p, first=0x%p\n", k,
444                                         ix, EXT_FIRST_INDEX(eh));
445                                 printk("%u <= %u\n",
446                                        le32_to_cpu(ix->ei_block),
447                                        le32_to_cpu(ix[-1].ei_block));
448                         }
449                         BUG_ON(k && le32_to_cpu(ix->ei_block)
450                                            <= le32_to_cpu(ix[-1].ei_block));
451                         if (block < le32_to_cpu(ix->ei_block))
452                                 break;
453                         chix = ix;
454                 }
455                 BUG_ON(chix != path->p_idx);
456         }
457 #endif
458
459 }
460
461 /*
462  * ext4_ext_binsearch:
463  * binary search for closest extent of the given block
464  * the header must be checked before calling this
465  */
466 static void
467 ext4_ext_binsearch(struct inode *inode,
468                 struct ext4_ext_path *path, ext4_lblk_t block)
469 {
470         struct ext4_extent_header *eh = path->p_hdr;
471         struct ext4_extent *r, *l, *m;
472
473         if (eh->eh_entries == 0) {
474                 /*
475                  * this leaf is empty:
476                  * we get such a leaf in split/add case
477                  */
478                 return;
479         }
480
481         ext_debug("binsearch for %u:  ", block);
482
483         l = EXT_FIRST_EXTENT(eh) + 1;
484         r = EXT_LAST_EXTENT(eh);
485
486         while (l <= r) {
487                 m = l + (r - l) / 2;
488                 if (block < le32_to_cpu(m->ee_block))
489                         r = m - 1;
490                 else
491                         l = m + 1;
492                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
493                                 m, le32_to_cpu(m->ee_block),
494                                 r, le32_to_cpu(r->ee_block));
495         }
496
497         path->p_ext = l - 1;
498         ext_debug("  -> %d:%llu:%d ",
499                         le32_to_cpu(path->p_ext->ee_block),
500                         ext_pblock(path->p_ext),
501                         ext4_ext_get_actual_len(path->p_ext));
502
503 #ifdef CHECK_BINSEARCH
504         {
505                 struct ext4_extent *chex, *ex;
506                 int k;
507
508                 chex = ex = EXT_FIRST_EXTENT(eh);
509                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
510                         BUG_ON(k && le32_to_cpu(ex->ee_block)
511                                           <= le32_to_cpu(ex[-1].ee_block));
512                         if (block < le32_to_cpu(ex->ee_block))
513                                 break;
514                         chex = ex;
515                 }
516                 BUG_ON(chex != path->p_ext);
517         }
518 #endif
519
520 }
521
522 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
523 {
524         struct ext4_extent_header *eh;
525
526         eh = ext_inode_hdr(inode);
527         eh->eh_depth = 0;
528         eh->eh_entries = 0;
529         eh->eh_magic = EXT4_EXT_MAGIC;
530         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
531         ext4_mark_inode_dirty(handle, inode);
532         ext4_ext_invalidate_cache(inode);
533         return 0;
534 }
535
536 struct ext4_ext_path *
537 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
538                                         struct ext4_ext_path *path)
539 {
540         struct ext4_extent_header *eh;
541         struct buffer_head *bh;
542         short int depth, i, ppos = 0, alloc = 0;
543
544         eh = ext_inode_hdr(inode);
545         depth = ext_depth(inode);
546         if (ext4_ext_check_header(inode, eh, depth))
547                 return ERR_PTR(-EIO);
548
549
550         /* account possible depth increase */
551         if (!path) {
552                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
553                                 GFP_NOFS);
554                 if (!path)
555                         return ERR_PTR(-ENOMEM);
556                 alloc = 1;
557         }
558         path[0].p_hdr = eh;
559         path[0].p_bh = NULL;
560
561         i = depth;
562         /* walk through the tree */
563         while (i) {
564                 ext_debug("depth %d: num %d, max %d\n",
565                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
566
567                 ext4_ext_binsearch_idx(inode, path + ppos, block);
568                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
569                 path[ppos].p_depth = i;
570                 path[ppos].p_ext = NULL;
571
572                 bh = sb_bread(inode->i_sb, path[ppos].p_block);
573                 if (!bh)
574                         goto err;
575
576                 eh = ext_block_hdr(bh);
577                 ppos++;
578                 BUG_ON(ppos > depth);
579                 path[ppos].p_bh = bh;
580                 path[ppos].p_hdr = eh;
581                 i--;
582
583                 if (ext4_ext_check_header(inode, eh, i))
584                         goto err;
585         }
586
587         path[ppos].p_depth = i;
588         path[ppos].p_ext = NULL;
589         path[ppos].p_idx = NULL;
590
591         /* find extent */
592         ext4_ext_binsearch(inode, path + ppos, block);
593         /* if not an empty leaf */
594         if (path[ppos].p_ext)
595                 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
596
597         ext4_ext_show_path(inode, path);
598
599         return path;
600
601 err:
602         ext4_ext_drop_refs(path);
603         if (alloc)
604                 kfree(path);
605         return ERR_PTR(-EIO);
606 }
607
608 /*
609  * ext4_ext_insert_index:
610  * insert new index [@logical;@ptr] into the block at @curp;
611  * check where to insert: before @curp or after @curp
612  */
613 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
614                                 struct ext4_ext_path *curp,
615                                 int logical, ext4_fsblk_t ptr)
616 {
617         struct ext4_extent_idx *ix;
618         int len, err;
619
620         err = ext4_ext_get_access(handle, inode, curp);
621         if (err)
622                 return err;
623
624         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
625         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
626         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
627                 /* insert after */
628                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
629                         len = (len - 1) * sizeof(struct ext4_extent_idx);
630                         len = len < 0 ? 0 : len;
631                         ext_debug("insert new index %d after: %llu. "
632                                         "move %d from 0x%p to 0x%p\n",
633                                         logical, ptr, len,
634                                         (curp->p_idx + 1), (curp->p_idx + 2));
635                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
636                 }
637                 ix = curp->p_idx + 1;
638         } else {
639                 /* insert before */
640                 len = len * sizeof(struct ext4_extent_idx);
641                 len = len < 0 ? 0 : len;
642                 ext_debug("insert new index %d before: %llu. "
643                                 "move %d from 0x%p to 0x%p\n",
644                                 logical, ptr, len,
645                                 curp->p_idx, (curp->p_idx + 1));
646                 memmove(curp->p_idx + 1, curp->p_idx, len);
647                 ix = curp->p_idx;
648         }
649
650         ix->ei_block = cpu_to_le32(logical);
651         ext4_idx_store_pblock(ix, ptr);
652         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
653
654         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
655                              > le16_to_cpu(curp->p_hdr->eh_max));
656         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
657
658         err = ext4_ext_dirty(handle, inode, curp);
659         ext4_std_error(inode->i_sb, err);
660
661         return err;
662 }
663
664 /*
665  * ext4_ext_split:
666  * inserts new subtree into the path, using free index entry
667  * at depth @at:
668  * - allocates all needed blocks (new leaf and all intermediate index blocks)
669  * - makes decision where to split
670  * - moves remaining extents and index entries (right to the split point)
671  *   into the newly allocated blocks
672  * - initializes subtree
673  */
674 static int ext4_ext_split(handle_t *handle, struct inode *inode,
675                                 struct ext4_ext_path *path,
676                                 struct ext4_extent *newext, int at)
677 {
678         struct buffer_head *bh = NULL;
679         int depth = ext_depth(inode);
680         struct ext4_extent_header *neh;
681         struct ext4_extent_idx *fidx;
682         struct ext4_extent *ex;
683         int i = at, k, m, a;
684         ext4_fsblk_t newblock, oldblock;
685         __le32 border;
686         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
687         int err = 0;
688
689         /* make decision: where to split? */
690         /* FIXME: now decision is simplest: at current extent */
691
692         /* if current leaf will be split, then we should use
693          * border from split point */
694         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
695         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
696                 border = path[depth].p_ext[1].ee_block;
697                 ext_debug("leaf will be split."
698                                 " next leaf starts at %d\n",
699                                   le32_to_cpu(border));
700         } else {
701                 border = newext->ee_block;
702                 ext_debug("leaf will be added."
703                                 " next leaf starts at %d\n",
704                                 le32_to_cpu(border));
705         }
706
707         /*
708          * If error occurs, then we break processing
709          * and mark filesystem read-only. index won't
710          * be inserted and tree will be in consistent
711          * state. Next mount will repair buffers too.
712          */
713
714         /*
715          * Get array to track all allocated blocks.
716          * We need this to handle errors and free blocks
717          * upon them.
718          */
719         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
720         if (!ablocks)
721                 return -ENOMEM;
722
723         /* allocate all needed blocks */
724         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
725         for (a = 0; a < depth - at; a++) {
726                 newblock = ext4_ext_new_meta_block(handle, inode, path,
727                                                    newext, &err);
728                 if (newblock == 0)
729                         goto cleanup;
730                 ablocks[a] = newblock;
731         }
732
733         /* initialize new leaf */
734         newblock = ablocks[--a];
735         BUG_ON(newblock == 0);
736         bh = sb_getblk(inode->i_sb, newblock);
737         if (!bh) {
738                 err = -EIO;
739                 goto cleanup;
740         }
741         lock_buffer(bh);
742
743         err = ext4_journal_get_create_access(handle, bh);
744         if (err)
745                 goto cleanup;
746
747         neh = ext_block_hdr(bh);
748         neh->eh_entries = 0;
749         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
750         neh->eh_magic = EXT4_EXT_MAGIC;
751         neh->eh_depth = 0;
752         ex = EXT_FIRST_EXTENT(neh);
753
754         /* move remainder of path[depth] to the new leaf */
755         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
756         /* start copy from next extent */
757         /* TODO: we could do it by single memmove */
758         m = 0;
759         path[depth].p_ext++;
760         while (path[depth].p_ext <=
761                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
762                 ext_debug("move %d:%llu:%d in new leaf %llu\n",
763                                 le32_to_cpu(path[depth].p_ext->ee_block),
764                                 ext_pblock(path[depth].p_ext),
765                                 ext4_ext_get_actual_len(path[depth].p_ext),
766                                 newblock);
767                 /*memmove(ex++, path[depth].p_ext++,
768                                 sizeof(struct ext4_extent));
769                 neh->eh_entries++;*/
770                 path[depth].p_ext++;
771                 m++;
772         }
773         if (m) {
774                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
775                 le16_add_cpu(&neh->eh_entries, m);
776         }
777
778         set_buffer_uptodate(bh);
779         unlock_buffer(bh);
780
781         err = ext4_journal_dirty_metadata(handle, bh);
782         if (err)
783                 goto cleanup;
784         brelse(bh);
785         bh = NULL;
786
787         /* correct old leaf */
788         if (m) {
789                 err = ext4_ext_get_access(handle, inode, path + depth);
790                 if (err)
791                         goto cleanup;
792                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
793                 err = ext4_ext_dirty(handle, inode, path + depth);
794                 if (err)
795                         goto cleanup;
796
797         }
798
799         /* create intermediate indexes */
800         k = depth - at - 1;
801         BUG_ON(k < 0);
802         if (k)
803                 ext_debug("create %d intermediate indices\n", k);
804         /* insert new index into current index block */
805         /* current depth stored in i var */
806         i = depth - 1;
807         while (k--) {
808                 oldblock = newblock;
809                 newblock = ablocks[--a];
810                 bh = sb_getblk(inode->i_sb, newblock);
811                 if (!bh) {
812                         err = -EIO;
813                         goto cleanup;
814                 }
815                 lock_buffer(bh);
816
817                 err = ext4_journal_get_create_access(handle, bh);
818                 if (err)
819                         goto cleanup;
820
821                 neh = ext_block_hdr(bh);
822                 neh->eh_entries = cpu_to_le16(1);
823                 neh->eh_magic = EXT4_EXT_MAGIC;
824                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
825                 neh->eh_depth = cpu_to_le16(depth - i);
826                 fidx = EXT_FIRST_INDEX(neh);
827                 fidx->ei_block = border;
828                 ext4_idx_store_pblock(fidx, oldblock);
829
830                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
831                                 i, newblock, le32_to_cpu(border), oldblock);
832                 /* copy indexes */
833                 m = 0;
834                 path[i].p_idx++;
835
836                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
837                                 EXT_MAX_INDEX(path[i].p_hdr));
838                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
839                                 EXT_LAST_INDEX(path[i].p_hdr));
840                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
841                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
842                                         le32_to_cpu(path[i].p_idx->ei_block),
843                                         idx_pblock(path[i].p_idx),
844                                         newblock);
845                         /*memmove(++fidx, path[i].p_idx++,
846                                         sizeof(struct ext4_extent_idx));
847                         neh->eh_entries++;
848                         BUG_ON(neh->eh_entries > neh->eh_max);*/
849                         path[i].p_idx++;
850                         m++;
851                 }
852                 if (m) {
853                         memmove(++fidx, path[i].p_idx - m,
854                                 sizeof(struct ext4_extent_idx) * m);
855                         le16_add_cpu(&neh->eh_entries, m);
856                 }
857                 set_buffer_uptodate(bh);
858                 unlock_buffer(bh);
859
860                 err = ext4_journal_dirty_metadata(handle, bh);
861                 if (err)
862                         goto cleanup;
863                 brelse(bh);
864                 bh = NULL;
865
866                 /* correct old index */
867                 if (m) {
868                         err = ext4_ext_get_access(handle, inode, path + i);
869                         if (err)
870                                 goto cleanup;
871                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
872                         err = ext4_ext_dirty(handle, inode, path + i);
873                         if (err)
874                                 goto cleanup;
875                 }
876
877                 i--;
878         }
879
880         /* insert new index */
881         err = ext4_ext_insert_index(handle, inode, path + at,
882                                     le32_to_cpu(border), newblock);
883
884 cleanup:
885         if (bh) {
886                 if (buffer_locked(bh))
887                         unlock_buffer(bh);
888                 brelse(bh);
889         }
890
891         if (err) {
892                 /* free all allocated blocks in error case */
893                 for (i = 0; i < depth; i++) {
894                         if (!ablocks[i])
895                                 continue;
896                         ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
897                 }
898         }
899         kfree(ablocks);
900
901         return err;
902 }
903
904 /*
905  * ext4_ext_grow_indepth:
906  * implements tree growing procedure:
907  * - allocates new block
908  * - moves top-level data (index block or leaf) into the new block
909  * - initializes new top-level, creating index that points to the
910  *   just created block
911  */
912 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
913                                         struct ext4_ext_path *path,
914                                         struct ext4_extent *newext)
915 {
916         struct ext4_ext_path *curp = path;
917         struct ext4_extent_header *neh;
918         struct ext4_extent_idx *fidx;
919         struct buffer_head *bh;
920         ext4_fsblk_t newblock;
921         int err = 0;
922
923         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
924         if (newblock == 0)
925                 return err;
926
927         bh = sb_getblk(inode->i_sb, newblock);
928         if (!bh) {
929                 err = -EIO;
930                 ext4_std_error(inode->i_sb, err);
931                 return err;
932         }
933         lock_buffer(bh);
934
935         err = ext4_journal_get_create_access(handle, bh);
936         if (err) {
937                 unlock_buffer(bh);
938                 goto out;
939         }
940
941         /* move top-level index/leaf into new block */
942         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
943
944         /* set size of new block */
945         neh = ext_block_hdr(bh);
946         /* old root could have indexes or leaves
947          * so calculate e_max right way */
948         if (ext_depth(inode))
949           neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
950         else
951           neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
952         neh->eh_magic = EXT4_EXT_MAGIC;
953         set_buffer_uptodate(bh);
954         unlock_buffer(bh);
955
956         err = ext4_journal_dirty_metadata(handle, bh);
957         if (err)
958                 goto out;
959
960         /* create index in new top-level index: num,max,pointer */
961         err = ext4_ext_get_access(handle, inode, curp);
962         if (err)
963                 goto out;
964
965         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
966         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
967         curp->p_hdr->eh_entries = cpu_to_le16(1);
968         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
969
970         if (path[0].p_hdr->eh_depth)
971                 curp->p_idx->ei_block =
972                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
973         else
974                 curp->p_idx->ei_block =
975                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
976         ext4_idx_store_pblock(curp->p_idx, newblock);
977
978         neh = ext_inode_hdr(inode);
979         fidx = EXT_FIRST_INDEX(neh);
980         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
981                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
982                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
983
984         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
985         err = ext4_ext_dirty(handle, inode, curp);
986 out:
987         brelse(bh);
988
989         return err;
990 }
991
992 /*
993  * ext4_ext_create_new_leaf:
994  * finds empty index and adds new leaf.
995  * if no free index is found, then it requests in-depth growing.
996  */
997 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
998                                         struct ext4_ext_path *path,
999                                         struct ext4_extent *newext)
1000 {
1001         struct ext4_ext_path *curp;
1002         int depth, i, err = 0;
1003
1004 repeat:
1005         i = depth = ext_depth(inode);
1006
1007         /* walk up to the tree and look for free index entry */
1008         curp = path + depth;
1009         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1010                 i--;
1011                 curp--;
1012         }
1013
1014         /* we use already allocated block for index block,
1015          * so subsequent data blocks should be contiguous */
1016         if (EXT_HAS_FREE_INDEX(curp)) {
1017                 /* if we found index with free entry, then use that
1018                  * entry: create all needed subtree and add new leaf */
1019                 err = ext4_ext_split(handle, inode, path, newext, i);
1020                 if (err)
1021                         goto out;
1022
1023                 /* refill path */
1024                 ext4_ext_drop_refs(path);
1025                 path = ext4_ext_find_extent(inode,
1026                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1027                                     path);
1028                 if (IS_ERR(path))
1029                         err = PTR_ERR(path);
1030         } else {
1031                 /* tree is full, time to grow in depth */
1032                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1033                 if (err)
1034                         goto out;
1035
1036                 /* refill path */
1037                 ext4_ext_drop_refs(path);
1038                 path = ext4_ext_find_extent(inode,
1039                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1040                                     path);
1041                 if (IS_ERR(path)) {
1042                         err = PTR_ERR(path);
1043                         goto out;
1044                 }
1045
1046                 /*
1047                  * only first (depth 0 -> 1) produces free space;
1048                  * in all other cases we have to split the grown tree
1049                  */
1050                 depth = ext_depth(inode);
1051                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1052                         /* now we need to split */
1053                         goto repeat;
1054                 }
1055         }
1056
1057 out:
1058         return err;
1059 }
1060
1061 /*
1062  * search the closest allocated block to the left for *logical
1063  * and returns it at @logical + it's physical address at @phys
1064  * if *logical is the smallest allocated block, the function
1065  * returns 0 at @phys
1066  * return value contains 0 (success) or error code
1067  */
1068 int
1069 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1070                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1071 {
1072         struct ext4_extent_idx *ix;
1073         struct ext4_extent *ex;
1074         int depth, ee_len;
1075
1076         BUG_ON(path == NULL);
1077         depth = path->p_depth;
1078         *phys = 0;
1079
1080         if (depth == 0 && path->p_ext == NULL)
1081                 return 0;
1082
1083         /* usually extent in the path covers blocks smaller
1084          * then *logical, but it can be that extent is the
1085          * first one in the file */
1086
1087         ex = path[depth].p_ext;
1088         ee_len = ext4_ext_get_actual_len(ex);
1089         if (*logical < le32_to_cpu(ex->ee_block)) {
1090                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1091                 while (--depth >= 0) {
1092                         ix = path[depth].p_idx;
1093                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1094                 }
1095                 return 0;
1096         }
1097
1098         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1099
1100         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1101         *phys = ext_pblock(ex) + ee_len - 1;
1102         return 0;
1103 }
1104
1105 /*
1106  * search the closest allocated block to the right for *logical
1107  * and returns it at @logical + it's physical address at @phys
1108  * if *logical is the smallest allocated block, the function
1109  * returns 0 at @phys
1110  * return value contains 0 (success) or error code
1111  */
1112 int
1113 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1114                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1115 {
1116         struct buffer_head *bh = NULL;
1117         struct ext4_extent_header *eh;
1118         struct ext4_extent_idx *ix;
1119         struct ext4_extent *ex;
1120         ext4_fsblk_t block;
1121         int depth, ee_len;
1122
1123         BUG_ON(path == NULL);
1124         depth = path->p_depth;
1125         *phys = 0;
1126
1127         if (depth == 0 && path->p_ext == NULL)
1128                 return 0;
1129
1130         /* usually extent in the path covers blocks smaller
1131          * then *logical, but it can be that extent is the
1132          * first one in the file */
1133
1134         ex = path[depth].p_ext;
1135         ee_len = ext4_ext_get_actual_len(ex);
1136         if (*logical < le32_to_cpu(ex->ee_block)) {
1137                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1138                 while (--depth >= 0) {
1139                         ix = path[depth].p_idx;
1140                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1141                 }
1142                 *logical = le32_to_cpu(ex->ee_block);
1143                 *phys = ext_pblock(ex);
1144                 return 0;
1145         }
1146
1147         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1148
1149         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1150                 /* next allocated block in this leaf */
1151                 ex++;
1152                 *logical = le32_to_cpu(ex->ee_block);
1153                 *phys = ext_pblock(ex);
1154                 return 0;
1155         }
1156
1157         /* go up and search for index to the right */
1158         while (--depth >= 0) {
1159                 ix = path[depth].p_idx;
1160                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1161                         break;
1162         }
1163
1164         if (depth < 0) {
1165                 /* we've gone up to the root and
1166                  * found no index to the right */
1167                 return 0;
1168         }
1169
1170         /* we've found index to the right, let's
1171          * follow it and find the closest allocated
1172          * block to the right */
1173         ix++;
1174         block = idx_pblock(ix);
1175         while (++depth < path->p_depth) {
1176                 bh = sb_bread(inode->i_sb, block);
1177                 if (bh == NULL)
1178                         return -EIO;
1179                 eh = ext_block_hdr(bh);
1180                 if (ext4_ext_check_header(inode, eh, depth)) {
1181                         put_bh(bh);
1182                         return -EIO;
1183                 }
1184                 ix = EXT_FIRST_INDEX(eh);
1185                 block = idx_pblock(ix);
1186                 put_bh(bh);
1187         }
1188
1189         bh = sb_bread(inode->i_sb, block);
1190         if (bh == NULL)
1191                 return -EIO;
1192         eh = ext_block_hdr(bh);
1193         if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1194                 put_bh(bh);
1195                 return -EIO;
1196         }
1197         ex = EXT_FIRST_EXTENT(eh);
1198         *logical = le32_to_cpu(ex->ee_block);
1199         *phys = ext_pblock(ex);
1200         put_bh(bh);
1201         return 0;
1202
1203 }
1204
1205 /*
1206  * ext4_ext_next_allocated_block:
1207  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1208  * NOTE: it considers block number from index entry as
1209  * allocated block. Thus, index entries have to be consistent
1210  * with leaves.
1211  */
1212 static ext4_lblk_t
1213 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1214 {
1215         int depth;
1216
1217         BUG_ON(path == NULL);
1218         depth = path->p_depth;
1219
1220         if (depth == 0 && path->p_ext == NULL)
1221                 return EXT_MAX_BLOCK;
1222
1223         while (depth >= 0) {
1224                 if (depth == path->p_depth) {
1225                         /* leaf */
1226                         if (path[depth].p_ext !=
1227                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1228                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1229                 } else {
1230                         /* index */
1231                         if (path[depth].p_idx !=
1232                                         EXT_LAST_INDEX(path[depth].p_hdr))
1233                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1234                 }
1235                 depth--;
1236         }
1237
1238         return EXT_MAX_BLOCK;
1239 }
1240
1241 /*
1242  * ext4_ext_next_leaf_block:
1243  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1244  */
1245 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1246                                         struct ext4_ext_path *path)
1247 {
1248         int depth;
1249
1250         BUG_ON(path == NULL);
1251         depth = path->p_depth;
1252
1253         /* zero-tree has no leaf blocks at all */
1254         if (depth == 0)
1255                 return EXT_MAX_BLOCK;
1256
1257         /* go to index block */
1258         depth--;
1259
1260         while (depth >= 0) {
1261                 if (path[depth].p_idx !=
1262                                 EXT_LAST_INDEX(path[depth].p_hdr))
1263                         return (ext4_lblk_t)
1264                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1265                 depth--;
1266         }
1267
1268         return EXT_MAX_BLOCK;
1269 }
1270
1271 /*
1272  * ext4_ext_correct_indexes:
1273  * if leaf gets modified and modified extent is first in the leaf,
1274  * then we have to correct all indexes above.
1275  * TODO: do we need to correct tree in all cases?
1276  */
1277 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1278                                 struct ext4_ext_path *path)
1279 {
1280         struct ext4_extent_header *eh;
1281         int depth = ext_depth(inode);
1282         struct ext4_extent *ex;
1283         __le32 border;
1284         int k, err = 0;
1285
1286         eh = path[depth].p_hdr;
1287         ex = path[depth].p_ext;
1288         BUG_ON(ex == NULL);
1289         BUG_ON(eh == NULL);
1290
1291         if (depth == 0) {
1292                 /* there is no tree at all */
1293                 return 0;
1294         }
1295
1296         if (ex != EXT_FIRST_EXTENT(eh)) {
1297                 /* we correct tree if first leaf got modified only */
1298                 return 0;
1299         }
1300
1301         /*
1302          * TODO: we need correction if border is smaller than current one
1303          */
1304         k = depth - 1;
1305         border = path[depth].p_ext->ee_block;
1306         err = ext4_ext_get_access(handle, inode, path + k);
1307         if (err)
1308                 return err;
1309         path[k].p_idx->ei_block = border;
1310         err = ext4_ext_dirty(handle, inode, path + k);
1311         if (err)
1312                 return err;
1313
1314         while (k--) {
1315                 /* change all left-side indexes */
1316                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1317                         break;
1318                 err = ext4_ext_get_access(handle, inode, path + k);
1319                 if (err)
1320                         break;
1321                 path[k].p_idx->ei_block = border;
1322                 err = ext4_ext_dirty(handle, inode, path + k);
1323                 if (err)
1324                         break;
1325         }
1326
1327         return err;
1328 }
1329
1330 static int
1331 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1332                                 struct ext4_extent *ex2)
1333 {
1334         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1335
1336         /*
1337          * Make sure that either both extents are uninitialized, or
1338          * both are _not_.
1339          */
1340         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1341                 return 0;
1342
1343         if (ext4_ext_is_uninitialized(ex1))
1344                 max_len = EXT_UNINIT_MAX_LEN;
1345         else
1346                 max_len = EXT_INIT_MAX_LEN;
1347
1348         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1349         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1350
1351         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1352                         le32_to_cpu(ex2->ee_block))
1353                 return 0;
1354
1355         /*
1356          * To allow future support for preallocated extents to be added
1357          * as an RO_COMPAT feature, refuse to merge to extents if
1358          * this can result in the top bit of ee_len being set.
1359          */
1360         if (ext1_ee_len + ext2_ee_len > max_len)
1361                 return 0;
1362 #ifdef AGGRESSIVE_TEST
1363         if (ext1_ee_len >= 4)
1364                 return 0;
1365 #endif
1366
1367         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1368                 return 1;
1369         return 0;
1370 }
1371
1372 /*
1373  * This function tries to merge the "ex" extent to the next extent in the tree.
1374  * It always tries to merge towards right. If you want to merge towards
1375  * left, pass "ex - 1" as argument instead of "ex".
1376  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1377  * 1 if they got merged.
1378  */
1379 int ext4_ext_try_to_merge(struct inode *inode,
1380                           struct ext4_ext_path *path,
1381                           struct ext4_extent *ex)
1382 {
1383         struct ext4_extent_header *eh;
1384         unsigned int depth, len;
1385         int merge_done = 0;
1386         int uninitialized = 0;
1387
1388         depth = ext_depth(inode);
1389         BUG_ON(path[depth].p_hdr == NULL);
1390         eh = path[depth].p_hdr;
1391
1392         while (ex < EXT_LAST_EXTENT(eh)) {
1393                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1394                         break;
1395                 /* merge with next extent! */
1396                 if (ext4_ext_is_uninitialized(ex))
1397                         uninitialized = 1;
1398                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1399                                 + ext4_ext_get_actual_len(ex + 1));
1400                 if (uninitialized)
1401                         ext4_ext_mark_uninitialized(ex);
1402
1403                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1404                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1405                                 * sizeof(struct ext4_extent);
1406                         memmove(ex + 1, ex + 2, len);
1407                 }
1408                 le16_add_cpu(&eh->eh_entries, -1);
1409                 merge_done = 1;
1410                 WARN_ON(eh->eh_entries == 0);
1411                 if (!eh->eh_entries)
1412                         ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1413                            "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1414         }
1415
1416         return merge_done;
1417 }
1418
1419 /*
1420  * check if a portion of the "newext" extent overlaps with an
1421  * existing extent.
1422  *
1423  * If there is an overlap discovered, it updates the length of the newext
1424  * such that there will be no overlap, and then returns 1.
1425  * If there is no overlap found, it returns 0.
1426  */
1427 unsigned int ext4_ext_check_overlap(struct inode *inode,
1428                                     struct ext4_extent *newext,
1429                                     struct ext4_ext_path *path)
1430 {
1431         ext4_lblk_t b1, b2;
1432         unsigned int depth, len1;
1433         unsigned int ret = 0;
1434
1435         b1 = le32_to_cpu(newext->ee_block);
1436         len1 = ext4_ext_get_actual_len(newext);
1437         depth = ext_depth(inode);
1438         if (!path[depth].p_ext)
1439                 goto out;
1440         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1441
1442         /*
1443          * get the next allocated block if the extent in the path
1444          * is before the requested block(s)
1445          */
1446         if (b2 < b1) {
1447                 b2 = ext4_ext_next_allocated_block(path);
1448                 if (b2 == EXT_MAX_BLOCK)
1449                         goto out;
1450         }
1451
1452         /* check for wrap through zero on extent logical start block*/
1453         if (b1 + len1 < b1) {
1454                 len1 = EXT_MAX_BLOCK - b1;
1455                 newext->ee_len = cpu_to_le16(len1);
1456                 ret = 1;
1457         }
1458
1459         /* check for overlap */
1460         if (b1 + len1 > b2) {
1461                 newext->ee_len = cpu_to_le16(b2 - b1);
1462                 ret = 1;
1463         }
1464 out:
1465         return ret;
1466 }
1467
1468 /*
1469  * ext4_ext_insert_extent:
1470  * tries to merge requsted extent into the existing extent or
1471  * inserts requested extent as new one into the tree,
1472  * creating new leaf in the no-space case.
1473  */
1474 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1475                                 struct ext4_ext_path *path,
1476                                 struct ext4_extent *newext)
1477 {
1478         struct ext4_extent_header * eh;
1479         struct ext4_extent *ex, *fex;
1480         struct ext4_extent *nearex; /* nearest extent */
1481         struct ext4_ext_path *npath = NULL;
1482         int depth, len, err;
1483         ext4_lblk_t next;
1484         unsigned uninitialized = 0;
1485
1486         BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1487         depth = ext_depth(inode);
1488         ex = path[depth].p_ext;
1489         BUG_ON(path[depth].p_hdr == NULL);
1490
1491         /* try to insert block into found extent and return */
1492         if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1493                 ext_debug("append %d block to %d:%d (from %llu)\n",
1494                                 ext4_ext_get_actual_len(newext),
1495                                 le32_to_cpu(ex->ee_block),
1496                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1497                 err = ext4_ext_get_access(handle, inode, path + depth);
1498                 if (err)
1499                         return err;
1500
1501                 /*
1502                  * ext4_can_extents_be_merged should have checked that either
1503                  * both extents are uninitialized, or both aren't. Thus we
1504                  * need to check only one of them here.
1505                  */
1506                 if (ext4_ext_is_uninitialized(ex))
1507                         uninitialized = 1;
1508                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1509                                         + ext4_ext_get_actual_len(newext));
1510                 if (uninitialized)
1511                         ext4_ext_mark_uninitialized(ex);
1512                 eh = path[depth].p_hdr;
1513                 nearex = ex;
1514                 goto merge;
1515         }
1516
1517 repeat:
1518         depth = ext_depth(inode);
1519         eh = path[depth].p_hdr;
1520         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1521                 goto has_space;
1522
1523         /* probably next leaf has space for us? */
1524         fex = EXT_LAST_EXTENT(eh);
1525         next = ext4_ext_next_leaf_block(inode, path);
1526         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1527             && next != EXT_MAX_BLOCK) {
1528                 ext_debug("next leaf block - %d\n", next);
1529                 BUG_ON(npath != NULL);
1530                 npath = ext4_ext_find_extent(inode, next, NULL);
1531                 if (IS_ERR(npath))
1532                         return PTR_ERR(npath);
1533                 BUG_ON(npath->p_depth != path->p_depth);
1534                 eh = npath[depth].p_hdr;
1535                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1536                         ext_debug("next leaf isnt full(%d)\n",
1537                                   le16_to_cpu(eh->eh_entries));
1538                         path = npath;
1539                         goto repeat;
1540                 }
1541                 ext_debug("next leaf has no free space(%d,%d)\n",
1542                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1543         }
1544
1545         /*
1546          * There is no free space in the found leaf.
1547          * We're gonna add a new leaf in the tree.
1548          */
1549         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1550         if (err)
1551                 goto cleanup;
1552         depth = ext_depth(inode);
1553         eh = path[depth].p_hdr;
1554
1555 has_space:
1556         nearex = path[depth].p_ext;
1557
1558         err = ext4_ext_get_access(handle, inode, path + depth);
1559         if (err)
1560                 goto cleanup;
1561
1562         if (!nearex) {
1563                 /* there is no extent in this leaf, create first one */
1564                 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1565                                 le32_to_cpu(newext->ee_block),
1566                                 ext_pblock(newext),
1567                                 ext4_ext_get_actual_len(newext));
1568                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1569         } else if (le32_to_cpu(newext->ee_block)
1570                            > le32_to_cpu(nearex->ee_block)) {
1571 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1572                 if (nearex != EXT_LAST_EXTENT(eh)) {
1573                         len = EXT_MAX_EXTENT(eh) - nearex;
1574                         len = (len - 1) * sizeof(struct ext4_extent);
1575                         len = len < 0 ? 0 : len;
1576                         ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1577                                         "move %d from 0x%p to 0x%p\n",
1578                                         le32_to_cpu(newext->ee_block),
1579                                         ext_pblock(newext),
1580                                         ext4_ext_get_actual_len(newext),
1581                                         nearex, len, nearex + 1, nearex + 2);
1582                         memmove(nearex + 2, nearex + 1, len);
1583                 }
1584                 path[depth].p_ext = nearex + 1;
1585         } else {
1586                 BUG_ON(newext->ee_block == nearex->ee_block);
1587                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1588                 len = len < 0 ? 0 : len;
1589                 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1590                                 "move %d from 0x%p to 0x%p\n",
1591                                 le32_to_cpu(newext->ee_block),
1592                                 ext_pblock(newext),
1593                                 ext4_ext_get_actual_len(newext),
1594                                 nearex, len, nearex + 1, nearex + 2);
1595                 memmove(nearex + 1, nearex, len);
1596                 path[depth].p_ext = nearex;
1597         }
1598
1599         le16_add_cpu(&eh->eh_entries, 1);
1600         nearex = path[depth].p_ext;
1601         nearex->ee_block = newext->ee_block;
1602         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1603         nearex->ee_len = newext->ee_len;
1604
1605 merge:
1606         /* try to merge extents to the right */
1607         ext4_ext_try_to_merge(inode, path, nearex);
1608
1609         /* try to merge extents to the left */
1610
1611         /* time to correct all indexes above */
1612         err = ext4_ext_correct_indexes(handle, inode, path);
1613         if (err)
1614                 goto cleanup;
1615
1616         err = ext4_ext_dirty(handle, inode, path + depth);
1617
1618 cleanup:
1619         if (npath) {
1620                 ext4_ext_drop_refs(npath);
1621                 kfree(npath);
1622         }
1623         ext4_ext_tree_changed(inode);
1624         ext4_ext_invalidate_cache(inode);
1625         return err;
1626 }
1627
1628 static void
1629 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1630                         __u32 len, ext4_fsblk_t start, int type)
1631 {
1632         struct ext4_ext_cache *cex;
1633         BUG_ON(len == 0);
1634         cex = &EXT4_I(inode)->i_cached_extent;
1635         cex->ec_type = type;
1636         cex->ec_block = block;
1637         cex->ec_len = len;
1638         cex->ec_start = start;
1639 }
1640
1641 /*
1642  * ext4_ext_put_gap_in_cache:
1643  * calculate boundaries of the gap that the requested block fits into
1644  * and cache this gap
1645  */
1646 static void
1647 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1648                                 ext4_lblk_t block)
1649 {
1650         int depth = ext_depth(inode);
1651         unsigned long len;
1652         ext4_lblk_t lblock;
1653         struct ext4_extent *ex;
1654
1655         ex = path[depth].p_ext;
1656         if (ex == NULL) {
1657                 /* there is no extent yet, so gap is [0;-] */
1658                 lblock = 0;
1659                 len = EXT_MAX_BLOCK;
1660                 ext_debug("cache gap(whole file):");
1661         } else if (block < le32_to_cpu(ex->ee_block)) {
1662                 lblock = block;
1663                 len = le32_to_cpu(ex->ee_block) - block;
1664                 ext_debug("cache gap(before): %u [%u:%u]",
1665                                 block,
1666                                 le32_to_cpu(ex->ee_block),
1667                                  ext4_ext_get_actual_len(ex));
1668         } else if (block >= le32_to_cpu(ex->ee_block)
1669                         + ext4_ext_get_actual_len(ex)) {
1670                 ext4_lblk_t next;
1671                 lblock = le32_to_cpu(ex->ee_block)
1672                         + ext4_ext_get_actual_len(ex);
1673
1674                 next = ext4_ext_next_allocated_block(path);
1675                 ext_debug("cache gap(after): [%u:%u] %u",
1676                                 le32_to_cpu(ex->ee_block),
1677                                 ext4_ext_get_actual_len(ex),
1678                                 block);
1679                 BUG_ON(next == lblock);
1680                 len = next - lblock;
1681         } else {
1682                 lblock = len = 0;
1683                 BUG();
1684         }
1685
1686         ext_debug(" -> %u:%lu\n", lblock, len);
1687         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1688 }
1689
1690 static int
1691 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1692                         struct ext4_extent *ex)
1693 {
1694         struct ext4_ext_cache *cex;
1695
1696         cex = &EXT4_I(inode)->i_cached_extent;
1697
1698         /* has cache valid data? */
1699         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1700                 return EXT4_EXT_CACHE_NO;
1701
1702         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1703                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1704         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1705                 ex->ee_block = cpu_to_le32(cex->ec_block);
1706                 ext4_ext_store_pblock(ex, cex->ec_start);
1707                 ex->ee_len = cpu_to_le16(cex->ec_len);
1708                 ext_debug("%u cached by %u:%u:%llu\n",
1709                                 block,
1710                                 cex->ec_block, cex->ec_len, cex->ec_start);
1711                 return cex->ec_type;
1712         }
1713
1714         /* not in cache */
1715         return EXT4_EXT_CACHE_NO;
1716 }
1717
1718 /*
1719  * ext4_ext_rm_idx:
1720  * removes index from the index block.
1721  * It's used in truncate case only, thus all requests are for
1722  * last index in the block only.
1723  */
1724 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1725                         struct ext4_ext_path *path)
1726 {
1727         struct buffer_head *bh;
1728         int err;
1729         ext4_fsblk_t leaf;
1730
1731         /* free index block */
1732         path--;
1733         leaf = idx_pblock(path->p_idx);
1734         BUG_ON(path->p_hdr->eh_entries == 0);
1735         err = ext4_ext_get_access(handle, inode, path);
1736         if (err)
1737                 return err;
1738         le16_add_cpu(&path->p_hdr->eh_entries, -1);
1739         err = ext4_ext_dirty(handle, inode, path);
1740         if (err)
1741                 return err;
1742         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1743         bh = sb_find_get_block(inode->i_sb, leaf);
1744         ext4_forget(handle, 1, inode, bh, leaf);
1745         ext4_free_blocks(handle, inode, leaf, 1, 1);
1746         return err;
1747 }
1748
1749 /*
1750  * ext4_ext_calc_credits_for_insert:
1751  * This routine returns max. credits that the extent tree can consume.
1752  * It should be OK for low-performance paths like ->writepage()
1753  * To allow many writing processes to fit into a single transaction,
1754  * the caller should calculate credits under i_data_sem and
1755  * pass the actual path.
1756  */
1757 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1758                                                 struct ext4_ext_path *path)
1759 {
1760         int depth, needed;
1761
1762         if (path) {
1763                 /* probably there is space in leaf? */
1764                 depth = ext_depth(inode);
1765                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1766                                 < le16_to_cpu(path[depth].p_hdr->eh_max))
1767                         return 1;
1768         }
1769
1770         /*
1771          * given 32-bit logical block (4294967296 blocks), max. tree
1772          * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1773          * Let's also add one more level for imbalance.
1774          */
1775         depth = 5;
1776
1777         /* allocation of new data block(s) */
1778         needed = 2;
1779
1780         /*
1781          * tree can be full, so it would need to grow in depth:
1782          * we need one credit to modify old root, credits for
1783          * new root will be added in split accounting
1784          */
1785         needed += 1;
1786
1787         /*
1788          * Index split can happen, we would need:
1789          *    allocate intermediate indexes (bitmap + group)
1790          *  + change two blocks at each level, but root (already included)
1791          */
1792         needed += (depth * 2) + (depth * 2);
1793
1794         /* any allocation modifies superblock */
1795         needed += 1;
1796
1797         return needed;
1798 }
1799
1800 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1801                                 struct ext4_extent *ex,
1802                                 ext4_lblk_t from, ext4_lblk_t to)
1803 {
1804         struct buffer_head *bh;
1805         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
1806         int i, metadata = 0;
1807
1808         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1809                 metadata = 1;
1810 #ifdef EXTENTS_STATS
1811         {
1812                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1813                 spin_lock(&sbi->s_ext_stats_lock);
1814                 sbi->s_ext_blocks += ee_len;
1815                 sbi->s_ext_extents++;
1816                 if (ee_len < sbi->s_ext_min)
1817                         sbi->s_ext_min = ee_len;
1818                 if (ee_len > sbi->s_ext_max)
1819                         sbi->s_ext_max = ee_len;
1820                 if (ext_depth(inode) > sbi->s_depth_max)
1821                         sbi->s_depth_max = ext_depth(inode);
1822                 spin_unlock(&sbi->s_ext_stats_lock);
1823         }
1824 #endif
1825         if (from >= le32_to_cpu(ex->ee_block)
1826             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
1827                 /* tail removal */
1828                 ext4_lblk_t num;
1829                 ext4_fsblk_t start;
1830
1831                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1832                 start = ext_pblock(ex) + ee_len - num;
1833                 ext_debug("free last %u blocks starting %llu\n", num, start);
1834                 for (i = 0; i < num; i++) {
1835                         bh = sb_find_get_block(inode->i_sb, start + i);
1836                         ext4_forget(handle, 0, inode, bh, start + i);
1837                 }
1838                 ext4_free_blocks(handle, inode, start, num, metadata);
1839         } else if (from == le32_to_cpu(ex->ee_block)
1840                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
1841                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
1842                         from, to, le32_to_cpu(ex->ee_block), ee_len);
1843         } else {
1844                 printk(KERN_INFO "strange request: removal(2) "
1845                                 "%u-%u from %u:%u\n",
1846                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
1847         }
1848         return 0;
1849 }
1850
1851 static int
1852 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1853                 struct ext4_ext_path *path, ext4_lblk_t start)
1854 {
1855         int err = 0, correct_index = 0;
1856         int depth = ext_depth(inode), credits;
1857         struct ext4_extent_header *eh;
1858         ext4_lblk_t a, b, block;
1859         unsigned num;
1860         ext4_lblk_t ex_ee_block;
1861         unsigned short ex_ee_len;
1862         unsigned uninitialized = 0;
1863         struct ext4_extent *ex;
1864
1865         /* the header must be checked already in ext4_ext_remove_space() */
1866         ext_debug("truncate since %u in leaf\n", start);
1867         if (!path[depth].p_hdr)
1868                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1869         eh = path[depth].p_hdr;
1870         BUG_ON(eh == NULL);
1871
1872         /* find where to start removing */
1873         ex = EXT_LAST_EXTENT(eh);
1874
1875         ex_ee_block = le32_to_cpu(ex->ee_block);
1876         if (ext4_ext_is_uninitialized(ex))
1877                 uninitialized = 1;
1878         ex_ee_len = ext4_ext_get_actual_len(ex);
1879
1880         while (ex >= EXT_FIRST_EXTENT(eh) &&
1881                         ex_ee_block + ex_ee_len > start) {
1882                 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1883                 path[depth].p_ext = ex;
1884
1885                 a = ex_ee_block > start ? ex_ee_block : start;
1886                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1887                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1888
1889                 ext_debug("  border %u:%u\n", a, b);
1890
1891                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1892                         block = 0;
1893                         num = 0;
1894                         BUG();
1895                 } else if (a != ex_ee_block) {
1896                         /* remove tail of the extent */
1897                         block = ex_ee_block;
1898                         num = a - block;
1899                 } else if (b != ex_ee_block + ex_ee_len - 1) {
1900                         /* remove head of the extent */
1901                         block = a;
1902                         num = b - a;
1903                         /* there is no "make a hole" API yet */
1904                         BUG();
1905                 } else {
1906                         /* remove whole extent: excellent! */
1907                         block = ex_ee_block;
1908                         num = 0;
1909                         BUG_ON(a != ex_ee_block);
1910                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1911                 }
1912
1913                 /*
1914                  * 3 for leaf, sb, and inode plus 2 (bmap and group
1915                  * descriptor) for each block group; assume two block
1916                  * groups plus ex_ee_len/blocks_per_block_group for
1917                  * the worst case
1918                  */
1919                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
1920                 if (ex == EXT_FIRST_EXTENT(eh)) {
1921                         correct_index = 1;
1922                         credits += (ext_depth(inode)) + 1;
1923                 }
1924 #ifdef CONFIG_QUOTA
1925                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1926 #endif
1927
1928                 err = ext4_ext_journal_restart(handle, credits);
1929                 if (err)
1930                         goto out;
1931
1932                 err = ext4_ext_get_access(handle, inode, path + depth);
1933                 if (err)
1934                         goto out;
1935
1936                 err = ext4_remove_blocks(handle, inode, ex, a, b);
1937                 if (err)
1938                         goto out;
1939
1940                 if (num == 0) {
1941                         /* this extent is removed; mark slot entirely unused */
1942                         ext4_ext_store_pblock(ex, 0);
1943                         le16_add_cpu(&eh->eh_entries, -1);
1944                 }
1945
1946                 ex->ee_block = cpu_to_le32(block);
1947                 ex->ee_len = cpu_to_le16(num);
1948                 /*
1949                  * Do not mark uninitialized if all the blocks in the
1950                  * extent have been removed.
1951                  */
1952                 if (uninitialized && num)
1953                         ext4_ext_mark_uninitialized(ex);
1954
1955                 err = ext4_ext_dirty(handle, inode, path + depth);
1956                 if (err)
1957                         goto out;
1958
1959                 ext_debug("new extent: %u:%u:%llu\n", block, num,
1960                                 ext_pblock(ex));
1961                 ex--;
1962                 ex_ee_block = le32_to_cpu(ex->ee_block);
1963                 ex_ee_len = ext4_ext_get_actual_len(ex);
1964         }
1965
1966         if (correct_index && eh->eh_entries)
1967                 err = ext4_ext_correct_indexes(handle, inode, path);
1968
1969         /* if this leaf is free, then we should
1970          * remove it from index block above */
1971         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1972                 err = ext4_ext_rm_idx(handle, inode, path + depth);
1973
1974 out:
1975         return err;
1976 }
1977
1978 /*
1979  * ext4_ext_more_to_rm:
1980  * returns 1 if current index has to be freed (even partial)
1981  */
1982 static int
1983 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1984 {
1985         BUG_ON(path->p_idx == NULL);
1986
1987         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1988                 return 0;
1989
1990         /*
1991          * if truncate on deeper level happened, it wasn't partial,
1992          * so we have to consider current index for truncation
1993          */
1994         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1995                 return 0;
1996         return 1;
1997 }
1998
1999 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2000 {
2001         struct super_block *sb = inode->i_sb;
2002         int depth = ext_depth(inode);
2003         struct ext4_ext_path *path;
2004         handle_t *handle;
2005         int i = 0, err = 0;
2006
2007         ext_debug("truncate since %u\n", start);
2008
2009         /* probably first extent we're gonna free will be last in block */
2010         handle = ext4_journal_start(inode, depth + 1);
2011         if (IS_ERR(handle))
2012                 return PTR_ERR(handle);
2013
2014         ext4_ext_invalidate_cache(inode);
2015
2016         /*
2017          * We start scanning from right side, freeing all the blocks
2018          * after i_size and walking into the tree depth-wise.
2019          */
2020         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2021         if (path == NULL) {
2022                 ext4_journal_stop(handle);
2023                 return -ENOMEM;
2024         }
2025         path[0].p_hdr = ext_inode_hdr(inode);
2026         if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
2027                 err = -EIO;
2028                 goto out;
2029         }
2030         path[0].p_depth = depth;
2031
2032         while (i >= 0 && err == 0) {
2033                 if (i == depth) {
2034                         /* this is leaf block */
2035                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2036                         /* root level has p_bh == NULL, brelse() eats this */
2037                         brelse(path[i].p_bh);
2038                         path[i].p_bh = NULL;
2039                         i--;
2040                         continue;
2041                 }
2042
2043                 /* this is index block */
2044                 if (!path[i].p_hdr) {
2045                         ext_debug("initialize header\n");
2046                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2047                 }
2048
2049                 if (!path[i].p_idx) {
2050                         /* this level hasn't been touched yet */
2051                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2052                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2053                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2054                                   path[i].p_hdr,
2055                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2056                 } else {
2057                         /* we were already here, see at next index */
2058                         path[i].p_idx--;
2059                 }
2060
2061                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2062                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2063                                 path[i].p_idx);
2064                 if (ext4_ext_more_to_rm(path + i)) {
2065                         struct buffer_head *bh;
2066                         /* go to the next level */
2067                         ext_debug("move to level %d (block %llu)\n",
2068                                   i + 1, idx_pblock(path[i].p_idx));
2069                         memset(path + i + 1, 0, sizeof(*path));
2070                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2071                         if (!bh) {
2072                                 /* should we reset i_size? */
2073                                 err = -EIO;
2074                                 break;
2075                         }
2076                         if (WARN_ON(i + 1 > depth)) {
2077                                 err = -EIO;
2078                                 break;
2079                         }
2080                         if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2081                                                         depth - i - 1)) {
2082                                 err = -EIO;
2083                                 break;
2084                         }
2085                         path[i + 1].p_bh = bh;
2086
2087                         /* save actual number of indexes since this
2088                          * number is changed at the next iteration */
2089                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2090                         i++;
2091                 } else {
2092                         /* we finished processing this index, go up */
2093                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2094                                 /* index is empty, remove it;
2095                                  * handle must be already prepared by the
2096                                  * truncatei_leaf() */
2097                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2098                         }
2099                         /* root level has p_bh == NULL, brelse() eats this */
2100                         brelse(path[i].p_bh);
2101                         path[i].p_bh = NULL;
2102                         i--;
2103                         ext_debug("return to level %d\n", i);
2104                 }
2105         }
2106
2107         /* TODO: flexible tree reduction should be here */
2108         if (path->p_hdr->eh_entries == 0) {
2109                 /*
2110                  * truncate to zero freed all the tree,
2111                  * so we need to correct eh_depth
2112                  */
2113                 err = ext4_ext_get_access(handle, inode, path);
2114                 if (err == 0) {
2115                         ext_inode_hdr(inode)->eh_depth = 0;
2116                         ext_inode_hdr(inode)->eh_max =
2117                                 cpu_to_le16(ext4_ext_space_root(inode));
2118                         err = ext4_ext_dirty(handle, inode, path);
2119                 }
2120         }
2121 out:
2122         ext4_ext_tree_changed(inode);
2123         ext4_ext_drop_refs(path);
2124         kfree(path);
2125         ext4_journal_stop(handle);
2126
2127         return err;
2128 }
2129
2130 /*
2131  * called at mount time
2132  */
2133 void ext4_ext_init(struct super_block *sb)
2134 {
2135         /*
2136          * possible initialization would be here
2137          */
2138
2139         if (test_opt(sb, EXTENTS)) {
2140                 printk("EXT4-fs: file extents enabled");
2141 #ifdef AGGRESSIVE_TEST
2142                 printk(", aggressive tests");
2143 #endif
2144 #ifdef CHECK_BINSEARCH
2145                 printk(", check binsearch");
2146 #endif
2147 #ifdef EXTENTS_STATS
2148                 printk(", stats");
2149 #endif
2150                 printk("\n");
2151 #ifdef EXTENTS_STATS
2152                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2153                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2154                 EXT4_SB(sb)->s_ext_max = 0;
2155 #endif
2156         }
2157 }
2158
2159 /*
2160  * called at umount time
2161  */
2162 void ext4_ext_release(struct super_block *sb)
2163 {
2164         if (!test_opt(sb, EXTENTS))
2165                 return;
2166
2167 #ifdef EXTENTS_STATS
2168         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2169                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2170                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2171                         sbi->s_ext_blocks, sbi->s_ext_extents,
2172                         sbi->s_ext_blocks / sbi->s_ext_extents);
2173                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2174                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2175         }
2176 #endif
2177 }
2178
2179 static void bi_complete(struct bio *bio, int error)
2180 {
2181         complete((struct completion *)bio->bi_private);
2182 }
2183
2184 /* FIXME!! we need to try to merge to left or right after zero-out  */
2185 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2186 {
2187         int ret = -EIO;
2188         struct bio *bio;
2189         int blkbits, blocksize;
2190         sector_t ee_pblock;
2191         struct completion event;
2192         unsigned int ee_len, len, done, offset;
2193
2194
2195         blkbits   = inode->i_blkbits;
2196         blocksize = inode->i_sb->s_blocksize;
2197         ee_len    = ext4_ext_get_actual_len(ex);
2198         ee_pblock = ext_pblock(ex);
2199
2200         /* convert ee_pblock to 512 byte sectors */
2201         ee_pblock = ee_pblock << (blkbits - 9);
2202
2203         while (ee_len > 0) {
2204
2205                 if (ee_len > BIO_MAX_PAGES)
2206                         len = BIO_MAX_PAGES;
2207                 else
2208                         len = ee_len;
2209
2210                 bio = bio_alloc(GFP_NOIO, len);
2211                 if (!bio)
2212                         return -ENOMEM;
2213                 bio->bi_sector = ee_pblock;
2214                 bio->bi_bdev   = inode->i_sb->s_bdev;
2215
2216                 done = 0;
2217                 offset = 0;
2218                 while (done < len) {
2219                         ret = bio_add_page(bio, ZERO_PAGE(0),
2220                                                         blocksize, offset);
2221                         if (ret != blocksize) {
2222                                 /*
2223                                  * We can't add any more pages because of
2224                                  * hardware limitations.  Start a new bio.
2225                                  */
2226                                 break;
2227                         }
2228                         done++;
2229                         offset += blocksize;
2230                         if (offset >= PAGE_CACHE_SIZE)
2231                                 offset = 0;
2232                 }
2233
2234                 init_completion(&event);
2235                 bio->bi_private = &event;
2236                 bio->bi_end_io = bi_complete;
2237                 submit_bio(WRITE, bio);
2238                 wait_for_completion(&event);
2239
2240                 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2241                         ret = 0;
2242                 else {
2243                         ret = -EIO;
2244                         break;
2245                 }
2246                 bio_put(bio);
2247                 ee_len    -= done;
2248                 ee_pblock += done  << (blkbits - 9);
2249         }
2250         return ret;
2251 }
2252
2253 #define EXT4_EXT_ZERO_LEN 7
2254
2255 /*
2256  * This function is called by ext4_ext_get_blocks() if someone tries to write
2257  * to an uninitialized extent. It may result in splitting the uninitialized
2258  * extent into multiple extents (upto three - one initialized and two
2259  * uninitialized).
2260  * There are three possibilities:
2261  *   a> There is no split required: Entire extent should be initialized
2262  *   b> Splits in two extents: Write is happening at either end of the extent
2263  *   c> Splits in three extents: Somone is writing in middle of the extent
2264  */
2265 static int ext4_ext_convert_to_initialized(handle_t *handle,
2266                                                 struct inode *inode,
2267                                                 struct ext4_ext_path *path,
2268                                                 ext4_lblk_t iblock,
2269                                                 unsigned long max_blocks)
2270 {
2271         struct ext4_extent *ex, newex, orig_ex;
2272         struct ext4_extent *ex1 = NULL;
2273         struct ext4_extent *ex2 = NULL;
2274         struct ext4_extent *ex3 = NULL;
2275         struct ext4_extent_header *eh;
2276         ext4_lblk_t ee_block;
2277         unsigned int allocated, ee_len, depth;
2278         ext4_fsblk_t newblock;
2279         int err = 0;
2280         int ret = 0;
2281
2282         depth = ext_depth(inode);
2283         eh = path[depth].p_hdr;
2284         ex = path[depth].p_ext;
2285         ee_block = le32_to_cpu(ex->ee_block);
2286         ee_len = ext4_ext_get_actual_len(ex);
2287         allocated = ee_len - (iblock - ee_block);
2288         newblock = iblock - ee_block + ext_pblock(ex);
2289         ex2 = ex;
2290         orig_ex.ee_block = ex->ee_block;
2291         orig_ex.ee_len   = cpu_to_le16(ee_len);
2292         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2293
2294         err = ext4_ext_get_access(handle, inode, path + depth);
2295         if (err)
2296                 goto out;
2297         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2298         if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2299                 err =  ext4_ext_zeroout(inode, &orig_ex);
2300                 if (err)
2301                         goto fix_extent_len;
2302                 /* update the extent length and mark as initialized */
2303                 ex->ee_block = orig_ex.ee_block;
2304                 ex->ee_len   = orig_ex.ee_len;
2305                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2306                 ext4_ext_dirty(handle, inode, path + depth);
2307                 /* zeroed the full extent */
2308                 return allocated;
2309         }
2310
2311         /* ex1: ee_block to iblock - 1 : uninitialized */
2312         if (iblock > ee_block) {
2313                 ex1 = ex;
2314                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2315                 ext4_ext_mark_uninitialized(ex1);
2316                 ex2 = &newex;
2317         }
2318         /*
2319          * for sanity, update the length of the ex2 extent before
2320          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2321          * overlap of blocks.
2322          */
2323         if (!ex1 && allocated > max_blocks)
2324                 ex2->ee_len = cpu_to_le16(max_blocks);
2325         /* ex3: to ee_block + ee_len : uninitialised */
2326         if (allocated > max_blocks) {
2327                 unsigned int newdepth;
2328                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2329                 if (allocated <= EXT4_EXT_ZERO_LEN) {
2330                         /*
2331                          * iblock == ee_block is handled by the zerouout
2332                          * at the beginning.
2333                          * Mark first half uninitialized.
2334                          * Mark second half initialized and zero out the
2335                          * initialized extent
2336                          */
2337                         ex->ee_block = orig_ex.ee_block;
2338                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2339                         ext4_ext_mark_uninitialized(ex);
2340                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2341                         ext4_ext_dirty(handle, inode, path + depth);
2342
2343                         ex3 = &newex;
2344                         ex3->ee_block = cpu_to_le32(iblock);
2345                         ext4_ext_store_pblock(ex3, newblock);
2346                         ex3->ee_len = cpu_to_le16(allocated);
2347                         err = ext4_ext_insert_extent(handle, inode, path, ex3);
2348                         if (err == -ENOSPC) {
2349                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2350                                 if (err)
2351                                         goto fix_extent_len;
2352                                 ex->ee_block = orig_ex.ee_block;
2353                                 ex->ee_len   = orig_ex.ee_len;
2354                                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2355                                 ext4_ext_dirty(handle, inode, path + depth);
2356                                 /* blocks available from iblock */
2357                                 return allocated;
2358
2359                         } else if (err)
2360                                 goto fix_extent_len;
2361
2362                         /*
2363                          * We need to zero out the second half because
2364                          * an fallocate request can update file size and
2365                          * converting the second half to initialized extent
2366                          * implies that we can leak some junk data to user
2367                          * space.
2368                          */
2369                         err =  ext4_ext_zeroout(inode, ex3);
2370                         if (err) {
2371                                 /*
2372                                  * We should actually mark the
2373                                  * second half as uninit and return error
2374                                  * Insert would have changed the extent
2375                                  */
2376                                 depth = ext_depth(inode);
2377                                 ext4_ext_drop_refs(path);
2378                                 path = ext4_ext_find_extent(inode,
2379                                                                 iblock, path);
2380                                 if (IS_ERR(path)) {
2381                                         err = PTR_ERR(path);
2382                                         return err;
2383                                 }
2384                                 /* get the second half extent details */
2385                                 ex = path[depth].p_ext;
2386                                 err = ext4_ext_get_access(handle, inode,
2387                                                                 path + depth);
2388                                 if (err)
2389                                         return err;
2390                                 ext4_ext_mark_uninitialized(ex);
2391                                 ext4_ext_dirty(handle, inode, path + depth);
2392                                 return err;
2393                         }
2394
2395                         /* zeroed the second half */
2396                         return allocated;
2397                 }
2398                 ex3 = &newex;
2399                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2400                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2401                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2402                 ext4_ext_mark_uninitialized(ex3);
2403                 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2404                 if (err == -ENOSPC) {
2405                         err =  ext4_ext_zeroout(inode, &orig_ex);
2406                         if (err)
2407                                 goto fix_extent_len;
2408                         /* update the extent length and mark as initialized */
2409                         ex->ee_block = orig_ex.ee_block;
2410                         ex->ee_len   = orig_ex.ee_len;
2411                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2412                         ext4_ext_dirty(handle, inode, path + depth);
2413                         /* zeroed the full extent */
2414                         /* blocks available from iblock */
2415                         return allocated;
2416
2417                 } else if (err)
2418                         goto fix_extent_len;
2419                 /*
2420                  * The depth, and hence eh & ex might change
2421                  * as part of the insert above.
2422                  */
2423                 newdepth = ext_depth(inode);
2424                 /*
2425                  * update the extent length after successfull insert of the
2426                  * split extent
2427                  */
2428                 orig_ex.ee_len = cpu_to_le16(ee_len -
2429                                                 ext4_ext_get_actual_len(ex3));
2430                 depth = newdepth;
2431                 ext4_ext_drop_refs(path);
2432                 path = ext4_ext_find_extent(inode, iblock, path);
2433                 if (IS_ERR(path)) {
2434                         err = PTR_ERR(path);
2435                         goto out;
2436                 }
2437                 eh = path[depth].p_hdr;
2438                 ex = path[depth].p_ext;
2439                 if (ex2 != &newex)
2440                         ex2 = ex;
2441
2442                 err = ext4_ext_get_access(handle, inode, path + depth);
2443                 if (err)
2444                         goto out;
2445
2446                 allocated = max_blocks;
2447
2448                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2449                  * to insert a extent in the middle zerout directly
2450                  * otherwise give the extent a chance to merge to left
2451                  */
2452                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2453                                                         iblock != ee_block) {
2454                         err =  ext4_ext_zeroout(inode, &orig_ex);
2455                         if (err)
2456                                 goto fix_extent_len;
2457                         /* update the extent length and mark as initialized */
2458                         ex->ee_block = orig_ex.ee_block;
2459                         ex->ee_len   = orig_ex.ee_len;
2460                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2461                         ext4_ext_dirty(handle, inode, path + depth);
2462                         /* zero out the first half */
2463                         /* blocks available from iblock */
2464                         return allocated;
2465                 }
2466         }
2467         /*
2468          * If there was a change of depth as part of the
2469          * insertion of ex3 above, we need to update the length
2470          * of the ex1 extent again here
2471          */
2472         if (ex1 && ex1 != ex) {
2473                 ex1 = ex;
2474                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2475                 ext4_ext_mark_uninitialized(ex1);
2476                 ex2 = &newex;
2477         }
2478         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2479         ex2->ee_block = cpu_to_le32(iblock);
2480         ext4_ext_store_pblock(ex2, newblock);
2481         ex2->ee_len = cpu_to_le16(allocated);
2482         if (ex2 != ex)
2483                 goto insert;
2484         /*
2485          * New (initialized) extent starts from the first block
2486          * in the current extent. i.e., ex2 == ex
2487          * We have to see if it can be merged with the extent
2488          * on the left.
2489          */
2490         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2491                 /*
2492                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2493                  * since it merges towards right _only_.
2494                  */
2495                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2496                 if (ret) {
2497                         err = ext4_ext_correct_indexes(handle, inode, path);
2498                         if (err)
2499                                 goto out;
2500                         depth = ext_depth(inode);
2501                         ex2--;
2502                 }
2503         }
2504         /*
2505          * Try to Merge towards right. This might be required
2506          * only when the whole extent is being written to.
2507          * i.e. ex2 == ex and ex3 == NULL.
2508          */
2509         if (!ex3) {
2510                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2511                 if (ret) {
2512                         err = ext4_ext_correct_indexes(handle, inode, path);
2513                         if (err)
2514                                 goto out;
2515                 }
2516         }
2517         /* Mark modified extent as dirty */
2518         err = ext4_ext_dirty(handle, inode, path + depth);
2519         goto out;
2520 insert:
2521         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2522         if (err == -ENOSPC) {
2523                 err =  ext4_ext_zeroout(inode, &orig_ex);
2524                 if (err)
2525                         goto fix_extent_len;
2526                 /* update the extent length and mark as initialized */
2527                 ex->ee_block = orig_ex.ee_block;
2528                 ex->ee_len   = orig_ex.ee_len;
2529                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2530                 ext4_ext_dirty(handle, inode, path + depth);
2531                 /* zero out the first half */
2532                 return allocated;
2533         } else if (err)
2534                 goto fix_extent_len;
2535 out:
2536         return err ? err : allocated;
2537
2538 fix_extent_len:
2539         ex->ee_block = orig_ex.ee_block;
2540         ex->ee_len   = orig_ex.ee_len;
2541         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2542         ext4_ext_mark_uninitialized(ex);
2543         ext4_ext_dirty(handle, inode, path + depth);
2544         return err;
2545 }
2546
2547 /*
2548  * Block allocation/map/preallocation routine for extents based files
2549  *
2550  *
2551  * Need to be called with
2552  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2553  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2554  *
2555  * return > 0, number of of blocks already mapped/allocated
2556  *          if create == 0 and these are pre-allocated blocks
2557  *              buffer head is unmapped
2558  *          otherwise blocks are mapped
2559  *
2560  * return = 0, if plain look up failed (blocks have not been allocated)
2561  *          buffer head is unmapped
2562  *
2563  * return < 0, error case.
2564  */
2565 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2566                         ext4_lblk_t iblock,
2567                         unsigned long max_blocks, struct buffer_head *bh_result,
2568                         int create, int extend_disksize)
2569 {
2570         struct ext4_ext_path *path = NULL;
2571         struct ext4_extent_header *eh;
2572         struct ext4_extent newex, *ex;
2573         ext4_fsblk_t goal, newblock;
2574         int err = 0, depth, ret;
2575         unsigned long allocated = 0;
2576         struct ext4_allocation_request ar;
2577         loff_t disksize;
2578
2579         __clear_bit(BH_New, &bh_result->b_state);
2580         ext_debug("blocks %u/%lu requested for inode %u\n",
2581                         iblock, max_blocks, inode->i_ino);
2582
2583         /* check in cache */
2584         goal = ext4_ext_in_cache(inode, iblock, &newex);
2585         if (goal) {
2586                 if (goal == EXT4_EXT_CACHE_GAP) {
2587                         if (!create) {
2588                                 /*
2589                                  * block isn't allocated yet and
2590                                  * user doesn't want to allocate it
2591                                  */
2592                                 goto out2;
2593                         }
2594                         /* we should allocate requested block */
2595                 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2596                         /* block is already allocated */
2597                         newblock = iblock
2598                                    - le32_to_cpu(newex.ee_block)
2599                                    + ext_pblock(&newex);
2600                         /* number of remaining blocks in the extent */
2601                         allocated = ext4_ext_get_actual_len(&newex) -
2602                                         (iblock - le32_to_cpu(newex.ee_block));
2603                         goto out;
2604                 } else {
2605                         BUG();
2606                 }
2607         }
2608
2609         /* find extent for this block */
2610         path = ext4_ext_find_extent(inode, iblock, NULL);
2611         if (IS_ERR(path)) {
2612                 err = PTR_ERR(path);
2613                 path = NULL;
2614                 goto out2;
2615         }
2616
2617         depth = ext_depth(inode);
2618
2619         /*
2620          * consistent leaf must not be empty;
2621          * this situation is possible, though, _during_ tree modification;
2622          * this is why assert can't be put in ext4_ext_find_extent()
2623          */
2624         BUG_ON(path[depth].p_ext == NULL && depth != 0);
2625         eh = path[depth].p_hdr;
2626
2627         ex = path[depth].p_ext;
2628         if (ex) {
2629                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2630                 ext4_fsblk_t ee_start = ext_pblock(ex);
2631                 unsigned short ee_len;
2632
2633                 /*
2634                  * Uninitialized extents are treated as holes, except that
2635                  * we split out initialized portions during a write.
2636                  */
2637                 ee_len = ext4_ext_get_actual_len(ex);
2638                 /* if found extent covers block, simply return it */
2639                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2640                         newblock = iblock - ee_block + ee_start;
2641                         /* number of remaining blocks in the extent */
2642                         allocated = ee_len - (iblock - ee_block);
2643                         ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2644                                         ee_block, ee_len, newblock);
2645
2646                         /* Do not put uninitialized extent in the cache */
2647                         if (!ext4_ext_is_uninitialized(ex)) {
2648                                 ext4_ext_put_in_cache(inode, ee_block,
2649                                                         ee_len, ee_start,
2650                                                         EXT4_EXT_CACHE_EXTENT);
2651                                 goto out;
2652                         }
2653                         if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2654                                 goto out;
2655                         if (!create) {
2656                                 /*
2657                                  * We have blocks reserved already.  We
2658                                  * return allocated blocks so that delalloc
2659                                  * won't do block reservation for us.  But
2660                                  * the buffer head will be unmapped so that
2661                                  * a read from the block returns 0s.
2662                                  */
2663                                 if (allocated > max_blocks)
2664                                         allocated = max_blocks;
2665                                 set_buffer_unwritten(bh_result);
2666                                 goto out2;
2667                         }
2668
2669                         ret = ext4_ext_convert_to_initialized(handle, inode,
2670                                                                 path, iblock,
2671                                                                 max_blocks);
2672                         if (ret <= 0) {
2673                                 err = ret;
2674                                 goto out2;
2675                         } else
2676                                 allocated = ret;
2677                         goto outnew;
2678                 }
2679         }
2680
2681         /*
2682          * requested block isn't allocated yet;
2683          * we couldn't try to create block if create flag is zero
2684          */
2685         if (!create) {
2686                 /*
2687                  * put just found gap into cache to speed up
2688                  * subsequent requests
2689                  */
2690                 ext4_ext_put_gap_in_cache(inode, path, iblock);
2691                 goto out2;
2692         }
2693         /*
2694          * Okay, we need to do block allocation.  Lazily initialize the block
2695          * allocation info here if necessary.
2696          */
2697         if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2698                 ext4_init_block_alloc_info(inode);
2699
2700         /* find neighbour allocated blocks */
2701         ar.lleft = iblock;
2702         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2703         if (err)
2704                 goto out2;
2705         ar.lright = iblock;
2706         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2707         if (err)
2708                 goto out2;
2709
2710         /*
2711          * See if request is beyond maximum number of blocks we can have in
2712          * a single extent. For an initialized extent this limit is
2713          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2714          * EXT_UNINIT_MAX_LEN.
2715          */
2716         if (max_blocks > EXT_INIT_MAX_LEN &&
2717             create != EXT4_CREATE_UNINITIALIZED_EXT)
2718                 max_blocks = EXT_INIT_MAX_LEN;
2719         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2720                  create == EXT4_CREATE_UNINITIALIZED_EXT)
2721                 max_blocks = EXT_UNINIT_MAX_LEN;
2722
2723         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2724         newex.ee_block = cpu_to_le32(iblock);
2725         newex.ee_len = cpu_to_le16(max_blocks);
2726         err = ext4_ext_check_overlap(inode, &newex, path);
2727         if (err)
2728                 allocated = ext4_ext_get_actual_len(&newex);
2729         else
2730                 allocated = max_blocks;
2731
2732         /* allocate new block */
2733         ar.inode = inode;
2734         ar.goal = ext4_ext_find_goal(inode, path, iblock);
2735         ar.logical = iblock;
2736         ar.len = allocated;
2737         if (S_ISREG(inode->i_mode))
2738                 ar.flags = EXT4_MB_HINT_DATA;
2739         else
2740                 /* disable in-core preallocation for non-regular files */
2741                 ar.flags = 0;
2742         newblock = ext4_mb_new_blocks(handle, &ar, &err);
2743         if (!newblock)
2744                 goto out2;
2745         ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2746                         goal, newblock, allocated);
2747
2748         /* try to insert new extent into found leaf and return */
2749         ext4_ext_store_pblock(&newex, newblock);
2750         newex.ee_len = cpu_to_le16(ar.len);
2751         if (create == EXT4_CREATE_UNINITIALIZED_EXT)  /* Mark uninitialized */
2752                 ext4_ext_mark_uninitialized(&newex);
2753         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2754         if (err) {
2755                 /* free data blocks we just allocated */
2756                 /* not a good idea to call discard here directly,
2757                  * but otherwise we'd need to call it every free() */
2758                 ext4_mb_discard_inode_preallocations(inode);
2759                 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2760                                         ext4_ext_get_actual_len(&newex), 0);
2761                 goto out2;
2762         }
2763
2764         /* previous routine could use block we allocated */
2765         newblock = ext_pblock(&newex);
2766         allocated = ext4_ext_get_actual_len(&newex);
2767 outnew:
2768         if (extend_disksize) {
2769                 disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits;
2770                 if (disksize > i_size_read(inode))
2771                         disksize = i_size_read(inode);
2772                 if (disksize > EXT4_I(inode)->i_disksize)
2773                         EXT4_I(inode)->i_disksize = disksize;
2774         }
2775
2776         set_buffer_new(bh_result);
2777
2778         /* Cache only when it is _not_ an uninitialized extent */
2779         if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2780                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2781                                                 EXT4_EXT_CACHE_EXTENT);
2782 out:
2783         if (allocated > max_blocks)
2784                 allocated = max_blocks;
2785         ext4_ext_show_leaf(inode, path);
2786         set_buffer_mapped(bh_result);
2787         bh_result->b_bdev = inode->i_sb->s_bdev;
2788         bh_result->b_blocknr = newblock;
2789 out2:
2790         if (path) {
2791                 ext4_ext_drop_refs(path);
2792                 kfree(path);
2793         }
2794         return err ? err : allocated;
2795 }
2796
2797 void ext4_ext_truncate(struct inode *inode)
2798 {
2799         struct address_space *mapping = inode->i_mapping;
2800         struct super_block *sb = inode->i_sb;
2801         ext4_lblk_t last_block;
2802         handle_t *handle;
2803         int err = 0;
2804
2805         /*
2806          * probably first extent we're gonna free will be last in block
2807          */
2808         err = ext4_writepage_trans_blocks(inode) + 3;
2809         handle = ext4_journal_start(inode, err);
2810         if (IS_ERR(handle))
2811                 return;
2812
2813         if (inode->i_size & (sb->s_blocksize - 1))
2814                 ext4_block_truncate_page(handle, mapping, inode->i_size);
2815
2816         if (ext4_orphan_add(handle, inode))
2817                 goto out_stop;
2818
2819         down_write(&EXT4_I(inode)->i_data_sem);
2820         ext4_ext_invalidate_cache(inode);
2821
2822         ext4_mb_discard_inode_preallocations(inode);
2823
2824         /*
2825          * TODO: optimization is possible here.
2826          * Probably we need not scan at all,
2827          * because page truncation is enough.
2828          */
2829
2830         /* we have to know where to truncate from in crash case */
2831         EXT4_I(inode)->i_disksize = inode->i_size;
2832         ext4_mark_inode_dirty(handle, inode);
2833
2834         last_block = (inode->i_size + sb->s_blocksize - 1)
2835                         >> EXT4_BLOCK_SIZE_BITS(sb);
2836         err = ext4_ext_remove_space(inode, last_block);
2837
2838         /* In a multi-transaction truncate, we only make the final
2839          * transaction synchronous.
2840          */
2841         if (IS_SYNC(inode))
2842                 handle->h_sync = 1;
2843
2844 out_stop:
2845         up_write(&EXT4_I(inode)->i_data_sem);
2846         /*
2847          * If this was a simple ftruncate() and the file will remain alive,
2848          * then we need to clear up the orphan record which we created above.
2849          * However, if this was a real unlink then we were called by
2850          * ext4_delete_inode(), and we allow that function to clean up the
2851          * orphan info for us.
2852          */
2853         if (inode->i_nlink)
2854                 ext4_orphan_del(handle, inode);
2855
2856         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2857         ext4_mark_inode_dirty(handle, inode);
2858         ext4_journal_stop(handle);
2859 }
2860
2861 /*
2862  * ext4_ext_writepage_trans_blocks:
2863  * calculate max number of blocks we could modify
2864  * in order to allocate new block for an inode
2865  */
2866 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2867 {
2868         int needed;
2869
2870         needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2871
2872         /* caller wants to allocate num blocks, but note it includes sb */
2873         needed = needed * num - (num - 1);
2874
2875 #ifdef CONFIG_QUOTA
2876         needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2877 #endif
2878
2879         return needed;
2880 }
2881
2882 static void ext4_falloc_update_inode(struct inode *inode,
2883                                 int mode, loff_t new_size, int update_ctime)
2884 {
2885         struct timespec now;
2886
2887         if (update_ctime) {
2888                 now = current_fs_time(inode->i_sb);
2889                 if (!timespec_equal(&inode->i_ctime, &now))
2890                         inode->i_ctime = now;
2891         }
2892         /*
2893          * Update only when preallocation was requested beyond
2894          * the file size.
2895          */
2896         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2897                                 new_size > i_size_read(inode)) {
2898                 i_size_write(inode, new_size);
2899                 EXT4_I(inode)->i_disksize = new_size;
2900         }
2901
2902 }
2903
2904 /*
2905  * preallocate space for a file. This implements ext4's fallocate inode
2906  * operation, which gets called from sys_fallocate system call.
2907  * For block-mapped files, posix_fallocate should fall back to the method
2908  * of writing zeroes to the required new blocks (the same behavior which is
2909  * expected for file systems which do not support fallocate() system call).
2910  */
2911 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2912 {
2913         handle_t *handle;
2914         ext4_lblk_t block;
2915         loff_t new_size;
2916         unsigned long max_blocks;
2917         int ret = 0;
2918         int ret2 = 0;
2919         int retries = 0;
2920         struct buffer_head map_bh;
2921         unsigned int credits, blkbits = inode->i_blkbits;
2922
2923         /*
2924          * currently supporting (pre)allocate mode for extent-based
2925          * files _only_
2926          */
2927         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2928                 return -EOPNOTSUPP;
2929
2930         /* preallocation to directories is currently not supported */
2931         if (S_ISDIR(inode->i_mode))
2932                 return -ENODEV;
2933
2934         block = offset >> blkbits;
2935         /*
2936          * We can't just convert len to max_blocks because
2937          * If blocksize = 4096 offset = 3072 and len = 2048
2938          */
2939         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2940                                                         - block;
2941         /*
2942          * credits to insert 1 extent into extent tree + buffers to be able to
2943          * modify 1 super block, 1 block bitmap and 1 group descriptor.
2944          */
2945         credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
2946         mutex_lock(&inode->i_mutex);
2947 retry:
2948         while (ret >= 0 && ret < max_blocks) {
2949                 block = block + ret;
2950                 max_blocks = max_blocks - ret;
2951                 handle = ext4_journal_start(inode, credits);
2952                 if (IS_ERR(handle)) {
2953                         ret = PTR_ERR(handle);
2954                         break;
2955                 }
2956                 ret = ext4_get_blocks_wrap(handle, inode, block,
2957                                           max_blocks, &map_bh,
2958                                           EXT4_CREATE_UNINITIALIZED_EXT, 0, 0);
2959                 if (ret <= 0) {
2960 #ifdef EXT4FS_DEBUG
2961                         WARN_ON(ret <= 0);
2962                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
2963                                     "returned error inode#%lu, block=%u, "
2964                                     "max_blocks=%lu", __func__,
2965                                     inode->i_ino, block, max_blocks);
2966 #endif
2967                         ext4_mark_inode_dirty(handle, inode);
2968                         ret2 = ext4_journal_stop(handle);
2969                         break;
2970                 }
2971                 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
2972                                                 blkbits) >> blkbits))
2973                         new_size = offset + len;
2974                 else
2975                         new_size = (block + ret) << blkbits;
2976
2977                 ext4_falloc_update_inode(inode, mode, new_size,
2978                                                 buffer_new(&map_bh));
2979                 ext4_mark_inode_dirty(handle, inode);
2980                 ret2 = ext4_journal_stop(handle);
2981                 if (ret2)
2982                         break;
2983         }
2984         if (ret == -ENOSPC &&
2985                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
2986                 ret = 0;
2987                 goto retry;
2988         }
2989         mutex_unlock(&inode->i_mutex);
2990         return ret > 0 ? ret2 : ret;
2991 }