Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[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/ext4_jbd2.h>
36 #include <linux/jbd.h>
37 #include <linux/highuid.h>
38 #include <linux/pagemap.h>
39 #include <linux/quotaops.h>
40 #include <linux/string.h>
41 #include <linux/slab.h>
42 #include <linux/ext4_fs_extents.h>
43 #include <asm/uaccess.h>
44
45
46 /*
47  * ext_pblock:
48  * combine low and high parts of physical block number into ext4_fsblk_t
49  */
50 static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
51 {
52         ext4_fsblk_t block;
53
54         block = le32_to_cpu(ex->ee_start);
55         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
56         return block;
57 }
58
59 /*
60  * idx_pblock:
61  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
62  */
63 static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
64 {
65         ext4_fsblk_t block;
66
67         block = le32_to_cpu(ix->ei_leaf);
68         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
69         return block;
70 }
71
72 /*
73  * ext4_ext_store_pblock:
74  * stores a large physical block number into an extent struct,
75  * breaking it into parts
76  */
77 static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
78 {
79         ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff));
80         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
81 }
82
83 /*
84  * ext4_idx_store_pblock:
85  * stores a large physical block number into an index struct,
86  * breaking it into parts
87  */
88 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
89 {
90         ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff));
91         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
92 }
93
94 static int ext4_ext_check_header(const char *function, struct inode *inode,
95                                 struct ext4_extent_header *eh)
96 {
97         const char *error_msg = NULL;
98
99         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
100                 error_msg = "invalid magic";
101                 goto corrupted;
102         }
103         if (unlikely(eh->eh_max == 0)) {
104                 error_msg = "invalid eh_max";
105                 goto corrupted;
106         }
107         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
108                 error_msg = "invalid eh_entries";
109                 goto corrupted;
110         }
111         return 0;
112
113 corrupted:
114         ext4_error(inode->i_sb, function,
115                         "bad header in inode #%lu: %s - magic %x, "
116                         "entries %u, max %u, depth %u",
117                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
118                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
119                         le16_to_cpu(eh->eh_depth));
120
121         return -EIO;
122 }
123
124 static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
125 {
126         int err;
127
128         if (handle->h_buffer_credits > needed)
129                 return handle;
130         if (!ext4_journal_extend(handle, needed))
131                 return handle;
132         err = ext4_journal_restart(handle, needed);
133
134         return handle;
135 }
136
137 /*
138  * could return:
139  *  - EROFS
140  *  - ENOMEM
141  */
142 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
143                                 struct ext4_ext_path *path)
144 {
145         if (path->p_bh) {
146                 /* path points to block */
147                 return ext4_journal_get_write_access(handle, path->p_bh);
148         }
149         /* path points to leaf/index in inode body */
150         /* we use in-core data, no need to protect them */
151         return 0;
152 }
153
154 /*
155  * could return:
156  *  - EROFS
157  *  - ENOMEM
158  *  - EIO
159  */
160 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
161                                 struct ext4_ext_path *path)
162 {
163         int err;
164         if (path->p_bh) {
165                 /* path points to block */
166                 err = ext4_journal_dirty_metadata(handle, path->p_bh);
167         } else {
168                 /* path points to leaf/index in inode body */
169                 err = ext4_mark_inode_dirty(handle, inode);
170         }
171         return err;
172 }
173
174 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
175                               struct ext4_ext_path *path,
176                               ext4_fsblk_t block)
177 {
178         struct ext4_inode_info *ei = EXT4_I(inode);
179         ext4_fsblk_t bg_start;
180         ext4_grpblk_t colour;
181         int depth;
182
183         if (path) {
184                 struct ext4_extent *ex;
185                 depth = path->p_depth;
186
187                 /* try to predict block placement */
188                 ex = path[depth].p_ext;
189                 if (ex)
190                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
191
192                 /* it looks like index is empty;
193                  * try to find starting block from index itself */
194                 if (path[depth].p_bh)
195                         return path[depth].p_bh->b_blocknr;
196         }
197
198         /* OK. use inode's group */
199         bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
200                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
201         colour = (current->pid % 16) *
202                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
203         return bg_start + colour + block;
204 }
205
206 static ext4_fsblk_t
207 ext4_ext_new_block(handle_t *handle, struct inode *inode,
208                         struct ext4_ext_path *path,
209                         struct ext4_extent *ex, int *err)
210 {
211         ext4_fsblk_t goal, newblock;
212
213         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
214         newblock = ext4_new_block(handle, inode, goal, err);
215         return newblock;
216 }
217
218 static int ext4_ext_space_block(struct inode *inode)
219 {
220         int size;
221
222         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
223                         / sizeof(struct ext4_extent);
224 #ifdef AGGRESSIVE_TEST
225         if (size > 6)
226                 size = 6;
227 #endif
228         return size;
229 }
230
231 static int ext4_ext_space_block_idx(struct inode *inode)
232 {
233         int size;
234
235         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
236                         / sizeof(struct ext4_extent_idx);
237 #ifdef AGGRESSIVE_TEST
238         if (size > 5)
239                 size = 5;
240 #endif
241         return size;
242 }
243
244 static int ext4_ext_space_root(struct inode *inode)
245 {
246         int size;
247
248         size = sizeof(EXT4_I(inode)->i_data);
249         size -= sizeof(struct ext4_extent_header);
250         size /= sizeof(struct ext4_extent);
251 #ifdef AGGRESSIVE_TEST
252         if (size > 3)
253                 size = 3;
254 #endif
255         return size;
256 }
257
258 static int ext4_ext_space_root_idx(struct inode *inode)
259 {
260         int size;
261
262         size = sizeof(EXT4_I(inode)->i_data);
263         size -= sizeof(struct ext4_extent_header);
264         size /= sizeof(struct ext4_extent_idx);
265 #ifdef AGGRESSIVE_TEST
266         if (size > 4)
267                 size = 4;
268 #endif
269         return size;
270 }
271
272 #ifdef EXT_DEBUG
273 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
274 {
275         int k, l = path->p_depth;
276
277         ext_debug("path:");
278         for (k = 0; k <= l; k++, path++) {
279                 if (path->p_idx) {
280                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
281                             idx_pblock(path->p_idx));
282                 } else if (path->p_ext) {
283                         ext_debug("  %d:%d:%llu ",
284                                   le32_to_cpu(path->p_ext->ee_block),
285                                   le16_to_cpu(path->p_ext->ee_len),
286                                   ext_pblock(path->p_ext));
287                 } else
288                         ext_debug("  []");
289         }
290         ext_debug("\n");
291 }
292
293 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
294 {
295         int depth = ext_depth(inode);
296         struct ext4_extent_header *eh;
297         struct ext4_extent *ex;
298         int i;
299
300         if (!path)
301                 return;
302
303         eh = path[depth].p_hdr;
304         ex = EXT_FIRST_EXTENT(eh);
305
306         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
307                 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
308                           le16_to_cpu(ex->ee_len), ext_pblock(ex));
309         }
310         ext_debug("\n");
311 }
312 #else
313 #define ext4_ext_show_path(inode,path)
314 #define ext4_ext_show_leaf(inode,path)
315 #endif
316
317 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
318 {
319         int depth = path->p_depth;
320         int i;
321
322         for (i = 0; i <= depth; i++, path++)
323                 if (path->p_bh) {
324                         brelse(path->p_bh);
325                         path->p_bh = NULL;
326                 }
327 }
328
329 /*
330  * ext4_ext_binsearch_idx:
331  * binary search for the closest index of the given block
332  */
333 static void
334 ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block)
335 {
336         struct ext4_extent_header *eh = path->p_hdr;
337         struct ext4_extent_idx *r, *l, *m;
338
339         BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
340         BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
341         BUG_ON(le16_to_cpu(eh->eh_entries) <= 0);
342
343         ext_debug("binsearch for %d(idx):  ", block);
344
345         l = EXT_FIRST_INDEX(eh) + 1;
346         r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
347         while (l <= r) {
348                 m = l + (r - l) / 2;
349                 if (block < le32_to_cpu(m->ei_block))
350                         r = m - 1;
351                 else
352                         l = m + 1;
353                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block,
354                                 m, m->ei_block, r, r->ei_block);
355         }
356
357         path->p_idx = l - 1;
358         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
359                   idx_block(path->p_idx));
360
361 #ifdef CHECK_BINSEARCH
362         {
363                 struct ext4_extent_idx *chix, *ix;
364                 int k;
365
366                 chix = ix = EXT_FIRST_INDEX(eh);
367                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
368                   if (k != 0 &&
369                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
370                                 printk("k=%d, ix=0x%p, first=0x%p\n", k,
371                                         ix, EXT_FIRST_INDEX(eh));
372                                 printk("%u <= %u\n",
373                                        le32_to_cpu(ix->ei_block),
374                                        le32_to_cpu(ix[-1].ei_block));
375                         }
376                         BUG_ON(k && le32_to_cpu(ix->ei_block)
377                                            <= le32_to_cpu(ix[-1].ei_block));
378                         if (block < le32_to_cpu(ix->ei_block))
379                                 break;
380                         chix = ix;
381                 }
382                 BUG_ON(chix != path->p_idx);
383         }
384 #endif
385
386 }
387
388 /*
389  * ext4_ext_binsearch:
390  * binary search for closest extent of the given block
391  */
392 static void
393 ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
394 {
395         struct ext4_extent_header *eh = path->p_hdr;
396         struct ext4_extent *r, *l, *m;
397
398         BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
399         BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
400
401         if (eh->eh_entries == 0) {
402                 /*
403                  * this leaf is empty:
404                  * we get such a leaf in split/add case
405                  */
406                 return;
407         }
408
409         ext_debug("binsearch for %d:  ", block);
410
411         l = EXT_FIRST_EXTENT(eh) + 1;
412         r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
413
414         while (l <= r) {
415                 m = l + (r - l) / 2;
416                 if (block < le32_to_cpu(m->ee_block))
417                         r = m - 1;
418                 else
419                         l = m + 1;
420                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block,
421                                 m, m->ee_block, r, r->ee_block);
422         }
423
424         path->p_ext = l - 1;
425         ext_debug("  -> %d:%llu:%d ",
426                         le32_to_cpu(path->p_ext->ee_block),
427                         ext_pblock(path->p_ext),
428                         le16_to_cpu(path->p_ext->ee_len));
429
430 #ifdef CHECK_BINSEARCH
431         {
432                 struct ext4_extent *chex, *ex;
433                 int k;
434
435                 chex = ex = EXT_FIRST_EXTENT(eh);
436                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
437                         BUG_ON(k && le32_to_cpu(ex->ee_block)
438                                           <= le32_to_cpu(ex[-1].ee_block));
439                         if (block < le32_to_cpu(ex->ee_block))
440                                 break;
441                         chex = ex;
442                 }
443                 BUG_ON(chex != path->p_ext);
444         }
445 #endif
446
447 }
448
449 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
450 {
451         struct ext4_extent_header *eh;
452
453         eh = ext_inode_hdr(inode);
454         eh->eh_depth = 0;
455         eh->eh_entries = 0;
456         eh->eh_magic = EXT4_EXT_MAGIC;
457         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
458         ext4_mark_inode_dirty(handle, inode);
459         ext4_ext_invalidate_cache(inode);
460         return 0;
461 }
462
463 struct ext4_ext_path *
464 ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
465 {
466         struct ext4_extent_header *eh;
467         struct buffer_head *bh;
468         short int depth, i, ppos = 0, alloc = 0;
469
470         eh = ext_inode_hdr(inode);
471         BUG_ON(eh == NULL);
472         if (ext4_ext_check_header(__FUNCTION__, inode, eh))
473                 return ERR_PTR(-EIO);
474
475         i = depth = ext_depth(inode);
476
477         /* account possible depth increase */
478         if (!path) {
479                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
480                                 GFP_NOFS);
481                 if (!path)
482                         return ERR_PTR(-ENOMEM);
483                 alloc = 1;
484         }
485         path[0].p_hdr = eh;
486
487         /* walk through the tree */
488         while (i) {
489                 ext_debug("depth %d: num %d, max %d\n",
490                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
491                 ext4_ext_binsearch_idx(inode, path + ppos, block);
492                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
493                 path[ppos].p_depth = i;
494                 path[ppos].p_ext = NULL;
495
496                 bh = sb_bread(inode->i_sb, path[ppos].p_block);
497                 if (!bh)
498                         goto err;
499
500                 eh = ext_block_hdr(bh);
501                 ppos++;
502                 BUG_ON(ppos > depth);
503                 path[ppos].p_bh = bh;
504                 path[ppos].p_hdr = eh;
505                 i--;
506
507                 if (ext4_ext_check_header(__FUNCTION__, inode, eh))
508                         goto err;
509         }
510
511         path[ppos].p_depth = i;
512         path[ppos].p_hdr = eh;
513         path[ppos].p_ext = NULL;
514         path[ppos].p_idx = NULL;
515
516         if (ext4_ext_check_header(__FUNCTION__, inode, eh))
517                 goto err;
518
519         /* find extent */
520         ext4_ext_binsearch(inode, path + ppos, block);
521
522         ext4_ext_show_path(inode, path);
523
524         return path;
525
526 err:
527         ext4_ext_drop_refs(path);
528         if (alloc)
529                 kfree(path);
530         return ERR_PTR(-EIO);
531 }
532
533 /*
534  * ext4_ext_insert_index:
535  * insert new index [@logical;@ptr] into the block at @curp;
536  * check where to insert: before @curp or after @curp
537  */
538 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
539                                 struct ext4_ext_path *curp,
540                                 int logical, ext4_fsblk_t ptr)
541 {
542         struct ext4_extent_idx *ix;
543         int len, err;
544
545         err = ext4_ext_get_access(handle, inode, curp);
546         if (err)
547                 return err;
548
549         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
550         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
551         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
552                 /* insert after */
553                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
554                         len = (len - 1) * sizeof(struct ext4_extent_idx);
555                         len = len < 0 ? 0 : len;
556                         ext_debug("insert new index %d after: %d. "
557                                         "move %d from 0x%p to 0x%p\n",
558                                         logical, ptr, len,
559                                         (curp->p_idx + 1), (curp->p_idx + 2));
560                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
561                 }
562                 ix = curp->p_idx + 1;
563         } else {
564                 /* insert before */
565                 len = len * sizeof(struct ext4_extent_idx);
566                 len = len < 0 ? 0 : len;
567                 ext_debug("insert new index %d before: %d. "
568                                 "move %d from 0x%p to 0x%p\n",
569                                 logical, ptr, len,
570                                 curp->p_idx, (curp->p_idx + 1));
571                 memmove(curp->p_idx + 1, curp->p_idx, len);
572                 ix = curp->p_idx;
573         }
574
575         ix->ei_block = cpu_to_le32(logical);
576         ext4_idx_store_pblock(ix, ptr);
577         curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
578
579         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
580                              > le16_to_cpu(curp->p_hdr->eh_max));
581         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
582
583         err = ext4_ext_dirty(handle, inode, curp);
584         ext4_std_error(inode->i_sb, err);
585
586         return err;
587 }
588
589 /*
590  * ext4_ext_split:
591  * inserts new subtree into the path, using free index entry
592  * at depth @at:
593  * - allocates all needed blocks (new leaf and all intermediate index blocks)
594  * - makes decision where to split
595  * - moves remaining extents and index entries (right to the split point)
596  *   into the newly allocated blocks
597  * - initializes subtree
598  */
599 static int ext4_ext_split(handle_t *handle, struct inode *inode,
600                                 struct ext4_ext_path *path,
601                                 struct ext4_extent *newext, int at)
602 {
603         struct buffer_head *bh = NULL;
604         int depth = ext_depth(inode);
605         struct ext4_extent_header *neh;
606         struct ext4_extent_idx *fidx;
607         struct ext4_extent *ex;
608         int i = at, k, m, a;
609         ext4_fsblk_t newblock, oldblock;
610         __le32 border;
611         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
612         int err = 0;
613
614         /* make decision: where to split? */
615         /* FIXME: now decision is simplest: at current extent */
616
617         /* if current leaf will be split, then we should use
618          * border from split point */
619         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
620         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
621                 border = path[depth].p_ext[1].ee_block;
622                 ext_debug("leaf will be split."
623                                 " next leaf starts at %d\n",
624                                   le32_to_cpu(border));
625         } else {
626                 border = newext->ee_block;
627                 ext_debug("leaf will be added."
628                                 " next leaf starts at %d\n",
629                                 le32_to_cpu(border));
630         }
631
632         /*
633          * If error occurs, then we break processing
634          * and mark filesystem read-only. index won't
635          * be inserted and tree will be in consistent
636          * state. Next mount will repair buffers too.
637          */
638
639         /*
640          * Get array to track all allocated blocks.
641          * We need this to handle errors and free blocks
642          * upon them.
643          */
644         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
645         if (!ablocks)
646                 return -ENOMEM;
647
648         /* allocate all needed blocks */
649         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
650         for (a = 0; a < depth - at; a++) {
651                 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
652                 if (newblock == 0)
653                         goto cleanup;
654                 ablocks[a] = newblock;
655         }
656
657         /* initialize new leaf */
658         newblock = ablocks[--a];
659         BUG_ON(newblock == 0);
660         bh = sb_getblk(inode->i_sb, newblock);
661         if (!bh) {
662                 err = -EIO;
663                 goto cleanup;
664         }
665         lock_buffer(bh);
666
667         err = ext4_journal_get_create_access(handle, bh);
668         if (err)
669                 goto cleanup;
670
671         neh = ext_block_hdr(bh);
672         neh->eh_entries = 0;
673         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
674         neh->eh_magic = EXT4_EXT_MAGIC;
675         neh->eh_depth = 0;
676         ex = EXT_FIRST_EXTENT(neh);
677
678         /* move remainder of path[depth] to the new leaf */
679         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
680         /* start copy from next extent */
681         /* TODO: we could do it by single memmove */
682         m = 0;
683         path[depth].p_ext++;
684         while (path[depth].p_ext <=
685                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
686                 ext_debug("move %d:%llu:%d in new leaf %llu\n",
687                                 le32_to_cpu(path[depth].p_ext->ee_block),
688                                 ext_pblock(path[depth].p_ext),
689                                 le16_to_cpu(path[depth].p_ext->ee_len),
690                                 newblock);
691                 /*memmove(ex++, path[depth].p_ext++,
692                                 sizeof(struct ext4_extent));
693                 neh->eh_entries++;*/
694                 path[depth].p_ext++;
695                 m++;
696         }
697         if (m) {
698                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
699                 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
700         }
701
702         set_buffer_uptodate(bh);
703         unlock_buffer(bh);
704
705         err = ext4_journal_dirty_metadata(handle, bh);
706         if (err)
707                 goto cleanup;
708         brelse(bh);
709         bh = NULL;
710
711         /* correct old leaf */
712         if (m) {
713                 err = ext4_ext_get_access(handle, inode, path + depth);
714                 if (err)
715                         goto cleanup;
716                 path[depth].p_hdr->eh_entries =
717                      cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
718                 err = ext4_ext_dirty(handle, inode, path + depth);
719                 if (err)
720                         goto cleanup;
721
722         }
723
724         /* create intermediate indexes */
725         k = depth - at - 1;
726         BUG_ON(k < 0);
727         if (k)
728                 ext_debug("create %d intermediate indices\n", k);
729         /* insert new index into current index block */
730         /* current depth stored in i var */
731         i = depth - 1;
732         while (k--) {
733                 oldblock = newblock;
734                 newblock = ablocks[--a];
735                 bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock);
736                 if (!bh) {
737                         err = -EIO;
738                         goto cleanup;
739                 }
740                 lock_buffer(bh);
741
742                 err = ext4_journal_get_create_access(handle, bh);
743                 if (err)
744                         goto cleanup;
745
746                 neh = ext_block_hdr(bh);
747                 neh->eh_entries = cpu_to_le16(1);
748                 neh->eh_magic = EXT4_EXT_MAGIC;
749                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
750                 neh->eh_depth = cpu_to_le16(depth - i);
751                 fidx = EXT_FIRST_INDEX(neh);
752                 fidx->ei_block = border;
753                 ext4_idx_store_pblock(fidx, oldblock);
754
755                 ext_debug("int.index at %d (block %llu): %lu -> %llu\n", i,
756                                 newblock, (unsigned long) le32_to_cpu(border),
757                                 oldblock);
758                 /* copy indexes */
759                 m = 0;
760                 path[i].p_idx++;
761
762                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
763                                 EXT_MAX_INDEX(path[i].p_hdr));
764                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
765                                 EXT_LAST_INDEX(path[i].p_hdr));
766                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
767                         ext_debug("%d: move %d:%d in new index %llu\n", i,
768                                         le32_to_cpu(path[i].p_idx->ei_block),
769                                         idx_pblock(path[i].p_idx),
770                                         newblock);
771                         /*memmove(++fidx, path[i].p_idx++,
772                                         sizeof(struct ext4_extent_idx));
773                         neh->eh_entries++;
774                         BUG_ON(neh->eh_entries > neh->eh_max);*/
775                         path[i].p_idx++;
776                         m++;
777                 }
778                 if (m) {
779                         memmove(++fidx, path[i].p_idx - m,
780                                 sizeof(struct ext4_extent_idx) * m);
781                         neh->eh_entries =
782                                 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
783                 }
784                 set_buffer_uptodate(bh);
785                 unlock_buffer(bh);
786
787                 err = ext4_journal_dirty_metadata(handle, bh);
788                 if (err)
789                         goto cleanup;
790                 brelse(bh);
791                 bh = NULL;
792
793                 /* correct old index */
794                 if (m) {
795                         err = ext4_ext_get_access(handle, inode, path + i);
796                         if (err)
797                                 goto cleanup;
798                         path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
799                         err = ext4_ext_dirty(handle, inode, path + i);
800                         if (err)
801                                 goto cleanup;
802                 }
803
804                 i--;
805         }
806
807         /* insert new index */
808         err = ext4_ext_insert_index(handle, inode, path + at,
809                                     le32_to_cpu(border), newblock);
810
811 cleanup:
812         if (bh) {
813                 if (buffer_locked(bh))
814                         unlock_buffer(bh);
815                 brelse(bh);
816         }
817
818         if (err) {
819                 /* free all allocated blocks in error case */
820                 for (i = 0; i < depth; i++) {
821                         if (!ablocks[i])
822                                 continue;
823                         ext4_free_blocks(handle, inode, ablocks[i], 1);
824                 }
825         }
826         kfree(ablocks);
827
828         return err;
829 }
830
831 /*
832  * ext4_ext_grow_indepth:
833  * implements tree growing procedure:
834  * - allocates new block
835  * - moves top-level data (index block or leaf) into the new block
836  * - initializes new top-level, creating index that points to the
837  *   just created block
838  */
839 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
840                                         struct ext4_ext_path *path,
841                                         struct ext4_extent *newext)
842 {
843         struct ext4_ext_path *curp = path;
844         struct ext4_extent_header *neh;
845         struct ext4_extent_idx *fidx;
846         struct buffer_head *bh;
847         ext4_fsblk_t newblock;
848         int err = 0;
849
850         newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
851         if (newblock == 0)
852                 return err;
853
854         bh = sb_getblk(inode->i_sb, newblock);
855         if (!bh) {
856                 err = -EIO;
857                 ext4_std_error(inode->i_sb, err);
858                 return err;
859         }
860         lock_buffer(bh);
861
862         err = ext4_journal_get_create_access(handle, bh);
863         if (err) {
864                 unlock_buffer(bh);
865                 goto out;
866         }
867
868         /* move top-level index/leaf into new block */
869         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
870
871         /* set size of new block */
872         neh = ext_block_hdr(bh);
873         /* old root could have indexes or leaves
874          * so calculate e_max right way */
875         if (ext_depth(inode))
876           neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
877         else
878           neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
879         neh->eh_magic = EXT4_EXT_MAGIC;
880         set_buffer_uptodate(bh);
881         unlock_buffer(bh);
882
883         err = ext4_journal_dirty_metadata(handle, bh);
884         if (err)
885                 goto out;
886
887         /* create index in new top-level index: num,max,pointer */
888         err = ext4_ext_get_access(handle, inode, curp);
889         if (err)
890                 goto out;
891
892         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
893         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
894         curp->p_hdr->eh_entries = cpu_to_le16(1);
895         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
896         /* FIXME: it works, but actually path[0] can be index */
897         curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
898         ext4_idx_store_pblock(curp->p_idx, newblock);
899
900         neh = ext_inode_hdr(inode);
901         fidx = EXT_FIRST_INDEX(neh);
902         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
903                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
904                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
905
906         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
907         err = ext4_ext_dirty(handle, inode, curp);
908 out:
909         brelse(bh);
910
911         return err;
912 }
913
914 /*
915  * ext4_ext_create_new_leaf:
916  * finds empty index and adds new leaf.
917  * if no free index is found, then it requests in-depth growing.
918  */
919 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
920                                         struct ext4_ext_path *path,
921                                         struct ext4_extent *newext)
922 {
923         struct ext4_ext_path *curp;
924         int depth, i, err = 0;
925
926 repeat:
927         i = depth = ext_depth(inode);
928
929         /* walk up to the tree and look for free index entry */
930         curp = path + depth;
931         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
932                 i--;
933                 curp--;
934         }
935
936         /* we use already allocated block for index block,
937          * so subsequent data blocks should be contiguous */
938         if (EXT_HAS_FREE_INDEX(curp)) {
939                 /* if we found index with free entry, then use that
940                  * entry: create all needed subtree and add new leaf */
941                 err = ext4_ext_split(handle, inode, path, newext, i);
942
943                 /* refill path */
944                 ext4_ext_drop_refs(path);
945                 path = ext4_ext_find_extent(inode,
946                                             le32_to_cpu(newext->ee_block),
947                                             path);
948                 if (IS_ERR(path))
949                         err = PTR_ERR(path);
950         } else {
951                 /* tree is full, time to grow in depth */
952                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
953                 if (err)
954                         goto out;
955
956                 /* refill path */
957                 ext4_ext_drop_refs(path);
958                 path = ext4_ext_find_extent(inode,
959                                             le32_to_cpu(newext->ee_block),
960                                             path);
961                 if (IS_ERR(path)) {
962                         err = PTR_ERR(path);
963                         goto out;
964                 }
965
966                 /*
967                  * only first (depth 0 -> 1) produces free space;
968                  * in all other cases we have to split the grown tree
969                  */
970                 depth = ext_depth(inode);
971                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
972                         /* now we need to split */
973                         goto repeat;
974                 }
975         }
976
977 out:
978         return err;
979 }
980
981 /*
982  * ext4_ext_next_allocated_block:
983  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
984  * NOTE: it considers block number from index entry as
985  * allocated block. Thus, index entries have to be consistent
986  * with leaves.
987  */
988 static unsigned long
989 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
990 {
991         int depth;
992
993         BUG_ON(path == NULL);
994         depth = path->p_depth;
995
996         if (depth == 0 && path->p_ext == NULL)
997                 return EXT_MAX_BLOCK;
998
999         while (depth >= 0) {
1000                 if (depth == path->p_depth) {
1001                         /* leaf */
1002                         if (path[depth].p_ext !=
1003                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1004                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1005                 } else {
1006                         /* index */
1007                         if (path[depth].p_idx !=
1008                                         EXT_LAST_INDEX(path[depth].p_hdr))
1009                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1010                 }
1011                 depth--;
1012         }
1013
1014         return EXT_MAX_BLOCK;
1015 }
1016
1017 /*
1018  * ext4_ext_next_leaf_block:
1019  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1020  */
1021 static unsigned ext4_ext_next_leaf_block(struct inode *inode,
1022                                         struct ext4_ext_path *path)
1023 {
1024         int depth;
1025
1026         BUG_ON(path == NULL);
1027         depth = path->p_depth;
1028
1029         /* zero-tree has no leaf blocks at all */
1030         if (depth == 0)
1031                 return EXT_MAX_BLOCK;
1032
1033         /* go to index block */
1034         depth--;
1035
1036         while (depth >= 0) {
1037                 if (path[depth].p_idx !=
1038                                 EXT_LAST_INDEX(path[depth].p_hdr))
1039                   return le32_to_cpu(path[depth].p_idx[1].ei_block);
1040                 depth--;
1041         }
1042
1043         return EXT_MAX_BLOCK;
1044 }
1045
1046 /*
1047  * ext4_ext_correct_indexes:
1048  * if leaf gets modified and modified extent is first in the leaf,
1049  * then we have to correct all indexes above.
1050  * TODO: do we need to correct tree in all cases?
1051  */
1052 int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1053                                 struct ext4_ext_path *path)
1054 {
1055         struct ext4_extent_header *eh;
1056         int depth = ext_depth(inode);
1057         struct ext4_extent *ex;
1058         __le32 border;
1059         int k, err = 0;
1060
1061         eh = path[depth].p_hdr;
1062         ex = path[depth].p_ext;
1063         BUG_ON(ex == NULL);
1064         BUG_ON(eh == NULL);
1065
1066         if (depth == 0) {
1067                 /* there is no tree at all */
1068                 return 0;
1069         }
1070
1071         if (ex != EXT_FIRST_EXTENT(eh)) {
1072                 /* we correct tree if first leaf got modified only */
1073                 return 0;
1074         }
1075
1076         /*
1077          * TODO: we need correction if border is smaller than current one
1078          */
1079         k = depth - 1;
1080         border = path[depth].p_ext->ee_block;
1081         err = ext4_ext_get_access(handle, inode, path + k);
1082         if (err)
1083                 return err;
1084         path[k].p_idx->ei_block = border;
1085         err = ext4_ext_dirty(handle, inode, path + k);
1086         if (err)
1087                 return err;
1088
1089         while (k--) {
1090                 /* change all left-side indexes */
1091                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1092                         break;
1093                 err = ext4_ext_get_access(handle, inode, path + k);
1094                 if (err)
1095                         break;
1096                 path[k].p_idx->ei_block = border;
1097                 err = ext4_ext_dirty(handle, inode, path + k);
1098                 if (err)
1099                         break;
1100         }
1101
1102         return err;
1103 }
1104
1105 static int
1106 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1107                                 struct ext4_extent *ex2)
1108 {
1109         if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len) !=
1110                         le32_to_cpu(ex2->ee_block))
1111                 return 0;
1112
1113         /*
1114          * To allow future support for preallocated extents to be added
1115          * as an RO_COMPAT feature, refuse to merge to extents if
1116          * this can result in the top bit of ee_len being set.
1117          */
1118         if (le16_to_cpu(ex1->ee_len) + le16_to_cpu(ex2->ee_len) > EXT_MAX_LEN)
1119                 return 0;
1120 #ifdef AGGRESSIVE_TEST
1121         if (le16_to_cpu(ex1->ee_len) >= 4)
1122                 return 0;
1123 #endif
1124
1125         if (ext_pblock(ex1) + le16_to_cpu(ex1->ee_len) == ext_pblock(ex2))
1126                 return 1;
1127         return 0;
1128 }
1129
1130 /*
1131  * check if a portion of the "newext" extent overlaps with an
1132  * existing extent.
1133  *
1134  * If there is an overlap discovered, it updates the length of the newext
1135  * such that there will be no overlap, and then returns 1.
1136  * If there is no overlap found, it returns 0.
1137  */
1138 unsigned int ext4_ext_check_overlap(struct inode *inode,
1139                                     struct ext4_extent *newext,
1140                                     struct ext4_ext_path *path)
1141 {
1142         unsigned long b1, b2;
1143         unsigned int depth, len1;
1144         unsigned int ret = 0;
1145
1146         b1 = le32_to_cpu(newext->ee_block);
1147         len1 = le16_to_cpu(newext->ee_len);
1148         depth = ext_depth(inode);
1149         if (!path[depth].p_ext)
1150                 goto out;
1151         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1152
1153         /*
1154          * get the next allocated block if the extent in the path
1155          * is before the requested block(s) 
1156          */
1157         if (b2 < b1) {
1158                 b2 = ext4_ext_next_allocated_block(path);
1159                 if (b2 == EXT_MAX_BLOCK)
1160                         goto out;
1161         }
1162
1163         /* check for wrap through zero */
1164         if (b1 + len1 < b1) {
1165                 len1 = EXT_MAX_BLOCK - b1;
1166                 newext->ee_len = cpu_to_le16(len1);
1167                 ret = 1;
1168         }
1169
1170         /* check for overlap */
1171         if (b1 + len1 > b2) {
1172                 newext->ee_len = cpu_to_le16(b2 - b1);
1173                 ret = 1;
1174         }
1175 out:
1176         return ret;
1177 }
1178
1179 /*
1180  * ext4_ext_insert_extent:
1181  * tries to merge requsted extent into the existing extent or
1182  * inserts requested extent as new one into the tree,
1183  * creating new leaf in the no-space case.
1184  */
1185 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1186                                 struct ext4_ext_path *path,
1187                                 struct ext4_extent *newext)
1188 {
1189         struct ext4_extent_header * eh;
1190         struct ext4_extent *ex, *fex;
1191         struct ext4_extent *nearex; /* nearest extent */
1192         struct ext4_ext_path *npath = NULL;
1193         int depth, len, err, next;
1194
1195         BUG_ON(newext->ee_len == 0);
1196         depth = ext_depth(inode);
1197         ex = path[depth].p_ext;
1198         BUG_ON(path[depth].p_hdr == NULL);
1199
1200         /* try to insert block into found extent and return */
1201         if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1202                 ext_debug("append %d block to %d:%d (from %llu)\n",
1203                                 le16_to_cpu(newext->ee_len),
1204                                 le32_to_cpu(ex->ee_block),
1205                                 le16_to_cpu(ex->ee_len), ext_pblock(ex));
1206                 err = ext4_ext_get_access(handle, inode, path + depth);
1207                 if (err)
1208                         return err;
1209                 ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len)
1210                                          + le16_to_cpu(newext->ee_len));
1211                 eh = path[depth].p_hdr;
1212                 nearex = ex;
1213                 goto merge;
1214         }
1215
1216 repeat:
1217         depth = ext_depth(inode);
1218         eh = path[depth].p_hdr;
1219         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1220                 goto has_space;
1221
1222         /* probably next leaf has space for us? */
1223         fex = EXT_LAST_EXTENT(eh);
1224         next = ext4_ext_next_leaf_block(inode, path);
1225         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1226             && next != EXT_MAX_BLOCK) {
1227                 ext_debug("next leaf block - %d\n", next);
1228                 BUG_ON(npath != NULL);
1229                 npath = ext4_ext_find_extent(inode, next, NULL);
1230                 if (IS_ERR(npath))
1231                         return PTR_ERR(npath);
1232                 BUG_ON(npath->p_depth != path->p_depth);
1233                 eh = npath[depth].p_hdr;
1234                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1235                         ext_debug("next leaf isnt full(%d)\n",
1236                                   le16_to_cpu(eh->eh_entries));
1237                         path = npath;
1238                         goto repeat;
1239                 }
1240                 ext_debug("next leaf has no free space(%d,%d)\n",
1241                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1242         }
1243
1244         /*
1245          * There is no free space in the found leaf.
1246          * We're gonna add a new leaf in the tree.
1247          */
1248         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1249         if (err)
1250                 goto cleanup;
1251         depth = ext_depth(inode);
1252         eh = path[depth].p_hdr;
1253
1254 has_space:
1255         nearex = path[depth].p_ext;
1256
1257         err = ext4_ext_get_access(handle, inode, path + depth);
1258         if (err)
1259                 goto cleanup;
1260
1261         if (!nearex) {
1262                 /* there is no extent in this leaf, create first one */
1263                 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1264                                 le32_to_cpu(newext->ee_block),
1265                                 ext_pblock(newext),
1266                                 le16_to_cpu(newext->ee_len));
1267                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1268         } else if (le32_to_cpu(newext->ee_block)
1269                            > le32_to_cpu(nearex->ee_block)) {
1270 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1271                 if (nearex != EXT_LAST_EXTENT(eh)) {
1272                         len = EXT_MAX_EXTENT(eh) - nearex;
1273                         len = (len - 1) * sizeof(struct ext4_extent);
1274                         len = len < 0 ? 0 : len;
1275                         ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1276                                         "move %d from 0x%p to 0x%p\n",
1277                                         le32_to_cpu(newext->ee_block),
1278                                         ext_pblock(newext),
1279                                         le16_to_cpu(newext->ee_len),
1280                                         nearex, len, nearex + 1, nearex + 2);
1281                         memmove(nearex + 2, nearex + 1, len);
1282                 }
1283                 path[depth].p_ext = nearex + 1;
1284         } else {
1285                 BUG_ON(newext->ee_block == nearex->ee_block);
1286                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1287                 len = len < 0 ? 0 : len;
1288                 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1289                                 "move %d from 0x%p to 0x%p\n",
1290                                 le32_to_cpu(newext->ee_block),
1291                                 ext_pblock(newext),
1292                                 le16_to_cpu(newext->ee_len),
1293                                 nearex, len, nearex + 1, nearex + 2);
1294                 memmove(nearex + 1, nearex, len);
1295                 path[depth].p_ext = nearex;
1296         }
1297
1298         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1299         nearex = path[depth].p_ext;
1300         nearex->ee_block = newext->ee_block;
1301         nearex->ee_start = newext->ee_start;
1302         nearex->ee_start_hi = newext->ee_start_hi;
1303         nearex->ee_len = newext->ee_len;
1304
1305 merge:
1306         /* try to merge extents to the right */
1307         while (nearex < EXT_LAST_EXTENT(eh)) {
1308                 if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1))
1309                         break;
1310                 /* merge with next extent! */
1311                 nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len)
1312                                              + le16_to_cpu(nearex[1].ee_len));
1313                 if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1314                         len = (EXT_LAST_EXTENT(eh) - nearex - 1)
1315                                         * sizeof(struct ext4_extent);
1316                         memmove(nearex + 1, nearex + 2, len);
1317                 }
1318                 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1319                 BUG_ON(eh->eh_entries == 0);
1320         }
1321
1322         /* try to merge extents to the left */
1323
1324         /* time to correct all indexes above */
1325         err = ext4_ext_correct_indexes(handle, inode, path);
1326         if (err)
1327                 goto cleanup;
1328
1329         err = ext4_ext_dirty(handle, inode, path + depth);
1330
1331 cleanup:
1332         if (npath) {
1333                 ext4_ext_drop_refs(npath);
1334                 kfree(npath);
1335         }
1336         ext4_ext_tree_changed(inode);
1337         ext4_ext_invalidate_cache(inode);
1338         return err;
1339 }
1340
1341 int ext4_ext_walk_space(struct inode *inode, unsigned long block,
1342                         unsigned long num, ext_prepare_callback func,
1343                         void *cbdata)
1344 {
1345         struct ext4_ext_path *path = NULL;
1346         struct ext4_ext_cache cbex;
1347         struct ext4_extent *ex;
1348         unsigned long next, start = 0, end = 0;
1349         unsigned long last = block + num;
1350         int depth, exists, err = 0;
1351
1352         BUG_ON(func == NULL);
1353         BUG_ON(inode == NULL);
1354
1355         while (block < last && block != EXT_MAX_BLOCK) {
1356                 num = last - block;
1357                 /* find extent for this block */
1358                 path = ext4_ext_find_extent(inode, block, path);
1359                 if (IS_ERR(path)) {
1360                         err = PTR_ERR(path);
1361                         path = NULL;
1362                         break;
1363                 }
1364
1365                 depth = ext_depth(inode);
1366                 BUG_ON(path[depth].p_hdr == NULL);
1367                 ex = path[depth].p_ext;
1368                 next = ext4_ext_next_allocated_block(path);
1369
1370                 exists = 0;
1371                 if (!ex) {
1372                         /* there is no extent yet, so try to allocate
1373                          * all requested space */
1374                         start = block;
1375                         end = block + num;
1376                 } else if (le32_to_cpu(ex->ee_block) > block) {
1377                         /* need to allocate space before found extent */
1378                         start = block;
1379                         end = le32_to_cpu(ex->ee_block);
1380                         if (block + num < end)
1381                                 end = block + num;
1382                 } else if (block >=
1383                              le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) {
1384                         /* need to allocate space after found extent */
1385                         start = block;
1386                         end = block + num;
1387                         if (end >= next)
1388                                 end = next;
1389                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1390                         /*
1391                          * some part of requested space is covered
1392                          * by found extent
1393                          */
1394                         start = block;
1395                         end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len);
1396                         if (block + num < end)
1397                                 end = block + num;
1398                         exists = 1;
1399                 } else {
1400                         BUG();
1401                 }
1402                 BUG_ON(end <= start);
1403
1404                 if (!exists) {
1405                         cbex.ec_block = start;
1406                         cbex.ec_len = end - start;
1407                         cbex.ec_start = 0;
1408                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1409                 } else {
1410                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1411                         cbex.ec_len = le16_to_cpu(ex->ee_len);
1412                         cbex.ec_start = ext_pblock(ex);
1413                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1414                 }
1415
1416                 BUG_ON(cbex.ec_len == 0);
1417                 err = func(inode, path, &cbex, cbdata);
1418                 ext4_ext_drop_refs(path);
1419
1420                 if (err < 0)
1421                         break;
1422                 if (err == EXT_REPEAT)
1423                         continue;
1424                 else if (err == EXT_BREAK) {
1425                         err = 0;
1426                         break;
1427                 }
1428
1429                 if (ext_depth(inode) != depth) {
1430                         /* depth was changed. we have to realloc path */
1431                         kfree(path);
1432                         path = NULL;
1433                 }
1434
1435                 block = cbex.ec_block + cbex.ec_len;
1436         }
1437
1438         if (path) {
1439                 ext4_ext_drop_refs(path);
1440                 kfree(path);
1441         }
1442
1443         return err;
1444 }
1445
1446 static void
1447 ext4_ext_put_in_cache(struct inode *inode, __u32 block,
1448                         __u32 len, __u32 start, int type)
1449 {
1450         struct ext4_ext_cache *cex;
1451         BUG_ON(len == 0);
1452         cex = &EXT4_I(inode)->i_cached_extent;
1453         cex->ec_type = type;
1454         cex->ec_block = block;
1455         cex->ec_len = len;
1456         cex->ec_start = start;
1457 }
1458
1459 /*
1460  * ext4_ext_put_gap_in_cache:
1461  * calculate boundaries of the gap that the requested block fits into
1462  * and cache this gap
1463  */
1464 static void
1465 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1466                                 unsigned long block)
1467 {
1468         int depth = ext_depth(inode);
1469         unsigned long lblock, len;
1470         struct ext4_extent *ex;
1471
1472         ex = path[depth].p_ext;
1473         if (ex == NULL) {
1474                 /* there is no extent yet, so gap is [0;-] */
1475                 lblock = 0;
1476                 len = EXT_MAX_BLOCK;
1477                 ext_debug("cache gap(whole file):");
1478         } else if (block < le32_to_cpu(ex->ee_block)) {
1479                 lblock = block;
1480                 len = le32_to_cpu(ex->ee_block) - block;
1481                 ext_debug("cache gap(before): %lu [%lu:%lu]",
1482                                 (unsigned long) block,
1483                                 (unsigned long) le32_to_cpu(ex->ee_block),
1484                                 (unsigned long) le16_to_cpu(ex->ee_len));
1485         } else if (block >= le32_to_cpu(ex->ee_block)
1486                             + le16_to_cpu(ex->ee_len)) {
1487                 lblock = le32_to_cpu(ex->ee_block)
1488                          + le16_to_cpu(ex->ee_len);
1489                 len = ext4_ext_next_allocated_block(path);
1490                 ext_debug("cache gap(after): [%lu:%lu] %lu",
1491                                 (unsigned long) le32_to_cpu(ex->ee_block),
1492                                 (unsigned long) le16_to_cpu(ex->ee_len),
1493                                 (unsigned long) block);
1494                 BUG_ON(len == lblock);
1495                 len = len - lblock;
1496         } else {
1497                 lblock = len = 0;
1498                 BUG();
1499         }
1500
1501         ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len);
1502         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1503 }
1504
1505 static int
1506 ext4_ext_in_cache(struct inode *inode, unsigned long block,
1507                         struct ext4_extent *ex)
1508 {
1509         struct ext4_ext_cache *cex;
1510
1511         cex = &EXT4_I(inode)->i_cached_extent;
1512
1513         /* has cache valid data? */
1514         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1515                 return EXT4_EXT_CACHE_NO;
1516
1517         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1518                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1519         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1520                 ex->ee_block = cpu_to_le32(cex->ec_block);
1521                 ext4_ext_store_pblock(ex, cex->ec_start);
1522                 ex->ee_len = cpu_to_le16(cex->ec_len);
1523                 ext_debug("%lu cached by %lu:%lu:%llu\n",
1524                                 (unsigned long) block,
1525                                 (unsigned long) cex->ec_block,
1526                                 (unsigned long) cex->ec_len,
1527                                 cex->ec_start);
1528                 return cex->ec_type;
1529         }
1530
1531         /* not in cache */
1532         return EXT4_EXT_CACHE_NO;
1533 }
1534
1535 /*
1536  * ext4_ext_rm_idx:
1537  * removes index from the index block.
1538  * It's used in truncate case only, thus all requests are for
1539  * last index in the block only.
1540  */
1541 int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1542                         struct ext4_ext_path *path)
1543 {
1544         struct buffer_head *bh;
1545         int err;
1546         ext4_fsblk_t leaf;
1547
1548         /* free index block */
1549         path--;
1550         leaf = idx_pblock(path->p_idx);
1551         BUG_ON(path->p_hdr->eh_entries == 0);
1552         err = ext4_ext_get_access(handle, inode, path);
1553         if (err)
1554                 return err;
1555         path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1556         err = ext4_ext_dirty(handle, inode, path);
1557         if (err)
1558                 return err;
1559         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1560         bh = sb_find_get_block(inode->i_sb, leaf);
1561         ext4_forget(handle, 1, inode, bh, leaf);
1562         ext4_free_blocks(handle, inode, leaf, 1);
1563         return err;
1564 }
1565
1566 /*
1567  * ext4_ext_calc_credits_for_insert:
1568  * This routine returns max. credits that the extent tree can consume.
1569  * It should be OK for low-performance paths like ->writepage()
1570  * To allow many writing processes to fit into a single transaction,
1571  * the caller should calculate credits under truncate_mutex and
1572  * pass the actual path.
1573  */
1574 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1575                                                 struct ext4_ext_path *path)
1576 {
1577         int depth, needed;
1578
1579         if (path) {
1580                 /* probably there is space in leaf? */
1581                 depth = ext_depth(inode);
1582                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1583                                 < le16_to_cpu(path[depth].p_hdr->eh_max))
1584                         return 1;
1585         }
1586
1587         /*
1588          * given 32-bit logical block (4294967296 blocks), max. tree
1589          * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1590          * Let's also add one more level for imbalance.
1591          */
1592         depth = 5;
1593
1594         /* allocation of new data block(s) */
1595         needed = 2;
1596
1597         /*
1598          * tree can be full, so it would need to grow in depth:
1599          * we need one credit to modify old root, credits for
1600          * new root will be added in split accounting
1601          */
1602         needed += 1;
1603
1604         /*
1605          * Index split can happen, we would need:
1606          *    allocate intermediate indexes (bitmap + group)
1607          *  + change two blocks at each level, but root (already included)
1608          */
1609         needed += (depth * 2) + (depth * 2);
1610
1611         /* any allocation modifies superblock */
1612         needed += 1;
1613
1614         return needed;
1615 }
1616
1617 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1618                                 struct ext4_extent *ex,
1619                                 unsigned long from, unsigned long to)
1620 {
1621         struct buffer_head *bh;
1622         int i;
1623
1624 #ifdef EXTENTS_STATS
1625         {
1626                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1627                 unsigned short ee_len =  le16_to_cpu(ex->ee_len);
1628                 spin_lock(&sbi->s_ext_stats_lock);
1629                 sbi->s_ext_blocks += ee_len;
1630                 sbi->s_ext_extents++;
1631                 if (ee_len < sbi->s_ext_min)
1632                         sbi->s_ext_min = ee_len;
1633                 if (ee_len > sbi->s_ext_max)
1634                         sbi->s_ext_max = ee_len;
1635                 if (ext_depth(inode) > sbi->s_depth_max)
1636                         sbi->s_depth_max = ext_depth(inode);
1637                 spin_unlock(&sbi->s_ext_stats_lock);
1638         }
1639 #endif
1640         if (from >= le32_to_cpu(ex->ee_block)
1641             && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1642                 /* tail removal */
1643                 unsigned long num;
1644                 ext4_fsblk_t start;
1645                 num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from;
1646                 start = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - num;
1647                 ext_debug("free last %lu blocks starting %llu\n", num, start);
1648                 for (i = 0; i < num; i++) {
1649                         bh = sb_find_get_block(inode->i_sb, start + i);
1650                         ext4_forget(handle, 0, inode, bh, start + i);
1651                 }
1652                 ext4_free_blocks(handle, inode, start, num);
1653         } else if (from == le32_to_cpu(ex->ee_block)
1654                    && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1655                 printk("strange request: removal %lu-%lu from %u:%u\n",
1656                        from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1657         } else {
1658                 printk("strange request: removal(2) %lu-%lu from %u:%u\n",
1659                        from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1660         }
1661         return 0;
1662 }
1663
1664 static int
1665 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1666                 struct ext4_ext_path *path, unsigned long start)
1667 {
1668         int err = 0, correct_index = 0;
1669         int depth = ext_depth(inode), credits;
1670         struct ext4_extent_header *eh;
1671         unsigned a, b, block, num;
1672         unsigned long ex_ee_block;
1673         unsigned short ex_ee_len;
1674         struct ext4_extent *ex;
1675
1676         ext_debug("truncate since %lu in leaf\n", start);
1677         if (!path[depth].p_hdr)
1678                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1679         eh = path[depth].p_hdr;
1680         BUG_ON(eh == NULL);
1681         BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
1682         BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
1683
1684         /* find where to start removing */
1685         ex = EXT_LAST_EXTENT(eh);
1686
1687         ex_ee_block = le32_to_cpu(ex->ee_block);
1688         ex_ee_len = le16_to_cpu(ex->ee_len);
1689
1690         while (ex >= EXT_FIRST_EXTENT(eh) &&
1691                         ex_ee_block + ex_ee_len > start) {
1692                 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1693                 path[depth].p_ext = ex;
1694
1695                 a = ex_ee_block > start ? ex_ee_block : start;
1696                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1697                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1698
1699                 ext_debug("  border %u:%u\n", a, b);
1700
1701                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1702                         block = 0;
1703                         num = 0;
1704                         BUG();
1705                 } else if (a != ex_ee_block) {
1706                         /* remove tail of the extent */
1707                         block = ex_ee_block;
1708                         num = a - block;
1709                 } else if (b != ex_ee_block + ex_ee_len - 1) {
1710                         /* remove head of the extent */
1711                         block = a;
1712                         num = b - a;
1713                         /* there is no "make a hole" API yet */
1714                         BUG();
1715                 } else {
1716                         /* remove whole extent: excellent! */
1717                         block = ex_ee_block;
1718                         num = 0;
1719                         BUG_ON(a != ex_ee_block);
1720                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1721                 }
1722
1723                 /* at present, extent can't cross block group: */
1724                 /* leaf + bitmap + group desc + sb + inode */
1725                 credits = 5;
1726                 if (ex == EXT_FIRST_EXTENT(eh)) {
1727                         correct_index = 1;
1728                         credits += (ext_depth(inode)) + 1;
1729                 }
1730 #ifdef CONFIG_QUOTA
1731                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1732 #endif
1733
1734                 handle = ext4_ext_journal_restart(handle, credits);
1735                 if (IS_ERR(handle)) {
1736                         err = PTR_ERR(handle);
1737                         goto out;
1738                 }
1739
1740                 err = ext4_ext_get_access(handle, inode, path + depth);
1741                 if (err)
1742                         goto out;
1743
1744                 err = ext4_remove_blocks(handle, inode, ex, a, b);
1745                 if (err)
1746                         goto out;
1747
1748                 if (num == 0) {
1749                         /* this extent is removed; mark slot entirely unused */
1750                         ext4_ext_store_pblock(ex, 0);
1751                         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1752                 }
1753
1754                 ex->ee_block = cpu_to_le32(block);
1755                 ex->ee_len = cpu_to_le16(num);
1756
1757                 err = ext4_ext_dirty(handle, inode, path + depth);
1758                 if (err)
1759                         goto out;
1760
1761                 ext_debug("new extent: %u:%u:%llu\n", block, num,
1762                                 ext_pblock(ex));
1763                 ex--;
1764                 ex_ee_block = le32_to_cpu(ex->ee_block);
1765                 ex_ee_len = le16_to_cpu(ex->ee_len);
1766         }
1767
1768         if (correct_index && eh->eh_entries)
1769                 err = ext4_ext_correct_indexes(handle, inode, path);
1770
1771         /* if this leaf is free, then we should
1772          * remove it from index block above */
1773         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1774                 err = ext4_ext_rm_idx(handle, inode, path + depth);
1775
1776 out:
1777         return err;
1778 }
1779
1780 /*
1781  * ext4_ext_more_to_rm:
1782  * returns 1 if current index has to be freed (even partial)
1783  */
1784 static int
1785 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1786 {
1787         BUG_ON(path->p_idx == NULL);
1788
1789         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1790                 return 0;
1791
1792         /*
1793          * if truncate on deeper level happened, it wasn't partial,
1794          * so we have to consider current index for truncation
1795          */
1796         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1797                 return 0;
1798         return 1;
1799 }
1800
1801 int ext4_ext_remove_space(struct inode *inode, unsigned long start)
1802 {
1803         struct super_block *sb = inode->i_sb;
1804         int depth = ext_depth(inode);
1805         struct ext4_ext_path *path;
1806         handle_t *handle;
1807         int i = 0, err = 0;
1808
1809         ext_debug("truncate since %lu\n", start);
1810
1811         /* probably first extent we're gonna free will be last in block */
1812         handle = ext4_journal_start(inode, depth + 1);
1813         if (IS_ERR(handle))
1814                 return PTR_ERR(handle);
1815
1816         ext4_ext_invalidate_cache(inode);
1817
1818         /*
1819          * We start scanning from right side, freeing all the blocks
1820          * after i_size and walking into the tree depth-wise.
1821          */
1822         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1823         if (path == NULL) {
1824                 ext4_journal_stop(handle);
1825                 return -ENOMEM;
1826         }
1827         path[0].p_hdr = ext_inode_hdr(inode);
1828         if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) {
1829                 err = -EIO;
1830                 goto out;
1831         }
1832         path[0].p_depth = depth;
1833
1834         while (i >= 0 && err == 0) {
1835                 if (i == depth) {
1836                         /* this is leaf block */
1837                         err = ext4_ext_rm_leaf(handle, inode, path, start);
1838                         /* root level has p_bh == NULL, brelse() eats this */
1839                         brelse(path[i].p_bh);
1840                         path[i].p_bh = NULL;
1841                         i--;
1842                         continue;
1843                 }
1844
1845                 /* this is index block */
1846                 if (!path[i].p_hdr) {
1847                         ext_debug("initialize header\n");
1848                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1849                         if (ext4_ext_check_header(__FUNCTION__, inode,
1850                                                         path[i].p_hdr)) {
1851                                 err = -EIO;
1852                                 goto out;
1853                         }
1854                 }
1855
1856                 BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries)
1857                            > le16_to_cpu(path[i].p_hdr->eh_max));
1858                 BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC);
1859
1860                 if (!path[i].p_idx) {
1861                         /* this level hasn't been touched yet */
1862                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1863                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1864                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
1865                                   path[i].p_hdr,
1866                                   le16_to_cpu(path[i].p_hdr->eh_entries));
1867                 } else {
1868                         /* we were already here, see at next index */
1869                         path[i].p_idx--;
1870                 }
1871
1872                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
1873                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
1874                                 path[i].p_idx);
1875                 if (ext4_ext_more_to_rm(path + i)) {
1876                         /* go to the next level */
1877                         ext_debug("move to level %d (block %llu)\n",
1878                                   i + 1, idx_pblock(path[i].p_idx));
1879                         memset(path + i + 1, 0, sizeof(*path));
1880                         path[i+1].p_bh =
1881                                 sb_bread(sb, idx_pblock(path[i].p_idx));
1882                         if (!path[i+1].p_bh) {
1883                                 /* should we reset i_size? */
1884                                 err = -EIO;
1885                                 break;
1886                         }
1887
1888                         /* save actual number of indexes since this
1889                          * number is changed at the next iteration */
1890                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1891                         i++;
1892                 } else {
1893                         /* we finished processing this index, go up */
1894                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
1895                                 /* index is empty, remove it;
1896                                  * handle must be already prepared by the
1897                                  * truncatei_leaf() */
1898                                 err = ext4_ext_rm_idx(handle, inode, path + i);
1899                         }
1900                         /* root level has p_bh == NULL, brelse() eats this */
1901                         brelse(path[i].p_bh);
1902                         path[i].p_bh = NULL;
1903                         i--;
1904                         ext_debug("return to level %d\n", i);
1905                 }
1906         }
1907
1908         /* TODO: flexible tree reduction should be here */
1909         if (path->p_hdr->eh_entries == 0) {
1910                 /*
1911                  * truncate to zero freed all the tree,
1912                  * so we need to correct eh_depth
1913                  */
1914                 err = ext4_ext_get_access(handle, inode, path);
1915                 if (err == 0) {
1916                         ext_inode_hdr(inode)->eh_depth = 0;
1917                         ext_inode_hdr(inode)->eh_max =
1918                                 cpu_to_le16(ext4_ext_space_root(inode));
1919                         err = ext4_ext_dirty(handle, inode, path);
1920                 }
1921         }
1922 out:
1923         ext4_ext_tree_changed(inode);
1924         ext4_ext_drop_refs(path);
1925         kfree(path);
1926         ext4_journal_stop(handle);
1927
1928         return err;
1929 }
1930
1931 /*
1932  * called at mount time
1933  */
1934 void ext4_ext_init(struct super_block *sb)
1935 {
1936         /*
1937          * possible initialization would be here
1938          */
1939
1940         if (test_opt(sb, EXTENTS)) {
1941                 printk("EXT4-fs: file extents enabled");
1942 #ifdef AGGRESSIVE_TEST
1943                 printk(", aggressive tests");
1944 #endif
1945 #ifdef CHECK_BINSEARCH
1946                 printk(", check binsearch");
1947 #endif
1948 #ifdef EXTENTS_STATS
1949                 printk(", stats");
1950 #endif
1951                 printk("\n");
1952 #ifdef EXTENTS_STATS
1953                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
1954                 EXT4_SB(sb)->s_ext_min = 1 << 30;
1955                 EXT4_SB(sb)->s_ext_max = 0;
1956 #endif
1957         }
1958 }
1959
1960 /*
1961  * called at umount time
1962  */
1963 void ext4_ext_release(struct super_block *sb)
1964 {
1965         if (!test_opt(sb, EXTENTS))
1966                 return;
1967
1968 #ifdef EXTENTS_STATS
1969         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
1970                 struct ext4_sb_info *sbi = EXT4_SB(sb);
1971                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
1972                         sbi->s_ext_blocks, sbi->s_ext_extents,
1973                         sbi->s_ext_blocks / sbi->s_ext_extents);
1974                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
1975                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
1976         }
1977 #endif
1978 }
1979
1980 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
1981                         ext4_fsblk_t iblock,
1982                         unsigned long max_blocks, struct buffer_head *bh_result,
1983                         int create, int extend_disksize)
1984 {
1985         struct ext4_ext_path *path = NULL;
1986         struct ext4_extent newex, *ex;
1987         ext4_fsblk_t goal, newblock;
1988         int err = 0, depth;
1989         unsigned long allocated = 0;
1990
1991         __clear_bit(BH_New, &bh_result->b_state);
1992         ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock,
1993                         max_blocks, (unsigned) inode->i_ino);
1994         mutex_lock(&EXT4_I(inode)->truncate_mutex);
1995
1996         /* check in cache */
1997         goal = ext4_ext_in_cache(inode, iblock, &newex);
1998         if (goal) {
1999                 if (goal == EXT4_EXT_CACHE_GAP) {
2000                         if (!create) {
2001                                 /* block isn't allocated yet and
2002                                  * user doesn't want to allocate it */
2003                                 goto out2;
2004                         }
2005                         /* we should allocate requested block */
2006                 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2007                         /* block is already allocated */
2008                         newblock = iblock
2009                                    - le32_to_cpu(newex.ee_block)
2010                                    + ext_pblock(&newex);
2011                         /* number of remaining blocks in the extent */
2012                         allocated = le16_to_cpu(newex.ee_len) -
2013                                         (iblock - le32_to_cpu(newex.ee_block));
2014                         goto out;
2015                 } else {
2016                         BUG();
2017                 }
2018         }
2019
2020         /* find extent for this block */
2021         path = ext4_ext_find_extent(inode, iblock, NULL);
2022         if (IS_ERR(path)) {
2023                 err = PTR_ERR(path);
2024                 path = NULL;
2025                 goto out2;
2026         }
2027
2028         depth = ext_depth(inode);
2029
2030         /*
2031          * consistent leaf must not be empty;
2032          * this situation is possible, though, _during_ tree modification;
2033          * this is why assert can't be put in ext4_ext_find_extent()
2034          */
2035         BUG_ON(path[depth].p_ext == NULL && depth != 0);
2036
2037         ex = path[depth].p_ext;
2038         if (ex) {
2039                 unsigned long ee_block = le32_to_cpu(ex->ee_block);
2040                 ext4_fsblk_t ee_start = ext_pblock(ex);
2041                 unsigned short ee_len  = le16_to_cpu(ex->ee_len);
2042
2043                 /*
2044                  * Allow future support for preallocated extents to be added
2045                  * as an RO_COMPAT feature:
2046                  * Uninitialized extents are treated as holes, except that
2047                  * we avoid (fail) allocating new blocks during a write.
2048                  */
2049                 if (ee_len > EXT_MAX_LEN)
2050                         goto out2;
2051                 /* if found extent covers block, simply return it */
2052                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2053                         newblock = iblock - ee_block + ee_start;
2054                         /* number of remaining blocks in the extent */
2055                         allocated = ee_len - (iblock - ee_block);
2056                         ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock,
2057                                         ee_block, ee_len, newblock);
2058                         ext4_ext_put_in_cache(inode, ee_block, ee_len,
2059                                                 ee_start, EXT4_EXT_CACHE_EXTENT);
2060                         goto out;
2061                 }
2062         }
2063
2064         /*
2065          * requested block isn't allocated yet;
2066          * we couldn't try to create block if create flag is zero
2067          */
2068         if (!create) {
2069                 /* put just found gap into cache to speed up
2070                  * subsequent requests */
2071                 ext4_ext_put_gap_in_cache(inode, path, iblock);
2072                 goto out2;
2073         }
2074         /*
2075          * Okay, we need to do block allocation.  Lazily initialize the block
2076          * allocation info here if necessary.
2077          */
2078         if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2079                 ext4_init_block_alloc_info(inode);
2080
2081         /* allocate new block */
2082         goal = ext4_ext_find_goal(inode, path, iblock);
2083
2084         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2085         newex.ee_block = cpu_to_le32(iblock);
2086         newex.ee_len = cpu_to_le16(max_blocks);
2087         err = ext4_ext_check_overlap(inode, &newex, path);
2088         if (err)
2089                 allocated = le16_to_cpu(newex.ee_len);
2090         else
2091                 allocated = max_blocks;
2092         newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2093         if (!newblock)
2094                 goto out2;
2095         ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2096                         goal, newblock, allocated);
2097
2098         /* try to insert new extent into found leaf and return */
2099         ext4_ext_store_pblock(&newex, newblock);
2100         newex.ee_len = cpu_to_le16(allocated);
2101         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2102         if (err) {
2103                 /* free data blocks we just allocated */
2104                 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2105                                         le16_to_cpu(newex.ee_len));
2106                 goto out2;
2107         }
2108
2109         if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2110                 EXT4_I(inode)->i_disksize = inode->i_size;
2111
2112         /* previous routine could use block we allocated */
2113         newblock = ext_pblock(&newex);
2114         __set_bit(BH_New, &bh_result->b_state);
2115
2116         ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2117                                 EXT4_EXT_CACHE_EXTENT);
2118 out:
2119         if (allocated > max_blocks)
2120                 allocated = max_blocks;
2121         ext4_ext_show_leaf(inode, path);
2122         __set_bit(BH_Mapped, &bh_result->b_state);
2123         bh_result->b_bdev = inode->i_sb->s_bdev;
2124         bh_result->b_blocknr = newblock;
2125 out2:
2126         if (path) {
2127                 ext4_ext_drop_refs(path);
2128                 kfree(path);
2129         }
2130         mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2131
2132         return err ? err : allocated;
2133 }
2134
2135 void ext4_ext_truncate(struct inode * inode, struct page *page)
2136 {
2137         struct address_space *mapping = inode->i_mapping;
2138         struct super_block *sb = inode->i_sb;
2139         unsigned long last_block;
2140         handle_t *handle;
2141         int err = 0;
2142
2143         /*
2144          * probably first extent we're gonna free will be last in block
2145          */
2146         err = ext4_writepage_trans_blocks(inode) + 3;
2147         handle = ext4_journal_start(inode, err);
2148         if (IS_ERR(handle)) {
2149                 if (page) {
2150                         clear_highpage(page);
2151                         flush_dcache_page(page);
2152                         unlock_page(page);
2153                         page_cache_release(page);
2154                 }
2155                 return;
2156         }
2157
2158         if (page)
2159                 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2160
2161         mutex_lock(&EXT4_I(inode)->truncate_mutex);
2162         ext4_ext_invalidate_cache(inode);
2163
2164         /*
2165          * TODO: optimization is possible here.
2166          * Probably we need not scan at all,
2167          * because page truncation is enough.
2168          */
2169         if (ext4_orphan_add(handle, inode))
2170                 goto out_stop;
2171
2172         /* we have to know where to truncate from in crash case */
2173         EXT4_I(inode)->i_disksize = inode->i_size;
2174         ext4_mark_inode_dirty(handle, inode);
2175
2176         last_block = (inode->i_size + sb->s_blocksize - 1)
2177                         >> EXT4_BLOCK_SIZE_BITS(sb);
2178         err = ext4_ext_remove_space(inode, last_block);
2179
2180         /* In a multi-transaction truncate, we only make the final
2181          * transaction synchronous. */
2182         if (IS_SYNC(inode))
2183                 handle->h_sync = 1;
2184
2185 out_stop:
2186         /*
2187          * If this was a simple ftruncate() and the file will remain alive,
2188          * then we need to clear up the orphan record which we created above.
2189          * However, if this was a real unlink then we were called by
2190          * ext4_delete_inode(), and we allow that function to clean up the
2191          * orphan info for us.
2192          */
2193         if (inode->i_nlink)
2194                 ext4_orphan_del(handle, inode);
2195
2196         mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2197         ext4_journal_stop(handle);
2198 }
2199
2200 /*
2201  * ext4_ext_writepage_trans_blocks:
2202  * calculate max number of blocks we could modify
2203  * in order to allocate new block for an inode
2204  */
2205 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2206 {
2207         int needed;
2208
2209         needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2210
2211         /* caller wants to allocate num blocks, but note it includes sb */
2212         needed = needed * num - (num - 1);
2213
2214 #ifdef CONFIG_QUOTA
2215         needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2216 #endif
2217
2218         return needed;
2219 }