ext4: Add new functions for searching extent tree
[safe/jmp/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_grpblk_t colour;
152         int depth;
153
154         if (path) {
155                 struct ext4_extent *ex;
156                 depth = path->p_depth;
157
158                 /* try to predict block placement */
159                 ex = path[depth].p_ext;
160                 if (ex)
161                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
162
163                 /* it looks like index is empty;
164                  * try to find starting block from index itself */
165                 if (path[depth].p_bh)
166                         return path[depth].p_bh->b_blocknr;
167         }
168
169         /* OK. use inode's group */
170         bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
171                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
172         colour = (current->pid % 16) *
173                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
174         return bg_start + colour + block;
175 }
176
177 static ext4_fsblk_t
178 ext4_ext_new_block(handle_t *handle, struct inode *inode,
179                         struct ext4_ext_path *path,
180                         struct ext4_extent *ex, int *err)
181 {
182         ext4_fsblk_t goal, newblock;
183
184         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
185         newblock = ext4_new_block(handle, inode, goal, err);
186         return newblock;
187 }
188
189 static int ext4_ext_space_block(struct inode *inode)
190 {
191         int size;
192
193         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
194                         / sizeof(struct ext4_extent);
195 #ifdef AGGRESSIVE_TEST
196         if (size > 6)
197                 size = 6;
198 #endif
199         return size;
200 }
201
202 static int ext4_ext_space_block_idx(struct inode *inode)
203 {
204         int size;
205
206         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
207                         / sizeof(struct ext4_extent_idx);
208 #ifdef AGGRESSIVE_TEST
209         if (size > 5)
210                 size = 5;
211 #endif
212         return size;
213 }
214
215 static int ext4_ext_space_root(struct inode *inode)
216 {
217         int size;
218
219         size = sizeof(EXT4_I(inode)->i_data);
220         size -= sizeof(struct ext4_extent_header);
221         size /= sizeof(struct ext4_extent);
222 #ifdef AGGRESSIVE_TEST
223         if (size > 3)
224                 size = 3;
225 #endif
226         return size;
227 }
228
229 static int ext4_ext_space_root_idx(struct inode *inode)
230 {
231         int size;
232
233         size = sizeof(EXT4_I(inode)->i_data);
234         size -= sizeof(struct ext4_extent_header);
235         size /= sizeof(struct ext4_extent_idx);
236 #ifdef AGGRESSIVE_TEST
237         if (size > 4)
238                 size = 4;
239 #endif
240         return size;
241 }
242
243 static int
244 ext4_ext_max_entries(struct inode *inode, int depth)
245 {
246         int max;
247
248         if (depth == ext_depth(inode)) {
249                 if (depth == 0)
250                         max = ext4_ext_space_root(inode);
251                 else
252                         max = ext4_ext_space_root_idx(inode);
253         } else {
254                 if (depth == 0)
255                         max = ext4_ext_space_block(inode);
256                 else
257                         max = ext4_ext_space_block_idx(inode);
258         }
259
260         return max;
261 }
262
263 static int __ext4_ext_check_header(const char *function, struct inode *inode,
264                                         struct ext4_extent_header *eh,
265                                         int depth)
266 {
267         const char *error_msg;
268         int max = 0;
269
270         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
271                 error_msg = "invalid magic";
272                 goto corrupted;
273         }
274         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
275                 error_msg = "unexpected eh_depth";
276                 goto corrupted;
277         }
278         if (unlikely(eh->eh_max == 0)) {
279                 error_msg = "invalid eh_max";
280                 goto corrupted;
281         }
282         max = ext4_ext_max_entries(inode, depth);
283         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
284                 error_msg = "too large eh_max";
285                 goto corrupted;
286         }
287         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
288                 error_msg = "invalid eh_entries";
289                 goto corrupted;
290         }
291         return 0;
292
293 corrupted:
294         ext4_error(inode->i_sb, function,
295                         "bad header in inode #%lu: %s - magic %x, "
296                         "entries %u, max %u(%u), depth %u(%u)",
297                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
298                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
299                         max, le16_to_cpu(eh->eh_depth), depth);
300
301         return -EIO;
302 }
303
304 #define ext4_ext_check_header(inode, eh, depth) \
305         __ext4_ext_check_header(__FUNCTION__, inode, eh, depth)
306
307 #ifdef EXT_DEBUG
308 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
309 {
310         int k, l = path->p_depth;
311
312         ext_debug("path:");
313         for (k = 0; k <= l; k++, path++) {
314                 if (path->p_idx) {
315                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
316                             idx_pblock(path->p_idx));
317                 } else if (path->p_ext) {
318                         ext_debug("  %d:%d:%llu ",
319                                   le32_to_cpu(path->p_ext->ee_block),
320                                   ext4_ext_get_actual_len(path->p_ext),
321                                   ext_pblock(path->p_ext));
322                 } else
323                         ext_debug("  []");
324         }
325         ext_debug("\n");
326 }
327
328 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
329 {
330         int depth = ext_depth(inode);
331         struct ext4_extent_header *eh;
332         struct ext4_extent *ex;
333         int i;
334
335         if (!path)
336                 return;
337
338         eh = path[depth].p_hdr;
339         ex = EXT_FIRST_EXTENT(eh);
340
341         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
342                 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
343                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
344         }
345         ext_debug("\n");
346 }
347 #else
348 #define ext4_ext_show_path(inode,path)
349 #define ext4_ext_show_leaf(inode,path)
350 #endif
351
352 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
353 {
354         int depth = path->p_depth;
355         int i;
356
357         for (i = 0; i <= depth; i++, path++)
358                 if (path->p_bh) {
359                         brelse(path->p_bh);
360                         path->p_bh = NULL;
361                 }
362 }
363
364 /*
365  * ext4_ext_binsearch_idx:
366  * binary search for the closest index of the given block
367  * the header must be checked before calling this
368  */
369 static void
370 ext4_ext_binsearch_idx(struct inode *inode,
371                         struct ext4_ext_path *path, ext4_lblk_t block)
372 {
373         struct ext4_extent_header *eh = path->p_hdr;
374         struct ext4_extent_idx *r, *l, *m;
375
376
377         ext_debug("binsearch for %u(idx):  ", block);
378
379         l = EXT_FIRST_INDEX(eh) + 1;
380         r = EXT_LAST_INDEX(eh);
381         while (l <= r) {
382                 m = l + (r - l) / 2;
383                 if (block < le32_to_cpu(m->ei_block))
384                         r = m - 1;
385                 else
386                         l = m + 1;
387                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
388                                 m, le32_to_cpu(m->ei_block),
389                                 r, le32_to_cpu(r->ei_block));
390         }
391
392         path->p_idx = l - 1;
393         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
394                   idx_pblock(path->p_idx));
395
396 #ifdef CHECK_BINSEARCH
397         {
398                 struct ext4_extent_idx *chix, *ix;
399                 int k;
400
401                 chix = ix = EXT_FIRST_INDEX(eh);
402                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
403                   if (k != 0 &&
404                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
405                                 printk("k=%d, ix=0x%p, first=0x%p\n", k,
406                                         ix, EXT_FIRST_INDEX(eh));
407                                 printk("%u <= %u\n",
408                                        le32_to_cpu(ix->ei_block),
409                                        le32_to_cpu(ix[-1].ei_block));
410                         }
411                         BUG_ON(k && le32_to_cpu(ix->ei_block)
412                                            <= le32_to_cpu(ix[-1].ei_block));
413                         if (block < le32_to_cpu(ix->ei_block))
414                                 break;
415                         chix = ix;
416                 }
417                 BUG_ON(chix != path->p_idx);
418         }
419 #endif
420
421 }
422
423 /*
424  * ext4_ext_binsearch:
425  * binary search for closest extent of the given block
426  * the header must be checked before calling this
427  */
428 static void
429 ext4_ext_binsearch(struct inode *inode,
430                 struct ext4_ext_path *path, ext4_lblk_t block)
431 {
432         struct ext4_extent_header *eh = path->p_hdr;
433         struct ext4_extent *r, *l, *m;
434
435         if (eh->eh_entries == 0) {
436                 /*
437                  * this leaf is empty:
438                  * we get such a leaf in split/add case
439                  */
440                 return;
441         }
442
443         ext_debug("binsearch for %u:  ", block);
444
445         l = EXT_FIRST_EXTENT(eh) + 1;
446         r = EXT_LAST_EXTENT(eh);
447
448         while (l <= r) {
449                 m = l + (r - l) / 2;
450                 if (block < le32_to_cpu(m->ee_block))
451                         r = m - 1;
452                 else
453                         l = m + 1;
454                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
455                                 m, le32_to_cpu(m->ee_block),
456                                 r, le32_to_cpu(r->ee_block));
457         }
458
459         path->p_ext = l - 1;
460         ext_debug("  -> %d:%llu:%d ",
461                         le32_to_cpu(path->p_ext->ee_block),
462                         ext_pblock(path->p_ext),
463                         ext4_ext_get_actual_len(path->p_ext));
464
465 #ifdef CHECK_BINSEARCH
466         {
467                 struct ext4_extent *chex, *ex;
468                 int k;
469
470                 chex = ex = EXT_FIRST_EXTENT(eh);
471                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
472                         BUG_ON(k && le32_to_cpu(ex->ee_block)
473                                           <= le32_to_cpu(ex[-1].ee_block));
474                         if (block < le32_to_cpu(ex->ee_block))
475                                 break;
476                         chex = ex;
477                 }
478                 BUG_ON(chex != path->p_ext);
479         }
480 #endif
481
482 }
483
484 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
485 {
486         struct ext4_extent_header *eh;
487
488         eh = ext_inode_hdr(inode);
489         eh->eh_depth = 0;
490         eh->eh_entries = 0;
491         eh->eh_magic = EXT4_EXT_MAGIC;
492         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
493         ext4_mark_inode_dirty(handle, inode);
494         ext4_ext_invalidate_cache(inode);
495         return 0;
496 }
497
498 struct ext4_ext_path *
499 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
500                                         struct ext4_ext_path *path)
501 {
502         struct ext4_extent_header *eh;
503         struct buffer_head *bh;
504         short int depth, i, ppos = 0, alloc = 0;
505
506         eh = ext_inode_hdr(inode);
507         depth = ext_depth(inode);
508         if (ext4_ext_check_header(inode, eh, depth))
509                 return ERR_PTR(-EIO);
510
511
512         /* account possible depth increase */
513         if (!path) {
514                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
515                                 GFP_NOFS);
516                 if (!path)
517                         return ERR_PTR(-ENOMEM);
518                 alloc = 1;
519         }
520         path[0].p_hdr = eh;
521
522         i = depth;
523         /* walk through the tree */
524         while (i) {
525                 ext_debug("depth %d: num %d, max %d\n",
526                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
527
528                 ext4_ext_binsearch_idx(inode, path + ppos, block);
529                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
530                 path[ppos].p_depth = i;
531                 path[ppos].p_ext = NULL;
532
533                 bh = sb_bread(inode->i_sb, path[ppos].p_block);
534                 if (!bh)
535                         goto err;
536
537                 eh = ext_block_hdr(bh);
538                 ppos++;
539                 BUG_ON(ppos > depth);
540                 path[ppos].p_bh = bh;
541                 path[ppos].p_hdr = eh;
542                 i--;
543
544                 if (ext4_ext_check_header(inode, eh, i))
545                         goto err;
546         }
547
548         path[ppos].p_depth = i;
549         path[ppos].p_hdr = eh;
550         path[ppos].p_ext = NULL;
551         path[ppos].p_idx = NULL;
552
553         /* find extent */
554         ext4_ext_binsearch(inode, path + ppos, block);
555
556         ext4_ext_show_path(inode, path);
557
558         return path;
559
560 err:
561         ext4_ext_drop_refs(path);
562         if (alloc)
563                 kfree(path);
564         return ERR_PTR(-EIO);
565 }
566
567 /*
568  * ext4_ext_insert_index:
569  * insert new index [@logical;@ptr] into the block at @curp;
570  * check where to insert: before @curp or after @curp
571  */
572 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
573                                 struct ext4_ext_path *curp,
574                                 int logical, ext4_fsblk_t ptr)
575 {
576         struct ext4_extent_idx *ix;
577         int len, err;
578
579         err = ext4_ext_get_access(handle, inode, curp);
580         if (err)
581                 return err;
582
583         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
584         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
585         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
586                 /* insert after */
587                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
588                         len = (len - 1) * sizeof(struct ext4_extent_idx);
589                         len = len < 0 ? 0 : len;
590                         ext_debug("insert new index %d after: %llu. "
591                                         "move %d from 0x%p to 0x%p\n",
592                                         logical, ptr, len,
593                                         (curp->p_idx + 1), (curp->p_idx + 2));
594                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
595                 }
596                 ix = curp->p_idx + 1;
597         } else {
598                 /* insert before */
599                 len = len * sizeof(struct ext4_extent_idx);
600                 len = len < 0 ? 0 : len;
601                 ext_debug("insert new index %d before: %llu. "
602                                 "move %d from 0x%p to 0x%p\n",
603                                 logical, ptr, len,
604                                 curp->p_idx, (curp->p_idx + 1));
605                 memmove(curp->p_idx + 1, curp->p_idx, len);
606                 ix = curp->p_idx;
607         }
608
609         ix->ei_block = cpu_to_le32(logical);
610         ext4_idx_store_pblock(ix, ptr);
611         curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
612
613         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
614                              > le16_to_cpu(curp->p_hdr->eh_max));
615         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
616
617         err = ext4_ext_dirty(handle, inode, curp);
618         ext4_std_error(inode->i_sb, err);
619
620         return err;
621 }
622
623 /*
624  * ext4_ext_split:
625  * inserts new subtree into the path, using free index entry
626  * at depth @at:
627  * - allocates all needed blocks (new leaf and all intermediate index blocks)
628  * - makes decision where to split
629  * - moves remaining extents and index entries (right to the split point)
630  *   into the newly allocated blocks
631  * - initializes subtree
632  */
633 static int ext4_ext_split(handle_t *handle, struct inode *inode,
634                                 struct ext4_ext_path *path,
635                                 struct ext4_extent *newext, int at)
636 {
637         struct buffer_head *bh = NULL;
638         int depth = ext_depth(inode);
639         struct ext4_extent_header *neh;
640         struct ext4_extent_idx *fidx;
641         struct ext4_extent *ex;
642         int i = at, k, m, a;
643         ext4_fsblk_t newblock, oldblock;
644         __le32 border;
645         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
646         int err = 0;
647
648         /* make decision: where to split? */
649         /* FIXME: now decision is simplest: at current extent */
650
651         /* if current leaf will be split, then we should use
652          * border from split point */
653         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
654         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
655                 border = path[depth].p_ext[1].ee_block;
656                 ext_debug("leaf will be split."
657                                 " next leaf starts at %d\n",
658                                   le32_to_cpu(border));
659         } else {
660                 border = newext->ee_block;
661                 ext_debug("leaf will be added."
662                                 " next leaf starts at %d\n",
663                                 le32_to_cpu(border));
664         }
665
666         /*
667          * If error occurs, then we break processing
668          * and mark filesystem read-only. index won't
669          * be inserted and tree will be in consistent
670          * state. Next mount will repair buffers too.
671          */
672
673         /*
674          * Get array to track all allocated blocks.
675          * We need this to handle errors and free blocks
676          * upon them.
677          */
678         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
679         if (!ablocks)
680                 return -ENOMEM;
681
682         /* allocate all needed blocks */
683         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
684         for (a = 0; a < depth - at; a++) {
685                 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
686                 if (newblock == 0)
687                         goto cleanup;
688                 ablocks[a] = newblock;
689         }
690
691         /* initialize new leaf */
692         newblock = ablocks[--a];
693         BUG_ON(newblock == 0);
694         bh = sb_getblk(inode->i_sb, newblock);
695         if (!bh) {
696                 err = -EIO;
697                 goto cleanup;
698         }
699         lock_buffer(bh);
700
701         err = ext4_journal_get_create_access(handle, bh);
702         if (err)
703                 goto cleanup;
704
705         neh = ext_block_hdr(bh);
706         neh->eh_entries = 0;
707         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
708         neh->eh_magic = EXT4_EXT_MAGIC;
709         neh->eh_depth = 0;
710         ex = EXT_FIRST_EXTENT(neh);
711
712         /* move remainder of path[depth] to the new leaf */
713         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
714         /* start copy from next extent */
715         /* TODO: we could do it by single memmove */
716         m = 0;
717         path[depth].p_ext++;
718         while (path[depth].p_ext <=
719                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
720                 ext_debug("move %d:%llu:%d in new leaf %llu\n",
721                                 le32_to_cpu(path[depth].p_ext->ee_block),
722                                 ext_pblock(path[depth].p_ext),
723                                 ext4_ext_get_actual_len(path[depth].p_ext),
724                                 newblock);
725                 /*memmove(ex++, path[depth].p_ext++,
726                                 sizeof(struct ext4_extent));
727                 neh->eh_entries++;*/
728                 path[depth].p_ext++;
729                 m++;
730         }
731         if (m) {
732                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
733                 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
734         }
735
736         set_buffer_uptodate(bh);
737         unlock_buffer(bh);
738
739         err = ext4_journal_dirty_metadata(handle, bh);
740         if (err)
741                 goto cleanup;
742         brelse(bh);
743         bh = NULL;
744
745         /* correct old leaf */
746         if (m) {
747                 err = ext4_ext_get_access(handle, inode, path + depth);
748                 if (err)
749                         goto cleanup;
750                 path[depth].p_hdr->eh_entries =
751                      cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
752                 err = ext4_ext_dirty(handle, inode, path + depth);
753                 if (err)
754                         goto cleanup;
755
756         }
757
758         /* create intermediate indexes */
759         k = depth - at - 1;
760         BUG_ON(k < 0);
761         if (k)
762                 ext_debug("create %d intermediate indices\n", k);
763         /* insert new index into current index block */
764         /* current depth stored in i var */
765         i = depth - 1;
766         while (k--) {
767                 oldblock = newblock;
768                 newblock = ablocks[--a];
769                 bh = sb_getblk(inode->i_sb, newblock);
770                 if (!bh) {
771                         err = -EIO;
772                         goto cleanup;
773                 }
774                 lock_buffer(bh);
775
776                 err = ext4_journal_get_create_access(handle, bh);
777                 if (err)
778                         goto cleanup;
779
780                 neh = ext_block_hdr(bh);
781                 neh->eh_entries = cpu_to_le16(1);
782                 neh->eh_magic = EXT4_EXT_MAGIC;
783                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
784                 neh->eh_depth = cpu_to_le16(depth - i);
785                 fidx = EXT_FIRST_INDEX(neh);
786                 fidx->ei_block = border;
787                 ext4_idx_store_pblock(fidx, oldblock);
788
789                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
790                                 i, newblock, le32_to_cpu(border), oldblock);
791                 /* copy indexes */
792                 m = 0;
793                 path[i].p_idx++;
794
795                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
796                                 EXT_MAX_INDEX(path[i].p_hdr));
797                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
798                                 EXT_LAST_INDEX(path[i].p_hdr));
799                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
800                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
801                                         le32_to_cpu(path[i].p_idx->ei_block),
802                                         idx_pblock(path[i].p_idx),
803                                         newblock);
804                         /*memmove(++fidx, path[i].p_idx++,
805                                         sizeof(struct ext4_extent_idx));
806                         neh->eh_entries++;
807                         BUG_ON(neh->eh_entries > neh->eh_max);*/
808                         path[i].p_idx++;
809                         m++;
810                 }
811                 if (m) {
812                         memmove(++fidx, path[i].p_idx - m,
813                                 sizeof(struct ext4_extent_idx) * m);
814                         neh->eh_entries =
815                                 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
816                 }
817                 set_buffer_uptodate(bh);
818                 unlock_buffer(bh);
819
820                 err = ext4_journal_dirty_metadata(handle, bh);
821                 if (err)
822                         goto cleanup;
823                 brelse(bh);
824                 bh = NULL;
825
826                 /* correct old index */
827                 if (m) {
828                         err = ext4_ext_get_access(handle, inode, path + i);
829                         if (err)
830                                 goto cleanup;
831                         path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
832                         err = ext4_ext_dirty(handle, inode, path + i);
833                         if (err)
834                                 goto cleanup;
835                 }
836
837                 i--;
838         }
839
840         /* insert new index */
841         err = ext4_ext_insert_index(handle, inode, path + at,
842                                     le32_to_cpu(border), newblock);
843
844 cleanup:
845         if (bh) {
846                 if (buffer_locked(bh))
847                         unlock_buffer(bh);
848                 brelse(bh);
849         }
850
851         if (err) {
852                 /* free all allocated blocks in error case */
853                 for (i = 0; i < depth; i++) {
854                         if (!ablocks[i])
855                                 continue;
856                         ext4_free_blocks(handle, inode, ablocks[i], 1);
857                 }
858         }
859         kfree(ablocks);
860
861         return err;
862 }
863
864 /*
865  * ext4_ext_grow_indepth:
866  * implements tree growing procedure:
867  * - allocates new block
868  * - moves top-level data (index block or leaf) into the new block
869  * - initializes new top-level, creating index that points to the
870  *   just created block
871  */
872 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
873                                         struct ext4_ext_path *path,
874                                         struct ext4_extent *newext)
875 {
876         struct ext4_ext_path *curp = path;
877         struct ext4_extent_header *neh;
878         struct ext4_extent_idx *fidx;
879         struct buffer_head *bh;
880         ext4_fsblk_t newblock;
881         int err = 0;
882
883         newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
884         if (newblock == 0)
885                 return err;
886
887         bh = sb_getblk(inode->i_sb, newblock);
888         if (!bh) {
889                 err = -EIO;
890                 ext4_std_error(inode->i_sb, err);
891                 return err;
892         }
893         lock_buffer(bh);
894
895         err = ext4_journal_get_create_access(handle, bh);
896         if (err) {
897                 unlock_buffer(bh);
898                 goto out;
899         }
900
901         /* move top-level index/leaf into new block */
902         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
903
904         /* set size of new block */
905         neh = ext_block_hdr(bh);
906         /* old root could have indexes or leaves
907          * so calculate e_max right way */
908         if (ext_depth(inode))
909           neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
910         else
911           neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
912         neh->eh_magic = EXT4_EXT_MAGIC;
913         set_buffer_uptodate(bh);
914         unlock_buffer(bh);
915
916         err = ext4_journal_dirty_metadata(handle, bh);
917         if (err)
918                 goto out;
919
920         /* create index in new top-level index: num,max,pointer */
921         err = ext4_ext_get_access(handle, inode, curp);
922         if (err)
923                 goto out;
924
925         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
926         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
927         curp->p_hdr->eh_entries = cpu_to_le16(1);
928         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
929
930         if (path[0].p_hdr->eh_depth)
931                 curp->p_idx->ei_block =
932                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
933         else
934                 curp->p_idx->ei_block =
935                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
936         ext4_idx_store_pblock(curp->p_idx, newblock);
937
938         neh = ext_inode_hdr(inode);
939         fidx = EXT_FIRST_INDEX(neh);
940         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
941                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
942                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
943
944         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
945         err = ext4_ext_dirty(handle, inode, curp);
946 out:
947         brelse(bh);
948
949         return err;
950 }
951
952 /*
953  * ext4_ext_create_new_leaf:
954  * finds empty index and adds new leaf.
955  * if no free index is found, then it requests in-depth growing.
956  */
957 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
958                                         struct ext4_ext_path *path,
959                                         struct ext4_extent *newext)
960 {
961         struct ext4_ext_path *curp;
962         int depth, i, err = 0;
963
964 repeat:
965         i = depth = ext_depth(inode);
966
967         /* walk up to the tree and look for free index entry */
968         curp = path + depth;
969         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
970                 i--;
971                 curp--;
972         }
973
974         /* we use already allocated block for index block,
975          * so subsequent data blocks should be contiguous */
976         if (EXT_HAS_FREE_INDEX(curp)) {
977                 /* if we found index with free entry, then use that
978                  * entry: create all needed subtree and add new leaf */
979                 err = ext4_ext_split(handle, inode, path, newext, i);
980
981                 /* refill path */
982                 ext4_ext_drop_refs(path);
983                 path = ext4_ext_find_extent(inode,
984                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
985                                     path);
986                 if (IS_ERR(path))
987                         err = PTR_ERR(path);
988         } else {
989                 /* tree is full, time to grow in depth */
990                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
991                 if (err)
992                         goto out;
993
994                 /* refill path */
995                 ext4_ext_drop_refs(path);
996                 path = ext4_ext_find_extent(inode,
997                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
998                                     path);
999                 if (IS_ERR(path)) {
1000                         err = PTR_ERR(path);
1001                         goto out;
1002                 }
1003
1004                 /*
1005                  * only first (depth 0 -> 1) produces free space;
1006                  * in all other cases we have to split the grown tree
1007                  */
1008                 depth = ext_depth(inode);
1009                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1010                         /* now we need to split */
1011                         goto repeat;
1012                 }
1013         }
1014
1015 out:
1016         return err;
1017 }
1018
1019 /*
1020  * search the closest allocated block to the left for *logical
1021  * and returns it at @logical + it's physical address at @phys
1022  * if *logical is the smallest allocated block, the function
1023  * returns 0 at @phys
1024  * return value contains 0 (success) or error code
1025  */
1026 int
1027 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1028                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1029 {
1030         struct ext4_extent_idx *ix;
1031         struct ext4_extent *ex;
1032         int depth;
1033
1034         BUG_ON(path == NULL);
1035         depth = path->p_depth;
1036         *phys = 0;
1037
1038         if (depth == 0 && path->p_ext == NULL)
1039                 return 0;
1040
1041         /* usually extent in the path covers blocks smaller
1042          * then *logical, but it can be that extent is the
1043          * first one in the file */
1044
1045         ex = path[depth].p_ext;
1046         if (*logical < le32_to_cpu(ex->ee_block)) {
1047                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1048                 while (--depth >= 0) {
1049                         ix = path[depth].p_idx;
1050                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1051                 }
1052                 return 0;
1053         }
1054
1055         BUG_ON(*logical < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
1056
1057         *logical = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1;
1058         *phys = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - 1;
1059         return 0;
1060 }
1061
1062 /*
1063  * search the closest allocated block to the right for *logical
1064  * and returns it at @logical + it's physical address at @phys
1065  * if *logical is the smallest allocated block, the function
1066  * returns 0 at @phys
1067  * return value contains 0 (success) or error code
1068  */
1069 int
1070 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1071                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1072 {
1073         struct buffer_head *bh = NULL;
1074         struct ext4_extent_header *eh;
1075         struct ext4_extent_idx *ix;
1076         struct ext4_extent *ex;
1077         ext4_fsblk_t block;
1078         int depth;
1079
1080         BUG_ON(path == NULL);
1081         depth = path->p_depth;
1082         *phys = 0;
1083
1084         if (depth == 0 && path->p_ext == NULL)
1085                 return 0;
1086
1087         /* usually extent in the path covers blocks smaller
1088          * then *logical, but it can be that extent is the
1089          * first one in the file */
1090
1091         ex = path[depth].p_ext;
1092         if (*logical < le32_to_cpu(ex->ee_block)) {
1093                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1094                 while (--depth >= 0) {
1095                         ix = path[depth].p_idx;
1096                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1097                 }
1098                 *logical = le32_to_cpu(ex->ee_block);
1099                 *phys = ext_pblock(ex);
1100                 return 0;
1101         }
1102
1103         BUG_ON(*logical < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
1104
1105         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1106                 /* next allocated block in this leaf */
1107                 ex++;
1108                 *logical = le32_to_cpu(ex->ee_block);
1109                 *phys = ext_pblock(ex);
1110                 return 0;
1111         }
1112
1113         /* go up and search for index to the right */
1114         while (--depth >= 0) {
1115                 ix = path[depth].p_idx;
1116                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1117                         break;
1118         }
1119
1120         if (depth < 0) {
1121                 /* we've gone up to the root and
1122                  * found no index to the right */
1123                 return 0;
1124         }
1125
1126         /* we've found index to the right, let's
1127          * follow it and find the closest allocated
1128          * block to the right */
1129         ix++;
1130         block = idx_pblock(ix);
1131         while (++depth < path->p_depth) {
1132                 bh = sb_bread(inode->i_sb, block);
1133                 if (bh == NULL)
1134                         return -EIO;
1135                 eh = ext_block_hdr(bh);
1136                 if (ext4_ext_check_header(inode, eh, depth)) {
1137                         put_bh(bh);
1138                         return -EIO;
1139                 }
1140                 ix = EXT_FIRST_INDEX(eh);
1141                 block = idx_pblock(ix);
1142                 put_bh(bh);
1143         }
1144
1145         bh = sb_bread(inode->i_sb, block);
1146         if (bh == NULL)
1147                 return -EIO;
1148         eh = ext_block_hdr(bh);
1149         if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1150                 put_bh(bh);
1151                 return -EIO;
1152         }
1153         ex = EXT_FIRST_EXTENT(eh);
1154         *logical = le32_to_cpu(ex->ee_block);
1155         *phys = ext_pblock(ex);
1156         put_bh(bh);
1157         return 0;
1158
1159 }
1160
1161 /*
1162  * ext4_ext_next_allocated_block:
1163  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1164  * NOTE: it considers block number from index entry as
1165  * allocated block. Thus, index entries have to be consistent
1166  * with leaves.
1167  */
1168 static ext4_lblk_t
1169 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1170 {
1171         int depth;
1172
1173         BUG_ON(path == NULL);
1174         depth = path->p_depth;
1175
1176         if (depth == 0 && path->p_ext == NULL)
1177                 return EXT_MAX_BLOCK;
1178
1179         while (depth >= 0) {
1180                 if (depth == path->p_depth) {
1181                         /* leaf */
1182                         if (path[depth].p_ext !=
1183                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1184                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1185                 } else {
1186                         /* index */
1187                         if (path[depth].p_idx !=
1188                                         EXT_LAST_INDEX(path[depth].p_hdr))
1189                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1190                 }
1191                 depth--;
1192         }
1193
1194         return EXT_MAX_BLOCK;
1195 }
1196
1197 /*
1198  * ext4_ext_next_leaf_block:
1199  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1200  */
1201 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1202                                         struct ext4_ext_path *path)
1203 {
1204         int depth;
1205
1206         BUG_ON(path == NULL);
1207         depth = path->p_depth;
1208
1209         /* zero-tree has no leaf blocks at all */
1210         if (depth == 0)
1211                 return EXT_MAX_BLOCK;
1212
1213         /* go to index block */
1214         depth--;
1215
1216         while (depth >= 0) {
1217                 if (path[depth].p_idx !=
1218                                 EXT_LAST_INDEX(path[depth].p_hdr))
1219                         return (ext4_lblk_t)
1220                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1221                 depth--;
1222         }
1223
1224         return EXT_MAX_BLOCK;
1225 }
1226
1227 /*
1228  * ext4_ext_correct_indexes:
1229  * if leaf gets modified and modified extent is first in the leaf,
1230  * then we have to correct all indexes above.
1231  * TODO: do we need to correct tree in all cases?
1232  */
1233 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1234                                 struct ext4_ext_path *path)
1235 {
1236         struct ext4_extent_header *eh;
1237         int depth = ext_depth(inode);
1238         struct ext4_extent *ex;
1239         __le32 border;
1240         int k, err = 0;
1241
1242         eh = path[depth].p_hdr;
1243         ex = path[depth].p_ext;
1244         BUG_ON(ex == NULL);
1245         BUG_ON(eh == NULL);
1246
1247         if (depth == 0) {
1248                 /* there is no tree at all */
1249                 return 0;
1250         }
1251
1252         if (ex != EXT_FIRST_EXTENT(eh)) {
1253                 /* we correct tree if first leaf got modified only */
1254                 return 0;
1255         }
1256
1257         /*
1258          * TODO: we need correction if border is smaller than current one
1259          */
1260         k = depth - 1;
1261         border = path[depth].p_ext->ee_block;
1262         err = ext4_ext_get_access(handle, inode, path + k);
1263         if (err)
1264                 return err;
1265         path[k].p_idx->ei_block = border;
1266         err = ext4_ext_dirty(handle, inode, path + k);
1267         if (err)
1268                 return err;
1269
1270         while (k--) {
1271                 /* change all left-side indexes */
1272                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1273                         break;
1274                 err = ext4_ext_get_access(handle, inode, path + k);
1275                 if (err)
1276                         break;
1277                 path[k].p_idx->ei_block = border;
1278                 err = ext4_ext_dirty(handle, inode, path + k);
1279                 if (err)
1280                         break;
1281         }
1282
1283         return err;
1284 }
1285
1286 static int
1287 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1288                                 struct ext4_extent *ex2)
1289 {
1290         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1291
1292         /*
1293          * Make sure that either both extents are uninitialized, or
1294          * both are _not_.
1295          */
1296         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1297                 return 0;
1298
1299         if (ext4_ext_is_uninitialized(ex1))
1300                 max_len = EXT_UNINIT_MAX_LEN;
1301         else
1302                 max_len = EXT_INIT_MAX_LEN;
1303
1304         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1305         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1306
1307         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1308                         le32_to_cpu(ex2->ee_block))
1309                 return 0;
1310
1311         /*
1312          * To allow future support for preallocated extents to be added
1313          * as an RO_COMPAT feature, refuse to merge to extents if
1314          * this can result in the top bit of ee_len being set.
1315          */
1316         if (ext1_ee_len + ext2_ee_len > max_len)
1317                 return 0;
1318 #ifdef AGGRESSIVE_TEST
1319         if (le16_to_cpu(ex1->ee_len) >= 4)
1320                 return 0;
1321 #endif
1322
1323         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1324                 return 1;
1325         return 0;
1326 }
1327
1328 /*
1329  * This function tries to merge the "ex" extent to the next extent in the tree.
1330  * It always tries to merge towards right. If you want to merge towards
1331  * left, pass "ex - 1" as argument instead of "ex".
1332  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1333  * 1 if they got merged.
1334  */
1335 int ext4_ext_try_to_merge(struct inode *inode,
1336                           struct ext4_ext_path *path,
1337                           struct ext4_extent *ex)
1338 {
1339         struct ext4_extent_header *eh;
1340         unsigned int depth, len;
1341         int merge_done = 0;
1342         int uninitialized = 0;
1343
1344         depth = ext_depth(inode);
1345         BUG_ON(path[depth].p_hdr == NULL);
1346         eh = path[depth].p_hdr;
1347
1348         while (ex < EXT_LAST_EXTENT(eh)) {
1349                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1350                         break;
1351                 /* merge with next extent! */
1352                 if (ext4_ext_is_uninitialized(ex))
1353                         uninitialized = 1;
1354                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1355                                 + ext4_ext_get_actual_len(ex + 1));
1356                 if (uninitialized)
1357                         ext4_ext_mark_uninitialized(ex);
1358
1359                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1360                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1361                                 * sizeof(struct ext4_extent);
1362                         memmove(ex + 1, ex + 2, len);
1363                 }
1364                 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
1365                 merge_done = 1;
1366                 WARN_ON(eh->eh_entries == 0);
1367                 if (!eh->eh_entries)
1368                         ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1369                            "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1370         }
1371
1372         return merge_done;
1373 }
1374
1375 /*
1376  * check if a portion of the "newext" extent overlaps with an
1377  * existing extent.
1378  *
1379  * If there is an overlap discovered, it updates the length of the newext
1380  * such that there will be no overlap, and then returns 1.
1381  * If there is no overlap found, it returns 0.
1382  */
1383 unsigned int ext4_ext_check_overlap(struct inode *inode,
1384                                     struct ext4_extent *newext,
1385                                     struct ext4_ext_path *path)
1386 {
1387         ext4_lblk_t b1, b2;
1388         unsigned int depth, len1;
1389         unsigned int ret = 0;
1390
1391         b1 = le32_to_cpu(newext->ee_block);
1392         len1 = ext4_ext_get_actual_len(newext);
1393         depth = ext_depth(inode);
1394         if (!path[depth].p_ext)
1395                 goto out;
1396         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1397
1398         /*
1399          * get the next allocated block if the extent in the path
1400          * is before the requested block(s) 
1401          */
1402         if (b2 < b1) {
1403                 b2 = ext4_ext_next_allocated_block(path);
1404                 if (b2 == EXT_MAX_BLOCK)
1405                         goto out;
1406         }
1407
1408         /* check for wrap through zero on extent logical start block*/
1409         if (b1 + len1 < b1) {
1410                 len1 = EXT_MAX_BLOCK - b1;
1411                 newext->ee_len = cpu_to_le16(len1);
1412                 ret = 1;
1413         }
1414
1415         /* check for overlap */
1416         if (b1 + len1 > b2) {
1417                 newext->ee_len = cpu_to_le16(b2 - b1);
1418                 ret = 1;
1419         }
1420 out:
1421         return ret;
1422 }
1423
1424 /*
1425  * ext4_ext_insert_extent:
1426  * tries to merge requsted extent into the existing extent or
1427  * inserts requested extent as new one into the tree,
1428  * creating new leaf in the no-space case.
1429  */
1430 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1431                                 struct ext4_ext_path *path,
1432                                 struct ext4_extent *newext)
1433 {
1434         struct ext4_extent_header * eh;
1435         struct ext4_extent *ex, *fex;
1436         struct ext4_extent *nearex; /* nearest extent */
1437         struct ext4_ext_path *npath = NULL;
1438         int depth, len, err;
1439         ext4_lblk_t next;
1440         unsigned uninitialized = 0;
1441
1442         BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1443         depth = ext_depth(inode);
1444         ex = path[depth].p_ext;
1445         BUG_ON(path[depth].p_hdr == NULL);
1446
1447         /* try to insert block into found extent and return */
1448         if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1449                 ext_debug("append %d block to %d:%d (from %llu)\n",
1450                                 ext4_ext_get_actual_len(newext),
1451                                 le32_to_cpu(ex->ee_block),
1452                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1453                 err = ext4_ext_get_access(handle, inode, path + depth);
1454                 if (err)
1455                         return err;
1456
1457                 /*
1458                  * ext4_can_extents_be_merged should have checked that either
1459                  * both extents are uninitialized, or both aren't. Thus we
1460                  * need to check only one of them here.
1461                  */
1462                 if (ext4_ext_is_uninitialized(ex))
1463                         uninitialized = 1;
1464                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1465                                         + ext4_ext_get_actual_len(newext));
1466                 if (uninitialized)
1467                         ext4_ext_mark_uninitialized(ex);
1468                 eh = path[depth].p_hdr;
1469                 nearex = ex;
1470                 goto merge;
1471         }
1472
1473 repeat:
1474         depth = ext_depth(inode);
1475         eh = path[depth].p_hdr;
1476         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1477                 goto has_space;
1478
1479         /* probably next leaf has space for us? */
1480         fex = EXT_LAST_EXTENT(eh);
1481         next = ext4_ext_next_leaf_block(inode, path);
1482         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1483             && next != EXT_MAX_BLOCK) {
1484                 ext_debug("next leaf block - %d\n", next);
1485                 BUG_ON(npath != NULL);
1486                 npath = ext4_ext_find_extent(inode, next, NULL);
1487                 if (IS_ERR(npath))
1488                         return PTR_ERR(npath);
1489                 BUG_ON(npath->p_depth != path->p_depth);
1490                 eh = npath[depth].p_hdr;
1491                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1492                         ext_debug("next leaf isnt full(%d)\n",
1493                                   le16_to_cpu(eh->eh_entries));
1494                         path = npath;
1495                         goto repeat;
1496                 }
1497                 ext_debug("next leaf has no free space(%d,%d)\n",
1498                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1499         }
1500
1501         /*
1502          * There is no free space in the found leaf.
1503          * We're gonna add a new leaf in the tree.
1504          */
1505         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1506         if (err)
1507                 goto cleanup;
1508         depth = ext_depth(inode);
1509         eh = path[depth].p_hdr;
1510
1511 has_space:
1512         nearex = path[depth].p_ext;
1513
1514         err = ext4_ext_get_access(handle, inode, path + depth);
1515         if (err)
1516                 goto cleanup;
1517
1518         if (!nearex) {
1519                 /* there is no extent in this leaf, create first one */
1520                 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1521                                 le32_to_cpu(newext->ee_block),
1522                                 ext_pblock(newext),
1523                                 ext4_ext_get_actual_len(newext));
1524                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1525         } else if (le32_to_cpu(newext->ee_block)
1526                            > le32_to_cpu(nearex->ee_block)) {
1527 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1528                 if (nearex != EXT_LAST_EXTENT(eh)) {
1529                         len = EXT_MAX_EXTENT(eh) - nearex;
1530                         len = (len - 1) * sizeof(struct ext4_extent);
1531                         len = len < 0 ? 0 : len;
1532                         ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1533                                         "move %d from 0x%p to 0x%p\n",
1534                                         le32_to_cpu(newext->ee_block),
1535                                         ext_pblock(newext),
1536                                         ext4_ext_get_actual_len(newext),
1537                                         nearex, len, nearex + 1, nearex + 2);
1538                         memmove(nearex + 2, nearex + 1, len);
1539                 }
1540                 path[depth].p_ext = nearex + 1;
1541         } else {
1542                 BUG_ON(newext->ee_block == nearex->ee_block);
1543                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1544                 len = len < 0 ? 0 : len;
1545                 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1546                                 "move %d from 0x%p to 0x%p\n",
1547                                 le32_to_cpu(newext->ee_block),
1548                                 ext_pblock(newext),
1549                                 ext4_ext_get_actual_len(newext),
1550                                 nearex, len, nearex + 1, nearex + 2);
1551                 memmove(nearex + 1, nearex, len);
1552                 path[depth].p_ext = nearex;
1553         }
1554
1555         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1556         nearex = path[depth].p_ext;
1557         nearex->ee_block = newext->ee_block;
1558         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1559         nearex->ee_len = newext->ee_len;
1560
1561 merge:
1562         /* try to merge extents to the right */
1563         ext4_ext_try_to_merge(inode, path, nearex);
1564
1565         /* try to merge extents to the left */
1566
1567         /* time to correct all indexes above */
1568         err = ext4_ext_correct_indexes(handle, inode, path);
1569         if (err)
1570                 goto cleanup;
1571
1572         err = ext4_ext_dirty(handle, inode, path + depth);
1573
1574 cleanup:
1575         if (npath) {
1576                 ext4_ext_drop_refs(npath);
1577                 kfree(npath);
1578         }
1579         ext4_ext_tree_changed(inode);
1580         ext4_ext_invalidate_cache(inode);
1581         return err;
1582 }
1583
1584 static void
1585 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1586                         __u32 len, ext4_fsblk_t start, int type)
1587 {
1588         struct ext4_ext_cache *cex;
1589         BUG_ON(len == 0);
1590         cex = &EXT4_I(inode)->i_cached_extent;
1591         cex->ec_type = type;
1592         cex->ec_block = block;
1593         cex->ec_len = len;
1594         cex->ec_start = start;
1595 }
1596
1597 /*
1598  * ext4_ext_put_gap_in_cache:
1599  * calculate boundaries of the gap that the requested block fits into
1600  * and cache this gap
1601  */
1602 static void
1603 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1604                                 ext4_lblk_t block)
1605 {
1606         int depth = ext_depth(inode);
1607         unsigned long len;
1608         ext4_lblk_t lblock;
1609         struct ext4_extent *ex;
1610
1611         ex = path[depth].p_ext;
1612         if (ex == NULL) {
1613                 /* there is no extent yet, so gap is [0;-] */
1614                 lblock = 0;
1615                 len = EXT_MAX_BLOCK;
1616                 ext_debug("cache gap(whole file):");
1617         } else if (block < le32_to_cpu(ex->ee_block)) {
1618                 lblock = block;
1619                 len = le32_to_cpu(ex->ee_block) - block;
1620                 ext_debug("cache gap(before): %u [%u:%u]",
1621                                 block,
1622                                 le32_to_cpu(ex->ee_block),
1623                                  ext4_ext_get_actual_len(ex));
1624         } else if (block >= le32_to_cpu(ex->ee_block)
1625                         + ext4_ext_get_actual_len(ex)) {
1626                 ext4_lblk_t next;
1627                 lblock = le32_to_cpu(ex->ee_block)
1628                         + ext4_ext_get_actual_len(ex);
1629
1630                 next = ext4_ext_next_allocated_block(path);
1631                 ext_debug("cache gap(after): [%u:%u] %u",
1632                                 le32_to_cpu(ex->ee_block),
1633                                 ext4_ext_get_actual_len(ex),
1634                                 block);
1635                 BUG_ON(next == lblock);
1636                 len = next - lblock;
1637         } else {
1638                 lblock = len = 0;
1639                 BUG();
1640         }
1641
1642         ext_debug(" -> %u:%lu\n", lblock, len);
1643         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1644 }
1645
1646 static int
1647 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1648                         struct ext4_extent *ex)
1649 {
1650         struct ext4_ext_cache *cex;
1651
1652         cex = &EXT4_I(inode)->i_cached_extent;
1653
1654         /* has cache valid data? */
1655         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1656                 return EXT4_EXT_CACHE_NO;
1657
1658         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1659                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1660         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1661                 ex->ee_block = cpu_to_le32(cex->ec_block);
1662                 ext4_ext_store_pblock(ex, cex->ec_start);
1663                 ex->ee_len = cpu_to_le16(cex->ec_len);
1664                 ext_debug("%u cached by %u:%u:%llu\n",
1665                                 block,
1666                                 cex->ec_block, cex->ec_len, cex->ec_start);
1667                 return cex->ec_type;
1668         }
1669
1670         /* not in cache */
1671         return EXT4_EXT_CACHE_NO;
1672 }
1673
1674 /*
1675  * ext4_ext_rm_idx:
1676  * removes index from the index block.
1677  * It's used in truncate case only, thus all requests are for
1678  * last index in the block only.
1679  */
1680 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1681                         struct ext4_ext_path *path)
1682 {
1683         struct buffer_head *bh;
1684         int err;
1685         ext4_fsblk_t leaf;
1686
1687         /* free index block */
1688         path--;
1689         leaf = idx_pblock(path->p_idx);
1690         BUG_ON(path->p_hdr->eh_entries == 0);
1691         err = ext4_ext_get_access(handle, inode, path);
1692         if (err)
1693                 return err;
1694         path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1695         err = ext4_ext_dirty(handle, inode, path);
1696         if (err)
1697                 return err;
1698         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1699         bh = sb_find_get_block(inode->i_sb, leaf);
1700         ext4_forget(handle, 1, inode, bh, leaf);
1701         ext4_free_blocks(handle, inode, leaf, 1);
1702         return err;
1703 }
1704
1705 /*
1706  * ext4_ext_calc_credits_for_insert:
1707  * This routine returns max. credits that the extent tree can consume.
1708  * It should be OK for low-performance paths like ->writepage()
1709  * To allow many writing processes to fit into a single transaction,
1710  * the caller should calculate credits under i_data_sem and
1711  * pass the actual path.
1712  */
1713 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1714                                                 struct ext4_ext_path *path)
1715 {
1716         int depth, needed;
1717
1718         if (path) {
1719                 /* probably there is space in leaf? */
1720                 depth = ext_depth(inode);
1721                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1722                                 < le16_to_cpu(path[depth].p_hdr->eh_max))
1723                         return 1;
1724         }
1725
1726         /*
1727          * given 32-bit logical block (4294967296 blocks), max. tree
1728          * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1729          * Let's also add one more level for imbalance.
1730          */
1731         depth = 5;
1732
1733         /* allocation of new data block(s) */
1734         needed = 2;
1735
1736         /*
1737          * tree can be full, so it would need to grow in depth:
1738          * we need one credit to modify old root, credits for
1739          * new root will be added in split accounting
1740          */
1741         needed += 1;
1742
1743         /*
1744          * Index split can happen, we would need:
1745          *    allocate intermediate indexes (bitmap + group)
1746          *  + change two blocks at each level, but root (already included)
1747          */
1748         needed += (depth * 2) + (depth * 2);
1749
1750         /* any allocation modifies superblock */
1751         needed += 1;
1752
1753         return needed;
1754 }
1755
1756 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1757                                 struct ext4_extent *ex,
1758                                 ext4_lblk_t from, ext4_lblk_t to)
1759 {
1760         struct buffer_head *bh;
1761         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
1762         int i;
1763
1764 #ifdef EXTENTS_STATS
1765         {
1766                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1767                 spin_lock(&sbi->s_ext_stats_lock);
1768                 sbi->s_ext_blocks += ee_len;
1769                 sbi->s_ext_extents++;
1770                 if (ee_len < sbi->s_ext_min)
1771                         sbi->s_ext_min = ee_len;
1772                 if (ee_len > sbi->s_ext_max)
1773                         sbi->s_ext_max = ee_len;
1774                 if (ext_depth(inode) > sbi->s_depth_max)
1775                         sbi->s_depth_max = ext_depth(inode);
1776                 spin_unlock(&sbi->s_ext_stats_lock);
1777         }
1778 #endif
1779         if (from >= le32_to_cpu(ex->ee_block)
1780             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
1781                 /* tail removal */
1782                 ext4_lblk_t num;
1783                 ext4_fsblk_t start;
1784
1785                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1786                 start = ext_pblock(ex) + ee_len - num;
1787                 ext_debug("free last %u blocks starting %llu\n", num, start);
1788                 for (i = 0; i < num; i++) {
1789                         bh = sb_find_get_block(inode->i_sb, start + i);
1790                         ext4_forget(handle, 0, inode, bh, start + i);
1791                 }
1792                 ext4_free_blocks(handle, inode, start, num);
1793         } else if (from == le32_to_cpu(ex->ee_block)
1794                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
1795                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
1796                         from, to, le32_to_cpu(ex->ee_block), ee_len);
1797         } else {
1798                 printk(KERN_INFO "strange request: removal(2) "
1799                                 "%u-%u from %u:%u\n",
1800                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
1801         }
1802         return 0;
1803 }
1804
1805 static int
1806 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1807                 struct ext4_ext_path *path, ext4_lblk_t start)
1808 {
1809         int err = 0, correct_index = 0;
1810         int depth = ext_depth(inode), credits;
1811         struct ext4_extent_header *eh;
1812         ext4_lblk_t a, b, block;
1813         unsigned num;
1814         ext4_lblk_t ex_ee_block;
1815         unsigned short ex_ee_len;
1816         unsigned uninitialized = 0;
1817         struct ext4_extent *ex;
1818
1819         /* the header must be checked already in ext4_ext_remove_space() */
1820         ext_debug("truncate since %u in leaf\n", start);
1821         if (!path[depth].p_hdr)
1822                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1823         eh = path[depth].p_hdr;
1824         BUG_ON(eh == NULL);
1825
1826         /* find where to start removing */
1827         ex = EXT_LAST_EXTENT(eh);
1828
1829         ex_ee_block = le32_to_cpu(ex->ee_block);
1830         if (ext4_ext_is_uninitialized(ex))
1831                 uninitialized = 1;
1832         ex_ee_len = ext4_ext_get_actual_len(ex);
1833
1834         while (ex >= EXT_FIRST_EXTENT(eh) &&
1835                         ex_ee_block + ex_ee_len > start) {
1836                 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1837                 path[depth].p_ext = ex;
1838
1839                 a = ex_ee_block > start ? ex_ee_block : start;
1840                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1841                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1842
1843                 ext_debug("  border %u:%u\n", a, b);
1844
1845                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1846                         block = 0;
1847                         num = 0;
1848                         BUG();
1849                 } else if (a != ex_ee_block) {
1850                         /* remove tail of the extent */
1851                         block = ex_ee_block;
1852                         num = a - block;
1853                 } else if (b != ex_ee_block + ex_ee_len - 1) {
1854                         /* remove head of the extent */
1855                         block = a;
1856                         num = b - a;
1857                         /* there is no "make a hole" API yet */
1858                         BUG();
1859                 } else {
1860                         /* remove whole extent: excellent! */
1861                         block = ex_ee_block;
1862                         num = 0;
1863                         BUG_ON(a != ex_ee_block);
1864                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1865                 }
1866
1867                 /* at present, extent can't cross block group: */
1868                 /* leaf + bitmap + group desc + sb + inode */
1869                 credits = 5;
1870                 if (ex == EXT_FIRST_EXTENT(eh)) {
1871                         correct_index = 1;
1872                         credits += (ext_depth(inode)) + 1;
1873                 }
1874 #ifdef CONFIG_QUOTA
1875                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1876 #endif
1877
1878                 handle = ext4_ext_journal_restart(handle, credits);
1879                 if (IS_ERR(handle)) {
1880                         err = PTR_ERR(handle);
1881                         goto out;
1882                 }
1883
1884                 err = ext4_ext_get_access(handle, inode, path + depth);
1885                 if (err)
1886                         goto out;
1887
1888                 err = ext4_remove_blocks(handle, inode, ex, a, b);
1889                 if (err)
1890                         goto out;
1891
1892                 if (num == 0) {
1893                         /* this extent is removed; mark slot entirely unused */
1894                         ext4_ext_store_pblock(ex, 0);
1895                         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1896                 }
1897
1898                 ex->ee_block = cpu_to_le32(block);
1899                 ex->ee_len = cpu_to_le16(num);
1900                 /*
1901                  * Do not mark uninitialized if all the blocks in the
1902                  * extent have been removed.
1903                  */
1904                 if (uninitialized && num)
1905                         ext4_ext_mark_uninitialized(ex);
1906
1907                 err = ext4_ext_dirty(handle, inode, path + depth);
1908                 if (err)
1909                         goto out;
1910
1911                 ext_debug("new extent: %u:%u:%llu\n", block, num,
1912                                 ext_pblock(ex));
1913                 ex--;
1914                 ex_ee_block = le32_to_cpu(ex->ee_block);
1915                 ex_ee_len = ext4_ext_get_actual_len(ex);
1916         }
1917
1918         if (correct_index && eh->eh_entries)
1919                 err = ext4_ext_correct_indexes(handle, inode, path);
1920
1921         /* if this leaf is free, then we should
1922          * remove it from index block above */
1923         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1924                 err = ext4_ext_rm_idx(handle, inode, path + depth);
1925
1926 out:
1927         return err;
1928 }
1929
1930 /*
1931  * ext4_ext_more_to_rm:
1932  * returns 1 if current index has to be freed (even partial)
1933  */
1934 static int
1935 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1936 {
1937         BUG_ON(path->p_idx == NULL);
1938
1939         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1940                 return 0;
1941
1942         /*
1943          * if truncate on deeper level happened, it wasn't partial,
1944          * so we have to consider current index for truncation
1945          */
1946         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1947                 return 0;
1948         return 1;
1949 }
1950
1951 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
1952 {
1953         struct super_block *sb = inode->i_sb;
1954         int depth = ext_depth(inode);
1955         struct ext4_ext_path *path;
1956         handle_t *handle;
1957         int i = 0, err = 0;
1958
1959         ext_debug("truncate since %u\n", start);
1960
1961         /* probably first extent we're gonna free will be last in block */
1962         handle = ext4_journal_start(inode, depth + 1);
1963         if (IS_ERR(handle))
1964                 return PTR_ERR(handle);
1965
1966         ext4_ext_invalidate_cache(inode);
1967
1968         /*
1969          * We start scanning from right side, freeing all the blocks
1970          * after i_size and walking into the tree depth-wise.
1971          */
1972         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1973         if (path == NULL) {
1974                 ext4_journal_stop(handle);
1975                 return -ENOMEM;
1976         }
1977         path[0].p_hdr = ext_inode_hdr(inode);
1978         if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
1979                 err = -EIO;
1980                 goto out;
1981         }
1982         path[0].p_depth = depth;
1983
1984         while (i >= 0 && err == 0) {
1985                 if (i == depth) {
1986                         /* this is leaf block */
1987                         err = ext4_ext_rm_leaf(handle, inode, path, start);
1988                         /* root level has p_bh == NULL, brelse() eats this */
1989                         brelse(path[i].p_bh);
1990                         path[i].p_bh = NULL;
1991                         i--;
1992                         continue;
1993                 }
1994
1995                 /* this is index block */
1996                 if (!path[i].p_hdr) {
1997                         ext_debug("initialize header\n");
1998                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1999                 }
2000
2001                 if (!path[i].p_idx) {
2002                         /* this level hasn't been touched yet */
2003                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2004                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2005                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2006                                   path[i].p_hdr,
2007                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2008                 } else {
2009                         /* we were already here, see at next index */
2010                         path[i].p_idx--;
2011                 }
2012
2013                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2014                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2015                                 path[i].p_idx);
2016                 if (ext4_ext_more_to_rm(path + i)) {
2017                         struct buffer_head *bh;
2018                         /* go to the next level */
2019                         ext_debug("move to level %d (block %llu)\n",
2020                                   i + 1, idx_pblock(path[i].p_idx));
2021                         memset(path + i + 1, 0, sizeof(*path));
2022                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2023                         if (!bh) {
2024                                 /* should we reset i_size? */
2025                                 err = -EIO;
2026                                 break;
2027                         }
2028                         if (WARN_ON(i + 1 > depth)) {
2029                                 err = -EIO;
2030                                 break;
2031                         }
2032                         if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2033                                                         depth - i - 1)) {
2034                                 err = -EIO;
2035                                 break;
2036                         }
2037                         path[i + 1].p_bh = bh;
2038
2039                         /* save actual number of indexes since this
2040                          * number is changed at the next iteration */
2041                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2042                         i++;
2043                 } else {
2044                         /* we finished processing this index, go up */
2045                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2046                                 /* index is empty, remove it;
2047                                  * handle must be already prepared by the
2048                                  * truncatei_leaf() */
2049                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2050                         }
2051                         /* root level has p_bh == NULL, brelse() eats this */
2052                         brelse(path[i].p_bh);
2053                         path[i].p_bh = NULL;
2054                         i--;
2055                         ext_debug("return to level %d\n", i);
2056                 }
2057         }
2058
2059         /* TODO: flexible tree reduction should be here */
2060         if (path->p_hdr->eh_entries == 0) {
2061                 /*
2062                  * truncate to zero freed all the tree,
2063                  * so we need to correct eh_depth
2064                  */
2065                 err = ext4_ext_get_access(handle, inode, path);
2066                 if (err == 0) {
2067                         ext_inode_hdr(inode)->eh_depth = 0;
2068                         ext_inode_hdr(inode)->eh_max =
2069                                 cpu_to_le16(ext4_ext_space_root(inode));
2070                         err = ext4_ext_dirty(handle, inode, path);
2071                 }
2072         }
2073 out:
2074         ext4_ext_tree_changed(inode);
2075         ext4_ext_drop_refs(path);
2076         kfree(path);
2077         ext4_journal_stop(handle);
2078
2079         return err;
2080 }
2081
2082 /*
2083  * called at mount time
2084  */
2085 void ext4_ext_init(struct super_block *sb)
2086 {
2087         /*
2088          * possible initialization would be here
2089          */
2090
2091         if (test_opt(sb, EXTENTS)) {
2092                 printk("EXT4-fs: file extents enabled");
2093 #ifdef AGGRESSIVE_TEST
2094                 printk(", aggressive tests");
2095 #endif
2096 #ifdef CHECK_BINSEARCH
2097                 printk(", check binsearch");
2098 #endif
2099 #ifdef EXTENTS_STATS
2100                 printk(", stats");
2101 #endif
2102                 printk("\n");
2103 #ifdef EXTENTS_STATS
2104                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2105                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2106                 EXT4_SB(sb)->s_ext_max = 0;
2107 #endif
2108         }
2109 }
2110
2111 /*
2112  * called at umount time
2113  */
2114 void ext4_ext_release(struct super_block *sb)
2115 {
2116         if (!test_opt(sb, EXTENTS))
2117                 return;
2118
2119 #ifdef EXTENTS_STATS
2120         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2121                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2122                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2123                         sbi->s_ext_blocks, sbi->s_ext_extents,
2124                         sbi->s_ext_blocks / sbi->s_ext_extents);
2125                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2126                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2127         }
2128 #endif
2129 }
2130
2131 /*
2132  * This function is called by ext4_ext_get_blocks() if someone tries to write
2133  * to an uninitialized extent. It may result in splitting the uninitialized
2134  * extent into multiple extents (upto three - one initialized and two
2135  * uninitialized).
2136  * There are three possibilities:
2137  *   a> There is no split required: Entire extent should be initialized
2138  *   b> Splits in two extents: Write is happening at either end of the extent
2139  *   c> Splits in three extents: Somone is writing in middle of the extent
2140  */
2141 static int ext4_ext_convert_to_initialized(handle_t *handle,
2142                                                 struct inode *inode,
2143                                                 struct ext4_ext_path *path,
2144                                                 ext4_lblk_t iblock,
2145                                                 unsigned long max_blocks)
2146 {
2147         struct ext4_extent *ex, newex;
2148         struct ext4_extent *ex1 = NULL;
2149         struct ext4_extent *ex2 = NULL;
2150         struct ext4_extent *ex3 = NULL;
2151         struct ext4_extent_header *eh;
2152         ext4_lblk_t ee_block;
2153         unsigned int allocated, ee_len, depth;
2154         ext4_fsblk_t newblock;
2155         int err = 0;
2156         int ret = 0;
2157
2158         depth = ext_depth(inode);
2159         eh = path[depth].p_hdr;
2160         ex = path[depth].p_ext;
2161         ee_block = le32_to_cpu(ex->ee_block);
2162         ee_len = ext4_ext_get_actual_len(ex);
2163         allocated = ee_len - (iblock - ee_block);
2164         newblock = iblock - ee_block + ext_pblock(ex);
2165         ex2 = ex;
2166
2167         /* ex1: ee_block to iblock - 1 : uninitialized */
2168         if (iblock > ee_block) {
2169                 ex1 = ex;
2170                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2171                 ext4_ext_mark_uninitialized(ex1);
2172                 ex2 = &newex;
2173         }
2174         /*
2175          * for sanity, update the length of the ex2 extent before
2176          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2177          * overlap of blocks.
2178          */
2179         if (!ex1 && allocated > max_blocks)
2180                 ex2->ee_len = cpu_to_le16(max_blocks);
2181         /* ex3: to ee_block + ee_len : uninitialised */
2182         if (allocated > max_blocks) {
2183                 unsigned int newdepth;
2184                 ex3 = &newex;
2185                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2186                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2187                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2188                 ext4_ext_mark_uninitialized(ex3);
2189                 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2190                 if (err)
2191                         goto out;
2192                 /*
2193                  * The depth, and hence eh & ex might change
2194                  * as part of the insert above.
2195                  */
2196                 newdepth = ext_depth(inode);
2197                 if (newdepth != depth) {
2198                         depth = newdepth;
2199                         path = ext4_ext_find_extent(inode, iblock, NULL);
2200                         if (IS_ERR(path)) {
2201                                 err = PTR_ERR(path);
2202                                 path = NULL;
2203                                 goto out;
2204                         }
2205                         eh = path[depth].p_hdr;
2206                         ex = path[depth].p_ext;
2207                         if (ex2 != &newex)
2208                                 ex2 = ex;
2209                 }
2210                 allocated = max_blocks;
2211         }
2212         /*
2213          * If there was a change of depth as part of the
2214          * insertion of ex3 above, we need to update the length
2215          * of the ex1 extent again here
2216          */
2217         if (ex1 && ex1 != ex) {
2218                 ex1 = ex;
2219                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2220                 ext4_ext_mark_uninitialized(ex1);
2221                 ex2 = &newex;
2222         }
2223         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2224         ex2->ee_block = cpu_to_le32(iblock);
2225         ext4_ext_store_pblock(ex2, newblock);
2226         ex2->ee_len = cpu_to_le16(allocated);
2227         if (ex2 != ex)
2228                 goto insert;
2229         err = ext4_ext_get_access(handle, inode, path + depth);
2230         if (err)
2231                 goto out;
2232         /*
2233          * New (initialized) extent starts from the first block
2234          * in the current extent. i.e., ex2 == ex
2235          * We have to see if it can be merged with the extent
2236          * on the left.
2237          */
2238         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2239                 /*
2240                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2241                  * since it merges towards right _only_.
2242                  */
2243                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2244                 if (ret) {
2245                         err = ext4_ext_correct_indexes(handle, inode, path);
2246                         if (err)
2247                                 goto out;
2248                         depth = ext_depth(inode);
2249                         ex2--;
2250                 }
2251         }
2252         /*
2253          * Try to Merge towards right. This might be required
2254          * only when the whole extent is being written to.
2255          * i.e. ex2 == ex and ex3 == NULL.
2256          */
2257         if (!ex3) {
2258                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2259                 if (ret) {
2260                         err = ext4_ext_correct_indexes(handle, inode, path);
2261                         if (err)
2262                                 goto out;
2263                 }
2264         }
2265         /* Mark modified extent as dirty */
2266         err = ext4_ext_dirty(handle, inode, path + depth);
2267         goto out;
2268 insert:
2269         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2270 out:
2271         return err ? err : allocated;
2272 }
2273
2274 /*
2275  * Need to be called with
2276  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2277  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2278  */
2279 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2280                         ext4_lblk_t iblock,
2281                         unsigned long max_blocks, struct buffer_head *bh_result,
2282                         int create, int extend_disksize)
2283 {
2284         struct ext4_ext_path *path = NULL;
2285         struct ext4_extent_header *eh;
2286         struct ext4_extent newex, *ex;
2287         ext4_fsblk_t goal, newblock;
2288         int err = 0, depth, ret;
2289         unsigned long allocated = 0;
2290
2291         __clear_bit(BH_New, &bh_result->b_state);
2292         ext_debug("blocks %u/%lu requested for inode %u\n",
2293                         iblock, max_blocks, inode->i_ino);
2294
2295         /* check in cache */
2296         goal = ext4_ext_in_cache(inode, iblock, &newex);
2297         if (goal) {
2298                 if (goal == EXT4_EXT_CACHE_GAP) {
2299                         if (!create) {
2300                                 /*
2301                                  * block isn't allocated yet and
2302                                  * user doesn't want to allocate it
2303                                  */
2304                                 goto out2;
2305                         }
2306                         /* we should allocate requested block */
2307                 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2308                         /* block is already allocated */
2309                         newblock = iblock
2310                                    - le32_to_cpu(newex.ee_block)
2311                                    + ext_pblock(&newex);
2312                         /* number of remaining blocks in the extent */
2313                         allocated = le16_to_cpu(newex.ee_len) -
2314                                         (iblock - le32_to_cpu(newex.ee_block));
2315                         goto out;
2316                 } else {
2317                         BUG();
2318                 }
2319         }
2320
2321         /* find extent for this block */
2322         path = ext4_ext_find_extent(inode, iblock, NULL);
2323         if (IS_ERR(path)) {
2324                 err = PTR_ERR(path);
2325                 path = NULL;
2326                 goto out2;
2327         }
2328
2329         depth = ext_depth(inode);
2330
2331         /*
2332          * consistent leaf must not be empty;
2333          * this situation is possible, though, _during_ tree modification;
2334          * this is why assert can't be put in ext4_ext_find_extent()
2335          */
2336         BUG_ON(path[depth].p_ext == NULL && depth != 0);
2337         eh = path[depth].p_hdr;
2338
2339         ex = path[depth].p_ext;
2340         if (ex) {
2341                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2342                 ext4_fsblk_t ee_start = ext_pblock(ex);
2343                 unsigned short ee_len;
2344
2345                 /*
2346                  * Uninitialized extents are treated as holes, except that
2347                  * we split out initialized portions during a write.
2348                  */
2349                 ee_len = ext4_ext_get_actual_len(ex);
2350                 /* if found extent covers block, simply return it */
2351                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2352                         newblock = iblock - ee_block + ee_start;
2353                         /* number of remaining blocks in the extent */
2354                         allocated = ee_len - (iblock - ee_block);
2355                         ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2356                                         ee_block, ee_len, newblock);
2357
2358                         /* Do not put uninitialized extent in the cache */
2359                         if (!ext4_ext_is_uninitialized(ex)) {
2360                                 ext4_ext_put_in_cache(inode, ee_block,
2361                                                         ee_len, ee_start,
2362                                                         EXT4_EXT_CACHE_EXTENT);
2363                                 goto out;
2364                         }
2365                         if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2366                                 goto out;
2367                         if (!create)
2368                                 goto out2;
2369
2370                         ret = ext4_ext_convert_to_initialized(handle, inode,
2371                                                                 path, iblock,
2372                                                                 max_blocks);
2373                         if (ret <= 0)
2374                                 goto out2;
2375                         else
2376                                 allocated = ret;
2377                         goto outnew;
2378                 }
2379         }
2380
2381         /*
2382          * requested block isn't allocated yet;
2383          * we couldn't try to create block if create flag is zero
2384          */
2385         if (!create) {
2386                 /*
2387                  * put just found gap into cache to speed up
2388                  * subsequent requests
2389                  */
2390                 ext4_ext_put_gap_in_cache(inode, path, iblock);
2391                 goto out2;
2392         }
2393         /*
2394          * Okay, we need to do block allocation.  Lazily initialize the block
2395          * allocation info here if necessary.
2396          */
2397         if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2398                 ext4_init_block_alloc_info(inode);
2399
2400         /* allocate new block */
2401         goal = ext4_ext_find_goal(inode, path, iblock);
2402
2403         /*
2404          * See if request is beyond maximum number of blocks we can have in
2405          * a single extent. For an initialized extent this limit is
2406          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2407          * EXT_UNINIT_MAX_LEN.
2408          */
2409         if (max_blocks > EXT_INIT_MAX_LEN &&
2410             create != EXT4_CREATE_UNINITIALIZED_EXT)
2411                 max_blocks = EXT_INIT_MAX_LEN;
2412         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2413                  create == EXT4_CREATE_UNINITIALIZED_EXT)
2414                 max_blocks = EXT_UNINIT_MAX_LEN;
2415
2416         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2417         newex.ee_block = cpu_to_le32(iblock);
2418         newex.ee_len = cpu_to_le16(max_blocks);
2419         err = ext4_ext_check_overlap(inode, &newex, path);
2420         if (err)
2421                 allocated = le16_to_cpu(newex.ee_len);
2422         else
2423                 allocated = max_blocks;
2424         newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2425         if (!newblock)
2426                 goto out2;
2427         ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2428                         goal, newblock, allocated);
2429
2430         /* try to insert new extent into found leaf and return */
2431         ext4_ext_store_pblock(&newex, newblock);
2432         newex.ee_len = cpu_to_le16(allocated);
2433         if (create == EXT4_CREATE_UNINITIALIZED_EXT)  /* Mark uninitialized */
2434                 ext4_ext_mark_uninitialized(&newex);
2435         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2436         if (err) {
2437                 /* free data blocks we just allocated */
2438                 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2439                                         le16_to_cpu(newex.ee_len));
2440                 goto out2;
2441         }
2442
2443         if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2444                 EXT4_I(inode)->i_disksize = inode->i_size;
2445
2446         /* previous routine could use block we allocated */
2447         newblock = ext_pblock(&newex);
2448 outnew:
2449         __set_bit(BH_New, &bh_result->b_state);
2450
2451         /* Cache only when it is _not_ an uninitialized extent */
2452         if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2453                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2454                                                 EXT4_EXT_CACHE_EXTENT);
2455 out:
2456         if (allocated > max_blocks)
2457                 allocated = max_blocks;
2458         ext4_ext_show_leaf(inode, path);
2459         __set_bit(BH_Mapped, &bh_result->b_state);
2460         bh_result->b_bdev = inode->i_sb->s_bdev;
2461         bh_result->b_blocknr = newblock;
2462 out2:
2463         if (path) {
2464                 ext4_ext_drop_refs(path);
2465                 kfree(path);
2466         }
2467         return err ? err : allocated;
2468 }
2469
2470 void ext4_ext_truncate(struct inode * inode, struct page *page)
2471 {
2472         struct address_space *mapping = inode->i_mapping;
2473         struct super_block *sb = inode->i_sb;
2474         ext4_lblk_t last_block;
2475         handle_t *handle;
2476         int err = 0;
2477
2478         /*
2479          * probably first extent we're gonna free will be last in block
2480          */
2481         err = ext4_writepage_trans_blocks(inode) + 3;
2482         handle = ext4_journal_start(inode, err);
2483         if (IS_ERR(handle)) {
2484                 if (page) {
2485                         clear_highpage(page);
2486                         flush_dcache_page(page);
2487                         unlock_page(page);
2488                         page_cache_release(page);
2489                 }
2490                 return;
2491         }
2492
2493         if (page)
2494                 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2495
2496         down_write(&EXT4_I(inode)->i_data_sem);
2497         ext4_ext_invalidate_cache(inode);
2498
2499         /*
2500          * TODO: optimization is possible here.
2501          * Probably we need not scan at all,
2502          * because page truncation is enough.
2503          */
2504         if (ext4_orphan_add(handle, inode))
2505                 goto out_stop;
2506
2507         /* we have to know where to truncate from in crash case */
2508         EXT4_I(inode)->i_disksize = inode->i_size;
2509         ext4_mark_inode_dirty(handle, inode);
2510
2511         last_block = (inode->i_size + sb->s_blocksize - 1)
2512                         >> EXT4_BLOCK_SIZE_BITS(sb);
2513         err = ext4_ext_remove_space(inode, last_block);
2514
2515         /* In a multi-transaction truncate, we only make the final
2516          * transaction synchronous.
2517          */
2518         if (IS_SYNC(inode))
2519                 handle->h_sync = 1;
2520
2521 out_stop:
2522         /*
2523          * If this was a simple ftruncate() and the file will remain alive,
2524          * then we need to clear up the orphan record which we created above.
2525          * However, if this was a real unlink then we were called by
2526          * ext4_delete_inode(), and we allow that function to clean up the
2527          * orphan info for us.
2528          */
2529         if (inode->i_nlink)
2530                 ext4_orphan_del(handle, inode);
2531
2532         up_write(&EXT4_I(inode)->i_data_sem);
2533         ext4_journal_stop(handle);
2534 }
2535
2536 /*
2537  * ext4_ext_writepage_trans_blocks:
2538  * calculate max number of blocks we could modify
2539  * in order to allocate new block for an inode
2540  */
2541 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2542 {
2543         int needed;
2544
2545         needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2546
2547         /* caller wants to allocate num blocks, but note it includes sb */
2548         needed = needed * num - (num - 1);
2549
2550 #ifdef CONFIG_QUOTA
2551         needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2552 #endif
2553
2554         return needed;
2555 }
2556
2557 /*
2558  * preallocate space for a file. This implements ext4's fallocate inode
2559  * operation, which gets called from sys_fallocate system call.
2560  * For block-mapped files, posix_fallocate should fall back to the method
2561  * of writing zeroes to the required new blocks (the same behavior which is
2562  * expected for file systems which do not support fallocate() system call).
2563  */
2564 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2565 {
2566         handle_t *handle;
2567         ext4_lblk_t block;
2568         unsigned long max_blocks;
2569         ext4_fsblk_t nblocks = 0;
2570         int ret = 0;
2571         int ret2 = 0;
2572         int retries = 0;
2573         struct buffer_head map_bh;
2574         unsigned int credits, blkbits = inode->i_blkbits;
2575
2576         /*
2577          * currently supporting (pre)allocate mode for extent-based
2578          * files _only_
2579          */
2580         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2581                 return -EOPNOTSUPP;
2582
2583         /* preallocation to directories is currently not supported */
2584         if (S_ISDIR(inode->i_mode))
2585                 return -ENODEV;
2586
2587         block = offset >> blkbits;
2588         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2589                         - block;
2590
2591         /*
2592          * credits to insert 1 extent into extent tree + buffers to be able to
2593          * modify 1 super block, 1 block bitmap and 1 group descriptor.
2594          */
2595         credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
2596         down_write((&EXT4_I(inode)->i_data_sem));
2597 retry:
2598         while (ret >= 0 && ret < max_blocks) {
2599                 block = block + ret;
2600                 max_blocks = max_blocks - ret;
2601                 handle = ext4_journal_start(inode, credits);
2602                 if (IS_ERR(handle)) {
2603                         ret = PTR_ERR(handle);
2604                         break;
2605                 }
2606
2607                 ret = ext4_ext_get_blocks(handle, inode, block,
2608                                           max_blocks, &map_bh,
2609                                           EXT4_CREATE_UNINITIALIZED_EXT, 0);
2610                 WARN_ON(ret <= 0);
2611                 if (ret <= 0) {
2612                         ext4_error(inode->i_sb, "ext4_fallocate",
2613                                     "ext4_ext_get_blocks returned error: "
2614                                     "inode#%lu, block=%u, max_blocks=%lu",
2615                                     inode->i_ino, block, max_blocks);
2616                         ret = -EIO;
2617                         ext4_mark_inode_dirty(handle, inode);
2618                         ret2 = ext4_journal_stop(handle);
2619                         break;
2620                 }
2621                 if (ret > 0) {
2622                         /* check wrap through sign-bit/zero here */
2623                         if ((block + ret) < 0 || (block + ret) < block) {
2624                                 ret = -EIO;
2625                                 ext4_mark_inode_dirty(handle, inode);
2626                                 ret2 = ext4_journal_stop(handle);
2627                                 break;
2628                         }
2629                         if (buffer_new(&map_bh) && ((block + ret) >
2630                             (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits)
2631                             >> blkbits)))
2632                                         nblocks = nblocks + ret;
2633                 }
2634
2635                 /* Update ctime if new blocks get allocated */
2636                 if (nblocks) {
2637                         struct timespec now;
2638
2639                         now = current_fs_time(inode->i_sb);
2640                         if (!timespec_equal(&inode->i_ctime, &now))
2641                                 inode->i_ctime = now;
2642                 }
2643
2644                 ext4_mark_inode_dirty(handle, inode);
2645                 ret2 = ext4_journal_stop(handle);
2646                 if (ret2)
2647                         break;
2648         }
2649
2650         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2651                 goto retry;
2652
2653         up_write((&EXT4_I(inode)->i_data_sem));
2654         /*
2655          * Time to update the file size.
2656          * Update only when preallocation was requested beyond the file size.
2657          */
2658         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2659             (offset + len) > i_size_read(inode)) {
2660                 if (ret > 0) {
2661                         /*
2662                          * if no error, we assume preallocation succeeded
2663                          * completely
2664                          */
2665                         mutex_lock(&inode->i_mutex);
2666                         i_size_write(inode, offset + len);
2667                         EXT4_I(inode)->i_disksize = i_size_read(inode);
2668                         mutex_unlock(&inode->i_mutex);
2669                 } else if (ret < 0 && nblocks) {
2670                         /* Handle partial allocation scenario */
2671                         loff_t newsize;
2672
2673                         mutex_lock(&inode->i_mutex);
2674                         newsize  = (nblocks << blkbits) + i_size_read(inode);
2675                         i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits));
2676                         EXT4_I(inode)->i_disksize = i_size_read(inode);
2677                         mutex_unlock(&inode->i_mutex);
2678                 }
2679         }
2680
2681         return ret > 0 ? ret2 : ret;
2682 }