Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
[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/jbd2.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/falloc.h>
43 #include <linux/ext4_fs_extents.h>
44 #include <asm/uaccess.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 handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
96 {
97         int err;
98
99         if (handle->h_buffer_credits > needed)
100                 return handle;
101         if (!ext4_journal_extend(handle, needed))
102                 return handle;
103         err = ext4_journal_restart(handle, needed);
104
105         return handle;
106 }
107
108 /*
109  * could return:
110  *  - EROFS
111  *  - ENOMEM
112  */
113 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
114                                 struct ext4_ext_path *path)
115 {
116         if (path->p_bh) {
117                 /* path points to block */
118                 return ext4_journal_get_write_access(handle, path->p_bh);
119         }
120         /* path points to leaf/index in inode body */
121         /* we use in-core data, no need to protect them */
122         return 0;
123 }
124
125 /*
126  * could return:
127  *  - EROFS
128  *  - ENOMEM
129  *  - EIO
130  */
131 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
132                                 struct ext4_ext_path *path)
133 {
134         int err;
135         if (path->p_bh) {
136                 /* path points to block */
137                 err = ext4_journal_dirty_metadata(handle, path->p_bh);
138         } else {
139                 /* path points to leaf/index in inode body */
140                 err = ext4_mark_inode_dirty(handle, inode);
141         }
142         return err;
143 }
144
145 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
146                               struct ext4_ext_path *path,
147                               ext4_lblk_t block)
148 {
149         struct ext4_inode_info *ei = EXT4_I(inode);
150         ext4_fsblk_t bg_start;
151         ext4_fsblk_t last_block;
152         ext4_grpblk_t colour;
153         int depth;
154
155         if (path) {
156                 struct ext4_extent *ex;
157                 depth = path->p_depth;
158
159                 /* try to predict block placement */
160                 ex = path[depth].p_ext;
161                 if (ex)
162                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
163
164                 /* it looks like index is empty;
165                  * try to find starting block from index itself */
166                 if (path[depth].p_bh)
167                         return path[depth].p_bh->b_blocknr;
168         }
169
170         /* OK. use inode's group */
171         bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
172                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
173         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
174
175         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
176                 colour = (current->pid % 16) *
177                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
178         else
179                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
180         return bg_start + colour + block;
181 }
182
183 static ext4_fsblk_t
184 ext4_ext_new_block(handle_t *handle, struct inode *inode,
185                         struct ext4_ext_path *path,
186                         struct ext4_extent *ex, int *err)
187 {
188         ext4_fsblk_t goal, newblock;
189
190         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
191         newblock = ext4_new_block(handle, inode, goal, err);
192         return newblock;
193 }
194
195 static int ext4_ext_space_block(struct inode *inode)
196 {
197         int size;
198
199         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
200                         / sizeof(struct ext4_extent);
201 #ifdef AGGRESSIVE_TEST
202         if (size > 6)
203                 size = 6;
204 #endif
205         return size;
206 }
207
208 static int ext4_ext_space_block_idx(struct inode *inode)
209 {
210         int size;
211
212         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
213                         / sizeof(struct ext4_extent_idx);
214 #ifdef AGGRESSIVE_TEST
215         if (size > 5)
216                 size = 5;
217 #endif
218         return size;
219 }
220
221 static int ext4_ext_space_root(struct inode *inode)
222 {
223         int size;
224
225         size = sizeof(EXT4_I(inode)->i_data);
226         size -= sizeof(struct ext4_extent_header);
227         size /= sizeof(struct ext4_extent);
228 #ifdef AGGRESSIVE_TEST
229         if (size > 3)
230                 size = 3;
231 #endif
232         return size;
233 }
234
235 static int ext4_ext_space_root_idx(struct inode *inode)
236 {
237         int size;
238
239         size = sizeof(EXT4_I(inode)->i_data);
240         size -= sizeof(struct ext4_extent_header);
241         size /= sizeof(struct ext4_extent_idx);
242 #ifdef AGGRESSIVE_TEST
243         if (size > 4)
244                 size = 4;
245 #endif
246         return size;
247 }
248
249 static int
250 ext4_ext_max_entries(struct inode *inode, int depth)
251 {
252         int max;
253
254         if (depth == ext_depth(inode)) {
255                 if (depth == 0)
256                         max = ext4_ext_space_root(inode);
257                 else
258                         max = ext4_ext_space_root_idx(inode);
259         } else {
260                 if (depth == 0)
261                         max = ext4_ext_space_block(inode);
262                 else
263                         max = ext4_ext_space_block_idx(inode);
264         }
265
266         return max;
267 }
268
269 static int __ext4_ext_check_header(const char *function, struct inode *inode,
270                                         struct ext4_extent_header *eh,
271                                         int depth)
272 {
273         const char *error_msg;
274         int max = 0;
275
276         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
277                 error_msg = "invalid magic";
278                 goto corrupted;
279         }
280         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
281                 error_msg = "unexpected eh_depth";
282                 goto corrupted;
283         }
284         if (unlikely(eh->eh_max == 0)) {
285                 error_msg = "invalid eh_max";
286                 goto corrupted;
287         }
288         max = ext4_ext_max_entries(inode, depth);
289         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
290                 error_msg = "too large eh_max";
291                 goto corrupted;
292         }
293         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
294                 error_msg = "invalid eh_entries";
295                 goto corrupted;
296         }
297         return 0;
298
299 corrupted:
300         ext4_error(inode->i_sb, function,
301                         "bad header in inode #%lu: %s - magic %x, "
302                         "entries %u, max %u(%u), depth %u(%u)",
303                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
304                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
305                         max, le16_to_cpu(eh->eh_depth), depth);
306
307         return -EIO;
308 }
309
310 #define ext4_ext_check_header(inode, eh, depth) \
311         __ext4_ext_check_header(__FUNCTION__, inode, eh, depth)
312
313 #ifdef EXT_DEBUG
314 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
315 {
316         int k, l = path->p_depth;
317
318         ext_debug("path:");
319         for (k = 0; k <= l; k++, path++) {
320                 if (path->p_idx) {
321                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
322                             idx_pblock(path->p_idx));
323                 } else if (path->p_ext) {
324                         ext_debug("  %d:%d:%llu ",
325                                   le32_to_cpu(path->p_ext->ee_block),
326                                   ext4_ext_get_actual_len(path->p_ext),
327                                   ext_pblock(path->p_ext));
328                 } else
329                         ext_debug("  []");
330         }
331         ext_debug("\n");
332 }
333
334 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
335 {
336         int depth = ext_depth(inode);
337         struct ext4_extent_header *eh;
338         struct ext4_extent *ex;
339         int i;
340
341         if (!path)
342                 return;
343
344         eh = path[depth].p_hdr;
345         ex = EXT_FIRST_EXTENT(eh);
346
347         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
348                 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
349                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
350         }
351         ext_debug("\n");
352 }
353 #else
354 #define ext4_ext_show_path(inode,path)
355 #define ext4_ext_show_leaf(inode,path)
356 #endif
357
358 void ext4_ext_drop_refs(struct ext4_ext_path *path)
359 {
360         int depth = path->p_depth;
361         int i;
362
363         for (i = 0; i <= depth; i++, path++)
364                 if (path->p_bh) {
365                         brelse(path->p_bh);
366                         path->p_bh = NULL;
367                 }
368 }
369
370 /*
371  * ext4_ext_binsearch_idx:
372  * binary search for the closest index of the given block
373  * the header must be checked before calling this
374  */
375 static void
376 ext4_ext_binsearch_idx(struct inode *inode,
377                         struct ext4_ext_path *path, ext4_lblk_t block)
378 {
379         struct ext4_extent_header *eh = path->p_hdr;
380         struct ext4_extent_idx *r, *l, *m;
381
382
383         ext_debug("binsearch for %u(idx):  ", block);
384
385         l = EXT_FIRST_INDEX(eh) + 1;
386         r = EXT_LAST_INDEX(eh);
387         while (l <= r) {
388                 m = l + (r - l) / 2;
389                 if (block < le32_to_cpu(m->ei_block))
390                         r = m - 1;
391                 else
392                         l = m + 1;
393                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
394                                 m, le32_to_cpu(m->ei_block),
395                                 r, le32_to_cpu(r->ei_block));
396         }
397
398         path->p_idx = l - 1;
399         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
400                   idx_pblock(path->p_idx));
401
402 #ifdef CHECK_BINSEARCH
403         {
404                 struct ext4_extent_idx *chix, *ix;
405                 int k;
406
407                 chix = ix = EXT_FIRST_INDEX(eh);
408                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
409                   if (k != 0 &&
410                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
411                                 printk("k=%d, ix=0x%p, first=0x%p\n", k,
412                                         ix, EXT_FIRST_INDEX(eh));
413                                 printk("%u <= %u\n",
414                                        le32_to_cpu(ix->ei_block),
415                                        le32_to_cpu(ix[-1].ei_block));
416                         }
417                         BUG_ON(k && le32_to_cpu(ix->ei_block)
418                                            <= le32_to_cpu(ix[-1].ei_block));
419                         if (block < le32_to_cpu(ix->ei_block))
420                                 break;
421                         chix = ix;
422                 }
423                 BUG_ON(chix != path->p_idx);
424         }
425 #endif
426
427 }
428
429 /*
430  * ext4_ext_binsearch:
431  * binary search for closest extent of the given block
432  * the header must be checked before calling this
433  */
434 static void
435 ext4_ext_binsearch(struct inode *inode,
436                 struct ext4_ext_path *path, ext4_lblk_t block)
437 {
438         struct ext4_extent_header *eh = path->p_hdr;
439         struct ext4_extent *r, *l, *m;
440
441         if (eh->eh_entries == 0) {
442                 /*
443                  * this leaf is empty:
444                  * we get such a leaf in split/add case
445                  */
446                 return;
447         }
448
449         ext_debug("binsearch for %u:  ", block);
450
451         l = EXT_FIRST_EXTENT(eh) + 1;
452         r = EXT_LAST_EXTENT(eh);
453
454         while (l <= r) {
455                 m = l + (r - l) / 2;
456                 if (block < le32_to_cpu(m->ee_block))
457                         r = m - 1;
458                 else
459                         l = m + 1;
460                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
461                                 m, le32_to_cpu(m->ee_block),
462                                 r, le32_to_cpu(r->ee_block));
463         }
464
465         path->p_ext = l - 1;
466         ext_debug("  -> %d:%llu:%d ",
467                         le32_to_cpu(path->p_ext->ee_block),
468                         ext_pblock(path->p_ext),
469                         ext4_ext_get_actual_len(path->p_ext));
470
471 #ifdef CHECK_BINSEARCH
472         {
473                 struct ext4_extent *chex, *ex;
474                 int k;
475
476                 chex = ex = EXT_FIRST_EXTENT(eh);
477                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
478                         BUG_ON(k && le32_to_cpu(ex->ee_block)
479                                           <= le32_to_cpu(ex[-1].ee_block));
480                         if (block < le32_to_cpu(ex->ee_block))
481                                 break;
482                         chex = ex;
483                 }
484                 BUG_ON(chex != path->p_ext);
485         }
486 #endif
487
488 }
489
490 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
491 {
492         struct ext4_extent_header *eh;
493
494         eh = ext_inode_hdr(inode);
495         eh->eh_depth = 0;
496         eh->eh_entries = 0;
497         eh->eh_magic = EXT4_EXT_MAGIC;
498         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
499         ext4_mark_inode_dirty(handle, inode);
500         ext4_ext_invalidate_cache(inode);
501         return 0;
502 }
503
504 struct ext4_ext_path *
505 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
506                                         struct ext4_ext_path *path)
507 {
508         struct ext4_extent_header *eh;
509         struct buffer_head *bh;
510         short int depth, i, ppos = 0, alloc = 0;
511
512         eh = ext_inode_hdr(inode);
513         depth = ext_depth(inode);
514         if (ext4_ext_check_header(inode, eh, depth))
515                 return ERR_PTR(-EIO);
516
517
518         /* account possible depth increase */
519         if (!path) {
520                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
521                                 GFP_NOFS);
522                 if (!path)
523                         return ERR_PTR(-ENOMEM);
524                 alloc = 1;
525         }
526         path[0].p_hdr = eh;
527
528         i = depth;
529         /* walk through the tree */
530         while (i) {
531                 ext_debug("depth %d: num %d, max %d\n",
532                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
533
534                 ext4_ext_binsearch_idx(inode, path + ppos, block);
535                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
536                 path[ppos].p_depth = i;
537                 path[ppos].p_ext = NULL;
538
539                 bh = sb_bread(inode->i_sb, path[ppos].p_block);
540                 if (!bh)
541                         goto err;
542
543                 eh = ext_block_hdr(bh);
544                 ppos++;
545                 BUG_ON(ppos > depth);
546                 path[ppos].p_bh = bh;
547                 path[ppos].p_hdr = eh;
548                 i--;
549
550                 if (ext4_ext_check_header(inode, eh, i))
551                         goto err;
552         }
553
554         path[ppos].p_depth = i;
555         path[ppos].p_hdr = eh;
556         path[ppos].p_ext = NULL;
557         path[ppos].p_idx = NULL;
558
559         /* find extent */
560         ext4_ext_binsearch(inode, path + ppos, block);
561
562         ext4_ext_show_path(inode, path);
563
564         return path;
565
566 err:
567         ext4_ext_drop_refs(path);
568         if (alloc)
569                 kfree(path);
570         return ERR_PTR(-EIO);
571 }
572
573 /*
574  * ext4_ext_insert_index:
575  * insert new index [@logical;@ptr] into the block at @curp;
576  * check where to insert: before @curp or after @curp
577  */
578 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
579                                 struct ext4_ext_path *curp,
580                                 int logical, ext4_fsblk_t ptr)
581 {
582         struct ext4_extent_idx *ix;
583         int len, err;
584
585         err = ext4_ext_get_access(handle, inode, curp);
586         if (err)
587                 return err;
588
589         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
590         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
591         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
592                 /* insert after */
593                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
594                         len = (len - 1) * sizeof(struct ext4_extent_idx);
595                         len = len < 0 ? 0 : len;
596                         ext_debug("insert new index %d after: %llu. "
597                                         "move %d from 0x%p to 0x%p\n",
598                                         logical, ptr, len,
599                                         (curp->p_idx + 1), (curp->p_idx + 2));
600                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
601                 }
602                 ix = curp->p_idx + 1;
603         } else {
604                 /* insert before */
605                 len = len * sizeof(struct ext4_extent_idx);
606                 len = len < 0 ? 0 : len;
607                 ext_debug("insert new index %d before: %llu. "
608                                 "move %d from 0x%p to 0x%p\n",
609                                 logical, ptr, len,
610                                 curp->p_idx, (curp->p_idx + 1));
611                 memmove(curp->p_idx + 1, curp->p_idx, len);
612                 ix = curp->p_idx;
613         }
614
615         ix->ei_block = cpu_to_le32(logical);
616         ext4_idx_store_pblock(ix, ptr);
617         curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
618
619         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
620                              > le16_to_cpu(curp->p_hdr->eh_max));
621         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
622
623         err = ext4_ext_dirty(handle, inode, curp);
624         ext4_std_error(inode->i_sb, err);
625
626         return err;
627 }
628
629 /*
630  * ext4_ext_split:
631  * inserts new subtree into the path, using free index entry
632  * at depth @at:
633  * - allocates all needed blocks (new leaf and all intermediate index blocks)
634  * - makes decision where to split
635  * - moves remaining extents and index entries (right to the split point)
636  *   into the newly allocated blocks
637  * - initializes subtree
638  */
639 static int ext4_ext_split(handle_t *handle, struct inode *inode,
640                                 struct ext4_ext_path *path,
641                                 struct ext4_extent *newext, int at)
642 {
643         struct buffer_head *bh = NULL;
644         int depth = ext_depth(inode);
645         struct ext4_extent_header *neh;
646         struct ext4_extent_idx *fidx;
647         struct ext4_extent *ex;
648         int i = at, k, m, a;
649         ext4_fsblk_t newblock, oldblock;
650         __le32 border;
651         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
652         int err = 0;
653
654         /* make decision: where to split? */
655         /* FIXME: now decision is simplest: at current extent */
656
657         /* if current leaf will be split, then we should use
658          * border from split point */
659         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
660         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
661                 border = path[depth].p_ext[1].ee_block;
662                 ext_debug("leaf will be split."
663                                 " next leaf starts at %d\n",
664                                   le32_to_cpu(border));
665         } else {
666                 border = newext->ee_block;
667                 ext_debug("leaf will be added."
668                                 " next leaf starts at %d\n",
669                                 le32_to_cpu(border));
670         }
671
672         /*
673          * If error occurs, then we break processing
674          * and mark filesystem read-only. index won't
675          * be inserted and tree will be in consistent
676          * state. Next mount will repair buffers too.
677          */
678
679         /*
680          * Get array to track all allocated blocks.
681          * We need this to handle errors and free blocks
682          * upon them.
683          */
684         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
685         if (!ablocks)
686                 return -ENOMEM;
687
688         /* allocate all needed blocks */
689         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
690         for (a = 0; a < depth - at; a++) {
691                 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
692                 if (newblock == 0)
693                         goto cleanup;
694                 ablocks[a] = newblock;
695         }
696
697         /* initialize new leaf */
698         newblock = ablocks[--a];
699         BUG_ON(newblock == 0);
700         bh = sb_getblk(inode->i_sb, newblock);
701         if (!bh) {
702                 err = -EIO;
703                 goto cleanup;
704         }
705         lock_buffer(bh);
706
707         err = ext4_journal_get_create_access(handle, bh);
708         if (err)
709                 goto cleanup;
710
711         neh = ext_block_hdr(bh);
712         neh->eh_entries = 0;
713         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
714         neh->eh_magic = EXT4_EXT_MAGIC;
715         neh->eh_depth = 0;
716         ex = EXT_FIRST_EXTENT(neh);
717
718         /* move remainder of path[depth] to the new leaf */
719         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
720         /* start copy from next extent */
721         /* TODO: we could do it by single memmove */
722         m = 0;
723         path[depth].p_ext++;
724         while (path[depth].p_ext <=
725                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
726                 ext_debug("move %d:%llu:%d in new leaf %llu\n",
727                                 le32_to_cpu(path[depth].p_ext->ee_block),
728                                 ext_pblock(path[depth].p_ext),
729                                 ext4_ext_get_actual_len(path[depth].p_ext),
730                                 newblock);
731                 /*memmove(ex++, path[depth].p_ext++,
732                                 sizeof(struct ext4_extent));
733                 neh->eh_entries++;*/
734                 path[depth].p_ext++;
735                 m++;
736         }
737         if (m) {
738                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
739                 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
740         }
741
742         set_buffer_uptodate(bh);
743         unlock_buffer(bh);
744
745         err = ext4_journal_dirty_metadata(handle, bh);
746         if (err)
747                 goto cleanup;
748         brelse(bh);
749         bh = NULL;
750
751         /* correct old leaf */
752         if (m) {
753                 err = ext4_ext_get_access(handle, inode, path + depth);
754                 if (err)
755                         goto cleanup;
756                 path[depth].p_hdr->eh_entries =
757                      cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
758                 err = ext4_ext_dirty(handle, inode, path + depth);
759                 if (err)
760                         goto cleanup;
761
762         }
763
764         /* create intermediate indexes */
765         k = depth - at - 1;
766         BUG_ON(k < 0);
767         if (k)
768                 ext_debug("create %d intermediate indices\n", k);
769         /* insert new index into current index block */
770         /* current depth stored in i var */
771         i = depth - 1;
772         while (k--) {
773                 oldblock = newblock;
774                 newblock = ablocks[--a];
775                 bh = sb_getblk(inode->i_sb, newblock);
776                 if (!bh) {
777                         err = -EIO;
778                         goto cleanup;
779                 }
780                 lock_buffer(bh);
781
782                 err = ext4_journal_get_create_access(handle, bh);
783                 if (err)
784                         goto cleanup;
785
786                 neh = ext_block_hdr(bh);
787                 neh->eh_entries = cpu_to_le16(1);
788                 neh->eh_magic = EXT4_EXT_MAGIC;
789                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
790                 neh->eh_depth = cpu_to_le16(depth - i);
791                 fidx = EXT_FIRST_INDEX(neh);
792                 fidx->ei_block = border;
793                 ext4_idx_store_pblock(fidx, oldblock);
794
795                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
796                                 i, newblock, le32_to_cpu(border), oldblock);
797                 /* copy indexes */
798                 m = 0;
799                 path[i].p_idx++;
800
801                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
802                                 EXT_MAX_INDEX(path[i].p_hdr));
803                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
804                                 EXT_LAST_INDEX(path[i].p_hdr));
805                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
806                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
807                                         le32_to_cpu(path[i].p_idx->ei_block),
808                                         idx_pblock(path[i].p_idx),
809                                         newblock);
810                         /*memmove(++fidx, path[i].p_idx++,
811                                         sizeof(struct ext4_extent_idx));
812                         neh->eh_entries++;
813                         BUG_ON(neh->eh_entries > neh->eh_max);*/
814                         path[i].p_idx++;
815                         m++;
816                 }
817                 if (m) {
818                         memmove(++fidx, path[i].p_idx - m,
819                                 sizeof(struct ext4_extent_idx) * m);
820                         neh->eh_entries =
821                                 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
822                 }
823                 set_buffer_uptodate(bh);
824                 unlock_buffer(bh);
825
826                 err = ext4_journal_dirty_metadata(handle, bh);
827                 if (err)
828                         goto cleanup;
829                 brelse(bh);
830                 bh = NULL;
831
832                 /* correct old index */
833                 if (m) {
834                         err = ext4_ext_get_access(handle, inode, path + i);
835                         if (err)
836                                 goto cleanup;
837                         path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
838                         err = ext4_ext_dirty(handle, inode, path + i);
839                         if (err)
840                                 goto cleanup;
841                 }
842
843                 i--;
844         }
845
846         /* insert new index */
847         err = ext4_ext_insert_index(handle, inode, path + at,
848                                     le32_to_cpu(border), newblock);
849
850 cleanup:
851         if (bh) {
852                 if (buffer_locked(bh))
853                         unlock_buffer(bh);
854                 brelse(bh);
855         }
856
857         if (err) {
858                 /* free all allocated blocks in error case */
859                 for (i = 0; i < depth; i++) {
860                         if (!ablocks[i])
861                                 continue;
862                         ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
863                 }
864         }
865         kfree(ablocks);
866
867         return err;
868 }
869
870 /*
871  * ext4_ext_grow_indepth:
872  * implements tree growing procedure:
873  * - allocates new block
874  * - moves top-level data (index block or leaf) into the new block
875  * - initializes new top-level, creating index that points to the
876  *   just created block
877  */
878 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
879                                         struct ext4_ext_path *path,
880                                         struct ext4_extent *newext)
881 {
882         struct ext4_ext_path *curp = path;
883         struct ext4_extent_header *neh;
884         struct ext4_extent_idx *fidx;
885         struct buffer_head *bh;
886         ext4_fsblk_t newblock;
887         int err = 0;
888
889         newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
890         if (newblock == 0)
891                 return err;
892
893         bh = sb_getblk(inode->i_sb, newblock);
894         if (!bh) {
895                 err = -EIO;
896                 ext4_std_error(inode->i_sb, err);
897                 return err;
898         }
899         lock_buffer(bh);
900
901         err = ext4_journal_get_create_access(handle, bh);
902         if (err) {
903                 unlock_buffer(bh);
904                 goto out;
905         }
906
907         /* move top-level index/leaf into new block */
908         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
909
910         /* set size of new block */
911         neh = ext_block_hdr(bh);
912         /* old root could have indexes or leaves
913          * so calculate e_max right way */
914         if (ext_depth(inode))
915           neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
916         else
917           neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
918         neh->eh_magic = EXT4_EXT_MAGIC;
919         set_buffer_uptodate(bh);
920         unlock_buffer(bh);
921
922         err = ext4_journal_dirty_metadata(handle, bh);
923         if (err)
924                 goto out;
925
926         /* create index in new top-level index: num,max,pointer */
927         err = ext4_ext_get_access(handle, inode, curp);
928         if (err)
929                 goto out;
930
931         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
932         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
933         curp->p_hdr->eh_entries = cpu_to_le16(1);
934         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
935
936         if (path[0].p_hdr->eh_depth)
937                 curp->p_idx->ei_block =
938                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
939         else
940                 curp->p_idx->ei_block =
941                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
942         ext4_idx_store_pblock(curp->p_idx, newblock);
943
944         neh = ext_inode_hdr(inode);
945         fidx = EXT_FIRST_INDEX(neh);
946         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
947                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
948                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
949
950         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
951         err = ext4_ext_dirty(handle, inode, curp);
952 out:
953         brelse(bh);
954
955         return err;
956 }
957
958 /*
959  * ext4_ext_create_new_leaf:
960  * finds empty index and adds new leaf.
961  * if no free index is found, then it requests in-depth growing.
962  */
963 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
964                                         struct ext4_ext_path *path,
965                                         struct ext4_extent *newext)
966 {
967         struct ext4_ext_path *curp;
968         int depth, i, err = 0;
969
970 repeat:
971         i = depth = ext_depth(inode);
972
973         /* walk up to the tree and look for free index entry */
974         curp = path + depth;
975         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
976                 i--;
977                 curp--;
978         }
979
980         /* we use already allocated block for index block,
981          * so subsequent data blocks should be contiguous */
982         if (EXT_HAS_FREE_INDEX(curp)) {
983                 /* if we found index with free entry, then use that
984                  * entry: create all needed subtree and add new leaf */
985                 err = ext4_ext_split(handle, inode, path, newext, i);
986
987                 /* refill path */
988                 ext4_ext_drop_refs(path);
989                 path = ext4_ext_find_extent(inode,
990                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
991                                     path);
992                 if (IS_ERR(path))
993                         err = PTR_ERR(path);
994         } else {
995                 /* tree is full, time to grow in depth */
996                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
997                 if (err)
998                         goto out;
999
1000                 /* refill path */
1001                 ext4_ext_drop_refs(path);
1002                 path = ext4_ext_find_extent(inode,
1003                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1004                                     path);
1005                 if (IS_ERR(path)) {
1006                         err = PTR_ERR(path);
1007                         goto out;
1008                 }
1009
1010                 /*
1011                  * only first (depth 0 -> 1) produces free space;
1012                  * in all other cases we have to split the grown tree
1013                  */
1014                 depth = ext_depth(inode);
1015                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1016                         /* now we need to split */
1017                         goto repeat;
1018                 }
1019         }
1020
1021 out:
1022         return err;
1023 }
1024
1025 /*
1026  * search the closest allocated block to the left for *logical
1027  * and returns it at @logical + it's physical address at @phys
1028  * if *logical is the smallest allocated block, the function
1029  * returns 0 at @phys
1030  * return value contains 0 (success) or error code
1031  */
1032 int
1033 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1034                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1035 {
1036         struct ext4_extent_idx *ix;
1037         struct ext4_extent *ex;
1038         int depth, ee_len;
1039
1040         BUG_ON(path == NULL);
1041         depth = path->p_depth;
1042         *phys = 0;
1043
1044         if (depth == 0 && path->p_ext == NULL)
1045                 return 0;
1046
1047         /* usually extent in the path covers blocks smaller
1048          * then *logical, but it can be that extent is the
1049          * first one in the file */
1050
1051         ex = path[depth].p_ext;
1052         ee_len = ext4_ext_get_actual_len(ex);
1053         if (*logical < le32_to_cpu(ex->ee_block)) {
1054                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1055                 while (--depth >= 0) {
1056                         ix = path[depth].p_idx;
1057                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1058                 }
1059                 return 0;
1060         }
1061
1062         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1063
1064         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1065         *phys = ext_pblock(ex) + ee_len - 1;
1066         return 0;
1067 }
1068
1069 /*
1070  * search the closest allocated block to the right for *logical
1071  * and returns it at @logical + it's physical address at @phys
1072  * if *logical is the smallest allocated block, the function
1073  * returns 0 at @phys
1074  * return value contains 0 (success) or error code
1075  */
1076 int
1077 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1078                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1079 {
1080         struct buffer_head *bh = NULL;
1081         struct ext4_extent_header *eh;
1082         struct ext4_extent_idx *ix;
1083         struct ext4_extent *ex;
1084         ext4_fsblk_t block;
1085         int depth, ee_len;
1086
1087         BUG_ON(path == NULL);
1088         depth = path->p_depth;
1089         *phys = 0;
1090
1091         if (depth == 0 && path->p_ext == NULL)
1092                 return 0;
1093
1094         /* usually extent in the path covers blocks smaller
1095          * then *logical, but it can be that extent is the
1096          * first one in the file */
1097
1098         ex = path[depth].p_ext;
1099         ee_len = ext4_ext_get_actual_len(ex);
1100         if (*logical < le32_to_cpu(ex->ee_block)) {
1101                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1102                 while (--depth >= 0) {
1103                         ix = path[depth].p_idx;
1104                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1105                 }
1106                 *logical = le32_to_cpu(ex->ee_block);
1107                 *phys = ext_pblock(ex);
1108                 return 0;
1109         }
1110
1111         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1112
1113         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1114                 /* next allocated block in this leaf */
1115                 ex++;
1116                 *logical = le32_to_cpu(ex->ee_block);
1117                 *phys = ext_pblock(ex);
1118                 return 0;
1119         }
1120
1121         /* go up and search for index to the right */
1122         while (--depth >= 0) {
1123                 ix = path[depth].p_idx;
1124                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1125                         break;
1126         }
1127
1128         if (depth < 0) {
1129                 /* we've gone up to the root and
1130                  * found no index to the right */
1131                 return 0;
1132         }
1133
1134         /* we've found index to the right, let's
1135          * follow it and find the closest allocated
1136          * block to the right */
1137         ix++;
1138         block = idx_pblock(ix);
1139         while (++depth < path->p_depth) {
1140                 bh = sb_bread(inode->i_sb, block);
1141                 if (bh == NULL)
1142                         return -EIO;
1143                 eh = ext_block_hdr(bh);
1144                 if (ext4_ext_check_header(inode, eh, depth)) {
1145                         put_bh(bh);
1146                         return -EIO;
1147                 }
1148                 ix = EXT_FIRST_INDEX(eh);
1149                 block = idx_pblock(ix);
1150                 put_bh(bh);
1151         }
1152
1153         bh = sb_bread(inode->i_sb, block);
1154         if (bh == NULL)
1155                 return -EIO;
1156         eh = ext_block_hdr(bh);
1157         if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1158                 put_bh(bh);
1159                 return -EIO;
1160         }
1161         ex = EXT_FIRST_EXTENT(eh);
1162         *logical = le32_to_cpu(ex->ee_block);
1163         *phys = ext_pblock(ex);
1164         put_bh(bh);
1165         return 0;
1166
1167 }
1168
1169 /*
1170  * ext4_ext_next_allocated_block:
1171  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1172  * NOTE: it considers block number from index entry as
1173  * allocated block. Thus, index entries have to be consistent
1174  * with leaves.
1175  */
1176 static ext4_lblk_t
1177 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1178 {
1179         int depth;
1180
1181         BUG_ON(path == NULL);
1182         depth = path->p_depth;
1183
1184         if (depth == 0 && path->p_ext == NULL)
1185                 return EXT_MAX_BLOCK;
1186
1187         while (depth >= 0) {
1188                 if (depth == path->p_depth) {
1189                         /* leaf */
1190                         if (path[depth].p_ext !=
1191                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1192                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1193                 } else {
1194                         /* index */
1195                         if (path[depth].p_idx !=
1196                                         EXT_LAST_INDEX(path[depth].p_hdr))
1197                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1198                 }
1199                 depth--;
1200         }
1201
1202         return EXT_MAX_BLOCK;
1203 }
1204
1205 /*
1206  * ext4_ext_next_leaf_block:
1207  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1208  */
1209 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1210                                         struct ext4_ext_path *path)
1211 {
1212         int depth;
1213
1214         BUG_ON(path == NULL);
1215         depth = path->p_depth;
1216
1217         /* zero-tree has no leaf blocks at all */
1218         if (depth == 0)
1219                 return EXT_MAX_BLOCK;
1220
1221         /* go to index block */
1222         depth--;
1223
1224         while (depth >= 0) {
1225                 if (path[depth].p_idx !=
1226                                 EXT_LAST_INDEX(path[depth].p_hdr))
1227                         return (ext4_lblk_t)
1228                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1229                 depth--;
1230         }
1231
1232         return EXT_MAX_BLOCK;
1233 }
1234
1235 /*
1236  * ext4_ext_correct_indexes:
1237  * if leaf gets modified and modified extent is first in the leaf,
1238  * then we have to correct all indexes above.
1239  * TODO: do we need to correct tree in all cases?
1240  */
1241 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1242                                 struct ext4_ext_path *path)
1243 {
1244         struct ext4_extent_header *eh;
1245         int depth = ext_depth(inode);
1246         struct ext4_extent *ex;
1247         __le32 border;
1248         int k, err = 0;
1249
1250         eh = path[depth].p_hdr;
1251         ex = path[depth].p_ext;
1252         BUG_ON(ex == NULL);
1253         BUG_ON(eh == NULL);
1254
1255         if (depth == 0) {
1256                 /* there is no tree at all */
1257                 return 0;
1258         }
1259
1260         if (ex != EXT_FIRST_EXTENT(eh)) {
1261                 /* we correct tree if first leaf got modified only */
1262                 return 0;
1263         }
1264
1265         /*
1266          * TODO: we need correction if border is smaller than current one
1267          */
1268         k = depth - 1;
1269         border = path[depth].p_ext->ee_block;
1270         err = ext4_ext_get_access(handle, inode, path + k);
1271         if (err)
1272                 return err;
1273         path[k].p_idx->ei_block = border;
1274         err = ext4_ext_dirty(handle, inode, path + k);
1275         if (err)
1276                 return err;
1277
1278         while (k--) {
1279                 /* change all left-side indexes */
1280                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1281                         break;
1282                 err = ext4_ext_get_access(handle, inode, path + k);
1283                 if (err)
1284                         break;
1285                 path[k].p_idx->ei_block = border;
1286                 err = ext4_ext_dirty(handle, inode, path + k);
1287                 if (err)
1288                         break;
1289         }
1290
1291         return err;
1292 }
1293
1294 static int
1295 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1296                                 struct ext4_extent *ex2)
1297 {
1298         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1299
1300         /*
1301          * Make sure that either both extents are uninitialized, or
1302          * both are _not_.
1303          */
1304         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1305                 return 0;
1306
1307         if (ext4_ext_is_uninitialized(ex1))
1308                 max_len = EXT_UNINIT_MAX_LEN;
1309         else
1310                 max_len = EXT_INIT_MAX_LEN;
1311
1312         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1313         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1314
1315         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1316                         le32_to_cpu(ex2->ee_block))
1317                 return 0;
1318
1319         /*
1320          * To allow future support for preallocated extents to be added
1321          * as an RO_COMPAT feature, refuse to merge to extents if
1322          * this can result in the top bit of ee_len being set.
1323          */
1324         if (ext1_ee_len + ext2_ee_len > max_len)
1325                 return 0;
1326 #ifdef AGGRESSIVE_TEST
1327         if (ext1_ee_len >= 4)
1328                 return 0;
1329 #endif
1330
1331         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1332                 return 1;
1333         return 0;
1334 }
1335
1336 /*
1337  * This function tries to merge the "ex" extent to the next extent in the tree.
1338  * It always tries to merge towards right. If you want to merge towards
1339  * left, pass "ex - 1" as argument instead of "ex".
1340  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1341  * 1 if they got merged.
1342  */
1343 int ext4_ext_try_to_merge(struct inode *inode,
1344                           struct ext4_ext_path *path,
1345                           struct ext4_extent *ex)
1346 {
1347         struct ext4_extent_header *eh;
1348         unsigned int depth, len;
1349         int merge_done = 0;
1350         int uninitialized = 0;
1351
1352         depth = ext_depth(inode);
1353         BUG_ON(path[depth].p_hdr == NULL);
1354         eh = path[depth].p_hdr;
1355
1356         while (ex < EXT_LAST_EXTENT(eh)) {
1357                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1358                         break;
1359                 /* merge with next extent! */
1360                 if (ext4_ext_is_uninitialized(ex))
1361                         uninitialized = 1;
1362                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1363                                 + ext4_ext_get_actual_len(ex + 1));
1364                 if (uninitialized)
1365                         ext4_ext_mark_uninitialized(ex);
1366
1367                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1368                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1369                                 * sizeof(struct ext4_extent);
1370                         memmove(ex + 1, ex + 2, len);
1371                 }
1372                 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
1373                 merge_done = 1;
1374                 WARN_ON(eh->eh_entries == 0);
1375                 if (!eh->eh_entries)
1376                         ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1377                            "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1378         }
1379
1380         return merge_done;
1381 }
1382
1383 /*
1384  * check if a portion of the "newext" extent overlaps with an
1385  * existing extent.
1386  *
1387  * If there is an overlap discovered, it updates the length of the newext
1388  * such that there will be no overlap, and then returns 1.
1389  * If there is no overlap found, it returns 0.
1390  */
1391 unsigned int ext4_ext_check_overlap(struct inode *inode,
1392                                     struct ext4_extent *newext,
1393                                     struct ext4_ext_path *path)
1394 {
1395         ext4_lblk_t b1, b2;
1396         unsigned int depth, len1;
1397         unsigned int ret = 0;
1398
1399         b1 = le32_to_cpu(newext->ee_block);
1400         len1 = ext4_ext_get_actual_len(newext);
1401         depth = ext_depth(inode);
1402         if (!path[depth].p_ext)
1403                 goto out;
1404         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1405
1406         /*
1407          * get the next allocated block if the extent in the path
1408          * is before the requested block(s) 
1409          */
1410         if (b2 < b1) {
1411                 b2 = ext4_ext_next_allocated_block(path);
1412                 if (b2 == EXT_MAX_BLOCK)
1413                         goto out;
1414         }
1415
1416         /* check for wrap through zero on extent logical start block*/
1417         if (b1 + len1 < b1) {
1418                 len1 = EXT_MAX_BLOCK - b1;
1419                 newext->ee_len = cpu_to_le16(len1);
1420                 ret = 1;
1421         }
1422
1423         /* check for overlap */
1424         if (b1 + len1 > b2) {
1425                 newext->ee_len = cpu_to_le16(b2 - b1);
1426                 ret = 1;
1427         }
1428 out:
1429         return ret;
1430 }
1431
1432 /*
1433  * ext4_ext_insert_extent:
1434  * tries to merge requsted extent into the existing extent or
1435  * inserts requested extent as new one into the tree,
1436  * creating new leaf in the no-space case.
1437  */
1438 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1439                                 struct ext4_ext_path *path,
1440                                 struct ext4_extent *newext)
1441 {
1442         struct ext4_extent_header * eh;
1443         struct ext4_extent *ex, *fex;
1444         struct ext4_extent *nearex; /* nearest extent */
1445         struct ext4_ext_path *npath = NULL;
1446         int depth, len, err;
1447         ext4_lblk_t next;
1448         unsigned uninitialized = 0;
1449
1450         BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1451         depth = ext_depth(inode);
1452         ex = path[depth].p_ext;
1453         BUG_ON(path[depth].p_hdr == NULL);
1454
1455         /* try to insert block into found extent and return */
1456         if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1457                 ext_debug("append %d block to %d:%d (from %llu)\n",
1458                                 ext4_ext_get_actual_len(newext),
1459                                 le32_to_cpu(ex->ee_block),
1460                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1461                 err = ext4_ext_get_access(handle, inode, path + depth);
1462                 if (err)
1463                         return err;
1464
1465                 /*
1466                  * ext4_can_extents_be_merged should have checked that either
1467                  * both extents are uninitialized, or both aren't. Thus we
1468                  * need to check only one of them here.
1469                  */
1470                 if (ext4_ext_is_uninitialized(ex))
1471                         uninitialized = 1;
1472                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1473                                         + ext4_ext_get_actual_len(newext));
1474                 if (uninitialized)
1475                         ext4_ext_mark_uninitialized(ex);
1476                 eh = path[depth].p_hdr;
1477                 nearex = ex;
1478                 goto merge;
1479         }
1480
1481 repeat:
1482         depth = ext_depth(inode);
1483         eh = path[depth].p_hdr;
1484         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1485                 goto has_space;
1486
1487         /* probably next leaf has space for us? */
1488         fex = EXT_LAST_EXTENT(eh);
1489         next = ext4_ext_next_leaf_block(inode, path);
1490         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1491             && next != EXT_MAX_BLOCK) {
1492                 ext_debug("next leaf block - %d\n", next);
1493                 BUG_ON(npath != NULL);
1494                 npath = ext4_ext_find_extent(inode, next, NULL);
1495                 if (IS_ERR(npath))
1496                         return PTR_ERR(npath);
1497                 BUG_ON(npath->p_depth != path->p_depth);
1498                 eh = npath[depth].p_hdr;
1499                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1500                         ext_debug("next leaf isnt full(%d)\n",
1501                                   le16_to_cpu(eh->eh_entries));
1502                         path = npath;
1503                         goto repeat;
1504                 }
1505                 ext_debug("next leaf has no free space(%d,%d)\n",
1506                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1507         }
1508
1509         /*
1510          * There is no free space in the found leaf.
1511          * We're gonna add a new leaf in the tree.
1512          */
1513         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1514         if (err)
1515                 goto cleanup;
1516         depth = ext_depth(inode);
1517         eh = path[depth].p_hdr;
1518
1519 has_space:
1520         nearex = path[depth].p_ext;
1521
1522         err = ext4_ext_get_access(handle, inode, path + depth);
1523         if (err)
1524                 goto cleanup;
1525
1526         if (!nearex) {
1527                 /* there is no extent in this leaf, create first one */
1528                 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1529                                 le32_to_cpu(newext->ee_block),
1530                                 ext_pblock(newext),
1531                                 ext4_ext_get_actual_len(newext));
1532                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1533         } else if (le32_to_cpu(newext->ee_block)
1534                            > le32_to_cpu(nearex->ee_block)) {
1535 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1536                 if (nearex != EXT_LAST_EXTENT(eh)) {
1537                         len = EXT_MAX_EXTENT(eh) - nearex;
1538                         len = (len - 1) * sizeof(struct ext4_extent);
1539                         len = len < 0 ? 0 : len;
1540                         ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1541                                         "move %d from 0x%p to 0x%p\n",
1542                                         le32_to_cpu(newext->ee_block),
1543                                         ext_pblock(newext),
1544                                         ext4_ext_get_actual_len(newext),
1545                                         nearex, len, nearex + 1, nearex + 2);
1546                         memmove(nearex + 2, nearex + 1, len);
1547                 }
1548                 path[depth].p_ext = nearex + 1;
1549         } else {
1550                 BUG_ON(newext->ee_block == nearex->ee_block);
1551                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1552                 len = len < 0 ? 0 : len;
1553                 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1554                                 "move %d from 0x%p to 0x%p\n",
1555                                 le32_to_cpu(newext->ee_block),
1556                                 ext_pblock(newext),
1557                                 ext4_ext_get_actual_len(newext),
1558                                 nearex, len, nearex + 1, nearex + 2);
1559                 memmove(nearex + 1, nearex, len);
1560                 path[depth].p_ext = nearex;
1561         }
1562
1563         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1564         nearex = path[depth].p_ext;
1565         nearex->ee_block = newext->ee_block;
1566         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1567         nearex->ee_len = newext->ee_len;
1568
1569 merge:
1570         /* try to merge extents to the right */
1571         ext4_ext_try_to_merge(inode, path, nearex);
1572
1573         /* try to merge extents to the left */
1574
1575         /* time to correct all indexes above */
1576         err = ext4_ext_correct_indexes(handle, inode, path);
1577         if (err)
1578                 goto cleanup;
1579
1580         err = ext4_ext_dirty(handle, inode, path + depth);
1581
1582 cleanup:
1583         if (npath) {
1584                 ext4_ext_drop_refs(npath);
1585                 kfree(npath);
1586         }
1587         ext4_ext_tree_changed(inode);
1588         ext4_ext_invalidate_cache(inode);
1589         return err;
1590 }
1591
1592 static void
1593 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1594                         __u32 len, ext4_fsblk_t start, int type)
1595 {
1596         struct ext4_ext_cache *cex;
1597         BUG_ON(len == 0);
1598         cex = &EXT4_I(inode)->i_cached_extent;
1599         cex->ec_type = type;
1600         cex->ec_block = block;
1601         cex->ec_len = len;
1602         cex->ec_start = start;
1603 }
1604
1605 /*
1606  * ext4_ext_put_gap_in_cache:
1607  * calculate boundaries of the gap that the requested block fits into
1608  * and cache this gap
1609  */
1610 static void
1611 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1612                                 ext4_lblk_t block)
1613 {
1614         int depth = ext_depth(inode);
1615         unsigned long len;
1616         ext4_lblk_t lblock;
1617         struct ext4_extent *ex;
1618
1619         ex = path[depth].p_ext;
1620         if (ex == NULL) {
1621                 /* there is no extent yet, so gap is [0;-] */
1622                 lblock = 0;
1623                 len = EXT_MAX_BLOCK;
1624                 ext_debug("cache gap(whole file):");
1625         } else if (block < le32_to_cpu(ex->ee_block)) {
1626                 lblock = block;
1627                 len = le32_to_cpu(ex->ee_block) - block;
1628                 ext_debug("cache gap(before): %u [%u:%u]",
1629                                 block,
1630                                 le32_to_cpu(ex->ee_block),
1631                                  ext4_ext_get_actual_len(ex));
1632         } else if (block >= le32_to_cpu(ex->ee_block)
1633                         + ext4_ext_get_actual_len(ex)) {
1634                 ext4_lblk_t next;
1635                 lblock = le32_to_cpu(ex->ee_block)
1636                         + ext4_ext_get_actual_len(ex);
1637
1638                 next = ext4_ext_next_allocated_block(path);
1639                 ext_debug("cache gap(after): [%u:%u] %u",
1640                                 le32_to_cpu(ex->ee_block),
1641                                 ext4_ext_get_actual_len(ex),
1642                                 block);
1643                 BUG_ON(next == lblock);
1644                 len = next - lblock;
1645         } else {
1646                 lblock = len = 0;
1647                 BUG();
1648         }
1649
1650         ext_debug(" -> %u:%lu\n", lblock, len);
1651         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1652 }
1653
1654 static int
1655 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1656                         struct ext4_extent *ex)
1657 {
1658         struct ext4_ext_cache *cex;
1659
1660         cex = &EXT4_I(inode)->i_cached_extent;
1661
1662         /* has cache valid data? */
1663         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1664                 return EXT4_EXT_CACHE_NO;
1665
1666         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1667                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1668         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1669                 ex->ee_block = cpu_to_le32(cex->ec_block);
1670                 ext4_ext_store_pblock(ex, cex->ec_start);
1671                 ex->ee_len = cpu_to_le16(cex->ec_len);
1672                 ext_debug("%u cached by %u:%u:%llu\n",
1673                                 block,
1674                                 cex->ec_block, cex->ec_len, cex->ec_start);
1675                 return cex->ec_type;
1676         }
1677
1678         /* not in cache */
1679         return EXT4_EXT_CACHE_NO;
1680 }
1681
1682 /*
1683  * ext4_ext_rm_idx:
1684  * removes index from the index block.
1685  * It's used in truncate case only, thus all requests are for
1686  * last index in the block only.
1687  */
1688 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1689                         struct ext4_ext_path *path)
1690 {
1691         struct buffer_head *bh;
1692         int err;
1693         ext4_fsblk_t leaf;
1694
1695         /* free index block */
1696         path--;
1697         leaf = idx_pblock(path->p_idx);
1698         BUG_ON(path->p_hdr->eh_entries == 0);
1699         err = ext4_ext_get_access(handle, inode, path);
1700         if (err)
1701                 return err;
1702         path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1703         err = ext4_ext_dirty(handle, inode, path);
1704         if (err)
1705                 return err;
1706         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1707         bh = sb_find_get_block(inode->i_sb, leaf);
1708         ext4_forget(handle, 1, inode, bh, leaf);
1709         ext4_free_blocks(handle, inode, leaf, 1, 1);
1710         return err;
1711 }
1712
1713 /*
1714  * ext4_ext_calc_credits_for_insert:
1715  * This routine returns max. credits that the extent tree can consume.
1716  * It should be OK for low-performance paths like ->writepage()
1717  * To allow many writing processes to fit into a single transaction,
1718  * the caller should calculate credits under i_data_sem and
1719  * pass the actual path.
1720  */
1721 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1722                                                 struct ext4_ext_path *path)
1723 {
1724         int depth, needed;
1725
1726         if (path) {
1727                 /* probably there is space in leaf? */
1728                 depth = ext_depth(inode);
1729                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1730                                 < le16_to_cpu(path[depth].p_hdr->eh_max))
1731                         return 1;
1732         }
1733
1734         /*
1735          * given 32-bit logical block (4294967296 blocks), max. tree
1736          * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1737          * Let's also add one more level for imbalance.
1738          */
1739         depth = 5;
1740
1741         /* allocation of new data block(s) */
1742         needed = 2;
1743
1744         /*
1745          * tree can be full, so it would need to grow in depth:
1746          * we need one credit to modify old root, credits for
1747          * new root will be added in split accounting
1748          */
1749         needed += 1;
1750
1751         /*
1752          * Index split can happen, we would need:
1753          *    allocate intermediate indexes (bitmap + group)
1754          *  + change two blocks at each level, but root (already included)
1755          */
1756         needed += (depth * 2) + (depth * 2);
1757
1758         /* any allocation modifies superblock */
1759         needed += 1;
1760
1761         return needed;
1762 }
1763
1764 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1765                                 struct ext4_extent *ex,
1766                                 ext4_lblk_t from, ext4_lblk_t to)
1767 {
1768         struct buffer_head *bh;
1769         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
1770         int i, metadata = 0;
1771
1772         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1773                 metadata = 1;
1774 #ifdef EXTENTS_STATS
1775         {
1776                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1777                 spin_lock(&sbi->s_ext_stats_lock);
1778                 sbi->s_ext_blocks += ee_len;
1779                 sbi->s_ext_extents++;
1780                 if (ee_len < sbi->s_ext_min)
1781                         sbi->s_ext_min = ee_len;
1782                 if (ee_len > sbi->s_ext_max)
1783                         sbi->s_ext_max = ee_len;
1784                 if (ext_depth(inode) > sbi->s_depth_max)
1785                         sbi->s_depth_max = ext_depth(inode);
1786                 spin_unlock(&sbi->s_ext_stats_lock);
1787         }
1788 #endif
1789         if (from >= le32_to_cpu(ex->ee_block)
1790             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
1791                 /* tail removal */
1792                 ext4_lblk_t num;
1793                 ext4_fsblk_t start;
1794
1795                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1796                 start = ext_pblock(ex) + ee_len - num;
1797                 ext_debug("free last %u blocks starting %llu\n", num, start);
1798                 for (i = 0; i < num; i++) {
1799                         bh = sb_find_get_block(inode->i_sb, start + i);
1800                         ext4_forget(handle, 0, inode, bh, start + i);
1801                 }
1802                 ext4_free_blocks(handle, inode, start, num, metadata);
1803         } else if (from == le32_to_cpu(ex->ee_block)
1804                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
1805                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
1806                         from, to, le32_to_cpu(ex->ee_block), ee_len);
1807         } else {
1808                 printk(KERN_INFO "strange request: removal(2) "
1809                                 "%u-%u from %u:%u\n",
1810                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
1811         }
1812         return 0;
1813 }
1814
1815 static int
1816 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1817                 struct ext4_ext_path *path, ext4_lblk_t start)
1818 {
1819         int err = 0, correct_index = 0;
1820         int depth = ext_depth(inode), credits;
1821         struct ext4_extent_header *eh;
1822         ext4_lblk_t a, b, block;
1823         unsigned num;
1824         ext4_lblk_t ex_ee_block;
1825         unsigned short ex_ee_len;
1826         unsigned uninitialized = 0;
1827         struct ext4_extent *ex;
1828
1829         /* the header must be checked already in ext4_ext_remove_space() */
1830         ext_debug("truncate since %u in leaf\n", start);
1831         if (!path[depth].p_hdr)
1832                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1833         eh = path[depth].p_hdr;
1834         BUG_ON(eh == NULL);
1835
1836         /* find where to start removing */
1837         ex = EXT_LAST_EXTENT(eh);
1838
1839         ex_ee_block = le32_to_cpu(ex->ee_block);
1840         if (ext4_ext_is_uninitialized(ex))
1841                 uninitialized = 1;
1842         ex_ee_len = ext4_ext_get_actual_len(ex);
1843
1844         while (ex >= EXT_FIRST_EXTENT(eh) &&
1845                         ex_ee_block + ex_ee_len > start) {
1846                 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1847                 path[depth].p_ext = ex;
1848
1849                 a = ex_ee_block > start ? ex_ee_block : start;
1850                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1851                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1852
1853                 ext_debug("  border %u:%u\n", a, b);
1854
1855                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1856                         block = 0;
1857                         num = 0;
1858                         BUG();
1859                 } else if (a != ex_ee_block) {
1860                         /* remove tail of the extent */
1861                         block = ex_ee_block;
1862                         num = a - block;
1863                 } else if (b != ex_ee_block + ex_ee_len - 1) {
1864                         /* remove head of the extent */
1865                         block = a;
1866                         num = b - a;
1867                         /* there is no "make a hole" API yet */
1868                         BUG();
1869                 } else {
1870                         /* remove whole extent: excellent! */
1871                         block = ex_ee_block;
1872                         num = 0;
1873                         BUG_ON(a != ex_ee_block);
1874                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1875                 }
1876
1877                 /* at present, extent can't cross block group: */
1878                 /* leaf + bitmap + group desc + sb + inode */
1879                 credits = 5;
1880                 if (ex == EXT_FIRST_EXTENT(eh)) {
1881                         correct_index = 1;
1882                         credits += (ext_depth(inode)) + 1;
1883                 }
1884 #ifdef CONFIG_QUOTA
1885                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1886 #endif
1887
1888                 handle = ext4_ext_journal_restart(handle, credits);
1889                 if (IS_ERR(handle)) {
1890                         err = PTR_ERR(handle);
1891                         goto out;
1892                 }
1893
1894                 err = ext4_ext_get_access(handle, inode, path + depth);
1895                 if (err)
1896                         goto out;
1897
1898                 err = ext4_remove_blocks(handle, inode, ex, a, b);
1899                 if (err)
1900                         goto out;
1901
1902                 if (num == 0) {
1903                         /* this extent is removed; mark slot entirely unused */
1904                         ext4_ext_store_pblock(ex, 0);
1905                         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1906                 }
1907
1908                 ex->ee_block = cpu_to_le32(block);
1909                 ex->ee_len = cpu_to_le16(num);
1910                 /*
1911                  * Do not mark uninitialized if all the blocks in the
1912                  * extent have been removed.
1913                  */
1914                 if (uninitialized && num)
1915                         ext4_ext_mark_uninitialized(ex);
1916
1917                 err = ext4_ext_dirty(handle, inode, path + depth);
1918                 if (err)
1919                         goto out;
1920
1921                 ext_debug("new extent: %u:%u:%llu\n", block, num,
1922                                 ext_pblock(ex));
1923                 ex--;
1924                 ex_ee_block = le32_to_cpu(ex->ee_block);
1925                 ex_ee_len = ext4_ext_get_actual_len(ex);
1926         }
1927
1928         if (correct_index && eh->eh_entries)
1929                 err = ext4_ext_correct_indexes(handle, inode, path);
1930
1931         /* if this leaf is free, then we should
1932          * remove it from index block above */
1933         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1934                 err = ext4_ext_rm_idx(handle, inode, path + depth);
1935
1936 out:
1937         return err;
1938 }
1939
1940 /*
1941  * ext4_ext_more_to_rm:
1942  * returns 1 if current index has to be freed (even partial)
1943  */
1944 static int
1945 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1946 {
1947         BUG_ON(path->p_idx == NULL);
1948
1949         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1950                 return 0;
1951
1952         /*
1953          * if truncate on deeper level happened, it wasn't partial,
1954          * so we have to consider current index for truncation
1955          */
1956         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1957                 return 0;
1958         return 1;
1959 }
1960
1961 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
1962 {
1963         struct super_block *sb = inode->i_sb;
1964         int depth = ext_depth(inode);
1965         struct ext4_ext_path *path;
1966         handle_t *handle;
1967         int i = 0, err = 0;
1968
1969         ext_debug("truncate since %u\n", start);
1970
1971         /* probably first extent we're gonna free will be last in block */
1972         handle = ext4_journal_start(inode, depth + 1);
1973         if (IS_ERR(handle))
1974                 return PTR_ERR(handle);
1975
1976         ext4_ext_invalidate_cache(inode);
1977
1978         /*
1979          * We start scanning from right side, freeing all the blocks
1980          * after i_size and walking into the tree depth-wise.
1981          */
1982         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1983         if (path == NULL) {
1984                 ext4_journal_stop(handle);
1985                 return -ENOMEM;
1986         }
1987         path[0].p_hdr = ext_inode_hdr(inode);
1988         if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
1989                 err = -EIO;
1990                 goto out;
1991         }
1992         path[0].p_depth = depth;
1993
1994         while (i >= 0 && err == 0) {
1995                 if (i == depth) {
1996                         /* this is leaf block */
1997                         err = ext4_ext_rm_leaf(handle, inode, path, start);
1998                         /* root level has p_bh == NULL, brelse() eats this */
1999                         brelse(path[i].p_bh);
2000                         path[i].p_bh = NULL;
2001                         i--;
2002                         continue;
2003                 }
2004
2005                 /* this is index block */
2006                 if (!path[i].p_hdr) {
2007                         ext_debug("initialize header\n");
2008                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2009                 }
2010
2011                 if (!path[i].p_idx) {
2012                         /* this level hasn't been touched yet */
2013                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2014                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2015                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2016                                   path[i].p_hdr,
2017                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2018                 } else {
2019                         /* we were already here, see at next index */
2020                         path[i].p_idx--;
2021                 }
2022
2023                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2024                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2025                                 path[i].p_idx);
2026                 if (ext4_ext_more_to_rm(path + i)) {
2027                         struct buffer_head *bh;
2028                         /* go to the next level */
2029                         ext_debug("move to level %d (block %llu)\n",
2030                                   i + 1, idx_pblock(path[i].p_idx));
2031                         memset(path + i + 1, 0, sizeof(*path));
2032                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2033                         if (!bh) {
2034                                 /* should we reset i_size? */
2035                                 err = -EIO;
2036                                 break;
2037                         }
2038                         if (WARN_ON(i + 1 > depth)) {
2039                                 err = -EIO;
2040                                 break;
2041                         }
2042                         if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2043                                                         depth - i - 1)) {
2044                                 err = -EIO;
2045                                 break;
2046                         }
2047                         path[i + 1].p_bh = bh;
2048
2049                         /* save actual number of indexes since this
2050                          * number is changed at the next iteration */
2051                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2052                         i++;
2053                 } else {
2054                         /* we finished processing this index, go up */
2055                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2056                                 /* index is empty, remove it;
2057                                  * handle must be already prepared by the
2058                                  * truncatei_leaf() */
2059                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2060                         }
2061                         /* root level has p_bh == NULL, brelse() eats this */
2062                         brelse(path[i].p_bh);
2063                         path[i].p_bh = NULL;
2064                         i--;
2065                         ext_debug("return to level %d\n", i);
2066                 }
2067         }
2068
2069         /* TODO: flexible tree reduction should be here */
2070         if (path->p_hdr->eh_entries == 0) {
2071                 /*
2072                  * truncate to zero freed all the tree,
2073                  * so we need to correct eh_depth
2074                  */
2075                 err = ext4_ext_get_access(handle, inode, path);
2076                 if (err == 0) {
2077                         ext_inode_hdr(inode)->eh_depth = 0;
2078                         ext_inode_hdr(inode)->eh_max =
2079                                 cpu_to_le16(ext4_ext_space_root(inode));
2080                         err = ext4_ext_dirty(handle, inode, path);
2081                 }
2082         }
2083 out:
2084         ext4_ext_tree_changed(inode);
2085         ext4_ext_drop_refs(path);
2086         kfree(path);
2087         ext4_journal_stop(handle);
2088
2089         return err;
2090 }
2091
2092 /*
2093  * called at mount time
2094  */
2095 void ext4_ext_init(struct super_block *sb)
2096 {
2097         /*
2098          * possible initialization would be here
2099          */
2100
2101         if (test_opt(sb, EXTENTS)) {
2102                 printk("EXT4-fs: file extents enabled");
2103 #ifdef AGGRESSIVE_TEST
2104                 printk(", aggressive tests");
2105 #endif
2106 #ifdef CHECK_BINSEARCH
2107                 printk(", check binsearch");
2108 #endif
2109 #ifdef EXTENTS_STATS
2110                 printk(", stats");
2111 #endif
2112                 printk("\n");
2113 #ifdef EXTENTS_STATS
2114                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2115                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2116                 EXT4_SB(sb)->s_ext_max = 0;
2117 #endif
2118         }
2119 }
2120
2121 /*
2122  * called at umount time
2123  */
2124 void ext4_ext_release(struct super_block *sb)
2125 {
2126         if (!test_opt(sb, EXTENTS))
2127                 return;
2128
2129 #ifdef EXTENTS_STATS
2130         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2131                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2132                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2133                         sbi->s_ext_blocks, sbi->s_ext_extents,
2134                         sbi->s_ext_blocks / sbi->s_ext_extents);
2135                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2136                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2137         }
2138 #endif
2139 }
2140
2141 /*
2142  * This function is called by ext4_ext_get_blocks() if someone tries to write
2143  * to an uninitialized extent. It may result in splitting the uninitialized
2144  * extent into multiple extents (upto three - one initialized and two
2145  * uninitialized).
2146  * There are three possibilities:
2147  *   a> There is no split required: Entire extent should be initialized
2148  *   b> Splits in two extents: Write is happening at either end of the extent
2149  *   c> Splits in three extents: Somone is writing in middle of the extent
2150  */
2151 static int ext4_ext_convert_to_initialized(handle_t *handle,
2152                                                 struct inode *inode,
2153                                                 struct ext4_ext_path *path,
2154                                                 ext4_lblk_t iblock,
2155                                                 unsigned long max_blocks)
2156 {
2157         struct ext4_extent *ex, newex;
2158         struct ext4_extent *ex1 = NULL;
2159         struct ext4_extent *ex2 = NULL;
2160         struct ext4_extent *ex3 = NULL;
2161         struct ext4_extent_header *eh;
2162         ext4_lblk_t ee_block;
2163         unsigned int allocated, ee_len, depth;
2164         ext4_fsblk_t newblock;
2165         int err = 0;
2166         int ret = 0;
2167
2168         depth = ext_depth(inode);
2169         eh = path[depth].p_hdr;
2170         ex = path[depth].p_ext;
2171         ee_block = le32_to_cpu(ex->ee_block);
2172         ee_len = ext4_ext_get_actual_len(ex);
2173         allocated = ee_len - (iblock - ee_block);
2174         newblock = iblock - ee_block + ext_pblock(ex);
2175         ex2 = ex;
2176
2177         err = ext4_ext_get_access(handle, inode, path + depth);
2178         if (err)
2179                 goto out;
2180
2181         /* ex1: ee_block to iblock - 1 : uninitialized */
2182         if (iblock > ee_block) {
2183                 ex1 = ex;
2184                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2185                 ext4_ext_mark_uninitialized(ex1);
2186                 ex2 = &newex;
2187         }
2188         /*
2189          * for sanity, update the length of the ex2 extent before
2190          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2191          * overlap of blocks.
2192          */
2193         if (!ex1 && allocated > max_blocks)
2194                 ex2->ee_len = cpu_to_le16(max_blocks);
2195         /* ex3: to ee_block + ee_len : uninitialised */
2196         if (allocated > max_blocks) {
2197                 unsigned int newdepth;
2198                 ex3 = &newex;
2199                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2200                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2201                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2202                 ext4_ext_mark_uninitialized(ex3);
2203                 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2204                 if (err)
2205                         goto out;
2206                 /*
2207                  * The depth, and hence eh & ex might change
2208                  * as part of the insert above.
2209                  */
2210                 newdepth = ext_depth(inode);
2211                 if (newdepth != depth) {
2212                         depth = newdepth;
2213                         ext4_ext_drop_refs(path);
2214                         path = ext4_ext_find_extent(inode, iblock, path);
2215                         if (IS_ERR(path)) {
2216                                 err = PTR_ERR(path);
2217                                 goto out;
2218                         }
2219                         eh = path[depth].p_hdr;
2220                         ex = path[depth].p_ext;
2221                         if (ex2 != &newex)
2222                                 ex2 = ex;
2223
2224                         err = ext4_ext_get_access(handle, inode, path + depth);
2225                         if (err)
2226                                 goto out;
2227                 }
2228                 allocated = max_blocks;
2229         }
2230         /*
2231          * If there was a change of depth as part of the
2232          * insertion of ex3 above, we need to update the length
2233          * of the ex1 extent again here
2234          */
2235         if (ex1 && ex1 != ex) {
2236                 ex1 = ex;
2237                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2238                 ext4_ext_mark_uninitialized(ex1);
2239                 ex2 = &newex;
2240         }
2241         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2242         ex2->ee_block = cpu_to_le32(iblock);
2243         ext4_ext_store_pblock(ex2, newblock);
2244         ex2->ee_len = cpu_to_le16(allocated);
2245         if (ex2 != ex)
2246                 goto insert;
2247         /*
2248          * New (initialized) extent starts from the first block
2249          * in the current extent. i.e., ex2 == ex
2250          * We have to see if it can be merged with the extent
2251          * on the left.
2252          */
2253         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2254                 /*
2255                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2256                  * since it merges towards right _only_.
2257                  */
2258                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2259                 if (ret) {
2260                         err = ext4_ext_correct_indexes(handle, inode, path);
2261                         if (err)
2262                                 goto out;
2263                         depth = ext_depth(inode);
2264                         ex2--;
2265                 }
2266         }
2267         /*
2268          * Try to Merge towards right. This might be required
2269          * only when the whole extent is being written to.
2270          * i.e. ex2 == ex and ex3 == NULL.
2271          */
2272         if (!ex3) {
2273                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2274                 if (ret) {
2275                         err = ext4_ext_correct_indexes(handle, inode, path);
2276                         if (err)
2277                                 goto out;
2278                 }
2279         }
2280         /* Mark modified extent as dirty */
2281         err = ext4_ext_dirty(handle, inode, path + depth);
2282         goto out;
2283 insert:
2284         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2285 out:
2286         return err ? err : allocated;
2287 }
2288
2289 /*
2290  * Block allocation/map/preallocation routine for extents based files
2291  *
2292  *
2293  * Need to be called with
2294  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2295  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2296  *
2297  * return > 0, number of of blocks already mapped/allocated
2298  *          if create == 0 and these are pre-allocated blocks
2299  *              buffer head is unmapped
2300  *          otherwise blocks are mapped
2301  *
2302  * return = 0, if plain look up failed (blocks have not been allocated)
2303  *          buffer head is unmapped
2304  *
2305  * return < 0, error case.
2306  */
2307 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2308                         ext4_lblk_t iblock,
2309                         unsigned long max_blocks, struct buffer_head *bh_result,
2310                         int create, int extend_disksize)
2311 {
2312         struct ext4_ext_path *path = NULL;
2313         struct ext4_extent_header *eh;
2314         struct ext4_extent newex, *ex;
2315         ext4_fsblk_t goal, newblock;
2316         int err = 0, depth, ret;
2317         unsigned long allocated = 0;
2318         struct ext4_allocation_request ar;
2319
2320         __clear_bit(BH_New, &bh_result->b_state);
2321         ext_debug("blocks %u/%lu requested for inode %u\n",
2322                         iblock, max_blocks, inode->i_ino);
2323
2324         /* check in cache */
2325         goal = ext4_ext_in_cache(inode, iblock, &newex);
2326         if (goal) {
2327                 if (goal == EXT4_EXT_CACHE_GAP) {
2328                         if (!create) {
2329                                 /*
2330                                  * block isn't allocated yet and
2331                                  * user doesn't want to allocate it
2332                                  */
2333                                 goto out2;
2334                         }
2335                         /* we should allocate requested block */
2336                 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2337                         /* block is already allocated */
2338                         newblock = iblock
2339                                    - le32_to_cpu(newex.ee_block)
2340                                    + ext_pblock(&newex);
2341                         /* number of remaining blocks in the extent */
2342                         allocated = ext4_ext_get_actual_len(&newex) -
2343                                         (iblock - le32_to_cpu(newex.ee_block));
2344                         goto out;
2345                 } else {
2346                         BUG();
2347                 }
2348         }
2349
2350         /* find extent for this block */
2351         path = ext4_ext_find_extent(inode, iblock, NULL);
2352         if (IS_ERR(path)) {
2353                 err = PTR_ERR(path);
2354                 path = NULL;
2355                 goto out2;
2356         }
2357
2358         depth = ext_depth(inode);
2359
2360         /*
2361          * consistent leaf must not be empty;
2362          * this situation is possible, though, _during_ tree modification;
2363          * this is why assert can't be put in ext4_ext_find_extent()
2364          */
2365         BUG_ON(path[depth].p_ext == NULL && depth != 0);
2366         eh = path[depth].p_hdr;
2367
2368         ex = path[depth].p_ext;
2369         if (ex) {
2370                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2371                 ext4_fsblk_t ee_start = ext_pblock(ex);
2372                 unsigned short ee_len;
2373
2374                 /*
2375                  * Uninitialized extents are treated as holes, except that
2376                  * we split out initialized portions during a write.
2377                  */
2378                 ee_len = ext4_ext_get_actual_len(ex);
2379                 /* if found extent covers block, simply return it */
2380                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2381                         newblock = iblock - ee_block + ee_start;
2382                         /* number of remaining blocks in the extent */
2383                         allocated = ee_len - (iblock - ee_block);
2384                         ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2385                                         ee_block, ee_len, newblock);
2386
2387                         /* Do not put uninitialized extent in the cache */
2388                         if (!ext4_ext_is_uninitialized(ex)) {
2389                                 ext4_ext_put_in_cache(inode, ee_block,
2390                                                         ee_len, ee_start,
2391                                                         EXT4_EXT_CACHE_EXTENT);
2392                                 goto out;
2393                         }
2394                         if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2395                                 goto out;
2396                         if (!create)
2397                                 goto out2;
2398
2399                         ret = ext4_ext_convert_to_initialized(handle, inode,
2400                                                                 path, iblock,
2401                                                                 max_blocks);
2402                         if (ret <= 0) {
2403                                 err = ret;
2404                                 goto out2;
2405                         } else
2406                                 allocated = ret;
2407                         goto outnew;
2408                 }
2409         }
2410
2411         /*
2412          * requested block isn't allocated yet;
2413          * we couldn't try to create block if create flag is zero
2414          */
2415         if (!create) {
2416                 /*
2417                  * put just found gap into cache to speed up
2418                  * subsequent requests
2419                  */
2420                 ext4_ext_put_gap_in_cache(inode, path, iblock);
2421                 goto out2;
2422         }
2423         /*
2424          * Okay, we need to do block allocation.  Lazily initialize the block
2425          * allocation info here if necessary.
2426          */
2427         if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2428                 ext4_init_block_alloc_info(inode);
2429
2430         /* find neighbour allocated blocks */
2431         ar.lleft = iblock;
2432         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2433         if (err)
2434                 goto out2;
2435         ar.lright = iblock;
2436         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2437         if (err)
2438                 goto out2;
2439
2440         /*
2441          * See if request is beyond maximum number of blocks we can have in
2442          * a single extent. For an initialized extent this limit is
2443          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2444          * EXT_UNINIT_MAX_LEN.
2445          */
2446         if (max_blocks > EXT_INIT_MAX_LEN &&
2447             create != EXT4_CREATE_UNINITIALIZED_EXT)
2448                 max_blocks = EXT_INIT_MAX_LEN;
2449         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2450                  create == EXT4_CREATE_UNINITIALIZED_EXT)
2451                 max_blocks = EXT_UNINIT_MAX_LEN;
2452
2453         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2454         newex.ee_block = cpu_to_le32(iblock);
2455         newex.ee_len = cpu_to_le16(max_blocks);
2456         err = ext4_ext_check_overlap(inode, &newex, path);
2457         if (err)
2458                 allocated = ext4_ext_get_actual_len(&newex);
2459         else
2460                 allocated = max_blocks;
2461
2462         /* allocate new block */
2463         ar.inode = inode;
2464         ar.goal = ext4_ext_find_goal(inode, path, iblock);
2465         ar.logical = iblock;
2466         ar.len = allocated;
2467         if (S_ISREG(inode->i_mode))
2468                 ar.flags = EXT4_MB_HINT_DATA;
2469         else
2470                 /* disable in-core preallocation for non-regular files */
2471                 ar.flags = 0;
2472         newblock = ext4_mb_new_blocks(handle, &ar, &err);
2473         if (!newblock)
2474                 goto out2;
2475         ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2476                         goal, newblock, allocated);
2477
2478         /* try to insert new extent into found leaf and return */
2479         ext4_ext_store_pblock(&newex, newblock);
2480         newex.ee_len = cpu_to_le16(ar.len);
2481         if (create == EXT4_CREATE_UNINITIALIZED_EXT)  /* Mark uninitialized */
2482                 ext4_ext_mark_uninitialized(&newex);
2483         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2484         if (err) {
2485                 /* free data blocks we just allocated */
2486                 /* not a good idea to call discard here directly,
2487                  * but otherwise we'd need to call it every free() */
2488                 ext4_mb_discard_inode_preallocations(inode);
2489                 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2490                                         ext4_ext_get_actual_len(&newex), 0);
2491                 goto out2;
2492         }
2493
2494         if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2495                 EXT4_I(inode)->i_disksize = inode->i_size;
2496
2497         /* previous routine could use block we allocated */
2498         newblock = ext_pblock(&newex);
2499         allocated = ext4_ext_get_actual_len(&newex);
2500 outnew:
2501         __set_bit(BH_New, &bh_result->b_state);
2502
2503         /* Cache only when it is _not_ an uninitialized extent */
2504         if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2505                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2506                                                 EXT4_EXT_CACHE_EXTENT);
2507 out:
2508         if (allocated > max_blocks)
2509                 allocated = max_blocks;
2510         ext4_ext_show_leaf(inode, path);
2511         __set_bit(BH_Mapped, &bh_result->b_state);
2512         bh_result->b_bdev = inode->i_sb->s_bdev;
2513         bh_result->b_blocknr = newblock;
2514 out2:
2515         if (path) {
2516                 ext4_ext_drop_refs(path);
2517                 kfree(path);
2518         }
2519         return err ? err : allocated;
2520 }
2521
2522 void ext4_ext_truncate(struct inode * inode, struct page *page)
2523 {
2524         struct address_space *mapping = inode->i_mapping;
2525         struct super_block *sb = inode->i_sb;
2526         ext4_lblk_t last_block;
2527         handle_t *handle;
2528         int err = 0;
2529
2530         /*
2531          * probably first extent we're gonna free will be last in block
2532          */
2533         err = ext4_writepage_trans_blocks(inode) + 3;
2534         handle = ext4_journal_start(inode, err);
2535         if (IS_ERR(handle)) {
2536                 if (page) {
2537                         clear_highpage(page);
2538                         flush_dcache_page(page);
2539                         unlock_page(page);
2540                         page_cache_release(page);
2541                 }
2542                 return;
2543         }
2544
2545         if (page)
2546                 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2547
2548         down_write(&EXT4_I(inode)->i_data_sem);
2549         ext4_ext_invalidate_cache(inode);
2550
2551         ext4_mb_discard_inode_preallocations(inode);
2552
2553         /*
2554          * TODO: optimization is possible here.
2555          * Probably we need not scan at all,
2556          * because page truncation is enough.
2557          */
2558         if (ext4_orphan_add(handle, inode))
2559                 goto out_stop;
2560
2561         /* we have to know where to truncate from in crash case */
2562         EXT4_I(inode)->i_disksize = inode->i_size;
2563         ext4_mark_inode_dirty(handle, inode);
2564
2565         last_block = (inode->i_size + sb->s_blocksize - 1)
2566                         >> EXT4_BLOCK_SIZE_BITS(sb);
2567         err = ext4_ext_remove_space(inode, last_block);
2568
2569         /* In a multi-transaction truncate, we only make the final
2570          * transaction synchronous.
2571          */
2572         if (IS_SYNC(inode))
2573                 handle->h_sync = 1;
2574
2575 out_stop:
2576         /*
2577          * If this was a simple ftruncate() and the file will remain alive,
2578          * then we need to clear up the orphan record which we created above.
2579          * However, if this was a real unlink then we were called by
2580          * ext4_delete_inode(), and we allow that function to clean up the
2581          * orphan info for us.
2582          */
2583         if (inode->i_nlink)
2584                 ext4_orphan_del(handle, inode);
2585
2586         up_write(&EXT4_I(inode)->i_data_sem);
2587         ext4_journal_stop(handle);
2588 }
2589
2590 /*
2591  * ext4_ext_writepage_trans_blocks:
2592  * calculate max number of blocks we could modify
2593  * in order to allocate new block for an inode
2594  */
2595 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2596 {
2597         int needed;
2598
2599         needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2600
2601         /* caller wants to allocate num blocks, but note it includes sb */
2602         needed = needed * num - (num - 1);
2603
2604 #ifdef CONFIG_QUOTA
2605         needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2606 #endif
2607
2608         return needed;
2609 }
2610
2611 /*
2612  * preallocate space for a file. This implements ext4's fallocate inode
2613  * operation, which gets called from sys_fallocate system call.
2614  * For block-mapped files, posix_fallocate should fall back to the method
2615  * of writing zeroes to the required new blocks (the same behavior which is
2616  * expected for file systems which do not support fallocate() system call).
2617  */
2618 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2619 {
2620         handle_t *handle;
2621         ext4_lblk_t block;
2622         unsigned long max_blocks;
2623         ext4_fsblk_t nblocks = 0;
2624         int ret = 0;
2625         int ret2 = 0;
2626         int retries = 0;
2627         struct buffer_head map_bh;
2628         unsigned int credits, blkbits = inode->i_blkbits;
2629
2630         /*
2631          * currently supporting (pre)allocate mode for extent-based
2632          * files _only_
2633          */
2634         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2635                 return -EOPNOTSUPP;
2636
2637         /* preallocation to directories is currently not supported */
2638         if (S_ISDIR(inode->i_mode))
2639                 return -ENODEV;
2640
2641         block = offset >> blkbits;
2642         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2643                         - block;
2644
2645         /*
2646          * credits to insert 1 extent into extent tree + buffers to be able to
2647          * modify 1 super block, 1 block bitmap and 1 group descriptor.
2648          */
2649         credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
2650         mutex_lock(&inode->i_mutex);
2651 retry:
2652         while (ret >= 0 && ret < max_blocks) {
2653                 block = block + ret;
2654                 max_blocks = max_blocks - ret;
2655                 handle = ext4_journal_start(inode, credits);
2656                 if (IS_ERR(handle)) {
2657                         ret = PTR_ERR(handle);
2658                         break;
2659                 }
2660
2661                 ret = ext4_get_blocks_wrap(handle, inode, block,
2662                                           max_blocks, &map_bh,
2663                                           EXT4_CREATE_UNINITIALIZED_EXT, 0);
2664                 if (ret <= 0) {
2665 #ifdef EXT4FS_DEBUG
2666                         WARN_ON(ret <= 0);
2667                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
2668                                     "returned error inode#%lu, block=%u, "
2669                                     "max_blocks=%lu", __func__,
2670                                     inode->i_ino, block, max_blocks);
2671 #endif
2672                         ext4_mark_inode_dirty(handle, inode);
2673                         ret2 = ext4_journal_stop(handle);
2674                         break;
2675                 }
2676                 if (ret > 0) {
2677                         /* check wrap through sign-bit/zero here */
2678                         if ((block + ret) < 0 || (block + ret) < block) {
2679                                 ret = -EIO;
2680                                 ext4_mark_inode_dirty(handle, inode);
2681                                 ret2 = ext4_journal_stop(handle);
2682                                 break;
2683                         }
2684                         if (buffer_new(&map_bh) && ((block + ret) >
2685                             (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits)
2686                             >> blkbits)))
2687                                         nblocks = nblocks + ret;
2688                 }
2689
2690                 /* Update ctime if new blocks get allocated */
2691                 if (nblocks) {
2692                         struct timespec now;
2693
2694                         now = current_fs_time(inode->i_sb);
2695                         if (!timespec_equal(&inode->i_ctime, &now))
2696                                 inode->i_ctime = now;
2697                 }
2698
2699                 ext4_mark_inode_dirty(handle, inode);
2700                 ret2 = ext4_journal_stop(handle);
2701                 if (ret2)
2702                         break;
2703         }
2704
2705         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2706                 goto retry;
2707
2708         /*
2709          * Time to update the file size.
2710          * Update only when preallocation was requested beyond the file size.
2711          */
2712         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2713             (offset + len) > i_size_read(inode)) {
2714                 if (ret > 0) {
2715                         /*
2716                          * if no error, we assume preallocation succeeded
2717                          * completely
2718                          */
2719                         i_size_write(inode, offset + len);
2720                         EXT4_I(inode)->i_disksize = i_size_read(inode);
2721                 } else if (ret < 0 && nblocks) {
2722                         /* Handle partial allocation scenario */
2723                         loff_t newsize;
2724
2725                         newsize  = (nblocks << blkbits) + i_size_read(inode);
2726                         i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits));
2727                         EXT4_I(inode)->i_disksize = i_size_read(inode);
2728                 }
2729         }
2730
2731         mutex_unlock(&inode->i_mutex);
2732         return ret > 0 ? ret2 : ret;
2733 }