Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
[linux-2.6.git] / fs / ext4 / extents.c
1 /*
2  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3  * Written by Alex Tomas <alex@clusterfs.com>
4  *
5  * Architecture independence:
6  *   Copyright (c) 2005, Bull S.A.
7  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public Licens
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
21  */
22
23 /*
24  * Extents support for EXT4
25  *
26  * TODO:
27  *   - ext4*_error() should be used in some situations
28  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29  *   - smart tree reduction
30  */
31
32 #include <linux/module.h>
33 #include <linux/fs.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
46
47
48 /*
49  * ext_pblock:
50  * combine low and high parts of physical block number into ext4_fsblk_t
51  */
52 ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
53 {
54         ext4_fsblk_t block;
55
56         block = le32_to_cpu(ex->ee_start_lo);
57         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
58         return block;
59 }
60
61 /*
62  * idx_pblock:
63  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64  */
65 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
66 {
67         ext4_fsblk_t block;
68
69         block = le32_to_cpu(ix->ei_leaf_lo);
70         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
71         return block;
72 }
73
74 /*
75  * ext4_ext_store_pblock:
76  * stores a large physical block number into an extent struct,
77  * breaking it into parts
78  */
79 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
80 {
81         ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
82         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
83 }
84
85 /*
86  * ext4_idx_store_pblock:
87  * stores a large physical block number into an index struct,
88  * breaking it into parts
89  */
90 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
91 {
92         ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
93         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
94 }
95
96 static int ext4_ext_truncate_extend_restart(handle_t *handle,
97                                             struct inode *inode,
98                                             int needed)
99 {
100         int err;
101
102         if (!ext4_handle_valid(handle))
103                 return 0;
104         if (handle->h_buffer_credits > needed)
105                 return 0;
106         err = ext4_journal_extend(handle, needed);
107         if (err <= 0)
108                 return err;
109         err = ext4_truncate_restart_trans(handle, inode, needed);
110         /*
111          * We have dropped i_data_sem so someone might have cached again
112          * an extent we are going to truncate.
113          */
114         ext4_ext_invalidate_cache(inode);
115
116         return err;
117 }
118
119 /*
120  * could return:
121  *  - EROFS
122  *  - ENOMEM
123  */
124 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
125                                 struct ext4_ext_path *path)
126 {
127         if (path->p_bh) {
128                 /* path points to block */
129                 return ext4_journal_get_write_access(handle, path->p_bh);
130         }
131         /* path points to leaf/index in inode body */
132         /* we use in-core data, no need to protect them */
133         return 0;
134 }
135
136 /*
137  * could return:
138  *  - EROFS
139  *  - ENOMEM
140  *  - EIO
141  */
142 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
143                                 struct ext4_ext_path *path)
144 {
145         int err;
146         if (path->p_bh) {
147                 /* path points to block */
148                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
149         } else {
150                 /* path points to leaf/index in inode body */
151                 err = ext4_mark_inode_dirty(handle, inode);
152         }
153         return err;
154 }
155
156 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
157                               struct ext4_ext_path *path,
158                               ext4_lblk_t block)
159 {
160         struct ext4_inode_info *ei = EXT4_I(inode);
161         ext4_fsblk_t bg_start;
162         ext4_fsblk_t last_block;
163         ext4_grpblk_t colour;
164         ext4_group_t block_group;
165         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
166         int depth;
167
168         if (path) {
169                 struct ext4_extent *ex;
170                 depth = path->p_depth;
171
172                 /* try to predict block placement */
173                 ex = path[depth].p_ext;
174                 if (ex)
175                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
176
177                 /* it looks like index is empty;
178                  * try to find starting block from index itself */
179                 if (path[depth].p_bh)
180                         return path[depth].p_bh->b_blocknr;
181         }
182
183         /* OK. use inode's group */
184         block_group = ei->i_block_group;
185         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
186                 /*
187                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
188                  * block groups per flexgroup, reserve the first block 
189                  * group for directories and special files.  Regular 
190                  * files will start at the second block group.  This
191                  * tends to speed up directory access and improves 
192                  * fsck times.
193                  */
194                 block_group &= ~(flex_size-1);
195                 if (S_ISREG(inode->i_mode))
196                         block_group++;
197         }
198         bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
199                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
200         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
201
202         /*
203          * If we are doing delayed allocation, we don't need take
204          * colour into account.
205          */
206         if (test_opt(inode->i_sb, DELALLOC))
207                 return bg_start;
208
209         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
210                 colour = (current->pid % 16) *
211                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
212         else
213                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
214         return bg_start + colour + block;
215 }
216
217 /*
218  * Allocation for a meta data block
219  */
220 static ext4_fsblk_t
221 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
222                         struct ext4_ext_path *path,
223                         struct ext4_extent *ex, int *err)
224 {
225         ext4_fsblk_t goal, newblock;
226
227         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
228         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
229         return newblock;
230 }
231
232 static inline int ext4_ext_space_block(struct inode *inode, int check)
233 {
234         int size;
235
236         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
237                         / sizeof(struct ext4_extent);
238         if (!check) {
239 #ifdef AGGRESSIVE_TEST
240                 if (size > 6)
241                         size = 6;
242 #endif
243         }
244         return size;
245 }
246
247 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
248 {
249         int size;
250
251         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
252                         / sizeof(struct ext4_extent_idx);
253         if (!check) {
254 #ifdef AGGRESSIVE_TEST
255                 if (size > 5)
256                         size = 5;
257 #endif
258         }
259         return size;
260 }
261
262 static inline int ext4_ext_space_root(struct inode *inode, int check)
263 {
264         int size;
265
266         size = sizeof(EXT4_I(inode)->i_data);
267         size -= sizeof(struct ext4_extent_header);
268         size /= sizeof(struct ext4_extent);
269         if (!check) {
270 #ifdef AGGRESSIVE_TEST
271                 if (size > 3)
272                         size = 3;
273 #endif
274         }
275         return size;
276 }
277
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
279 {
280         int size;
281
282         size = sizeof(EXT4_I(inode)->i_data);
283         size -= sizeof(struct ext4_extent_header);
284         size /= sizeof(struct ext4_extent_idx);
285         if (!check) {
286 #ifdef AGGRESSIVE_TEST
287                 if (size > 4)
288                         size = 4;
289 #endif
290         }
291         return size;
292 }
293
294 /*
295  * Calculate the number of metadata blocks needed
296  * to allocate @blocks
297  * Worse case is one block per extent
298  */
299 int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
300 {
301         int lcap, icap, rcap, leafs, idxs, num;
302         int newextents = blocks;
303
304         rcap = ext4_ext_space_root_idx(inode, 0);
305         lcap = ext4_ext_space_block(inode, 0);
306         icap = ext4_ext_space_block_idx(inode, 0);
307
308         /* number of new leaf blocks needed */
309         num = leafs = (newextents + lcap - 1) / lcap;
310
311         /*
312          * Worse case, we need separate index block(s)
313          * to link all new leaf blocks
314          */
315         idxs = (leafs + icap - 1) / icap;
316         do {
317                 num += idxs;
318                 idxs = (idxs + icap - 1) / icap;
319         } while (idxs > rcap);
320
321         return num;
322 }
323
324 static int
325 ext4_ext_max_entries(struct inode *inode, int depth)
326 {
327         int max;
328
329         if (depth == ext_depth(inode)) {
330                 if (depth == 0)
331                         max = ext4_ext_space_root(inode, 1);
332                 else
333                         max = ext4_ext_space_root_idx(inode, 1);
334         } else {
335                 if (depth == 0)
336                         max = ext4_ext_space_block(inode, 1);
337                 else
338                         max = ext4_ext_space_block_idx(inode, 1);
339         }
340
341         return max;
342 }
343
344 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
345 {
346         ext4_fsblk_t block = ext_pblock(ext);
347         int len = ext4_ext_get_actual_len(ext);
348
349         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
350 }
351
352 static int ext4_valid_extent_idx(struct inode *inode,
353                                 struct ext4_extent_idx *ext_idx)
354 {
355         ext4_fsblk_t block = idx_pblock(ext_idx);
356
357         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
358 }
359
360 static int ext4_valid_extent_entries(struct inode *inode,
361                                 struct ext4_extent_header *eh,
362                                 int depth)
363 {
364         struct ext4_extent *ext;
365         struct ext4_extent_idx *ext_idx;
366         unsigned short entries;
367         if (eh->eh_entries == 0)
368                 return 1;
369
370         entries = le16_to_cpu(eh->eh_entries);
371
372         if (depth == 0) {
373                 /* leaf entries */
374                 ext = EXT_FIRST_EXTENT(eh);
375                 while (entries) {
376                         if (!ext4_valid_extent(inode, ext))
377                                 return 0;
378                         ext++;
379                         entries--;
380                 }
381         } else {
382                 ext_idx = EXT_FIRST_INDEX(eh);
383                 while (entries) {
384                         if (!ext4_valid_extent_idx(inode, ext_idx))
385                                 return 0;
386                         ext_idx++;
387                         entries--;
388                 }
389         }
390         return 1;
391 }
392
393 static int __ext4_ext_check(const char *function, struct inode *inode,
394                                         struct ext4_extent_header *eh,
395                                         int depth)
396 {
397         const char *error_msg;
398         int max = 0;
399
400         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
401                 error_msg = "invalid magic";
402                 goto corrupted;
403         }
404         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
405                 error_msg = "unexpected eh_depth";
406                 goto corrupted;
407         }
408         if (unlikely(eh->eh_max == 0)) {
409                 error_msg = "invalid eh_max";
410                 goto corrupted;
411         }
412         max = ext4_ext_max_entries(inode, depth);
413         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
414                 error_msg = "too large eh_max";
415                 goto corrupted;
416         }
417         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
418                 error_msg = "invalid eh_entries";
419                 goto corrupted;
420         }
421         if (!ext4_valid_extent_entries(inode, eh, depth)) {
422                 error_msg = "invalid extent entries";
423                 goto corrupted;
424         }
425         return 0;
426
427 corrupted:
428         ext4_error(inode->i_sb, function,
429                         "bad header/extent in inode #%lu: %s - magic %x, "
430                         "entries %u, max %u(%u), depth %u(%u)",
431                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
432                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
433                         max, le16_to_cpu(eh->eh_depth), depth);
434
435         return -EIO;
436 }
437
438 #define ext4_ext_check(inode, eh, depth)        \
439         __ext4_ext_check(__func__, inode, eh, depth)
440
441 int ext4_ext_check_inode(struct inode *inode)
442 {
443         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
444 }
445
446 #ifdef EXT_DEBUG
447 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
448 {
449         int k, l = path->p_depth;
450
451         ext_debug("path:");
452         for (k = 0; k <= l; k++, path++) {
453                 if (path->p_idx) {
454                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
455                             idx_pblock(path->p_idx));
456                 } else if (path->p_ext) {
457                         ext_debug("  %d:[%d]%d:%llu ",
458                                   le32_to_cpu(path->p_ext->ee_block),
459                                   ext4_ext_is_uninitialized(path->p_ext),
460                                   ext4_ext_get_actual_len(path->p_ext),
461                                   ext_pblock(path->p_ext));
462                 } else
463                         ext_debug("  []");
464         }
465         ext_debug("\n");
466 }
467
468 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
469 {
470         int depth = ext_depth(inode);
471         struct ext4_extent_header *eh;
472         struct ext4_extent *ex;
473         int i;
474
475         if (!path)
476                 return;
477
478         eh = path[depth].p_hdr;
479         ex = EXT_FIRST_EXTENT(eh);
480
481         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
482
483         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
484                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
485                           ext4_ext_is_uninitialized(ex),
486                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
487         }
488         ext_debug("\n");
489 }
490 #else
491 #define ext4_ext_show_path(inode, path)
492 #define ext4_ext_show_leaf(inode, path)
493 #endif
494
495 void ext4_ext_drop_refs(struct ext4_ext_path *path)
496 {
497         int depth = path->p_depth;
498         int i;
499
500         for (i = 0; i <= depth; i++, path++)
501                 if (path->p_bh) {
502                         brelse(path->p_bh);
503                         path->p_bh = NULL;
504                 }
505 }
506
507 /*
508  * ext4_ext_binsearch_idx:
509  * binary search for the closest index of the given block
510  * the header must be checked before calling this
511  */
512 static void
513 ext4_ext_binsearch_idx(struct inode *inode,
514                         struct ext4_ext_path *path, ext4_lblk_t block)
515 {
516         struct ext4_extent_header *eh = path->p_hdr;
517         struct ext4_extent_idx *r, *l, *m;
518
519
520         ext_debug("binsearch for %u(idx):  ", block);
521
522         l = EXT_FIRST_INDEX(eh) + 1;
523         r = EXT_LAST_INDEX(eh);
524         while (l <= r) {
525                 m = l + (r - l) / 2;
526                 if (block < le32_to_cpu(m->ei_block))
527                         r = m - 1;
528                 else
529                         l = m + 1;
530                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
531                                 m, le32_to_cpu(m->ei_block),
532                                 r, le32_to_cpu(r->ei_block));
533         }
534
535         path->p_idx = l - 1;
536         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
537                   idx_pblock(path->p_idx));
538
539 #ifdef CHECK_BINSEARCH
540         {
541                 struct ext4_extent_idx *chix, *ix;
542                 int k;
543
544                 chix = ix = EXT_FIRST_INDEX(eh);
545                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
546                   if (k != 0 &&
547                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
548                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
549                                        "first=0x%p\n", k,
550                                        ix, EXT_FIRST_INDEX(eh));
551                                 printk(KERN_DEBUG "%u <= %u\n",
552                                        le32_to_cpu(ix->ei_block),
553                                        le32_to_cpu(ix[-1].ei_block));
554                         }
555                         BUG_ON(k && le32_to_cpu(ix->ei_block)
556                                            <= le32_to_cpu(ix[-1].ei_block));
557                         if (block < le32_to_cpu(ix->ei_block))
558                                 break;
559                         chix = ix;
560                 }
561                 BUG_ON(chix != path->p_idx);
562         }
563 #endif
564
565 }
566
567 /*
568  * ext4_ext_binsearch:
569  * binary search for closest extent of the given block
570  * the header must be checked before calling this
571  */
572 static void
573 ext4_ext_binsearch(struct inode *inode,
574                 struct ext4_ext_path *path, ext4_lblk_t block)
575 {
576         struct ext4_extent_header *eh = path->p_hdr;
577         struct ext4_extent *r, *l, *m;
578
579         if (eh->eh_entries == 0) {
580                 /*
581                  * this leaf is empty:
582                  * we get such a leaf in split/add case
583                  */
584                 return;
585         }
586
587         ext_debug("binsearch for %u:  ", block);
588
589         l = EXT_FIRST_EXTENT(eh) + 1;
590         r = EXT_LAST_EXTENT(eh);
591
592         while (l <= r) {
593                 m = l + (r - l) / 2;
594                 if (block < le32_to_cpu(m->ee_block))
595                         r = m - 1;
596                 else
597                         l = m + 1;
598                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
599                                 m, le32_to_cpu(m->ee_block),
600                                 r, le32_to_cpu(r->ee_block));
601         }
602
603         path->p_ext = l - 1;
604         ext_debug("  -> %d:%llu:[%d]%d ",
605                         le32_to_cpu(path->p_ext->ee_block),
606                         ext_pblock(path->p_ext),
607                         ext4_ext_is_uninitialized(path->p_ext),
608                         ext4_ext_get_actual_len(path->p_ext));
609
610 #ifdef CHECK_BINSEARCH
611         {
612                 struct ext4_extent *chex, *ex;
613                 int k;
614
615                 chex = ex = EXT_FIRST_EXTENT(eh);
616                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
617                         BUG_ON(k && le32_to_cpu(ex->ee_block)
618                                           <= le32_to_cpu(ex[-1].ee_block));
619                         if (block < le32_to_cpu(ex->ee_block))
620                                 break;
621                         chex = ex;
622                 }
623                 BUG_ON(chex != path->p_ext);
624         }
625 #endif
626
627 }
628
629 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
630 {
631         struct ext4_extent_header *eh;
632
633         eh = ext_inode_hdr(inode);
634         eh->eh_depth = 0;
635         eh->eh_entries = 0;
636         eh->eh_magic = EXT4_EXT_MAGIC;
637         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
638         ext4_mark_inode_dirty(handle, inode);
639         ext4_ext_invalidate_cache(inode);
640         return 0;
641 }
642
643 struct ext4_ext_path *
644 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
645                                         struct ext4_ext_path *path)
646 {
647         struct ext4_extent_header *eh;
648         struct buffer_head *bh;
649         short int depth, i, ppos = 0, alloc = 0;
650
651         eh = ext_inode_hdr(inode);
652         depth = ext_depth(inode);
653
654         /* account possible depth increase */
655         if (!path) {
656                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
657                                 GFP_NOFS);
658                 if (!path)
659                         return ERR_PTR(-ENOMEM);
660                 alloc = 1;
661         }
662         path[0].p_hdr = eh;
663         path[0].p_bh = NULL;
664
665         i = depth;
666         /* walk through the tree */
667         while (i) {
668                 int need_to_validate = 0;
669
670                 ext_debug("depth %d: num %d, max %d\n",
671                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
672
673                 ext4_ext_binsearch_idx(inode, path + ppos, block);
674                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
675                 path[ppos].p_depth = i;
676                 path[ppos].p_ext = NULL;
677
678                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
679                 if (unlikely(!bh))
680                         goto err;
681                 if (!bh_uptodate_or_lock(bh)) {
682                         if (bh_submit_read(bh) < 0) {
683                                 put_bh(bh);
684                                 goto err;
685                         }
686                         /* validate the extent entries */
687                         need_to_validate = 1;
688                 }
689                 eh = ext_block_hdr(bh);
690                 ppos++;
691                 BUG_ON(ppos > depth);
692                 path[ppos].p_bh = bh;
693                 path[ppos].p_hdr = eh;
694                 i--;
695
696                 if (need_to_validate && ext4_ext_check(inode, eh, i))
697                         goto err;
698         }
699
700         path[ppos].p_depth = i;
701         path[ppos].p_ext = NULL;
702         path[ppos].p_idx = NULL;
703
704         /* find extent */
705         ext4_ext_binsearch(inode, path + ppos, block);
706         /* if not an empty leaf */
707         if (path[ppos].p_ext)
708                 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
709
710         ext4_ext_show_path(inode, path);
711
712         return path;
713
714 err:
715         ext4_ext_drop_refs(path);
716         if (alloc)
717                 kfree(path);
718         return ERR_PTR(-EIO);
719 }
720
721 /*
722  * ext4_ext_insert_index:
723  * insert new index [@logical;@ptr] into the block at @curp;
724  * check where to insert: before @curp or after @curp
725  */
726 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
727                                 struct ext4_ext_path *curp,
728                                 int logical, ext4_fsblk_t ptr)
729 {
730         struct ext4_extent_idx *ix;
731         int len, err;
732
733         err = ext4_ext_get_access(handle, inode, curp);
734         if (err)
735                 return err;
736
737         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
738         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
739         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
740                 /* insert after */
741                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
742                         len = (len - 1) * sizeof(struct ext4_extent_idx);
743                         len = len < 0 ? 0 : len;
744                         ext_debug("insert new index %d after: %llu. "
745                                         "move %d from 0x%p to 0x%p\n",
746                                         logical, ptr, len,
747                                         (curp->p_idx + 1), (curp->p_idx + 2));
748                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
749                 }
750                 ix = curp->p_idx + 1;
751         } else {
752                 /* insert before */
753                 len = len * sizeof(struct ext4_extent_idx);
754                 len = len < 0 ? 0 : len;
755                 ext_debug("insert new index %d before: %llu. "
756                                 "move %d from 0x%p to 0x%p\n",
757                                 logical, ptr, len,
758                                 curp->p_idx, (curp->p_idx + 1));
759                 memmove(curp->p_idx + 1, curp->p_idx, len);
760                 ix = curp->p_idx;
761         }
762
763         ix->ei_block = cpu_to_le32(logical);
764         ext4_idx_store_pblock(ix, ptr);
765         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
766
767         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
768                              > le16_to_cpu(curp->p_hdr->eh_max));
769         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
770
771         err = ext4_ext_dirty(handle, inode, curp);
772         ext4_std_error(inode->i_sb, err);
773
774         return err;
775 }
776
777 /*
778  * ext4_ext_split:
779  * inserts new subtree into the path, using free index entry
780  * at depth @at:
781  * - allocates all needed blocks (new leaf and all intermediate index blocks)
782  * - makes decision where to split
783  * - moves remaining extents and index entries (right to the split point)
784  *   into the newly allocated blocks
785  * - initializes subtree
786  */
787 static int ext4_ext_split(handle_t *handle, struct inode *inode,
788                                 struct ext4_ext_path *path,
789                                 struct ext4_extent *newext, int at)
790 {
791         struct buffer_head *bh = NULL;
792         int depth = ext_depth(inode);
793         struct ext4_extent_header *neh;
794         struct ext4_extent_idx *fidx;
795         struct ext4_extent *ex;
796         int i = at, k, m, a;
797         ext4_fsblk_t newblock, oldblock;
798         __le32 border;
799         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
800         int err = 0;
801
802         /* make decision: where to split? */
803         /* FIXME: now decision is simplest: at current extent */
804
805         /* if current leaf will be split, then we should use
806          * border from split point */
807         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
808         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
809                 border = path[depth].p_ext[1].ee_block;
810                 ext_debug("leaf will be split."
811                                 " next leaf starts at %d\n",
812                                   le32_to_cpu(border));
813         } else {
814                 border = newext->ee_block;
815                 ext_debug("leaf will be added."
816                                 " next leaf starts at %d\n",
817                                 le32_to_cpu(border));
818         }
819
820         /*
821          * If error occurs, then we break processing
822          * and mark filesystem read-only. index won't
823          * be inserted and tree will be in consistent
824          * state. Next mount will repair buffers too.
825          */
826
827         /*
828          * Get array to track all allocated blocks.
829          * We need this to handle errors and free blocks
830          * upon them.
831          */
832         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
833         if (!ablocks)
834                 return -ENOMEM;
835
836         /* allocate all needed blocks */
837         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
838         for (a = 0; a < depth - at; a++) {
839                 newblock = ext4_ext_new_meta_block(handle, inode, path,
840                                                    newext, &err);
841                 if (newblock == 0)
842                         goto cleanup;
843                 ablocks[a] = newblock;
844         }
845
846         /* initialize new leaf */
847         newblock = ablocks[--a];
848         BUG_ON(newblock == 0);
849         bh = sb_getblk(inode->i_sb, newblock);
850         if (!bh) {
851                 err = -EIO;
852                 goto cleanup;
853         }
854         lock_buffer(bh);
855
856         err = ext4_journal_get_create_access(handle, bh);
857         if (err)
858                 goto cleanup;
859
860         neh = ext_block_hdr(bh);
861         neh->eh_entries = 0;
862         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
863         neh->eh_magic = EXT4_EXT_MAGIC;
864         neh->eh_depth = 0;
865         ex = EXT_FIRST_EXTENT(neh);
866
867         /* move remainder of path[depth] to the new leaf */
868         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
869         /* start copy from next extent */
870         /* TODO: we could do it by single memmove */
871         m = 0;
872         path[depth].p_ext++;
873         while (path[depth].p_ext <=
874                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
875                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
876                                 le32_to_cpu(path[depth].p_ext->ee_block),
877                                 ext_pblock(path[depth].p_ext),
878                                 ext4_ext_is_uninitialized(path[depth].p_ext),
879                                 ext4_ext_get_actual_len(path[depth].p_ext),
880                                 newblock);
881                 /*memmove(ex++, path[depth].p_ext++,
882                                 sizeof(struct ext4_extent));
883                 neh->eh_entries++;*/
884                 path[depth].p_ext++;
885                 m++;
886         }
887         if (m) {
888                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
889                 le16_add_cpu(&neh->eh_entries, m);
890         }
891
892         set_buffer_uptodate(bh);
893         unlock_buffer(bh);
894
895         err = ext4_handle_dirty_metadata(handle, inode, bh);
896         if (err)
897                 goto cleanup;
898         brelse(bh);
899         bh = NULL;
900
901         /* correct old leaf */
902         if (m) {
903                 err = ext4_ext_get_access(handle, inode, path + depth);
904                 if (err)
905                         goto cleanup;
906                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
907                 err = ext4_ext_dirty(handle, inode, path + depth);
908                 if (err)
909                         goto cleanup;
910
911         }
912
913         /* create intermediate indexes */
914         k = depth - at - 1;
915         BUG_ON(k < 0);
916         if (k)
917                 ext_debug("create %d intermediate indices\n", k);
918         /* insert new index into current index block */
919         /* current depth stored in i var */
920         i = depth - 1;
921         while (k--) {
922                 oldblock = newblock;
923                 newblock = ablocks[--a];
924                 bh = sb_getblk(inode->i_sb, newblock);
925                 if (!bh) {
926                         err = -EIO;
927                         goto cleanup;
928                 }
929                 lock_buffer(bh);
930
931                 err = ext4_journal_get_create_access(handle, bh);
932                 if (err)
933                         goto cleanup;
934
935                 neh = ext_block_hdr(bh);
936                 neh->eh_entries = cpu_to_le16(1);
937                 neh->eh_magic = EXT4_EXT_MAGIC;
938                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
939                 neh->eh_depth = cpu_to_le16(depth - i);
940                 fidx = EXT_FIRST_INDEX(neh);
941                 fidx->ei_block = border;
942                 ext4_idx_store_pblock(fidx, oldblock);
943
944                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
945                                 i, newblock, le32_to_cpu(border), oldblock);
946                 /* copy indexes */
947                 m = 0;
948                 path[i].p_idx++;
949
950                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
951                                 EXT_MAX_INDEX(path[i].p_hdr));
952                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
953                                 EXT_LAST_INDEX(path[i].p_hdr));
954                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
955                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
956                                         le32_to_cpu(path[i].p_idx->ei_block),
957                                         idx_pblock(path[i].p_idx),
958                                         newblock);
959                         /*memmove(++fidx, path[i].p_idx++,
960                                         sizeof(struct ext4_extent_idx));
961                         neh->eh_entries++;
962                         BUG_ON(neh->eh_entries > neh->eh_max);*/
963                         path[i].p_idx++;
964                         m++;
965                 }
966                 if (m) {
967                         memmove(++fidx, path[i].p_idx - m,
968                                 sizeof(struct ext4_extent_idx) * m);
969                         le16_add_cpu(&neh->eh_entries, m);
970                 }
971                 set_buffer_uptodate(bh);
972                 unlock_buffer(bh);
973
974                 err = ext4_handle_dirty_metadata(handle, inode, bh);
975                 if (err)
976                         goto cleanup;
977                 brelse(bh);
978                 bh = NULL;
979
980                 /* correct old index */
981                 if (m) {
982                         err = ext4_ext_get_access(handle, inode, path + i);
983                         if (err)
984                                 goto cleanup;
985                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
986                         err = ext4_ext_dirty(handle, inode, path + i);
987                         if (err)
988                                 goto cleanup;
989                 }
990
991                 i--;
992         }
993
994         /* insert new index */
995         err = ext4_ext_insert_index(handle, inode, path + at,
996                                     le32_to_cpu(border), newblock);
997
998 cleanup:
999         if (bh) {
1000                 if (buffer_locked(bh))
1001                         unlock_buffer(bh);
1002                 brelse(bh);
1003         }
1004
1005         if (err) {
1006                 /* free all allocated blocks in error case */
1007                 for (i = 0; i < depth; i++) {
1008                         if (!ablocks[i])
1009                                 continue;
1010                         ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
1011                 }
1012         }
1013         kfree(ablocks);
1014
1015         return err;
1016 }
1017
1018 /*
1019  * ext4_ext_grow_indepth:
1020  * implements tree growing procedure:
1021  * - allocates new block
1022  * - moves top-level data (index block or leaf) into the new block
1023  * - initializes new top-level, creating index that points to the
1024  *   just created block
1025  */
1026 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1027                                         struct ext4_ext_path *path,
1028                                         struct ext4_extent *newext)
1029 {
1030         struct ext4_ext_path *curp = path;
1031         struct ext4_extent_header *neh;
1032         struct ext4_extent_idx *fidx;
1033         struct buffer_head *bh;
1034         ext4_fsblk_t newblock;
1035         int err = 0;
1036
1037         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1038         if (newblock == 0)
1039                 return err;
1040
1041         bh = sb_getblk(inode->i_sb, newblock);
1042         if (!bh) {
1043                 err = -EIO;
1044                 ext4_std_error(inode->i_sb, err);
1045                 return err;
1046         }
1047         lock_buffer(bh);
1048
1049         err = ext4_journal_get_create_access(handle, bh);
1050         if (err) {
1051                 unlock_buffer(bh);
1052                 goto out;
1053         }
1054
1055         /* move top-level index/leaf into new block */
1056         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1057
1058         /* set size of new block */
1059         neh = ext_block_hdr(bh);
1060         /* old root could have indexes or leaves
1061          * so calculate e_max right way */
1062         if (ext_depth(inode))
1063                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1064         else
1065                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1066         neh->eh_magic = EXT4_EXT_MAGIC;
1067         set_buffer_uptodate(bh);
1068         unlock_buffer(bh);
1069
1070         err = ext4_handle_dirty_metadata(handle, inode, bh);
1071         if (err)
1072                 goto out;
1073
1074         /* create index in new top-level index: num,max,pointer */
1075         err = ext4_ext_get_access(handle, inode, curp);
1076         if (err)
1077                 goto out;
1078
1079         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1080         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1081         curp->p_hdr->eh_entries = cpu_to_le16(1);
1082         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1083
1084         if (path[0].p_hdr->eh_depth)
1085                 curp->p_idx->ei_block =
1086                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1087         else
1088                 curp->p_idx->ei_block =
1089                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1090         ext4_idx_store_pblock(curp->p_idx, newblock);
1091
1092         neh = ext_inode_hdr(inode);
1093         fidx = EXT_FIRST_INDEX(neh);
1094         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1095                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1096                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1097
1098         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1099         err = ext4_ext_dirty(handle, inode, curp);
1100 out:
1101         brelse(bh);
1102
1103         return err;
1104 }
1105
1106 /*
1107  * ext4_ext_create_new_leaf:
1108  * finds empty index and adds new leaf.
1109  * if no free index is found, then it requests in-depth growing.
1110  */
1111 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1112                                         struct ext4_ext_path *path,
1113                                         struct ext4_extent *newext)
1114 {
1115         struct ext4_ext_path *curp;
1116         int depth, i, err = 0;
1117
1118 repeat:
1119         i = depth = ext_depth(inode);
1120
1121         /* walk up to the tree and look for free index entry */
1122         curp = path + depth;
1123         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1124                 i--;
1125                 curp--;
1126         }
1127
1128         /* we use already allocated block for index block,
1129          * so subsequent data blocks should be contiguous */
1130         if (EXT_HAS_FREE_INDEX(curp)) {
1131                 /* if we found index with free entry, then use that
1132                  * entry: create all needed subtree and add new leaf */
1133                 err = ext4_ext_split(handle, inode, path, newext, i);
1134                 if (err)
1135                         goto out;
1136
1137                 /* refill path */
1138                 ext4_ext_drop_refs(path);
1139                 path = ext4_ext_find_extent(inode,
1140                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1141                                     path);
1142                 if (IS_ERR(path))
1143                         err = PTR_ERR(path);
1144         } else {
1145                 /* tree is full, time to grow in depth */
1146                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1147                 if (err)
1148                         goto out;
1149
1150                 /* refill path */
1151                 ext4_ext_drop_refs(path);
1152                 path = ext4_ext_find_extent(inode,
1153                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1154                                     path);
1155                 if (IS_ERR(path)) {
1156                         err = PTR_ERR(path);
1157                         goto out;
1158                 }
1159
1160                 /*
1161                  * only first (depth 0 -> 1) produces free space;
1162                  * in all other cases we have to split the grown tree
1163                  */
1164                 depth = ext_depth(inode);
1165                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1166                         /* now we need to split */
1167                         goto repeat;
1168                 }
1169         }
1170
1171 out:
1172         return err;
1173 }
1174
1175 /*
1176  * search the closest allocated block to the left for *logical
1177  * and returns it at @logical + it's physical address at @phys
1178  * if *logical is the smallest allocated block, the function
1179  * returns 0 at @phys
1180  * return value contains 0 (success) or error code
1181  */
1182 int
1183 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1184                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1185 {
1186         struct ext4_extent_idx *ix;
1187         struct ext4_extent *ex;
1188         int depth, ee_len;
1189
1190         BUG_ON(path == NULL);
1191         depth = path->p_depth;
1192         *phys = 0;
1193
1194         if (depth == 0 && path->p_ext == NULL)
1195                 return 0;
1196
1197         /* usually extent in the path covers blocks smaller
1198          * then *logical, but it can be that extent is the
1199          * first one in the file */
1200
1201         ex = path[depth].p_ext;
1202         ee_len = ext4_ext_get_actual_len(ex);
1203         if (*logical < le32_to_cpu(ex->ee_block)) {
1204                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1205                 while (--depth >= 0) {
1206                         ix = path[depth].p_idx;
1207                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1208                 }
1209                 return 0;
1210         }
1211
1212         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1213
1214         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1215         *phys = ext_pblock(ex) + ee_len - 1;
1216         return 0;
1217 }
1218
1219 /*
1220  * search the closest allocated block to the right for *logical
1221  * and returns it at @logical + it's physical address at @phys
1222  * if *logical is the smallest allocated block, the function
1223  * returns 0 at @phys
1224  * return value contains 0 (success) or error code
1225  */
1226 int
1227 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1228                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1229 {
1230         struct buffer_head *bh = NULL;
1231         struct ext4_extent_header *eh;
1232         struct ext4_extent_idx *ix;
1233         struct ext4_extent *ex;
1234         ext4_fsblk_t block;
1235         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1236         int ee_len;
1237
1238         BUG_ON(path == NULL);
1239         depth = path->p_depth;
1240         *phys = 0;
1241
1242         if (depth == 0 && path->p_ext == NULL)
1243                 return 0;
1244
1245         /* usually extent in the path covers blocks smaller
1246          * then *logical, but it can be that extent is the
1247          * first one in the file */
1248
1249         ex = path[depth].p_ext;
1250         ee_len = ext4_ext_get_actual_len(ex);
1251         if (*logical < le32_to_cpu(ex->ee_block)) {
1252                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1253                 while (--depth >= 0) {
1254                         ix = path[depth].p_idx;
1255                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1256                 }
1257                 *logical = le32_to_cpu(ex->ee_block);
1258                 *phys = ext_pblock(ex);
1259                 return 0;
1260         }
1261
1262         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1263
1264         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1265                 /* next allocated block in this leaf */
1266                 ex++;
1267                 *logical = le32_to_cpu(ex->ee_block);
1268                 *phys = ext_pblock(ex);
1269                 return 0;
1270         }
1271
1272         /* go up and search for index to the right */
1273         while (--depth >= 0) {
1274                 ix = path[depth].p_idx;
1275                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1276                         goto got_index;
1277         }
1278
1279         /* we've gone up to the root and found no index to the right */
1280         return 0;
1281
1282 got_index:
1283         /* we've found index to the right, let's
1284          * follow it and find the closest allocated
1285          * block to the right */
1286         ix++;
1287         block = idx_pblock(ix);
1288         while (++depth < path->p_depth) {
1289                 bh = sb_bread(inode->i_sb, block);
1290                 if (bh == NULL)
1291                         return -EIO;
1292                 eh = ext_block_hdr(bh);
1293                 /* subtract from p_depth to get proper eh_depth */
1294                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1295                         put_bh(bh);
1296                         return -EIO;
1297                 }
1298                 ix = EXT_FIRST_INDEX(eh);
1299                 block = idx_pblock(ix);
1300                 put_bh(bh);
1301         }
1302
1303         bh = sb_bread(inode->i_sb, block);
1304         if (bh == NULL)
1305                 return -EIO;
1306         eh = ext_block_hdr(bh);
1307         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1308                 put_bh(bh);
1309                 return -EIO;
1310         }
1311         ex = EXT_FIRST_EXTENT(eh);
1312         *logical = le32_to_cpu(ex->ee_block);
1313         *phys = ext_pblock(ex);
1314         put_bh(bh);
1315         return 0;
1316 }
1317
1318 /*
1319  * ext4_ext_next_allocated_block:
1320  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1321  * NOTE: it considers block number from index entry as
1322  * allocated block. Thus, index entries have to be consistent
1323  * with leaves.
1324  */
1325 static ext4_lblk_t
1326 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1327 {
1328         int depth;
1329
1330         BUG_ON(path == NULL);
1331         depth = path->p_depth;
1332
1333         if (depth == 0 && path->p_ext == NULL)
1334                 return EXT_MAX_BLOCK;
1335
1336         while (depth >= 0) {
1337                 if (depth == path->p_depth) {
1338                         /* leaf */
1339                         if (path[depth].p_ext !=
1340                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1341                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1342                 } else {
1343                         /* index */
1344                         if (path[depth].p_idx !=
1345                                         EXT_LAST_INDEX(path[depth].p_hdr))
1346                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1347                 }
1348                 depth--;
1349         }
1350
1351         return EXT_MAX_BLOCK;
1352 }
1353
1354 /*
1355  * ext4_ext_next_leaf_block:
1356  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1357  */
1358 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1359                                         struct ext4_ext_path *path)
1360 {
1361         int depth;
1362
1363         BUG_ON(path == NULL);
1364         depth = path->p_depth;
1365
1366         /* zero-tree has no leaf blocks at all */
1367         if (depth == 0)
1368                 return EXT_MAX_BLOCK;
1369
1370         /* go to index block */
1371         depth--;
1372
1373         while (depth >= 0) {
1374                 if (path[depth].p_idx !=
1375                                 EXT_LAST_INDEX(path[depth].p_hdr))
1376                         return (ext4_lblk_t)
1377                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1378                 depth--;
1379         }
1380
1381         return EXT_MAX_BLOCK;
1382 }
1383
1384 /*
1385  * ext4_ext_correct_indexes:
1386  * if leaf gets modified and modified extent is first in the leaf,
1387  * then we have to correct all indexes above.
1388  * TODO: do we need to correct tree in all cases?
1389  */
1390 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1391                                 struct ext4_ext_path *path)
1392 {
1393         struct ext4_extent_header *eh;
1394         int depth = ext_depth(inode);
1395         struct ext4_extent *ex;
1396         __le32 border;
1397         int k, err = 0;
1398
1399         eh = path[depth].p_hdr;
1400         ex = path[depth].p_ext;
1401         BUG_ON(ex == NULL);
1402         BUG_ON(eh == NULL);
1403
1404         if (depth == 0) {
1405                 /* there is no tree at all */
1406                 return 0;
1407         }
1408
1409         if (ex != EXT_FIRST_EXTENT(eh)) {
1410                 /* we correct tree if first leaf got modified only */
1411                 return 0;
1412         }
1413
1414         /*
1415          * TODO: we need correction if border is smaller than current one
1416          */
1417         k = depth - 1;
1418         border = path[depth].p_ext->ee_block;
1419         err = ext4_ext_get_access(handle, inode, path + k);
1420         if (err)
1421                 return err;
1422         path[k].p_idx->ei_block = border;
1423         err = ext4_ext_dirty(handle, inode, path + k);
1424         if (err)
1425                 return err;
1426
1427         while (k--) {
1428                 /* change all left-side indexes */
1429                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1430                         break;
1431                 err = ext4_ext_get_access(handle, inode, path + k);
1432                 if (err)
1433                         break;
1434                 path[k].p_idx->ei_block = border;
1435                 err = ext4_ext_dirty(handle, inode, path + k);
1436                 if (err)
1437                         break;
1438         }
1439
1440         return err;
1441 }
1442
1443 int
1444 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1445                                 struct ext4_extent *ex2)
1446 {
1447         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1448
1449         /*
1450          * Make sure that either both extents are uninitialized, or
1451          * both are _not_.
1452          */
1453         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1454                 return 0;
1455
1456         if (ext4_ext_is_uninitialized(ex1))
1457                 max_len = EXT_UNINIT_MAX_LEN;
1458         else
1459                 max_len = EXT_INIT_MAX_LEN;
1460
1461         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1462         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1463
1464         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1465                         le32_to_cpu(ex2->ee_block))
1466                 return 0;
1467
1468         /*
1469          * To allow future support for preallocated extents to be added
1470          * as an RO_COMPAT feature, refuse to merge to extents if
1471          * this can result in the top bit of ee_len being set.
1472          */
1473         if (ext1_ee_len + ext2_ee_len > max_len)
1474                 return 0;
1475 #ifdef AGGRESSIVE_TEST
1476         if (ext1_ee_len >= 4)
1477                 return 0;
1478 #endif
1479
1480         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1481                 return 1;
1482         return 0;
1483 }
1484
1485 /*
1486  * This function tries to merge the "ex" extent to the next extent in the tree.
1487  * It always tries to merge towards right. If you want to merge towards
1488  * left, pass "ex - 1" as argument instead of "ex".
1489  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1490  * 1 if they got merged.
1491  */
1492 int ext4_ext_try_to_merge(struct inode *inode,
1493                           struct ext4_ext_path *path,
1494                           struct ext4_extent *ex)
1495 {
1496         struct ext4_extent_header *eh;
1497         unsigned int depth, len;
1498         int merge_done = 0;
1499         int uninitialized = 0;
1500
1501         depth = ext_depth(inode);
1502         BUG_ON(path[depth].p_hdr == NULL);
1503         eh = path[depth].p_hdr;
1504
1505         while (ex < EXT_LAST_EXTENT(eh)) {
1506                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1507                         break;
1508                 /* merge with next extent! */
1509                 if (ext4_ext_is_uninitialized(ex))
1510                         uninitialized = 1;
1511                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1512                                 + ext4_ext_get_actual_len(ex + 1));
1513                 if (uninitialized)
1514                         ext4_ext_mark_uninitialized(ex);
1515
1516                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1517                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1518                                 * sizeof(struct ext4_extent);
1519                         memmove(ex + 1, ex + 2, len);
1520                 }
1521                 le16_add_cpu(&eh->eh_entries, -1);
1522                 merge_done = 1;
1523                 WARN_ON(eh->eh_entries == 0);
1524                 if (!eh->eh_entries)
1525                         ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1526                            "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1527         }
1528
1529         return merge_done;
1530 }
1531
1532 /*
1533  * check if a portion of the "newext" extent overlaps with an
1534  * existing extent.
1535  *
1536  * If there is an overlap discovered, it updates the length of the newext
1537  * such that there will be no overlap, and then returns 1.
1538  * If there is no overlap found, it returns 0.
1539  */
1540 unsigned int ext4_ext_check_overlap(struct inode *inode,
1541                                     struct ext4_extent *newext,
1542                                     struct ext4_ext_path *path)
1543 {
1544         ext4_lblk_t b1, b2;
1545         unsigned int depth, len1;
1546         unsigned int ret = 0;
1547
1548         b1 = le32_to_cpu(newext->ee_block);
1549         len1 = ext4_ext_get_actual_len(newext);
1550         depth = ext_depth(inode);
1551         if (!path[depth].p_ext)
1552                 goto out;
1553         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1554
1555         /*
1556          * get the next allocated block if the extent in the path
1557          * is before the requested block(s)
1558          */
1559         if (b2 < b1) {
1560                 b2 = ext4_ext_next_allocated_block(path);
1561                 if (b2 == EXT_MAX_BLOCK)
1562                         goto out;
1563         }
1564
1565         /* check for wrap through zero on extent logical start block*/
1566         if (b1 + len1 < b1) {
1567                 len1 = EXT_MAX_BLOCK - b1;
1568                 newext->ee_len = cpu_to_le16(len1);
1569                 ret = 1;
1570         }
1571
1572         /* check for overlap */
1573         if (b1 + len1 > b2) {
1574                 newext->ee_len = cpu_to_le16(b2 - b1);
1575                 ret = 1;
1576         }
1577 out:
1578         return ret;
1579 }
1580
1581 /*
1582  * ext4_ext_insert_extent:
1583  * tries to merge requsted extent into the existing extent or
1584  * inserts requested extent as new one into the tree,
1585  * creating new leaf in the no-space case.
1586  */
1587 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1588                                 struct ext4_ext_path *path,
1589                                 struct ext4_extent *newext)
1590 {
1591         struct ext4_extent_header *eh;
1592         struct ext4_extent *ex, *fex;
1593         struct ext4_extent *nearex; /* nearest extent */
1594         struct ext4_ext_path *npath = NULL;
1595         int depth, len, err;
1596         ext4_lblk_t next;
1597         unsigned uninitialized = 0;
1598
1599         BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1600         depth = ext_depth(inode);
1601         ex = path[depth].p_ext;
1602         BUG_ON(path[depth].p_hdr == NULL);
1603
1604         /* try to insert block into found extent and return */
1605         if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1606                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1607                                 ext4_ext_is_uninitialized(newext),
1608                                 ext4_ext_get_actual_len(newext),
1609                                 le32_to_cpu(ex->ee_block),
1610                                 ext4_ext_is_uninitialized(ex),
1611                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1612                 err = ext4_ext_get_access(handle, inode, path + depth);
1613                 if (err)
1614                         return err;
1615
1616                 /*
1617                  * ext4_can_extents_be_merged should have checked that either
1618                  * both extents are uninitialized, or both aren't. Thus we
1619                  * need to check only one of them here.
1620                  */
1621                 if (ext4_ext_is_uninitialized(ex))
1622                         uninitialized = 1;
1623                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1624                                         + ext4_ext_get_actual_len(newext));
1625                 if (uninitialized)
1626                         ext4_ext_mark_uninitialized(ex);
1627                 eh = path[depth].p_hdr;
1628                 nearex = ex;
1629                 goto merge;
1630         }
1631
1632 repeat:
1633         depth = ext_depth(inode);
1634         eh = path[depth].p_hdr;
1635         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1636                 goto has_space;
1637
1638         /* probably next leaf has space for us? */
1639         fex = EXT_LAST_EXTENT(eh);
1640         next = ext4_ext_next_leaf_block(inode, path);
1641         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1642             && next != EXT_MAX_BLOCK) {
1643                 ext_debug("next leaf block - %d\n", next);
1644                 BUG_ON(npath != NULL);
1645                 npath = ext4_ext_find_extent(inode, next, NULL);
1646                 if (IS_ERR(npath))
1647                         return PTR_ERR(npath);
1648                 BUG_ON(npath->p_depth != path->p_depth);
1649                 eh = npath[depth].p_hdr;
1650                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1651                         ext_debug("next leaf isnt full(%d)\n",
1652                                   le16_to_cpu(eh->eh_entries));
1653                         path = npath;
1654                         goto repeat;
1655                 }
1656                 ext_debug("next leaf has no free space(%d,%d)\n",
1657                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1658         }
1659
1660         /*
1661          * There is no free space in the found leaf.
1662          * We're gonna add a new leaf in the tree.
1663          */
1664         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1665         if (err)
1666                 goto cleanup;
1667         depth = ext_depth(inode);
1668         eh = path[depth].p_hdr;
1669
1670 has_space:
1671         nearex = path[depth].p_ext;
1672
1673         err = ext4_ext_get_access(handle, inode, path + depth);
1674         if (err)
1675                 goto cleanup;
1676
1677         if (!nearex) {
1678                 /* there is no extent in this leaf, create first one */
1679                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1680                                 le32_to_cpu(newext->ee_block),
1681                                 ext_pblock(newext),
1682                                 ext4_ext_is_uninitialized(newext),
1683                                 ext4_ext_get_actual_len(newext));
1684                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1685         } else if (le32_to_cpu(newext->ee_block)
1686                            > le32_to_cpu(nearex->ee_block)) {
1687 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1688                 if (nearex != EXT_LAST_EXTENT(eh)) {
1689                         len = EXT_MAX_EXTENT(eh) - nearex;
1690                         len = (len - 1) * sizeof(struct ext4_extent);
1691                         len = len < 0 ? 0 : len;
1692                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1693                                         "move %d from 0x%p to 0x%p\n",
1694                                         le32_to_cpu(newext->ee_block),
1695                                         ext_pblock(newext),
1696                                         ext4_ext_is_uninitialized(newext),
1697                                         ext4_ext_get_actual_len(newext),
1698                                         nearex, len, nearex + 1, nearex + 2);
1699                         memmove(nearex + 2, nearex + 1, len);
1700                 }
1701                 path[depth].p_ext = nearex + 1;
1702         } else {
1703                 BUG_ON(newext->ee_block == nearex->ee_block);
1704                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1705                 len = len < 0 ? 0 : len;
1706                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1707                                 "move %d from 0x%p to 0x%p\n",
1708                                 le32_to_cpu(newext->ee_block),
1709                                 ext_pblock(newext),
1710                                 ext4_ext_is_uninitialized(newext),
1711                                 ext4_ext_get_actual_len(newext),
1712                                 nearex, len, nearex + 1, nearex + 2);
1713                 memmove(nearex + 1, nearex, len);
1714                 path[depth].p_ext = nearex;
1715         }
1716
1717         le16_add_cpu(&eh->eh_entries, 1);
1718         nearex = path[depth].p_ext;
1719         nearex->ee_block = newext->ee_block;
1720         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1721         nearex->ee_len = newext->ee_len;
1722
1723 merge:
1724         /* try to merge extents to the right */
1725         ext4_ext_try_to_merge(inode, path, nearex);
1726
1727         /* try to merge extents to the left */
1728
1729         /* time to correct all indexes above */
1730         err = ext4_ext_correct_indexes(handle, inode, path);
1731         if (err)
1732                 goto cleanup;
1733
1734         err = ext4_ext_dirty(handle, inode, path + depth);
1735
1736 cleanup:
1737         if (npath) {
1738                 ext4_ext_drop_refs(npath);
1739                 kfree(npath);
1740         }
1741         ext4_ext_invalidate_cache(inode);
1742         return err;
1743 }
1744
1745 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1746                         ext4_lblk_t num, ext_prepare_callback func,
1747                         void *cbdata)
1748 {
1749         struct ext4_ext_path *path = NULL;
1750         struct ext4_ext_cache cbex;
1751         struct ext4_extent *ex;
1752         ext4_lblk_t next, start = 0, end = 0;
1753         ext4_lblk_t last = block + num;
1754         int depth, exists, err = 0;
1755
1756         BUG_ON(func == NULL);
1757         BUG_ON(inode == NULL);
1758
1759         while (block < last && block != EXT_MAX_BLOCK) {
1760                 num = last - block;
1761                 /* find extent for this block */
1762                 path = ext4_ext_find_extent(inode, block, path);
1763                 if (IS_ERR(path)) {
1764                         err = PTR_ERR(path);
1765                         path = NULL;
1766                         break;
1767                 }
1768
1769                 depth = ext_depth(inode);
1770                 BUG_ON(path[depth].p_hdr == NULL);
1771                 ex = path[depth].p_ext;
1772                 next = ext4_ext_next_allocated_block(path);
1773
1774                 exists = 0;
1775                 if (!ex) {
1776                         /* there is no extent yet, so try to allocate
1777                          * all requested space */
1778                         start = block;
1779                         end = block + num;
1780                 } else if (le32_to_cpu(ex->ee_block) > block) {
1781                         /* need to allocate space before found extent */
1782                         start = block;
1783                         end = le32_to_cpu(ex->ee_block);
1784                         if (block + num < end)
1785                                 end = block + num;
1786                 } else if (block >= le32_to_cpu(ex->ee_block)
1787                                         + ext4_ext_get_actual_len(ex)) {
1788                         /* need to allocate space after found extent */
1789                         start = block;
1790                         end = block + num;
1791                         if (end >= next)
1792                                 end = next;
1793                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1794                         /*
1795                          * some part of requested space is covered
1796                          * by found extent
1797                          */
1798                         start = block;
1799                         end = le32_to_cpu(ex->ee_block)
1800                                 + ext4_ext_get_actual_len(ex);
1801                         if (block + num < end)
1802                                 end = block + num;
1803                         exists = 1;
1804                 } else {
1805                         BUG();
1806                 }
1807                 BUG_ON(end <= start);
1808
1809                 if (!exists) {
1810                         cbex.ec_block = start;
1811                         cbex.ec_len = end - start;
1812                         cbex.ec_start = 0;
1813                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1814                 } else {
1815                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1816                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1817                         cbex.ec_start = ext_pblock(ex);
1818                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1819                 }
1820
1821                 BUG_ON(cbex.ec_len == 0);
1822                 err = func(inode, path, &cbex, ex, cbdata);
1823                 ext4_ext_drop_refs(path);
1824
1825                 if (err < 0)
1826                         break;
1827
1828                 if (err == EXT_REPEAT)
1829                         continue;
1830                 else if (err == EXT_BREAK) {
1831                         err = 0;
1832                         break;
1833                 }
1834
1835                 if (ext_depth(inode) != depth) {
1836                         /* depth was changed. we have to realloc path */
1837                         kfree(path);
1838                         path = NULL;
1839                 }
1840
1841                 block = cbex.ec_block + cbex.ec_len;
1842         }
1843
1844         if (path) {
1845                 ext4_ext_drop_refs(path);
1846                 kfree(path);
1847         }
1848
1849         return err;
1850 }
1851
1852 static void
1853 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1854                         __u32 len, ext4_fsblk_t start, int type)
1855 {
1856         struct ext4_ext_cache *cex;
1857         BUG_ON(len == 0);
1858         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1859         cex = &EXT4_I(inode)->i_cached_extent;
1860         cex->ec_type = type;
1861         cex->ec_block = block;
1862         cex->ec_len = len;
1863         cex->ec_start = start;
1864         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1865 }
1866
1867 /*
1868  * ext4_ext_put_gap_in_cache:
1869  * calculate boundaries of the gap that the requested block fits into
1870  * and cache this gap
1871  */
1872 static void
1873 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1874                                 ext4_lblk_t block)
1875 {
1876         int depth = ext_depth(inode);
1877         unsigned long len;
1878         ext4_lblk_t lblock;
1879         struct ext4_extent *ex;
1880
1881         ex = path[depth].p_ext;
1882         if (ex == NULL) {
1883                 /* there is no extent yet, so gap is [0;-] */
1884                 lblock = 0;
1885                 len = EXT_MAX_BLOCK;
1886                 ext_debug("cache gap(whole file):");
1887         } else if (block < le32_to_cpu(ex->ee_block)) {
1888                 lblock = block;
1889                 len = le32_to_cpu(ex->ee_block) - block;
1890                 ext_debug("cache gap(before): %u [%u:%u]",
1891                                 block,
1892                                 le32_to_cpu(ex->ee_block),
1893                                  ext4_ext_get_actual_len(ex));
1894         } else if (block >= le32_to_cpu(ex->ee_block)
1895                         + ext4_ext_get_actual_len(ex)) {
1896                 ext4_lblk_t next;
1897                 lblock = le32_to_cpu(ex->ee_block)
1898                         + ext4_ext_get_actual_len(ex);
1899
1900                 next = ext4_ext_next_allocated_block(path);
1901                 ext_debug("cache gap(after): [%u:%u] %u",
1902                                 le32_to_cpu(ex->ee_block),
1903                                 ext4_ext_get_actual_len(ex),
1904                                 block);
1905                 BUG_ON(next == lblock);
1906                 len = next - lblock;
1907         } else {
1908                 lblock = len = 0;
1909                 BUG();
1910         }
1911
1912         ext_debug(" -> %u:%lu\n", lblock, len);
1913         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1914 }
1915
1916 static int
1917 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1918                         struct ext4_extent *ex)
1919 {
1920         struct ext4_ext_cache *cex;
1921         int ret = EXT4_EXT_CACHE_NO;
1922
1923         /* 
1924          * We borrow i_block_reservation_lock to protect i_cached_extent
1925          */
1926         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1927         cex = &EXT4_I(inode)->i_cached_extent;
1928
1929         /* has cache valid data? */
1930         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1931                 goto errout;
1932
1933         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1934                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1935         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1936                 ex->ee_block = cpu_to_le32(cex->ec_block);
1937                 ext4_ext_store_pblock(ex, cex->ec_start);
1938                 ex->ee_len = cpu_to_le16(cex->ec_len);
1939                 ext_debug("%u cached by %u:%u:%llu\n",
1940                                 block,
1941                                 cex->ec_block, cex->ec_len, cex->ec_start);
1942                 ret = cex->ec_type;
1943         }
1944 errout:
1945         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1946         return ret;
1947 }
1948
1949 /*
1950  * ext4_ext_rm_idx:
1951  * removes index from the index block.
1952  * It's used in truncate case only, thus all requests are for
1953  * last index in the block only.
1954  */
1955 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1956                         struct ext4_ext_path *path)
1957 {
1958         struct buffer_head *bh;
1959         int err;
1960         ext4_fsblk_t leaf;
1961
1962         /* free index block */
1963         path--;
1964         leaf = idx_pblock(path->p_idx);
1965         BUG_ON(path->p_hdr->eh_entries == 0);
1966         err = ext4_ext_get_access(handle, inode, path);
1967         if (err)
1968                 return err;
1969         le16_add_cpu(&path->p_hdr->eh_entries, -1);
1970         err = ext4_ext_dirty(handle, inode, path);
1971         if (err)
1972                 return err;
1973         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1974         bh = sb_find_get_block(inode->i_sb, leaf);
1975         ext4_forget(handle, 1, inode, bh, leaf);
1976         ext4_free_blocks(handle, inode, leaf, 1, 1);
1977         return err;
1978 }
1979
1980 /*
1981  * ext4_ext_calc_credits_for_single_extent:
1982  * This routine returns max. credits that needed to insert an extent
1983  * to the extent tree.
1984  * When pass the actual path, the caller should calculate credits
1985  * under i_data_sem.
1986  */
1987 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1988                                                 struct ext4_ext_path *path)
1989 {
1990         if (path) {
1991                 int depth = ext_depth(inode);
1992                 int ret = 0;
1993
1994                 /* probably there is space in leaf? */
1995                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1996                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
1997
1998                         /*
1999                          *  There are some space in the leaf tree, no
2000                          *  need to account for leaf block credit
2001                          *
2002                          *  bitmaps and block group descriptor blocks
2003                          *  and other metadat blocks still need to be
2004                          *  accounted.
2005                          */
2006                         /* 1 bitmap, 1 block group descriptor */
2007                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2008                         return ret;
2009                 }
2010         }
2011
2012         return ext4_chunk_trans_blocks(inode, nrblocks);
2013 }
2014
2015 /*
2016  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2017  *
2018  * if nrblocks are fit in a single extent (chunk flag is 1), then
2019  * in the worse case, each tree level index/leaf need to be changed
2020  * if the tree split due to insert a new extent, then the old tree
2021  * index/leaf need to be updated too
2022  *
2023  * If the nrblocks are discontiguous, they could cause
2024  * the whole tree split more than once, but this is really rare.
2025  */
2026 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2027 {
2028         int index;
2029         int depth = ext_depth(inode);
2030
2031         if (chunk)
2032                 index = depth * 2;
2033         else
2034                 index = depth * 3;
2035
2036         return index;
2037 }
2038
2039 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2040                                 struct ext4_extent *ex,
2041                                 ext4_lblk_t from, ext4_lblk_t to)
2042 {
2043         struct buffer_head *bh;
2044         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2045         int i, metadata = 0;
2046
2047         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2048                 metadata = 1;
2049 #ifdef EXTENTS_STATS
2050         {
2051                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2052                 spin_lock(&sbi->s_ext_stats_lock);
2053                 sbi->s_ext_blocks += ee_len;
2054                 sbi->s_ext_extents++;
2055                 if (ee_len < sbi->s_ext_min)
2056                         sbi->s_ext_min = ee_len;
2057                 if (ee_len > sbi->s_ext_max)
2058                         sbi->s_ext_max = ee_len;
2059                 if (ext_depth(inode) > sbi->s_depth_max)
2060                         sbi->s_depth_max = ext_depth(inode);
2061                 spin_unlock(&sbi->s_ext_stats_lock);
2062         }
2063 #endif
2064         if (from >= le32_to_cpu(ex->ee_block)
2065             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2066                 /* tail removal */
2067                 ext4_lblk_t num;
2068                 ext4_fsblk_t start;
2069
2070                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2071                 start = ext_pblock(ex) + ee_len - num;
2072                 ext_debug("free last %u blocks starting %llu\n", num, start);
2073                 for (i = 0; i < num; i++) {
2074                         bh = sb_find_get_block(inode->i_sb, start + i);
2075                         ext4_forget(handle, 0, inode, bh, start + i);
2076                 }
2077                 ext4_free_blocks(handle, inode, start, num, metadata);
2078         } else if (from == le32_to_cpu(ex->ee_block)
2079                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2080                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2081                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2082         } else {
2083                 printk(KERN_INFO "strange request: removal(2) "
2084                                 "%u-%u from %u:%u\n",
2085                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2086         }
2087         return 0;
2088 }
2089
2090 static int
2091 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2092                 struct ext4_ext_path *path, ext4_lblk_t start)
2093 {
2094         int err = 0, correct_index = 0;
2095         int depth = ext_depth(inode), credits;
2096         struct ext4_extent_header *eh;
2097         ext4_lblk_t a, b, block;
2098         unsigned num;
2099         ext4_lblk_t ex_ee_block;
2100         unsigned short ex_ee_len;
2101         unsigned uninitialized = 0;
2102         struct ext4_extent *ex;
2103
2104         /* the header must be checked already in ext4_ext_remove_space() */
2105         ext_debug("truncate since %u in leaf\n", start);
2106         if (!path[depth].p_hdr)
2107                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2108         eh = path[depth].p_hdr;
2109         BUG_ON(eh == NULL);
2110
2111         /* find where to start removing */
2112         ex = EXT_LAST_EXTENT(eh);
2113
2114         ex_ee_block = le32_to_cpu(ex->ee_block);
2115         ex_ee_len = ext4_ext_get_actual_len(ex);
2116
2117         while (ex >= EXT_FIRST_EXTENT(eh) &&
2118                         ex_ee_block + ex_ee_len > start) {
2119
2120                 if (ext4_ext_is_uninitialized(ex))
2121                         uninitialized = 1;
2122                 else
2123                         uninitialized = 0;
2124
2125                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2126                          uninitialized, ex_ee_len);
2127                 path[depth].p_ext = ex;
2128
2129                 a = ex_ee_block > start ? ex_ee_block : start;
2130                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2131                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2132
2133                 ext_debug("  border %u:%u\n", a, b);
2134
2135                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2136                         block = 0;
2137                         num = 0;
2138                         BUG();
2139                 } else if (a != ex_ee_block) {
2140                         /* remove tail of the extent */
2141                         block = ex_ee_block;
2142                         num = a - block;
2143                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2144                         /* remove head of the extent */
2145                         block = a;
2146                         num = b - a;
2147                         /* there is no "make a hole" API yet */
2148                         BUG();
2149                 } else {
2150                         /* remove whole extent: excellent! */
2151                         block = ex_ee_block;
2152                         num = 0;
2153                         BUG_ON(a != ex_ee_block);
2154                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2155                 }
2156
2157                 /*
2158                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2159                  * descriptor) for each block group; assume two block
2160                  * groups plus ex_ee_len/blocks_per_block_group for
2161                  * the worst case
2162                  */
2163                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2164                 if (ex == EXT_FIRST_EXTENT(eh)) {
2165                         correct_index = 1;
2166                         credits += (ext_depth(inode)) + 1;
2167                 }
2168                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2169
2170                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2171                 if (err)
2172                         goto out;
2173
2174                 err = ext4_ext_get_access(handle, inode, path + depth);
2175                 if (err)
2176                         goto out;
2177
2178                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2179                 if (err)
2180                         goto out;
2181
2182                 if (num == 0) {
2183                         /* this extent is removed; mark slot entirely unused */
2184                         ext4_ext_store_pblock(ex, 0);
2185                         le16_add_cpu(&eh->eh_entries, -1);
2186                 }
2187
2188                 ex->ee_block = cpu_to_le32(block);
2189                 ex->ee_len = cpu_to_le16(num);
2190                 /*
2191                  * Do not mark uninitialized if all the blocks in the
2192                  * extent have been removed.
2193                  */
2194                 if (uninitialized && num)
2195                         ext4_ext_mark_uninitialized(ex);
2196
2197                 err = ext4_ext_dirty(handle, inode, path + depth);
2198                 if (err)
2199                         goto out;
2200
2201                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2202                                 ext_pblock(ex));
2203                 ex--;
2204                 ex_ee_block = le32_to_cpu(ex->ee_block);
2205                 ex_ee_len = ext4_ext_get_actual_len(ex);
2206         }
2207
2208         if (correct_index && eh->eh_entries)
2209                 err = ext4_ext_correct_indexes(handle, inode, path);
2210
2211         /* if this leaf is free, then we should
2212          * remove it from index block above */
2213         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2214                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2215
2216 out:
2217         return err;
2218 }
2219
2220 /*
2221  * ext4_ext_more_to_rm:
2222  * returns 1 if current index has to be freed (even partial)
2223  */
2224 static int
2225 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2226 {
2227         BUG_ON(path->p_idx == NULL);
2228
2229         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2230                 return 0;
2231
2232         /*
2233          * if truncate on deeper level happened, it wasn't partial,
2234          * so we have to consider current index for truncation
2235          */
2236         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2237                 return 0;
2238         return 1;
2239 }
2240
2241 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2242 {
2243         struct super_block *sb = inode->i_sb;
2244         int depth = ext_depth(inode);
2245         struct ext4_ext_path *path;
2246         handle_t *handle;
2247         int i = 0, err = 0;
2248
2249         ext_debug("truncate since %u\n", start);
2250
2251         /* probably first extent we're gonna free will be last in block */
2252         handle = ext4_journal_start(inode, depth + 1);
2253         if (IS_ERR(handle))
2254                 return PTR_ERR(handle);
2255
2256         ext4_ext_invalidate_cache(inode);
2257
2258         /*
2259          * We start scanning from right side, freeing all the blocks
2260          * after i_size and walking into the tree depth-wise.
2261          */
2262         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2263         if (path == NULL) {
2264                 ext4_journal_stop(handle);
2265                 return -ENOMEM;
2266         }
2267         path[0].p_hdr = ext_inode_hdr(inode);
2268         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2269                 err = -EIO;
2270                 goto out;
2271         }
2272         path[0].p_depth = depth;
2273
2274         while (i >= 0 && err == 0) {
2275                 if (i == depth) {
2276                         /* this is leaf block */
2277                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2278                         /* root level has p_bh == NULL, brelse() eats this */
2279                         brelse(path[i].p_bh);
2280                         path[i].p_bh = NULL;
2281                         i--;
2282                         continue;
2283                 }
2284
2285                 /* this is index block */
2286                 if (!path[i].p_hdr) {
2287                         ext_debug("initialize header\n");
2288                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2289                 }
2290
2291                 if (!path[i].p_idx) {
2292                         /* this level hasn't been touched yet */
2293                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2294                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2295                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2296                                   path[i].p_hdr,
2297                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2298                 } else {
2299                         /* we were already here, see at next index */
2300                         path[i].p_idx--;
2301                 }
2302
2303                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2304                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2305                                 path[i].p_idx);
2306                 if (ext4_ext_more_to_rm(path + i)) {
2307                         struct buffer_head *bh;
2308                         /* go to the next level */
2309                         ext_debug("move to level %d (block %llu)\n",
2310                                   i + 1, idx_pblock(path[i].p_idx));
2311                         memset(path + i + 1, 0, sizeof(*path));
2312                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2313                         if (!bh) {
2314                                 /* should we reset i_size? */
2315                                 err = -EIO;
2316                                 break;
2317                         }
2318                         if (WARN_ON(i + 1 > depth)) {
2319                                 err = -EIO;
2320                                 break;
2321                         }
2322                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2323                                                         depth - i - 1)) {
2324                                 err = -EIO;
2325                                 break;
2326                         }
2327                         path[i + 1].p_bh = bh;
2328
2329                         /* save actual number of indexes since this
2330                          * number is changed at the next iteration */
2331                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2332                         i++;
2333                 } else {
2334                         /* we finished processing this index, go up */
2335                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2336                                 /* index is empty, remove it;
2337                                  * handle must be already prepared by the
2338                                  * truncatei_leaf() */
2339                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2340                         }
2341                         /* root level has p_bh == NULL, brelse() eats this */
2342                         brelse(path[i].p_bh);
2343                         path[i].p_bh = NULL;
2344                         i--;
2345                         ext_debug("return to level %d\n", i);
2346                 }
2347         }
2348
2349         /* TODO: flexible tree reduction should be here */
2350         if (path->p_hdr->eh_entries == 0) {
2351                 /*
2352                  * truncate to zero freed all the tree,
2353                  * so we need to correct eh_depth
2354                  */
2355                 err = ext4_ext_get_access(handle, inode, path);
2356                 if (err == 0) {
2357                         ext_inode_hdr(inode)->eh_depth = 0;
2358                         ext_inode_hdr(inode)->eh_max =
2359                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2360                         err = ext4_ext_dirty(handle, inode, path);
2361                 }
2362         }
2363 out:
2364         ext4_ext_drop_refs(path);
2365         kfree(path);
2366         ext4_journal_stop(handle);
2367
2368         return err;
2369 }
2370
2371 /*
2372  * called at mount time
2373  */
2374 void ext4_ext_init(struct super_block *sb)
2375 {
2376         /*
2377          * possible initialization would be here
2378          */
2379
2380         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2381                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2382 #ifdef AGGRESSIVE_TEST
2383                 printk(", aggressive tests");
2384 #endif
2385 #ifdef CHECK_BINSEARCH
2386                 printk(", check binsearch");
2387 #endif
2388 #ifdef EXTENTS_STATS
2389                 printk(", stats");
2390 #endif
2391                 printk("\n");
2392 #ifdef EXTENTS_STATS
2393                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2394                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2395                 EXT4_SB(sb)->s_ext_max = 0;
2396 #endif
2397         }
2398 }
2399
2400 /*
2401  * called at umount time
2402  */
2403 void ext4_ext_release(struct super_block *sb)
2404 {
2405         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2406                 return;
2407
2408 #ifdef EXTENTS_STATS
2409         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2410                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2411                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2412                         sbi->s_ext_blocks, sbi->s_ext_extents,
2413                         sbi->s_ext_blocks / sbi->s_ext_extents);
2414                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2415                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2416         }
2417 #endif
2418 }
2419
2420 static void bi_complete(struct bio *bio, int error)
2421 {
2422         complete((struct completion *)bio->bi_private);
2423 }
2424
2425 /* FIXME!! we need to try to merge to left or right after zero-out  */
2426 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2427 {
2428         int ret = -EIO;
2429         struct bio *bio;
2430         int blkbits, blocksize;
2431         sector_t ee_pblock;
2432         struct completion event;
2433         unsigned int ee_len, len, done, offset;
2434
2435
2436         blkbits   = inode->i_blkbits;
2437         blocksize = inode->i_sb->s_blocksize;
2438         ee_len    = ext4_ext_get_actual_len(ex);
2439         ee_pblock = ext_pblock(ex);
2440
2441         /* convert ee_pblock to 512 byte sectors */
2442         ee_pblock = ee_pblock << (blkbits - 9);
2443
2444         while (ee_len > 0) {
2445
2446                 if (ee_len > BIO_MAX_PAGES)
2447                         len = BIO_MAX_PAGES;
2448                 else
2449                         len = ee_len;
2450
2451                 bio = bio_alloc(GFP_NOIO, len);
2452                 bio->bi_sector = ee_pblock;
2453                 bio->bi_bdev   = inode->i_sb->s_bdev;
2454
2455                 done = 0;
2456                 offset = 0;
2457                 while (done < len) {
2458                         ret = bio_add_page(bio, ZERO_PAGE(0),
2459                                                         blocksize, offset);
2460                         if (ret != blocksize) {
2461                                 /*
2462                                  * We can't add any more pages because of
2463                                  * hardware limitations.  Start a new bio.
2464                                  */
2465                                 break;
2466                         }
2467                         done++;
2468                         offset += blocksize;
2469                         if (offset >= PAGE_CACHE_SIZE)
2470                                 offset = 0;
2471                 }
2472
2473                 init_completion(&event);
2474                 bio->bi_private = &event;
2475                 bio->bi_end_io = bi_complete;
2476                 submit_bio(WRITE, bio);
2477                 wait_for_completion(&event);
2478
2479                 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2480                         ret = 0;
2481                 else {
2482                         ret = -EIO;
2483                         break;
2484                 }
2485                 bio_put(bio);
2486                 ee_len    -= done;
2487                 ee_pblock += done  << (blkbits - 9);
2488         }
2489         return ret;
2490 }
2491
2492 #define EXT4_EXT_ZERO_LEN 7
2493
2494 /*
2495  * This function is called by ext4_ext_get_blocks() if someone tries to write
2496  * to an uninitialized extent. It may result in splitting the uninitialized
2497  * extent into multiple extents (upto three - one initialized and two
2498  * uninitialized).
2499  * There are three possibilities:
2500  *   a> There is no split required: Entire extent should be initialized
2501  *   b> Splits in two extents: Write is happening at either end of the extent
2502  *   c> Splits in three extents: Somone is writing in middle of the extent
2503  */
2504 static int ext4_ext_convert_to_initialized(handle_t *handle,
2505                                                 struct inode *inode,
2506                                                 struct ext4_ext_path *path,
2507                                                 ext4_lblk_t iblock,
2508                                                 unsigned int max_blocks)
2509 {
2510         struct ext4_extent *ex, newex, orig_ex;
2511         struct ext4_extent *ex1 = NULL;
2512         struct ext4_extent *ex2 = NULL;
2513         struct ext4_extent *ex3 = NULL;
2514         struct ext4_extent_header *eh;
2515         ext4_lblk_t ee_block;
2516         unsigned int allocated, ee_len, depth;
2517         ext4_fsblk_t newblock;
2518         int err = 0;
2519         int ret = 0;
2520
2521         depth = ext_depth(inode);
2522         eh = path[depth].p_hdr;
2523         ex = path[depth].p_ext;
2524         ee_block = le32_to_cpu(ex->ee_block);
2525         ee_len = ext4_ext_get_actual_len(ex);
2526         allocated = ee_len - (iblock - ee_block);
2527         newblock = iblock - ee_block + ext_pblock(ex);
2528         ex2 = ex;
2529         orig_ex.ee_block = ex->ee_block;
2530         orig_ex.ee_len   = cpu_to_le16(ee_len);
2531         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2532
2533         err = ext4_ext_get_access(handle, inode, path + depth);
2534         if (err)
2535                 goto out;
2536         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2537         if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2538                 err =  ext4_ext_zeroout(inode, &orig_ex);
2539                 if (err)
2540                         goto fix_extent_len;
2541                 /* update the extent length and mark as initialized */
2542                 ex->ee_block = orig_ex.ee_block;
2543                 ex->ee_len   = orig_ex.ee_len;
2544                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2545                 ext4_ext_dirty(handle, inode, path + depth);
2546                 /* zeroed the full extent */
2547                 return allocated;
2548         }
2549
2550         /* ex1: ee_block to iblock - 1 : uninitialized */
2551         if (iblock > ee_block) {
2552                 ex1 = ex;
2553                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2554                 ext4_ext_mark_uninitialized(ex1);
2555                 ex2 = &newex;
2556         }
2557         /*
2558          * for sanity, update the length of the ex2 extent before
2559          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2560          * overlap of blocks.
2561          */
2562         if (!ex1 && allocated > max_blocks)
2563                 ex2->ee_len = cpu_to_le16(max_blocks);
2564         /* ex3: to ee_block + ee_len : uninitialised */
2565         if (allocated > max_blocks) {
2566                 unsigned int newdepth;
2567                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2568                 if (allocated <= EXT4_EXT_ZERO_LEN) {
2569                         /*
2570                          * iblock == ee_block is handled by the zerouout
2571                          * at the beginning.
2572                          * Mark first half uninitialized.
2573                          * Mark second half initialized and zero out the
2574                          * initialized extent
2575                          */
2576                         ex->ee_block = orig_ex.ee_block;
2577                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2578                         ext4_ext_mark_uninitialized(ex);
2579                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2580                         ext4_ext_dirty(handle, inode, path + depth);
2581
2582                         ex3 = &newex;
2583                         ex3->ee_block = cpu_to_le32(iblock);
2584                         ext4_ext_store_pblock(ex3, newblock);
2585                         ex3->ee_len = cpu_to_le16(allocated);
2586                         err = ext4_ext_insert_extent(handle, inode, path, ex3);
2587                         if (err == -ENOSPC) {
2588                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2589                                 if (err)
2590                                         goto fix_extent_len;
2591                                 ex->ee_block = orig_ex.ee_block;
2592                                 ex->ee_len   = orig_ex.ee_len;
2593                                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2594                                 ext4_ext_dirty(handle, inode, path + depth);
2595                                 /* blocks available from iblock */
2596                                 return allocated;
2597
2598                         } else if (err)
2599                                 goto fix_extent_len;
2600
2601                         /*
2602                          * We need to zero out the second half because
2603                          * an fallocate request can update file size and
2604                          * converting the second half to initialized extent
2605                          * implies that we can leak some junk data to user
2606                          * space.
2607                          */
2608                         err =  ext4_ext_zeroout(inode, ex3);
2609                         if (err) {
2610                                 /*
2611                                  * We should actually mark the
2612                                  * second half as uninit and return error
2613                                  * Insert would have changed the extent
2614                                  */
2615                                 depth = ext_depth(inode);
2616                                 ext4_ext_drop_refs(path);
2617                                 path = ext4_ext_find_extent(inode,
2618                                                                 iblock, path);
2619                                 if (IS_ERR(path)) {
2620                                         err = PTR_ERR(path);
2621                                         return err;
2622                                 }
2623                                 /* get the second half extent details */
2624                                 ex = path[depth].p_ext;
2625                                 err = ext4_ext_get_access(handle, inode,
2626                                                                 path + depth);
2627                                 if (err)
2628                                         return err;
2629                                 ext4_ext_mark_uninitialized(ex);
2630                                 ext4_ext_dirty(handle, inode, path + depth);
2631                                 return err;
2632                         }
2633
2634                         /* zeroed the second half */
2635                         return allocated;
2636                 }
2637                 ex3 = &newex;
2638                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2639                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2640                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2641                 ext4_ext_mark_uninitialized(ex3);
2642                 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2643                 if (err == -ENOSPC) {
2644                         err =  ext4_ext_zeroout(inode, &orig_ex);
2645                         if (err)
2646                                 goto fix_extent_len;
2647                         /* update the extent length and mark as initialized */
2648                         ex->ee_block = orig_ex.ee_block;
2649                         ex->ee_len   = orig_ex.ee_len;
2650                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2651                         ext4_ext_dirty(handle, inode, path + depth);
2652                         /* zeroed the full extent */
2653                         /* blocks available from iblock */
2654                         return allocated;
2655
2656                 } else if (err)
2657                         goto fix_extent_len;
2658                 /*
2659                  * The depth, and hence eh & ex might change
2660                  * as part of the insert above.
2661                  */
2662                 newdepth = ext_depth(inode);
2663                 /*
2664                  * update the extent length after successful insert of the
2665                  * split extent
2666                  */
2667                 orig_ex.ee_len = cpu_to_le16(ee_len -
2668                                                 ext4_ext_get_actual_len(ex3));
2669                 depth = newdepth;
2670                 ext4_ext_drop_refs(path);
2671                 path = ext4_ext_find_extent(inode, iblock, path);
2672                 if (IS_ERR(path)) {
2673                         err = PTR_ERR(path);
2674                         goto out;
2675                 }
2676                 eh = path[depth].p_hdr;
2677                 ex = path[depth].p_ext;
2678                 if (ex2 != &newex)
2679                         ex2 = ex;
2680
2681                 err = ext4_ext_get_access(handle, inode, path + depth);
2682                 if (err)
2683                         goto out;
2684
2685                 allocated = max_blocks;
2686
2687                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2688                  * to insert a extent in the middle zerout directly
2689                  * otherwise give the extent a chance to merge to left
2690                  */
2691                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2692                                                         iblock != ee_block) {
2693                         err =  ext4_ext_zeroout(inode, &orig_ex);
2694                         if (err)
2695                                 goto fix_extent_len;
2696                         /* update the extent length and mark as initialized */
2697                         ex->ee_block = orig_ex.ee_block;
2698                         ex->ee_len   = orig_ex.ee_len;
2699                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2700                         ext4_ext_dirty(handle, inode, path + depth);
2701                         /* zero out the first half */
2702                         /* blocks available from iblock */
2703                         return allocated;
2704                 }
2705         }
2706         /*
2707          * If there was a change of depth as part of the
2708          * insertion of ex3 above, we need to update the length
2709          * of the ex1 extent again here
2710          */
2711         if (ex1 && ex1 != ex) {
2712                 ex1 = ex;
2713                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2714                 ext4_ext_mark_uninitialized(ex1);
2715                 ex2 = &newex;
2716         }
2717         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2718         ex2->ee_block = cpu_to_le32(iblock);
2719         ext4_ext_store_pblock(ex2, newblock);
2720         ex2->ee_len = cpu_to_le16(allocated);
2721         if (ex2 != ex)
2722                 goto insert;
2723         /*
2724          * New (initialized) extent starts from the first block
2725          * in the current extent. i.e., ex2 == ex
2726          * We have to see if it can be merged with the extent
2727          * on the left.
2728          */
2729         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2730                 /*
2731                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2732                  * since it merges towards right _only_.
2733                  */
2734                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2735                 if (ret) {
2736                         err = ext4_ext_correct_indexes(handle, inode, path);
2737                         if (err)
2738                                 goto out;
2739                         depth = ext_depth(inode);
2740                         ex2--;
2741                 }
2742         }
2743         /*
2744          * Try to Merge towards right. This might be required
2745          * only when the whole extent is being written to.
2746          * i.e. ex2 == ex and ex3 == NULL.
2747          */
2748         if (!ex3) {
2749                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2750                 if (ret) {
2751                         err = ext4_ext_correct_indexes(handle, inode, path);
2752                         if (err)
2753                                 goto out;
2754                 }
2755         }
2756         /* Mark modified extent as dirty */
2757         err = ext4_ext_dirty(handle, inode, path + depth);
2758         goto out;
2759 insert:
2760         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2761         if (err == -ENOSPC) {
2762                 err =  ext4_ext_zeroout(inode, &orig_ex);
2763                 if (err)
2764                         goto fix_extent_len;
2765                 /* update the extent length and mark as initialized */
2766                 ex->ee_block = orig_ex.ee_block;
2767                 ex->ee_len   = orig_ex.ee_len;
2768                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2769                 ext4_ext_dirty(handle, inode, path + depth);
2770                 /* zero out the first half */
2771                 return allocated;
2772         } else if (err)
2773                 goto fix_extent_len;
2774 out:
2775         ext4_ext_show_leaf(inode, path);
2776         return err ? err : allocated;
2777
2778 fix_extent_len:
2779         ex->ee_block = orig_ex.ee_block;
2780         ex->ee_len   = orig_ex.ee_len;
2781         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2782         ext4_ext_mark_uninitialized(ex);
2783         ext4_ext_dirty(handle, inode, path + depth);
2784         return err;
2785 }
2786
2787 /*
2788  * Block allocation/map/preallocation routine for extents based files
2789  *
2790  *
2791  * Need to be called with
2792  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2793  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2794  *
2795  * return > 0, number of of blocks already mapped/allocated
2796  *          if create == 0 and these are pre-allocated blocks
2797  *              buffer head is unmapped
2798  *          otherwise blocks are mapped
2799  *
2800  * return = 0, if plain look up failed (blocks have not been allocated)
2801  *          buffer head is unmapped
2802  *
2803  * return < 0, error case.
2804  */
2805 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2806                         ext4_lblk_t iblock,
2807                         unsigned int max_blocks, struct buffer_head *bh_result,
2808                         int flags)
2809 {
2810         struct ext4_ext_path *path = NULL;
2811         struct ext4_extent_header *eh;
2812         struct ext4_extent newex, *ex;
2813         ext4_fsblk_t newblock;
2814         int err = 0, depth, ret, cache_type;
2815         unsigned int allocated = 0;
2816         struct ext4_allocation_request ar;
2817
2818         __clear_bit(BH_New, &bh_result->b_state);
2819         ext_debug("blocks %u/%u requested for inode %lu\n",
2820                         iblock, max_blocks, inode->i_ino);
2821
2822         /* check in cache */
2823         cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2824         if (cache_type) {
2825                 if (cache_type == EXT4_EXT_CACHE_GAP) {
2826                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2827                                 /*
2828                                  * block isn't allocated yet and
2829                                  * user doesn't want to allocate it
2830                                  */
2831                                 goto out2;
2832                         }
2833                         /* we should allocate requested block */
2834                 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
2835                         /* block is already allocated */
2836                         newblock = iblock
2837                                    - le32_to_cpu(newex.ee_block)
2838                                    + ext_pblock(&newex);
2839                         /* number of remaining blocks in the extent */
2840                         allocated = ext4_ext_get_actual_len(&newex) -
2841                                         (iblock - le32_to_cpu(newex.ee_block));
2842                         goto out;
2843                 } else {
2844                         BUG();
2845                 }
2846         }
2847
2848         /* find extent for this block */
2849         path = ext4_ext_find_extent(inode, iblock, NULL);
2850         if (IS_ERR(path)) {
2851                 err = PTR_ERR(path);
2852                 path = NULL;
2853                 goto out2;
2854         }
2855
2856         depth = ext_depth(inode);
2857
2858         /*
2859          * consistent leaf must not be empty;
2860          * this situation is possible, though, _during_ tree modification;
2861          * this is why assert can't be put in ext4_ext_find_extent()
2862          */
2863         BUG_ON(path[depth].p_ext == NULL && depth != 0);
2864         eh = path[depth].p_hdr;
2865
2866         ex = path[depth].p_ext;
2867         if (ex) {
2868                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2869                 ext4_fsblk_t ee_start = ext_pblock(ex);
2870                 unsigned short ee_len;
2871
2872                 /*
2873                  * Uninitialized extents are treated as holes, except that
2874                  * we split out initialized portions during a write.
2875                  */
2876                 ee_len = ext4_ext_get_actual_len(ex);
2877                 /* if found extent covers block, simply return it */
2878                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2879                         newblock = iblock - ee_block + ee_start;
2880                         /* number of remaining blocks in the extent */
2881                         allocated = ee_len - (iblock - ee_block);
2882                         ext_debug("%u fit into %u:%d -> %llu\n", iblock,
2883                                         ee_block, ee_len, newblock);
2884
2885                         /* Do not put uninitialized extent in the cache */
2886                         if (!ext4_ext_is_uninitialized(ex)) {
2887                                 ext4_ext_put_in_cache(inode, ee_block,
2888                                                         ee_len, ee_start,
2889                                                         EXT4_EXT_CACHE_EXTENT);
2890                                 goto out;
2891                         }
2892                         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
2893                                 goto out;
2894                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2895                                 if (allocated > max_blocks)
2896                                         allocated = max_blocks;
2897                                 /*
2898                                  * We have blocks reserved already.  We
2899                                  * return allocated blocks so that delalloc
2900                                  * won't do block reservation for us.  But
2901                                  * the buffer head will be unmapped so that
2902                                  * a read from the block returns 0s.
2903                                  */
2904                                 set_buffer_unwritten(bh_result);
2905                                 bh_result->b_bdev = inode->i_sb->s_bdev;
2906                                 bh_result->b_blocknr = newblock;
2907                                 goto out2;
2908                         }
2909
2910                         ret = ext4_ext_convert_to_initialized(handle, inode,
2911                                                                 path, iblock,
2912                                                                 max_blocks);
2913                         if (ret <= 0) {
2914                                 err = ret;
2915                                 goto out2;
2916                         } else
2917                                 allocated = ret;
2918                         goto outnew;
2919                 }
2920         }
2921
2922         /*
2923          * requested block isn't allocated yet;
2924          * we couldn't try to create block if create flag is zero
2925          */
2926         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2927                 /*
2928                  * put just found gap into cache to speed up
2929                  * subsequent requests
2930                  */
2931                 ext4_ext_put_gap_in_cache(inode, path, iblock);
2932                 goto out2;
2933         }
2934         /*
2935          * Okay, we need to do block allocation.
2936          */
2937
2938         /* find neighbour allocated blocks */
2939         ar.lleft = iblock;
2940         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2941         if (err)
2942                 goto out2;
2943         ar.lright = iblock;
2944         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2945         if (err)
2946                 goto out2;
2947
2948         /*
2949          * See if request is beyond maximum number of blocks we can have in
2950          * a single extent. For an initialized extent this limit is
2951          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2952          * EXT_UNINIT_MAX_LEN.
2953          */
2954         if (max_blocks > EXT_INIT_MAX_LEN &&
2955             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
2956                 max_blocks = EXT_INIT_MAX_LEN;
2957         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2958                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
2959                 max_blocks = EXT_UNINIT_MAX_LEN;
2960
2961         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2962         newex.ee_block = cpu_to_le32(iblock);
2963         newex.ee_len = cpu_to_le16(max_blocks);
2964         err = ext4_ext_check_overlap(inode, &newex, path);
2965         if (err)
2966                 allocated = ext4_ext_get_actual_len(&newex);
2967         else
2968                 allocated = max_blocks;
2969
2970         /* allocate new block */
2971         ar.inode = inode;
2972         ar.goal = ext4_ext_find_goal(inode, path, iblock);
2973         ar.logical = iblock;
2974         ar.len = allocated;
2975         if (S_ISREG(inode->i_mode))
2976                 ar.flags = EXT4_MB_HINT_DATA;
2977         else
2978                 /* disable in-core preallocation for non-regular files */
2979                 ar.flags = 0;
2980         newblock = ext4_mb_new_blocks(handle, &ar, &err);
2981         if (!newblock)
2982                 goto out2;
2983         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
2984                   ar.goal, newblock, allocated);
2985
2986         /* try to insert new extent into found leaf and return */
2987         ext4_ext_store_pblock(&newex, newblock);
2988         newex.ee_len = cpu_to_le16(ar.len);
2989         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)  /* Mark uninitialized */
2990                 ext4_ext_mark_uninitialized(&newex);
2991         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2992         if (err) {
2993                 /* free data blocks we just allocated */
2994                 /* not a good idea to call discard here directly,
2995                  * but otherwise we'd need to call it every free() */
2996                 ext4_discard_preallocations(inode);
2997                 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2998                                         ext4_ext_get_actual_len(&newex), 0);
2999                 goto out2;
3000         }
3001
3002         /* previous routine could use block we allocated */
3003         newblock = ext_pblock(&newex);
3004         allocated = ext4_ext_get_actual_len(&newex);
3005 outnew:
3006         set_buffer_new(bh_result);
3007
3008         /* Cache only when it is _not_ an uninitialized extent */
3009         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
3010                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
3011                                                 EXT4_EXT_CACHE_EXTENT);
3012 out:
3013         if (allocated > max_blocks)
3014                 allocated = max_blocks;
3015         ext4_ext_show_leaf(inode, path);
3016         set_buffer_mapped(bh_result);
3017         bh_result->b_bdev = inode->i_sb->s_bdev;
3018         bh_result->b_blocknr = newblock;
3019 out2:
3020         if (path) {
3021                 ext4_ext_drop_refs(path);
3022                 kfree(path);
3023         }
3024         return err ? err : allocated;
3025 }
3026
3027 void ext4_ext_truncate(struct inode *inode)
3028 {
3029         struct address_space *mapping = inode->i_mapping;
3030         struct super_block *sb = inode->i_sb;
3031         ext4_lblk_t last_block;
3032         handle_t *handle;
3033         int err = 0;
3034
3035         /*
3036          * probably first extent we're gonna free will be last in block
3037          */
3038         err = ext4_writepage_trans_blocks(inode);
3039         handle = ext4_journal_start(inode, err);
3040         if (IS_ERR(handle))
3041                 return;
3042
3043         if (inode->i_size & (sb->s_blocksize - 1))
3044                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3045
3046         if (ext4_orphan_add(handle, inode))
3047                 goto out_stop;
3048
3049         down_write(&EXT4_I(inode)->i_data_sem);
3050         ext4_ext_invalidate_cache(inode);
3051
3052         ext4_discard_preallocations(inode);
3053
3054         /*
3055          * TODO: optimization is possible here.
3056          * Probably we need not scan at all,
3057          * because page truncation is enough.
3058          */
3059
3060         /* we have to know where to truncate from in crash case */
3061         EXT4_I(inode)->i_disksize = inode->i_size;
3062         ext4_mark_inode_dirty(handle, inode);
3063
3064         last_block = (inode->i_size + sb->s_blocksize - 1)
3065                         >> EXT4_BLOCK_SIZE_BITS(sb);
3066         err = ext4_ext_remove_space(inode, last_block);
3067
3068         /* In a multi-transaction truncate, we only make the final
3069          * transaction synchronous.
3070          */
3071         if (IS_SYNC(inode))
3072                 ext4_handle_sync(handle);
3073
3074 out_stop:
3075         up_write(&EXT4_I(inode)->i_data_sem);
3076         /*
3077          * If this was a simple ftruncate() and the file will remain alive,
3078          * then we need to clear up the orphan record which we created above.
3079          * However, if this was a real unlink then we were called by
3080          * ext4_delete_inode(), and we allow that function to clean up the
3081          * orphan info for us.
3082          */
3083         if (inode->i_nlink)
3084                 ext4_orphan_del(handle, inode);
3085
3086         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3087         ext4_mark_inode_dirty(handle, inode);
3088         ext4_journal_stop(handle);
3089 }
3090
3091 static void ext4_falloc_update_inode(struct inode *inode,
3092                                 int mode, loff_t new_size, int update_ctime)
3093 {
3094         struct timespec now;
3095
3096         if (update_ctime) {
3097                 now = current_fs_time(inode->i_sb);
3098                 if (!timespec_equal(&inode->i_ctime, &now))
3099                         inode->i_ctime = now;
3100         }
3101         /*
3102          * Update only when preallocation was requested beyond
3103          * the file size.
3104          */
3105         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3106                 if (new_size > i_size_read(inode))
3107                         i_size_write(inode, new_size);
3108                 if (new_size > EXT4_I(inode)->i_disksize)
3109                         ext4_update_i_disksize(inode, new_size);
3110         }
3111
3112 }
3113
3114 /*
3115  * preallocate space for a file. This implements ext4's fallocate inode
3116  * operation, which gets called from sys_fallocate system call.
3117  * For block-mapped files, posix_fallocate should fall back to the method
3118  * of writing zeroes to the required new blocks (the same behavior which is
3119  * expected for file systems which do not support fallocate() system call).
3120  */
3121 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3122 {
3123         handle_t *handle;
3124         ext4_lblk_t block;
3125         loff_t new_size;
3126         unsigned int max_blocks;
3127         int ret = 0;
3128         int ret2 = 0;
3129         int retries = 0;
3130         struct buffer_head map_bh;
3131         unsigned int credits, blkbits = inode->i_blkbits;
3132
3133         /*
3134          * currently supporting (pre)allocate mode for extent-based
3135          * files _only_
3136          */
3137         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3138                 return -EOPNOTSUPP;
3139
3140         /* preallocation to directories is currently not supported */
3141         if (S_ISDIR(inode->i_mode))
3142                 return -ENODEV;
3143
3144         block = offset >> blkbits;
3145         /*
3146          * We can't just convert len to max_blocks because
3147          * If blocksize = 4096 offset = 3072 and len = 2048
3148          */
3149         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3150                                                         - block;
3151         /*
3152          * credits to insert 1 extent into extent tree
3153          */
3154         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3155         mutex_lock(&inode->i_mutex);
3156 retry:
3157         while (ret >= 0 && ret < max_blocks) {
3158                 block = block + ret;
3159                 max_blocks = max_blocks - ret;
3160                 handle = ext4_journal_start(inode, credits);
3161                 if (IS_ERR(handle)) {
3162                         ret = PTR_ERR(handle);
3163                         break;
3164                 }
3165                 map_bh.b_state = 0;
3166                 ret = ext4_get_blocks(handle, inode, block,
3167                                       max_blocks, &map_bh,
3168                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3169                 if (ret <= 0) {
3170 #ifdef EXT4FS_DEBUG
3171                         WARN_ON(ret <= 0);
3172                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3173                                     "returned error inode#%lu, block=%u, "
3174                                     "max_blocks=%u", __func__,
3175                                     inode->i_ino, block, max_blocks);
3176 #endif
3177                         ext4_mark_inode_dirty(handle, inode);
3178                         ret2 = ext4_journal_stop(handle);
3179                         break;
3180                 }
3181                 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3182                                                 blkbits) >> blkbits))
3183                         new_size = offset + len;
3184                 else
3185                         new_size = (block + ret) << blkbits;
3186
3187                 ext4_falloc_update_inode(inode, mode, new_size,
3188                                                 buffer_new(&map_bh));
3189                 ext4_mark_inode_dirty(handle, inode);
3190                 ret2 = ext4_journal_stop(handle);
3191                 if (ret2)
3192                         break;
3193         }
3194         if (ret == -ENOSPC &&
3195                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3196                 ret = 0;
3197                 goto retry;
3198         }
3199         mutex_unlock(&inode->i_mutex);
3200         return ret > 0 ? ret2 : ret;
3201 }
3202
3203 /*
3204  * Callback function called for each extent to gather FIEMAP information.
3205  */
3206 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3207                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3208                        void *data)
3209 {
3210         struct fiemap_extent_info *fieinfo = data;
3211         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3212         __u64   logical;
3213         __u64   physical;
3214         __u64   length;
3215         __u32   flags = 0;
3216         int     error;
3217
3218         logical =  (__u64)newex->ec_block << blksize_bits;
3219
3220         if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3221                 pgoff_t offset;
3222                 struct page *page;
3223                 struct buffer_head *bh = NULL;
3224
3225                 offset = logical >> PAGE_SHIFT;
3226                 page = find_get_page(inode->i_mapping, offset);
3227                 if (!page || !page_has_buffers(page))
3228                         return EXT_CONTINUE;
3229
3230                 bh = page_buffers(page);
3231
3232                 if (!bh)
3233                         return EXT_CONTINUE;
3234
3235                 if (buffer_delay(bh)) {
3236                         flags |= FIEMAP_EXTENT_DELALLOC;
3237                         page_cache_release(page);
3238                 } else {
3239                         page_cache_release(page);
3240                         return EXT_CONTINUE;
3241                 }
3242         }
3243
3244         physical = (__u64)newex->ec_start << blksize_bits;
3245         length =   (__u64)newex->ec_len << blksize_bits;
3246
3247         if (ex && ext4_ext_is_uninitialized(ex))
3248                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3249
3250         /*
3251          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3252          *
3253          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3254          * this also indicates no more allocated blocks.
3255          *
3256          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3257          */
3258         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3259             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3260                 loff_t size = i_size_read(inode);
3261                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3262
3263                 flags |= FIEMAP_EXTENT_LAST;
3264                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3265                     logical+length > size)
3266                         length = (size - logical + bs - 1) & ~(bs-1);
3267         }
3268
3269         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3270                                         length, flags);
3271         if (error < 0)
3272                 return error;
3273         if (error == 1)
3274                 return EXT_BREAK;
3275
3276         return EXT_CONTINUE;
3277 }
3278
3279 /* fiemap flags we can handle specified here */
3280 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3281
3282 static int ext4_xattr_fiemap(struct inode *inode,
3283                                 struct fiemap_extent_info *fieinfo)
3284 {
3285         __u64 physical = 0;
3286         __u64 length;
3287         __u32 flags = FIEMAP_EXTENT_LAST;
3288         int blockbits = inode->i_sb->s_blocksize_bits;
3289         int error = 0;
3290
3291         /* in-inode? */
3292         if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3293                 struct ext4_iloc iloc;
3294                 int offset;     /* offset of xattr in inode */
3295
3296                 error = ext4_get_inode_loc(inode, &iloc);
3297                 if (error)
3298                         return error;
3299                 physical = iloc.bh->b_blocknr << blockbits;
3300                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3301                                 EXT4_I(inode)->i_extra_isize;
3302                 physical += offset;
3303                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3304                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3305         } else { /* external block */
3306                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3307                 length = inode->i_sb->s_blocksize;
3308         }
3309
3310         if (physical)
3311                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3312                                                 length, flags);
3313         return (error < 0 ? error : 0);
3314 }
3315
3316 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3317                 __u64 start, __u64 len)
3318 {
3319         ext4_lblk_t start_blk;
3320         ext4_lblk_t len_blks;
3321         int error = 0;
3322
3323         /* fallback to generic here if not in extents fmt */
3324         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3325                 return generic_block_fiemap(inode, fieinfo, start, len,
3326                         ext4_get_block);
3327
3328         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3329                 return -EBADR;
3330
3331         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3332                 error = ext4_xattr_fiemap(inode, fieinfo);
3333         } else {
3334                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3335                 len_blks = len >> inode->i_sb->s_blocksize_bits;
3336
3337                 /*
3338                  * Walk the extent tree gathering extent information.
3339                  * ext4_ext_fiemap_cb will push extents back to user.
3340                  */
3341                 down_read(&EXT4_I(inode)->i_data_sem);
3342                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3343                                           ext4_ext_fiemap_cb, fieinfo);
3344                 up_read(&EXT4_I(inode)->i_data_sem);
3345         }
3346
3347         return error;
3348 }
3349