Merge branch 'next' into upstream-merge
[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 static int ext4_ext_truncate_extend_restart(handle_t *handle,
48                                             struct inode *inode,
49                                             int needed)
50 {
51         int err;
52
53         if (!ext4_handle_valid(handle))
54                 return 0;
55         if (handle->h_buffer_credits > needed)
56                 return 0;
57         err = ext4_journal_extend(handle, needed);
58         if (err <= 0)
59                 return err;
60         err = ext4_truncate_restart_trans(handle, inode, needed);
61         if (err == 0)
62                 err = -EAGAIN;
63
64         return err;
65 }
66
67 /*
68  * could return:
69  *  - EROFS
70  *  - ENOMEM
71  */
72 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
73                                 struct ext4_ext_path *path)
74 {
75         if (path->p_bh) {
76                 /* path points to block */
77                 return ext4_journal_get_write_access(handle, path->p_bh);
78         }
79         /* path points to leaf/index in inode body */
80         /* we use in-core data, no need to protect them */
81         return 0;
82 }
83
84 /*
85  * could return:
86  *  - EROFS
87  *  - ENOMEM
88  *  - EIO
89  */
90 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
91                                 struct ext4_ext_path *path)
92 {
93         int err;
94         if (path->p_bh) {
95                 /* path points to block */
96                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
97         } else {
98                 /* path points to leaf/index in inode body */
99                 err = ext4_mark_inode_dirty(handle, inode);
100         }
101         return err;
102 }
103
104 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
105                               struct ext4_ext_path *path,
106                               ext4_lblk_t block)
107 {
108         struct ext4_inode_info *ei = EXT4_I(inode);
109         ext4_fsblk_t bg_start;
110         ext4_fsblk_t last_block;
111         ext4_grpblk_t colour;
112         ext4_group_t block_group;
113         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
114         int depth;
115
116         if (path) {
117                 struct ext4_extent *ex;
118                 depth = path->p_depth;
119
120                 /* try to predict block placement */
121                 ex = path[depth].p_ext;
122                 if (ex)
123                         return (ext4_ext_pblock(ex) +
124                                 (block - le32_to_cpu(ex->ee_block)));
125
126                 /* it looks like index is empty;
127                  * try to find starting block from index itself */
128                 if (path[depth].p_bh)
129                         return path[depth].p_bh->b_blocknr;
130         }
131
132         /* OK. use inode's group */
133         block_group = ei->i_block_group;
134         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
135                 /*
136                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
137                  * block groups per flexgroup, reserve the first block
138                  * group for directories and special files.  Regular
139                  * files will start at the second block group.  This
140                  * tends to speed up directory access and improves
141                  * fsck times.
142                  */
143                 block_group &= ~(flex_size-1);
144                 if (S_ISREG(inode->i_mode))
145                         block_group++;
146         }
147         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
148         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
149
150         /*
151          * If we are doing delayed allocation, we don't need take
152          * colour into account.
153          */
154         if (test_opt(inode->i_sb, DELALLOC))
155                 return bg_start;
156
157         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
158                 colour = (current->pid % 16) *
159                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
160         else
161                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
162         return bg_start + colour + block;
163 }
164
165 /*
166  * Allocation for a meta data block
167  */
168 static ext4_fsblk_t
169 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
170                         struct ext4_ext_path *path,
171                         struct ext4_extent *ex, int *err)
172 {
173         ext4_fsblk_t goal, newblock;
174
175         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
176         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
177         return newblock;
178 }
179
180 static inline int ext4_ext_space_block(struct inode *inode, int check)
181 {
182         int size;
183
184         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
185                         / sizeof(struct ext4_extent);
186         if (!check) {
187 #ifdef AGGRESSIVE_TEST
188                 if (size > 6)
189                         size = 6;
190 #endif
191         }
192         return size;
193 }
194
195 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
196 {
197         int size;
198
199         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
200                         / sizeof(struct ext4_extent_idx);
201         if (!check) {
202 #ifdef AGGRESSIVE_TEST
203                 if (size > 5)
204                         size = 5;
205 #endif
206         }
207         return size;
208 }
209
210 static inline int ext4_ext_space_root(struct inode *inode, int check)
211 {
212         int size;
213
214         size = sizeof(EXT4_I(inode)->i_data);
215         size -= sizeof(struct ext4_extent_header);
216         size /= sizeof(struct ext4_extent);
217         if (!check) {
218 #ifdef AGGRESSIVE_TEST
219                 if (size > 3)
220                         size = 3;
221 #endif
222         }
223         return size;
224 }
225
226 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
227 {
228         int size;
229
230         size = sizeof(EXT4_I(inode)->i_data);
231         size -= sizeof(struct ext4_extent_header);
232         size /= sizeof(struct ext4_extent_idx);
233         if (!check) {
234 #ifdef AGGRESSIVE_TEST
235                 if (size > 4)
236                         size = 4;
237 #endif
238         }
239         return size;
240 }
241
242 /*
243  * Calculate the number of metadata blocks needed
244  * to allocate @blocks
245  * Worse case is one block per extent
246  */
247 int ext4_ext_calc_metadata_amount(struct inode *inode, sector_t lblock)
248 {
249         struct ext4_inode_info *ei = EXT4_I(inode);
250         int idxs, num = 0;
251
252         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
253                 / sizeof(struct ext4_extent_idx));
254
255         /*
256          * If the new delayed allocation block is contiguous with the
257          * previous da block, it can share index blocks with the
258          * previous block, so we only need to allocate a new index
259          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
260          * an additional index block, and at ldxs**3 blocks, yet
261          * another index blocks.
262          */
263         if (ei->i_da_metadata_calc_len &&
264             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
265                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
266                         num++;
267                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
268                         num++;
269                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
270                         num++;
271                         ei->i_da_metadata_calc_len = 0;
272                 } else
273                         ei->i_da_metadata_calc_len++;
274                 ei->i_da_metadata_calc_last_lblock++;
275                 return num;
276         }
277
278         /*
279          * In the worst case we need a new set of index blocks at
280          * every level of the inode's extent tree.
281          */
282         ei->i_da_metadata_calc_len = 1;
283         ei->i_da_metadata_calc_last_lblock = lblock;
284         return ext_depth(inode) + 1;
285 }
286
287 static int
288 ext4_ext_max_entries(struct inode *inode, int depth)
289 {
290         int max;
291
292         if (depth == ext_depth(inode)) {
293                 if (depth == 0)
294                         max = ext4_ext_space_root(inode, 1);
295                 else
296                         max = ext4_ext_space_root_idx(inode, 1);
297         } else {
298                 if (depth == 0)
299                         max = ext4_ext_space_block(inode, 1);
300                 else
301                         max = ext4_ext_space_block_idx(inode, 1);
302         }
303
304         return max;
305 }
306
307 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
308 {
309         ext4_fsblk_t block = ext4_ext_pblock(ext);
310         int len = ext4_ext_get_actual_len(ext);
311
312         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
313 }
314
315 static int ext4_valid_extent_idx(struct inode *inode,
316                                 struct ext4_extent_idx *ext_idx)
317 {
318         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
319
320         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
321 }
322
323 static int ext4_valid_extent_entries(struct inode *inode,
324                                 struct ext4_extent_header *eh,
325                                 int depth)
326 {
327         struct ext4_extent *ext;
328         struct ext4_extent_idx *ext_idx;
329         unsigned short entries;
330         if (eh->eh_entries == 0)
331                 return 1;
332
333         entries = le16_to_cpu(eh->eh_entries);
334
335         if (depth == 0) {
336                 /* leaf entries */
337                 ext = EXT_FIRST_EXTENT(eh);
338                 while (entries) {
339                         if (!ext4_valid_extent(inode, ext))
340                                 return 0;
341                         ext++;
342                         entries--;
343                 }
344         } else {
345                 ext_idx = EXT_FIRST_INDEX(eh);
346                 while (entries) {
347                         if (!ext4_valid_extent_idx(inode, ext_idx))
348                                 return 0;
349                         ext_idx++;
350                         entries--;
351                 }
352         }
353         return 1;
354 }
355
356 static int __ext4_ext_check(const char *function, unsigned int line,
357                             struct inode *inode, struct ext4_extent_header *eh,
358                             int depth)
359 {
360         const char *error_msg;
361         int max = 0;
362
363         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
364                 error_msg = "invalid magic";
365                 goto corrupted;
366         }
367         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
368                 error_msg = "unexpected eh_depth";
369                 goto corrupted;
370         }
371         if (unlikely(eh->eh_max == 0)) {
372                 error_msg = "invalid eh_max";
373                 goto corrupted;
374         }
375         max = ext4_ext_max_entries(inode, depth);
376         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
377                 error_msg = "too large eh_max";
378                 goto corrupted;
379         }
380         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
381                 error_msg = "invalid eh_entries";
382                 goto corrupted;
383         }
384         if (!ext4_valid_extent_entries(inode, eh, depth)) {
385                 error_msg = "invalid extent entries";
386                 goto corrupted;
387         }
388         return 0;
389
390 corrupted:
391         ext4_error_inode(inode, function, line, 0,
392                         "bad header/extent: %s - magic %x, "
393                         "entries %u, max %u(%u), depth %u(%u)",
394                         error_msg, le16_to_cpu(eh->eh_magic),
395                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
396                         max, le16_to_cpu(eh->eh_depth), depth);
397
398         return -EIO;
399 }
400
401 #define ext4_ext_check(inode, eh, depth)        \
402         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
403
404 int ext4_ext_check_inode(struct inode *inode)
405 {
406         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
407 }
408
409 #ifdef EXT_DEBUG
410 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
411 {
412         int k, l = path->p_depth;
413
414         ext_debug("path:");
415         for (k = 0; k <= l; k++, path++) {
416                 if (path->p_idx) {
417                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
418                             ext4_idx_pblock(path->p_idx));
419                 } else if (path->p_ext) {
420                         ext_debug("  %d:[%d]%d:%llu ",
421                                   le32_to_cpu(path->p_ext->ee_block),
422                                   ext4_ext_is_uninitialized(path->p_ext),
423                                   ext4_ext_get_actual_len(path->p_ext),
424                                   ext4_ext_pblock(path->p_ext));
425                 } else
426                         ext_debug("  []");
427         }
428         ext_debug("\n");
429 }
430
431 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
432 {
433         int depth = ext_depth(inode);
434         struct ext4_extent_header *eh;
435         struct ext4_extent *ex;
436         int i;
437
438         if (!path)
439                 return;
440
441         eh = path[depth].p_hdr;
442         ex = EXT_FIRST_EXTENT(eh);
443
444         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
445
446         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
447                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
448                           ext4_ext_is_uninitialized(ex),
449                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
450         }
451         ext_debug("\n");
452 }
453 #else
454 #define ext4_ext_show_path(inode, path)
455 #define ext4_ext_show_leaf(inode, path)
456 #endif
457
458 void ext4_ext_drop_refs(struct ext4_ext_path *path)
459 {
460         int depth = path->p_depth;
461         int i;
462
463         for (i = 0; i <= depth; i++, path++)
464                 if (path->p_bh) {
465                         brelse(path->p_bh);
466                         path->p_bh = NULL;
467                 }
468 }
469
470 /*
471  * ext4_ext_binsearch_idx:
472  * binary search for the closest index of the given block
473  * the header must be checked before calling this
474  */
475 static void
476 ext4_ext_binsearch_idx(struct inode *inode,
477                         struct ext4_ext_path *path, ext4_lblk_t block)
478 {
479         struct ext4_extent_header *eh = path->p_hdr;
480         struct ext4_extent_idx *r, *l, *m;
481
482
483         ext_debug("binsearch for %u(idx):  ", block);
484
485         l = EXT_FIRST_INDEX(eh) + 1;
486         r = EXT_LAST_INDEX(eh);
487         while (l <= r) {
488                 m = l + (r - l) / 2;
489                 if (block < le32_to_cpu(m->ei_block))
490                         r = m - 1;
491                 else
492                         l = m + 1;
493                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
494                                 m, le32_to_cpu(m->ei_block),
495                                 r, le32_to_cpu(r->ei_block));
496         }
497
498         path->p_idx = l - 1;
499         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
500                   ext4_idx_pblock(path->p_idx));
501
502 #ifdef CHECK_BINSEARCH
503         {
504                 struct ext4_extent_idx *chix, *ix;
505                 int k;
506
507                 chix = ix = EXT_FIRST_INDEX(eh);
508                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
509                   if (k != 0 &&
510                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
511                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
512                                        "first=0x%p\n", k,
513                                        ix, EXT_FIRST_INDEX(eh));
514                                 printk(KERN_DEBUG "%u <= %u\n",
515                                        le32_to_cpu(ix->ei_block),
516                                        le32_to_cpu(ix[-1].ei_block));
517                         }
518                         BUG_ON(k && le32_to_cpu(ix->ei_block)
519                                            <= le32_to_cpu(ix[-1].ei_block));
520                         if (block < le32_to_cpu(ix->ei_block))
521                                 break;
522                         chix = ix;
523                 }
524                 BUG_ON(chix != path->p_idx);
525         }
526 #endif
527
528 }
529
530 /*
531  * ext4_ext_binsearch:
532  * binary search for closest extent of the given block
533  * the header must be checked before calling this
534  */
535 static void
536 ext4_ext_binsearch(struct inode *inode,
537                 struct ext4_ext_path *path, ext4_lblk_t block)
538 {
539         struct ext4_extent_header *eh = path->p_hdr;
540         struct ext4_extent *r, *l, *m;
541
542         if (eh->eh_entries == 0) {
543                 /*
544                  * this leaf is empty:
545                  * we get such a leaf in split/add case
546                  */
547                 return;
548         }
549
550         ext_debug("binsearch for %u:  ", block);
551
552         l = EXT_FIRST_EXTENT(eh) + 1;
553         r = EXT_LAST_EXTENT(eh);
554
555         while (l <= r) {
556                 m = l + (r - l) / 2;
557                 if (block < le32_to_cpu(m->ee_block))
558                         r = m - 1;
559                 else
560                         l = m + 1;
561                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
562                                 m, le32_to_cpu(m->ee_block),
563                                 r, le32_to_cpu(r->ee_block));
564         }
565
566         path->p_ext = l - 1;
567         ext_debug("  -> %d:%llu:[%d]%d ",
568                         le32_to_cpu(path->p_ext->ee_block),
569                         ext4_ext_pblock(path->p_ext),
570                         ext4_ext_is_uninitialized(path->p_ext),
571                         ext4_ext_get_actual_len(path->p_ext));
572
573 #ifdef CHECK_BINSEARCH
574         {
575                 struct ext4_extent *chex, *ex;
576                 int k;
577
578                 chex = ex = EXT_FIRST_EXTENT(eh);
579                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
580                         BUG_ON(k && le32_to_cpu(ex->ee_block)
581                                           <= le32_to_cpu(ex[-1].ee_block));
582                         if (block < le32_to_cpu(ex->ee_block))
583                                 break;
584                         chex = ex;
585                 }
586                 BUG_ON(chex != path->p_ext);
587         }
588 #endif
589
590 }
591
592 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
593 {
594         struct ext4_extent_header *eh;
595
596         eh = ext_inode_hdr(inode);
597         eh->eh_depth = 0;
598         eh->eh_entries = 0;
599         eh->eh_magic = EXT4_EXT_MAGIC;
600         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
601         ext4_mark_inode_dirty(handle, inode);
602         ext4_ext_invalidate_cache(inode);
603         return 0;
604 }
605
606 struct ext4_ext_path *
607 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
608                                         struct ext4_ext_path *path)
609 {
610         struct ext4_extent_header *eh;
611         struct buffer_head *bh;
612         short int depth, i, ppos = 0, alloc = 0;
613
614         eh = ext_inode_hdr(inode);
615         depth = ext_depth(inode);
616
617         /* account possible depth increase */
618         if (!path) {
619                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
620                                 GFP_NOFS);
621                 if (!path)
622                         return ERR_PTR(-ENOMEM);
623                 alloc = 1;
624         }
625         path[0].p_hdr = eh;
626         path[0].p_bh = NULL;
627
628         i = depth;
629         /* walk through the tree */
630         while (i) {
631                 int need_to_validate = 0;
632
633                 ext_debug("depth %d: num %d, max %d\n",
634                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
635
636                 ext4_ext_binsearch_idx(inode, path + ppos, block);
637                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
638                 path[ppos].p_depth = i;
639                 path[ppos].p_ext = NULL;
640
641                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
642                 if (unlikely(!bh))
643                         goto err;
644                 if (!bh_uptodate_or_lock(bh)) {
645                         if (bh_submit_read(bh) < 0) {
646                                 put_bh(bh);
647                                 goto err;
648                         }
649                         /* validate the extent entries */
650                         need_to_validate = 1;
651                 }
652                 eh = ext_block_hdr(bh);
653                 ppos++;
654                 if (unlikely(ppos > depth)) {
655                         put_bh(bh);
656                         EXT4_ERROR_INODE(inode,
657                                          "ppos %d > depth %d", ppos, depth);
658                         goto err;
659                 }
660                 path[ppos].p_bh = bh;
661                 path[ppos].p_hdr = eh;
662                 i--;
663
664                 if (need_to_validate && ext4_ext_check(inode, eh, i))
665                         goto err;
666         }
667
668         path[ppos].p_depth = i;
669         path[ppos].p_ext = NULL;
670         path[ppos].p_idx = NULL;
671
672         /* find extent */
673         ext4_ext_binsearch(inode, path + ppos, block);
674         /* if not an empty leaf */
675         if (path[ppos].p_ext)
676                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
677
678         ext4_ext_show_path(inode, path);
679
680         return path;
681
682 err:
683         ext4_ext_drop_refs(path);
684         if (alloc)
685                 kfree(path);
686         return ERR_PTR(-EIO);
687 }
688
689 /*
690  * ext4_ext_insert_index:
691  * insert new index [@logical;@ptr] into the block at @curp;
692  * check where to insert: before @curp or after @curp
693  */
694 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
695                                  struct ext4_ext_path *curp,
696                                  int logical, ext4_fsblk_t ptr)
697 {
698         struct ext4_extent_idx *ix;
699         int len, err;
700
701         err = ext4_ext_get_access(handle, inode, curp);
702         if (err)
703                 return err;
704
705         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
706                 EXT4_ERROR_INODE(inode,
707                                  "logical %d == ei_block %d!",
708                                  logical, le32_to_cpu(curp->p_idx->ei_block));
709                 return -EIO;
710         }
711         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
712         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
713                 /* insert after */
714                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
715                         len = (len - 1) * sizeof(struct ext4_extent_idx);
716                         len = len < 0 ? 0 : len;
717                         ext_debug("insert new index %d after: %llu. "
718                                         "move %d from 0x%p to 0x%p\n",
719                                         logical, ptr, len,
720                                         (curp->p_idx + 1), (curp->p_idx + 2));
721                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
722                 }
723                 ix = curp->p_idx + 1;
724         } else {
725                 /* insert before */
726                 len = len * sizeof(struct ext4_extent_idx);
727                 len = len < 0 ? 0 : len;
728                 ext_debug("insert new index %d before: %llu. "
729                                 "move %d from 0x%p to 0x%p\n",
730                                 logical, ptr, len,
731                                 curp->p_idx, (curp->p_idx + 1));
732                 memmove(curp->p_idx + 1, curp->p_idx, len);
733                 ix = curp->p_idx;
734         }
735
736         ix->ei_block = cpu_to_le32(logical);
737         ext4_idx_store_pblock(ix, ptr);
738         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
739
740         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
741                              > le16_to_cpu(curp->p_hdr->eh_max))) {
742                 EXT4_ERROR_INODE(inode,
743                                  "logical %d == ei_block %d!",
744                                  logical, le32_to_cpu(curp->p_idx->ei_block));
745                 return -EIO;
746         }
747         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
748                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
749                 return -EIO;
750         }
751
752         err = ext4_ext_dirty(handle, inode, curp);
753         ext4_std_error(inode->i_sb, err);
754
755         return err;
756 }
757
758 /*
759  * ext4_ext_split:
760  * inserts new subtree into the path, using free index entry
761  * at depth @at:
762  * - allocates all needed blocks (new leaf and all intermediate index blocks)
763  * - makes decision where to split
764  * - moves remaining extents and index entries (right to the split point)
765  *   into the newly allocated blocks
766  * - initializes subtree
767  */
768 static int ext4_ext_split(handle_t *handle, struct inode *inode,
769                                 struct ext4_ext_path *path,
770                                 struct ext4_extent *newext, int at)
771 {
772         struct buffer_head *bh = NULL;
773         int depth = ext_depth(inode);
774         struct ext4_extent_header *neh;
775         struct ext4_extent_idx *fidx;
776         struct ext4_extent *ex;
777         int i = at, k, m, a;
778         ext4_fsblk_t newblock, oldblock;
779         __le32 border;
780         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
781         int err = 0;
782
783         /* make decision: where to split? */
784         /* FIXME: now decision is simplest: at current extent */
785
786         /* if current leaf will be split, then we should use
787          * border from split point */
788         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
789                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
790                 return -EIO;
791         }
792         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
793                 border = path[depth].p_ext[1].ee_block;
794                 ext_debug("leaf will be split."
795                                 " next leaf starts at %d\n",
796                                   le32_to_cpu(border));
797         } else {
798                 border = newext->ee_block;
799                 ext_debug("leaf will be added."
800                                 " next leaf starts at %d\n",
801                                 le32_to_cpu(border));
802         }
803
804         /*
805          * If error occurs, then we break processing
806          * and mark filesystem read-only. index won't
807          * be inserted and tree will be in consistent
808          * state. Next mount will repair buffers too.
809          */
810
811         /*
812          * Get array to track all allocated blocks.
813          * We need this to handle errors and free blocks
814          * upon them.
815          */
816         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
817         if (!ablocks)
818                 return -ENOMEM;
819
820         /* allocate all needed blocks */
821         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
822         for (a = 0; a < depth - at; a++) {
823                 newblock = ext4_ext_new_meta_block(handle, inode, path,
824                                                    newext, &err);
825                 if (newblock == 0)
826                         goto cleanup;
827                 ablocks[a] = newblock;
828         }
829
830         /* initialize new leaf */
831         newblock = ablocks[--a];
832         if (unlikely(newblock == 0)) {
833                 EXT4_ERROR_INODE(inode, "newblock == 0!");
834                 err = -EIO;
835                 goto cleanup;
836         }
837         bh = sb_getblk(inode->i_sb, newblock);
838         if (!bh) {
839                 err = -EIO;
840                 goto cleanup;
841         }
842         lock_buffer(bh);
843
844         err = ext4_journal_get_create_access(handle, bh);
845         if (err)
846                 goto cleanup;
847
848         neh = ext_block_hdr(bh);
849         neh->eh_entries = 0;
850         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
851         neh->eh_magic = EXT4_EXT_MAGIC;
852         neh->eh_depth = 0;
853         ex = EXT_FIRST_EXTENT(neh);
854
855         /* move remainder of path[depth] to the new leaf */
856         if (unlikely(path[depth].p_hdr->eh_entries !=
857                      path[depth].p_hdr->eh_max)) {
858                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
859                                  path[depth].p_hdr->eh_entries,
860                                  path[depth].p_hdr->eh_max);
861                 err = -EIO;
862                 goto cleanup;
863         }
864         /* start copy from next extent */
865         /* TODO: we could do it by single memmove */
866         m = 0;
867         path[depth].p_ext++;
868         while (path[depth].p_ext <=
869                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
870                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
871                                 le32_to_cpu(path[depth].p_ext->ee_block),
872                                 ext4_ext_pblock(path[depth].p_ext),
873                                 ext4_ext_is_uninitialized(path[depth].p_ext),
874                                 ext4_ext_get_actual_len(path[depth].p_ext),
875                                 newblock);
876                 /*memmove(ex++, path[depth].p_ext++,
877                                 sizeof(struct ext4_extent));
878                 neh->eh_entries++;*/
879                 path[depth].p_ext++;
880                 m++;
881         }
882         if (m) {
883                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
884                 le16_add_cpu(&neh->eh_entries, m);
885         }
886
887         set_buffer_uptodate(bh);
888         unlock_buffer(bh);
889
890         err = ext4_handle_dirty_metadata(handle, inode, bh);
891         if (err)
892                 goto cleanup;
893         brelse(bh);
894         bh = NULL;
895
896         /* correct old leaf */
897         if (m) {
898                 err = ext4_ext_get_access(handle, inode, path + depth);
899                 if (err)
900                         goto cleanup;
901                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
902                 err = ext4_ext_dirty(handle, inode, path + depth);
903                 if (err)
904                         goto cleanup;
905
906         }
907
908         /* create intermediate indexes */
909         k = depth - at - 1;
910         if (unlikely(k < 0)) {
911                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
912                 err = -EIO;
913                 goto cleanup;
914         }
915         if (k)
916                 ext_debug("create %d intermediate indices\n", k);
917         /* insert new index into current index block */
918         /* current depth stored in i var */
919         i = depth - 1;
920         while (k--) {
921                 oldblock = newblock;
922                 newblock = ablocks[--a];
923                 bh = sb_getblk(inode->i_sb, newblock);
924                 if (!bh) {
925                         err = -EIO;
926                         goto cleanup;
927                 }
928                 lock_buffer(bh);
929
930                 err = ext4_journal_get_create_access(handle, bh);
931                 if (err)
932                         goto cleanup;
933
934                 neh = ext_block_hdr(bh);
935                 neh->eh_entries = cpu_to_le16(1);
936                 neh->eh_magic = EXT4_EXT_MAGIC;
937                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
938                 neh->eh_depth = cpu_to_le16(depth - i);
939                 fidx = EXT_FIRST_INDEX(neh);
940                 fidx->ei_block = border;
941                 ext4_idx_store_pblock(fidx, oldblock);
942
943                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
944                                 i, newblock, le32_to_cpu(border), oldblock);
945                 /* copy indexes */
946                 m = 0;
947                 path[i].p_idx++;
948
949                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
950                                 EXT_MAX_INDEX(path[i].p_hdr));
951                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
952                                         EXT_LAST_INDEX(path[i].p_hdr))) {
953                         EXT4_ERROR_INODE(inode,
954                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
955                                          le32_to_cpu(path[i].p_ext->ee_block));
956                         err = -EIO;
957                         goto cleanup;
958                 }
959                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
960                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
961                                         le32_to_cpu(path[i].p_idx->ei_block),
962                                         ext4_idx_pblock(path[i].p_idx),
963                                         newblock);
964                         /*memmove(++fidx, path[i].p_idx++,
965                                         sizeof(struct ext4_extent_idx));
966                         neh->eh_entries++;
967                         BUG_ON(neh->eh_entries > neh->eh_max);*/
968                         path[i].p_idx++;
969                         m++;
970                 }
971                 if (m) {
972                         memmove(++fidx, path[i].p_idx - m,
973                                 sizeof(struct ext4_extent_idx) * m);
974                         le16_add_cpu(&neh->eh_entries, m);
975                 }
976                 set_buffer_uptodate(bh);
977                 unlock_buffer(bh);
978
979                 err = ext4_handle_dirty_metadata(handle, inode, bh);
980                 if (err)
981                         goto cleanup;
982                 brelse(bh);
983                 bh = NULL;
984
985                 /* correct old index */
986                 if (m) {
987                         err = ext4_ext_get_access(handle, inode, path + i);
988                         if (err)
989                                 goto cleanup;
990                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
991                         err = ext4_ext_dirty(handle, inode, path + i);
992                         if (err)
993                                 goto cleanup;
994                 }
995
996                 i--;
997         }
998
999         /* insert new index */
1000         err = ext4_ext_insert_index(handle, inode, path + at,
1001                                     le32_to_cpu(border), newblock);
1002
1003 cleanup:
1004         if (bh) {
1005                 if (buffer_locked(bh))
1006                         unlock_buffer(bh);
1007                 brelse(bh);
1008         }
1009
1010         if (err) {
1011                 /* free all allocated blocks in error case */
1012                 for (i = 0; i < depth; i++) {
1013                         if (!ablocks[i])
1014                                 continue;
1015                         ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1016                                          EXT4_FREE_BLOCKS_METADATA);
1017                 }
1018         }
1019         kfree(ablocks);
1020
1021         return err;
1022 }
1023
1024 /*
1025  * ext4_ext_grow_indepth:
1026  * implements tree growing procedure:
1027  * - allocates new block
1028  * - moves top-level data (index block or leaf) into the new block
1029  * - initializes new top-level, creating index that points to the
1030  *   just created block
1031  */
1032 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1033                                         struct ext4_ext_path *path,
1034                                         struct ext4_extent *newext)
1035 {
1036         struct ext4_ext_path *curp = path;
1037         struct ext4_extent_header *neh;
1038         struct buffer_head *bh;
1039         ext4_fsblk_t newblock;
1040         int err = 0;
1041
1042         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1043         if (newblock == 0)
1044                 return err;
1045
1046         bh = sb_getblk(inode->i_sb, newblock);
1047         if (!bh) {
1048                 err = -EIO;
1049                 ext4_std_error(inode->i_sb, err);
1050                 return err;
1051         }
1052         lock_buffer(bh);
1053
1054         err = ext4_journal_get_create_access(handle, bh);
1055         if (err) {
1056                 unlock_buffer(bh);
1057                 goto out;
1058         }
1059
1060         /* move top-level index/leaf into new block */
1061         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1062
1063         /* set size of new block */
1064         neh = ext_block_hdr(bh);
1065         /* old root could have indexes or leaves
1066          * so calculate e_max right way */
1067         if (ext_depth(inode))
1068                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1069         else
1070                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1071         neh->eh_magic = EXT4_EXT_MAGIC;
1072         set_buffer_uptodate(bh);
1073         unlock_buffer(bh);
1074
1075         err = ext4_handle_dirty_metadata(handle, inode, bh);
1076         if (err)
1077                 goto out;
1078
1079         /* create index in new top-level index: num,max,pointer */
1080         err = ext4_ext_get_access(handle, inode, curp);
1081         if (err)
1082                 goto out;
1083
1084         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1085         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1086         curp->p_hdr->eh_entries = cpu_to_le16(1);
1087         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1088
1089         if (path[0].p_hdr->eh_depth)
1090                 curp->p_idx->ei_block =
1091                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1092         else
1093                 curp->p_idx->ei_block =
1094                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1095         ext4_idx_store_pblock(curp->p_idx, newblock);
1096
1097         neh = ext_inode_hdr(inode);
1098         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1099                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1100                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1101                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1102
1103         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1104         err = ext4_ext_dirty(handle, inode, curp);
1105 out:
1106         brelse(bh);
1107
1108         return err;
1109 }
1110
1111 /*
1112  * ext4_ext_create_new_leaf:
1113  * finds empty index and adds new leaf.
1114  * if no free index is found, then it requests in-depth growing.
1115  */
1116 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1117                                         struct ext4_ext_path *path,
1118                                         struct ext4_extent *newext)
1119 {
1120         struct ext4_ext_path *curp;
1121         int depth, i, err = 0;
1122
1123 repeat:
1124         i = depth = ext_depth(inode);
1125
1126         /* walk up to the tree and look for free index entry */
1127         curp = path + depth;
1128         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1129                 i--;
1130                 curp--;
1131         }
1132
1133         /* we use already allocated block for index block,
1134          * so subsequent data blocks should be contiguous */
1135         if (EXT_HAS_FREE_INDEX(curp)) {
1136                 /* if we found index with free entry, then use that
1137                  * entry: create all needed subtree and add new leaf */
1138                 err = ext4_ext_split(handle, inode, path, newext, i);
1139                 if (err)
1140                         goto out;
1141
1142                 /* refill path */
1143                 ext4_ext_drop_refs(path);
1144                 path = ext4_ext_find_extent(inode,
1145                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1146                                     path);
1147                 if (IS_ERR(path))
1148                         err = PTR_ERR(path);
1149         } else {
1150                 /* tree is full, time to grow in depth */
1151                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1152                 if (err)
1153                         goto out;
1154
1155                 /* refill path */
1156                 ext4_ext_drop_refs(path);
1157                 path = ext4_ext_find_extent(inode,
1158                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1159                                     path);
1160                 if (IS_ERR(path)) {
1161                         err = PTR_ERR(path);
1162                         goto out;
1163                 }
1164
1165                 /*
1166                  * only first (depth 0 -> 1) produces free space;
1167                  * in all other cases we have to split the grown tree
1168                  */
1169                 depth = ext_depth(inode);
1170                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1171                         /* now we need to split */
1172                         goto repeat;
1173                 }
1174         }
1175
1176 out:
1177         return err;
1178 }
1179
1180 /*
1181  * search the closest allocated block to the left for *logical
1182  * and returns it at @logical + it's physical address at @phys
1183  * if *logical is the smallest allocated block, the function
1184  * returns 0 at @phys
1185  * return value contains 0 (success) or error code
1186  */
1187 static int ext4_ext_search_left(struct inode *inode,
1188                                 struct ext4_ext_path *path,
1189                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1190 {
1191         struct ext4_extent_idx *ix;
1192         struct ext4_extent *ex;
1193         int depth, ee_len;
1194
1195         if (unlikely(path == NULL)) {
1196                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1197                 return -EIO;
1198         }
1199         depth = path->p_depth;
1200         *phys = 0;
1201
1202         if (depth == 0 && path->p_ext == NULL)
1203                 return 0;
1204
1205         /* usually extent in the path covers blocks smaller
1206          * then *logical, but it can be that extent is the
1207          * first one in the file */
1208
1209         ex = path[depth].p_ext;
1210         ee_len = ext4_ext_get_actual_len(ex);
1211         if (*logical < le32_to_cpu(ex->ee_block)) {
1212                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1213                         EXT4_ERROR_INODE(inode,
1214                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1215                                          *logical, le32_to_cpu(ex->ee_block));
1216                         return -EIO;
1217                 }
1218                 while (--depth >= 0) {
1219                         ix = path[depth].p_idx;
1220                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1221                                 EXT4_ERROR_INODE(inode,
1222                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1223                                   ix != NULL ? ix->ei_block : 0,
1224                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1225                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1226                                   depth);
1227                                 return -EIO;
1228                         }
1229                 }
1230                 return 0;
1231         }
1232
1233         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1234                 EXT4_ERROR_INODE(inode,
1235                                  "logical %d < ee_block %d + ee_len %d!",
1236                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1237                 return -EIO;
1238         }
1239
1240         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1241         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1242         return 0;
1243 }
1244
1245 /*
1246  * search the closest allocated block to the right for *logical
1247  * and returns it at @logical + it's physical address at @phys
1248  * if *logical is the smallest allocated block, the function
1249  * returns 0 at @phys
1250  * return value contains 0 (success) or error code
1251  */
1252 static int ext4_ext_search_right(struct inode *inode,
1253                                  struct ext4_ext_path *path,
1254                                  ext4_lblk_t *logical, ext4_fsblk_t *phys)
1255 {
1256         struct buffer_head *bh = NULL;
1257         struct ext4_extent_header *eh;
1258         struct ext4_extent_idx *ix;
1259         struct ext4_extent *ex;
1260         ext4_fsblk_t block;
1261         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1262         int ee_len;
1263
1264         if (unlikely(path == NULL)) {
1265                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1266                 return -EIO;
1267         }
1268         depth = path->p_depth;
1269         *phys = 0;
1270
1271         if (depth == 0 && path->p_ext == NULL)
1272                 return 0;
1273
1274         /* usually extent in the path covers blocks smaller
1275          * then *logical, but it can be that extent is the
1276          * first one in the file */
1277
1278         ex = path[depth].p_ext;
1279         ee_len = ext4_ext_get_actual_len(ex);
1280         if (*logical < le32_to_cpu(ex->ee_block)) {
1281                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1282                         EXT4_ERROR_INODE(inode,
1283                                          "first_extent(path[%d].p_hdr) != ex",
1284                                          depth);
1285                         return -EIO;
1286                 }
1287                 while (--depth >= 0) {
1288                         ix = path[depth].p_idx;
1289                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1290                                 EXT4_ERROR_INODE(inode,
1291                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1292                                                  *logical);
1293                                 return -EIO;
1294                         }
1295                 }
1296                 *logical = le32_to_cpu(ex->ee_block);
1297                 *phys = ext4_ext_pblock(ex);
1298                 return 0;
1299         }
1300
1301         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1302                 EXT4_ERROR_INODE(inode,
1303                                  "logical %d < ee_block %d + ee_len %d!",
1304                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1305                 return -EIO;
1306         }
1307
1308         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1309                 /* next allocated block in this leaf */
1310                 ex++;
1311                 *logical = le32_to_cpu(ex->ee_block);
1312                 *phys = ext4_ext_pblock(ex);
1313                 return 0;
1314         }
1315
1316         /* go up and search for index to the right */
1317         while (--depth >= 0) {
1318                 ix = path[depth].p_idx;
1319                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1320                         goto got_index;
1321         }
1322
1323         /* we've gone up to the root and found no index to the right */
1324         return 0;
1325
1326 got_index:
1327         /* we've found index to the right, let's
1328          * follow it and find the closest allocated
1329          * block to the right */
1330         ix++;
1331         block = ext4_idx_pblock(ix);
1332         while (++depth < path->p_depth) {
1333                 bh = sb_bread(inode->i_sb, block);
1334                 if (bh == NULL)
1335                         return -EIO;
1336                 eh = ext_block_hdr(bh);
1337                 /* subtract from p_depth to get proper eh_depth */
1338                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1339                         put_bh(bh);
1340                         return -EIO;
1341                 }
1342                 ix = EXT_FIRST_INDEX(eh);
1343                 block = ext4_idx_pblock(ix);
1344                 put_bh(bh);
1345         }
1346
1347         bh = sb_bread(inode->i_sb, block);
1348         if (bh == NULL)
1349                 return -EIO;
1350         eh = ext_block_hdr(bh);
1351         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1352                 put_bh(bh);
1353                 return -EIO;
1354         }
1355         ex = EXT_FIRST_EXTENT(eh);
1356         *logical = le32_to_cpu(ex->ee_block);
1357         *phys = ext4_ext_pblock(ex);
1358         put_bh(bh);
1359         return 0;
1360 }
1361
1362 /*
1363  * ext4_ext_next_allocated_block:
1364  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1365  * NOTE: it considers block number from index entry as
1366  * allocated block. Thus, index entries have to be consistent
1367  * with leaves.
1368  */
1369 static ext4_lblk_t
1370 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1371 {
1372         int depth;
1373
1374         BUG_ON(path == NULL);
1375         depth = path->p_depth;
1376
1377         if (depth == 0 && path->p_ext == NULL)
1378                 return EXT_MAX_BLOCK;
1379
1380         while (depth >= 0) {
1381                 if (depth == path->p_depth) {
1382                         /* leaf */
1383                         if (path[depth].p_ext !=
1384                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1385                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1386                 } else {
1387                         /* index */
1388                         if (path[depth].p_idx !=
1389                                         EXT_LAST_INDEX(path[depth].p_hdr))
1390                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1391                 }
1392                 depth--;
1393         }
1394
1395         return EXT_MAX_BLOCK;
1396 }
1397
1398 /*
1399  * ext4_ext_next_leaf_block:
1400  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1401  */
1402 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1403                                         struct ext4_ext_path *path)
1404 {
1405         int depth;
1406
1407         BUG_ON(path == NULL);
1408         depth = path->p_depth;
1409
1410         /* zero-tree has no leaf blocks at all */
1411         if (depth == 0)
1412                 return EXT_MAX_BLOCK;
1413
1414         /* go to index block */
1415         depth--;
1416
1417         while (depth >= 0) {
1418                 if (path[depth].p_idx !=
1419                                 EXT_LAST_INDEX(path[depth].p_hdr))
1420                         return (ext4_lblk_t)
1421                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1422                 depth--;
1423         }
1424
1425         return EXT_MAX_BLOCK;
1426 }
1427
1428 /*
1429  * ext4_ext_correct_indexes:
1430  * if leaf gets modified and modified extent is first in the leaf,
1431  * then we have to correct all indexes above.
1432  * TODO: do we need to correct tree in all cases?
1433  */
1434 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1435                                 struct ext4_ext_path *path)
1436 {
1437         struct ext4_extent_header *eh;
1438         int depth = ext_depth(inode);
1439         struct ext4_extent *ex;
1440         __le32 border;
1441         int k, err = 0;
1442
1443         eh = path[depth].p_hdr;
1444         ex = path[depth].p_ext;
1445
1446         if (unlikely(ex == NULL || eh == NULL)) {
1447                 EXT4_ERROR_INODE(inode,
1448                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1449                 return -EIO;
1450         }
1451
1452         if (depth == 0) {
1453                 /* there is no tree at all */
1454                 return 0;
1455         }
1456
1457         if (ex != EXT_FIRST_EXTENT(eh)) {
1458                 /* we correct tree if first leaf got modified only */
1459                 return 0;
1460         }
1461
1462         /*
1463          * TODO: we need correction if border is smaller than current one
1464          */
1465         k = depth - 1;
1466         border = path[depth].p_ext->ee_block;
1467         err = ext4_ext_get_access(handle, inode, path + k);
1468         if (err)
1469                 return err;
1470         path[k].p_idx->ei_block = border;
1471         err = ext4_ext_dirty(handle, inode, path + k);
1472         if (err)
1473                 return err;
1474
1475         while (k--) {
1476                 /* change all left-side indexes */
1477                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1478                         break;
1479                 err = ext4_ext_get_access(handle, inode, path + k);
1480                 if (err)
1481                         break;
1482                 path[k].p_idx->ei_block = border;
1483                 err = ext4_ext_dirty(handle, inode, path + k);
1484                 if (err)
1485                         break;
1486         }
1487
1488         return err;
1489 }
1490
1491 int
1492 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1493                                 struct ext4_extent *ex2)
1494 {
1495         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1496
1497         /*
1498          * Make sure that either both extents are uninitialized, or
1499          * both are _not_.
1500          */
1501         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1502                 return 0;
1503
1504         if (ext4_ext_is_uninitialized(ex1))
1505                 max_len = EXT_UNINIT_MAX_LEN;
1506         else
1507                 max_len = EXT_INIT_MAX_LEN;
1508
1509         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1510         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1511
1512         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1513                         le32_to_cpu(ex2->ee_block))
1514                 return 0;
1515
1516         /*
1517          * To allow future support for preallocated extents to be added
1518          * as an RO_COMPAT feature, refuse to merge to extents if
1519          * this can result in the top bit of ee_len being set.
1520          */
1521         if (ext1_ee_len + ext2_ee_len > max_len)
1522                 return 0;
1523 #ifdef AGGRESSIVE_TEST
1524         if (ext1_ee_len >= 4)
1525                 return 0;
1526 #endif
1527
1528         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1529                 return 1;
1530         return 0;
1531 }
1532
1533 /*
1534  * This function tries to merge the "ex" extent to the next extent in the tree.
1535  * It always tries to merge towards right. If you want to merge towards
1536  * left, pass "ex - 1" as argument instead of "ex".
1537  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1538  * 1 if they got merged.
1539  */
1540 static int ext4_ext_try_to_merge(struct inode *inode,
1541                                  struct ext4_ext_path *path,
1542                                  struct ext4_extent *ex)
1543 {
1544         struct ext4_extent_header *eh;
1545         unsigned int depth, len;
1546         int merge_done = 0;
1547         int uninitialized = 0;
1548
1549         depth = ext_depth(inode);
1550         BUG_ON(path[depth].p_hdr == NULL);
1551         eh = path[depth].p_hdr;
1552
1553         while (ex < EXT_LAST_EXTENT(eh)) {
1554                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1555                         break;
1556                 /* merge with next extent! */
1557                 if (ext4_ext_is_uninitialized(ex))
1558                         uninitialized = 1;
1559                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1560                                 + ext4_ext_get_actual_len(ex + 1));
1561                 if (uninitialized)
1562                         ext4_ext_mark_uninitialized(ex);
1563
1564                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1565                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1566                                 * sizeof(struct ext4_extent);
1567                         memmove(ex + 1, ex + 2, len);
1568                 }
1569                 le16_add_cpu(&eh->eh_entries, -1);
1570                 merge_done = 1;
1571                 WARN_ON(eh->eh_entries == 0);
1572                 if (!eh->eh_entries)
1573                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1574         }
1575
1576         return merge_done;
1577 }
1578
1579 /*
1580  * check if a portion of the "newext" extent overlaps with an
1581  * existing extent.
1582  *
1583  * If there is an overlap discovered, it updates the length of the newext
1584  * such that there will be no overlap, and then returns 1.
1585  * If there is no overlap found, it returns 0.
1586  */
1587 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1588                                            struct ext4_extent *newext,
1589                                            struct ext4_ext_path *path)
1590 {
1591         ext4_lblk_t b1, b2;
1592         unsigned int depth, len1;
1593         unsigned int ret = 0;
1594
1595         b1 = le32_to_cpu(newext->ee_block);
1596         len1 = ext4_ext_get_actual_len(newext);
1597         depth = ext_depth(inode);
1598         if (!path[depth].p_ext)
1599                 goto out;
1600         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1601
1602         /*
1603          * get the next allocated block if the extent in the path
1604          * is before the requested block(s)
1605          */
1606         if (b2 < b1) {
1607                 b2 = ext4_ext_next_allocated_block(path);
1608                 if (b2 == EXT_MAX_BLOCK)
1609                         goto out;
1610         }
1611
1612         /* check for wrap through zero on extent logical start block*/
1613         if (b1 + len1 < b1) {
1614                 len1 = EXT_MAX_BLOCK - b1;
1615                 newext->ee_len = cpu_to_le16(len1);
1616                 ret = 1;
1617         }
1618
1619         /* check for overlap */
1620         if (b1 + len1 > b2) {
1621                 newext->ee_len = cpu_to_le16(b2 - b1);
1622                 ret = 1;
1623         }
1624 out:
1625         return ret;
1626 }
1627
1628 /*
1629  * ext4_ext_insert_extent:
1630  * tries to merge requsted extent into the existing extent or
1631  * inserts requested extent as new one into the tree,
1632  * creating new leaf in the no-space case.
1633  */
1634 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1635                                 struct ext4_ext_path *path,
1636                                 struct ext4_extent *newext, int flag)
1637 {
1638         struct ext4_extent_header *eh;
1639         struct ext4_extent *ex, *fex;
1640         struct ext4_extent *nearex; /* nearest extent */
1641         struct ext4_ext_path *npath = NULL;
1642         int depth, len, err;
1643         ext4_lblk_t next;
1644         unsigned uninitialized = 0;
1645
1646         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1647                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1648                 return -EIO;
1649         }
1650         depth = ext_depth(inode);
1651         ex = path[depth].p_ext;
1652         if (unlikely(path[depth].p_hdr == NULL)) {
1653                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1654                 return -EIO;
1655         }
1656
1657         /* try to insert block into found extent and return */
1658         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1659                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1660                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1661                           ext4_ext_is_uninitialized(newext),
1662                           ext4_ext_get_actual_len(newext),
1663                           le32_to_cpu(ex->ee_block),
1664                           ext4_ext_is_uninitialized(ex),
1665                           ext4_ext_get_actual_len(ex),
1666                           ext4_ext_pblock(ex));
1667                 err = ext4_ext_get_access(handle, inode, path + depth);
1668                 if (err)
1669                         return err;
1670
1671                 /*
1672                  * ext4_can_extents_be_merged should have checked that either
1673                  * both extents are uninitialized, or both aren't. Thus we
1674                  * need to check only one of them here.
1675                  */
1676                 if (ext4_ext_is_uninitialized(ex))
1677                         uninitialized = 1;
1678                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1679                                         + ext4_ext_get_actual_len(newext));
1680                 if (uninitialized)
1681                         ext4_ext_mark_uninitialized(ex);
1682                 eh = path[depth].p_hdr;
1683                 nearex = ex;
1684                 goto merge;
1685         }
1686
1687 repeat:
1688         depth = ext_depth(inode);
1689         eh = path[depth].p_hdr;
1690         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1691                 goto has_space;
1692
1693         /* probably next leaf has space for us? */
1694         fex = EXT_LAST_EXTENT(eh);
1695         next = ext4_ext_next_leaf_block(inode, path);
1696         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1697             && next != EXT_MAX_BLOCK) {
1698                 ext_debug("next leaf block - %d\n", next);
1699                 BUG_ON(npath != NULL);
1700                 npath = ext4_ext_find_extent(inode, next, NULL);
1701                 if (IS_ERR(npath))
1702                         return PTR_ERR(npath);
1703                 BUG_ON(npath->p_depth != path->p_depth);
1704                 eh = npath[depth].p_hdr;
1705                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1706                         ext_debug("next leaf isnt full(%d)\n",
1707                                   le16_to_cpu(eh->eh_entries));
1708                         path = npath;
1709                         goto repeat;
1710                 }
1711                 ext_debug("next leaf has no free space(%d,%d)\n",
1712                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1713         }
1714
1715         /*
1716          * There is no free space in the found leaf.
1717          * We're gonna add a new leaf in the tree.
1718          */
1719         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1720         if (err)
1721                 goto cleanup;
1722         depth = ext_depth(inode);
1723         eh = path[depth].p_hdr;
1724
1725 has_space:
1726         nearex = path[depth].p_ext;
1727
1728         err = ext4_ext_get_access(handle, inode, path + depth);
1729         if (err)
1730                 goto cleanup;
1731
1732         if (!nearex) {
1733                 /* there is no extent in this leaf, create first one */
1734                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1735                                 le32_to_cpu(newext->ee_block),
1736                                 ext4_ext_pblock(newext),
1737                                 ext4_ext_is_uninitialized(newext),
1738                                 ext4_ext_get_actual_len(newext));
1739                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1740         } else if (le32_to_cpu(newext->ee_block)
1741                            > le32_to_cpu(nearex->ee_block)) {
1742 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1743                 if (nearex != EXT_LAST_EXTENT(eh)) {
1744                         len = EXT_MAX_EXTENT(eh) - nearex;
1745                         len = (len - 1) * sizeof(struct ext4_extent);
1746                         len = len < 0 ? 0 : len;
1747                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1748                                         "move %d from 0x%p to 0x%p\n",
1749                                         le32_to_cpu(newext->ee_block),
1750                                         ext4_ext_pblock(newext),
1751                                         ext4_ext_is_uninitialized(newext),
1752                                         ext4_ext_get_actual_len(newext),
1753                                         nearex, len, nearex + 1, nearex + 2);
1754                         memmove(nearex + 2, nearex + 1, len);
1755                 }
1756                 path[depth].p_ext = nearex + 1;
1757         } else {
1758                 BUG_ON(newext->ee_block == nearex->ee_block);
1759                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1760                 len = len < 0 ? 0 : len;
1761                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1762                                 "move %d from 0x%p to 0x%p\n",
1763                                 le32_to_cpu(newext->ee_block),
1764                                 ext4_ext_pblock(newext),
1765                                 ext4_ext_is_uninitialized(newext),
1766                                 ext4_ext_get_actual_len(newext),
1767                                 nearex, len, nearex + 1, nearex + 2);
1768                 memmove(nearex + 1, nearex, len);
1769                 path[depth].p_ext = nearex;
1770         }
1771
1772         le16_add_cpu(&eh->eh_entries, 1);
1773         nearex = path[depth].p_ext;
1774         nearex->ee_block = newext->ee_block;
1775         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1776         nearex->ee_len = newext->ee_len;
1777
1778 merge:
1779         /* try to merge extents to the right */
1780         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1781                 ext4_ext_try_to_merge(inode, path, nearex);
1782
1783         /* try to merge extents to the left */
1784
1785         /* time to correct all indexes above */
1786         err = ext4_ext_correct_indexes(handle, inode, path);
1787         if (err)
1788                 goto cleanup;
1789
1790         err = ext4_ext_dirty(handle, inode, path + depth);
1791
1792 cleanup:
1793         if (npath) {
1794                 ext4_ext_drop_refs(npath);
1795                 kfree(npath);
1796         }
1797         ext4_ext_invalidate_cache(inode);
1798         return err;
1799 }
1800
1801 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1802                                ext4_lblk_t num, ext_prepare_callback func,
1803                                void *cbdata)
1804 {
1805         struct ext4_ext_path *path = NULL;
1806         struct ext4_ext_cache cbex;
1807         struct ext4_extent *ex;
1808         ext4_lblk_t next, start = 0, end = 0;
1809         ext4_lblk_t last = block + num;
1810         int depth, exists, err = 0;
1811
1812         BUG_ON(func == NULL);
1813         BUG_ON(inode == NULL);
1814
1815         while (block < last && block != EXT_MAX_BLOCK) {
1816                 num = last - block;
1817                 /* find extent for this block */
1818                 down_read(&EXT4_I(inode)->i_data_sem);
1819                 path = ext4_ext_find_extent(inode, block, path);
1820                 up_read(&EXT4_I(inode)->i_data_sem);
1821                 if (IS_ERR(path)) {
1822                         err = PTR_ERR(path);
1823                         path = NULL;
1824                         break;
1825                 }
1826
1827                 depth = ext_depth(inode);
1828                 if (unlikely(path[depth].p_hdr == NULL)) {
1829                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1830                         err = -EIO;
1831                         break;
1832                 }
1833                 ex = path[depth].p_ext;
1834                 next = ext4_ext_next_allocated_block(path);
1835
1836                 exists = 0;
1837                 if (!ex) {
1838                         /* there is no extent yet, so try to allocate
1839                          * all requested space */
1840                         start = block;
1841                         end = block + num;
1842                 } else if (le32_to_cpu(ex->ee_block) > block) {
1843                         /* need to allocate space before found extent */
1844                         start = block;
1845                         end = le32_to_cpu(ex->ee_block);
1846                         if (block + num < end)
1847                                 end = block + num;
1848                 } else if (block >= le32_to_cpu(ex->ee_block)
1849                                         + ext4_ext_get_actual_len(ex)) {
1850                         /* need to allocate space after found extent */
1851                         start = block;
1852                         end = block + num;
1853                         if (end >= next)
1854                                 end = next;
1855                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1856                         /*
1857                          * some part of requested space is covered
1858                          * by found extent
1859                          */
1860                         start = block;
1861                         end = le32_to_cpu(ex->ee_block)
1862                                 + ext4_ext_get_actual_len(ex);
1863                         if (block + num < end)
1864                                 end = block + num;
1865                         exists = 1;
1866                 } else {
1867                         BUG();
1868                 }
1869                 BUG_ON(end <= start);
1870
1871                 if (!exists) {
1872                         cbex.ec_block = start;
1873                         cbex.ec_len = end - start;
1874                         cbex.ec_start = 0;
1875                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1876                 } else {
1877                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1878                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1879                         cbex.ec_start = ext4_ext_pblock(ex);
1880                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1881                 }
1882
1883                 if (unlikely(cbex.ec_len == 0)) {
1884                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1885                         err = -EIO;
1886                         break;
1887                 }
1888                 err = func(inode, path, &cbex, ex, cbdata);
1889                 ext4_ext_drop_refs(path);
1890
1891                 if (err < 0)
1892                         break;
1893
1894                 if (err == EXT_REPEAT)
1895                         continue;
1896                 else if (err == EXT_BREAK) {
1897                         err = 0;
1898                         break;
1899                 }
1900
1901                 if (ext_depth(inode) != depth) {
1902                         /* depth was changed. we have to realloc path */
1903                         kfree(path);
1904                         path = NULL;
1905                 }
1906
1907                 block = cbex.ec_block + cbex.ec_len;
1908         }
1909
1910         if (path) {
1911                 ext4_ext_drop_refs(path);
1912                 kfree(path);
1913         }
1914
1915         return err;
1916 }
1917
1918 static void
1919 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1920                         __u32 len, ext4_fsblk_t start, int type)
1921 {
1922         struct ext4_ext_cache *cex;
1923         BUG_ON(len == 0);
1924         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1925         cex = &EXT4_I(inode)->i_cached_extent;
1926         cex->ec_type = type;
1927         cex->ec_block = block;
1928         cex->ec_len = len;
1929         cex->ec_start = start;
1930         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1931 }
1932
1933 /*
1934  * ext4_ext_put_gap_in_cache:
1935  * calculate boundaries of the gap that the requested block fits into
1936  * and cache this gap
1937  */
1938 static void
1939 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1940                                 ext4_lblk_t block)
1941 {
1942         int depth = ext_depth(inode);
1943         unsigned long len;
1944         ext4_lblk_t lblock;
1945         struct ext4_extent *ex;
1946
1947         ex = path[depth].p_ext;
1948         if (ex == NULL) {
1949                 /* there is no extent yet, so gap is [0;-] */
1950                 lblock = 0;
1951                 len = EXT_MAX_BLOCK;
1952                 ext_debug("cache gap(whole file):");
1953         } else if (block < le32_to_cpu(ex->ee_block)) {
1954                 lblock = block;
1955                 len = le32_to_cpu(ex->ee_block) - block;
1956                 ext_debug("cache gap(before): %u [%u:%u]",
1957                                 block,
1958                                 le32_to_cpu(ex->ee_block),
1959                                  ext4_ext_get_actual_len(ex));
1960         } else if (block >= le32_to_cpu(ex->ee_block)
1961                         + ext4_ext_get_actual_len(ex)) {
1962                 ext4_lblk_t next;
1963                 lblock = le32_to_cpu(ex->ee_block)
1964                         + ext4_ext_get_actual_len(ex);
1965
1966                 next = ext4_ext_next_allocated_block(path);
1967                 ext_debug("cache gap(after): [%u:%u] %u",
1968                                 le32_to_cpu(ex->ee_block),
1969                                 ext4_ext_get_actual_len(ex),
1970                                 block);
1971                 BUG_ON(next == lblock);
1972                 len = next - lblock;
1973         } else {
1974                 lblock = len = 0;
1975                 BUG();
1976         }
1977
1978         ext_debug(" -> %u:%lu\n", lblock, len);
1979         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1980 }
1981
1982 static int
1983 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1984                         struct ext4_extent *ex)
1985 {
1986         struct ext4_ext_cache *cex;
1987         int ret = EXT4_EXT_CACHE_NO;
1988
1989         /*
1990          * We borrow i_block_reservation_lock to protect i_cached_extent
1991          */
1992         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1993         cex = &EXT4_I(inode)->i_cached_extent;
1994
1995         /* has cache valid data? */
1996         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1997                 goto errout;
1998
1999         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
2000                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
2001         if (in_range(block, cex->ec_block, cex->ec_len)) {
2002                 ex->ee_block = cpu_to_le32(cex->ec_block);
2003                 ext4_ext_store_pblock(ex, cex->ec_start);
2004                 ex->ee_len = cpu_to_le16(cex->ec_len);
2005                 ext_debug("%u cached by %u:%u:%llu\n",
2006                                 block,
2007                                 cex->ec_block, cex->ec_len, cex->ec_start);
2008                 ret = cex->ec_type;
2009         }
2010 errout:
2011         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2012         return ret;
2013 }
2014
2015 /*
2016  * ext4_ext_rm_idx:
2017  * removes index from the index block.
2018  * It's used in truncate case only, thus all requests are for
2019  * last index in the block only.
2020  */
2021 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2022                         struct ext4_ext_path *path)
2023 {
2024         int err;
2025         ext4_fsblk_t leaf;
2026
2027         /* free index block */
2028         path--;
2029         leaf = ext4_idx_pblock(path->p_idx);
2030         if (unlikely(path->p_hdr->eh_entries == 0)) {
2031                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2032                 return -EIO;
2033         }
2034         err = ext4_ext_get_access(handle, inode, path);
2035         if (err)
2036                 return err;
2037         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2038         err = ext4_ext_dirty(handle, inode, path);
2039         if (err)
2040                 return err;
2041         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2042         ext4_free_blocks(handle, inode, 0, leaf, 1,
2043                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2044         return err;
2045 }
2046
2047 /*
2048  * ext4_ext_calc_credits_for_single_extent:
2049  * This routine returns max. credits that needed to insert an extent
2050  * to the extent tree.
2051  * When pass the actual path, the caller should calculate credits
2052  * under i_data_sem.
2053  */
2054 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2055                                                 struct ext4_ext_path *path)
2056 {
2057         if (path) {
2058                 int depth = ext_depth(inode);
2059                 int ret = 0;
2060
2061                 /* probably there is space in leaf? */
2062                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2063                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2064
2065                         /*
2066                          *  There are some space in the leaf tree, no
2067                          *  need to account for leaf block credit
2068                          *
2069                          *  bitmaps and block group descriptor blocks
2070                          *  and other metadat blocks still need to be
2071                          *  accounted.
2072                          */
2073                         /* 1 bitmap, 1 block group descriptor */
2074                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2075                         return ret;
2076                 }
2077         }
2078
2079         return ext4_chunk_trans_blocks(inode, nrblocks);
2080 }
2081
2082 /*
2083  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2084  *
2085  * if nrblocks are fit in a single extent (chunk flag is 1), then
2086  * in the worse case, each tree level index/leaf need to be changed
2087  * if the tree split due to insert a new extent, then the old tree
2088  * index/leaf need to be updated too
2089  *
2090  * If the nrblocks are discontiguous, they could cause
2091  * the whole tree split more than once, but this is really rare.
2092  */
2093 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2094 {
2095         int index;
2096         int depth = ext_depth(inode);
2097
2098         if (chunk)
2099                 index = depth * 2;
2100         else
2101                 index = depth * 3;
2102
2103         return index;
2104 }
2105
2106 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2107                                 struct ext4_extent *ex,
2108                                 ext4_lblk_t from, ext4_lblk_t to)
2109 {
2110         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2111         int flags = EXT4_FREE_BLOCKS_FORGET;
2112
2113         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2114                 flags |= EXT4_FREE_BLOCKS_METADATA;
2115 #ifdef EXTENTS_STATS
2116         {
2117                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2118                 spin_lock(&sbi->s_ext_stats_lock);
2119                 sbi->s_ext_blocks += ee_len;
2120                 sbi->s_ext_extents++;
2121                 if (ee_len < sbi->s_ext_min)
2122                         sbi->s_ext_min = ee_len;
2123                 if (ee_len > sbi->s_ext_max)
2124                         sbi->s_ext_max = ee_len;
2125                 if (ext_depth(inode) > sbi->s_depth_max)
2126                         sbi->s_depth_max = ext_depth(inode);
2127                 spin_unlock(&sbi->s_ext_stats_lock);
2128         }
2129 #endif
2130         if (from >= le32_to_cpu(ex->ee_block)
2131             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2132                 /* tail removal */
2133                 ext4_lblk_t num;
2134                 ext4_fsblk_t start;
2135
2136                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2137                 start = ext4_ext_pblock(ex) + ee_len - num;
2138                 ext_debug("free last %u blocks starting %llu\n", num, start);
2139                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2140         } else if (from == le32_to_cpu(ex->ee_block)
2141                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2142                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2143                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2144         } else {
2145                 printk(KERN_INFO "strange request: removal(2) "
2146                                 "%u-%u from %u:%u\n",
2147                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2148         }
2149         return 0;
2150 }
2151
2152 static int
2153 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2154                 struct ext4_ext_path *path, ext4_lblk_t start)
2155 {
2156         int err = 0, correct_index = 0;
2157         int depth = ext_depth(inode), credits;
2158         struct ext4_extent_header *eh;
2159         ext4_lblk_t a, b, block;
2160         unsigned num;
2161         ext4_lblk_t ex_ee_block;
2162         unsigned short ex_ee_len;
2163         unsigned uninitialized = 0;
2164         struct ext4_extent *ex;
2165
2166         /* the header must be checked already in ext4_ext_remove_space() */
2167         ext_debug("truncate since %u in leaf\n", start);
2168         if (!path[depth].p_hdr)
2169                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2170         eh = path[depth].p_hdr;
2171         if (unlikely(path[depth].p_hdr == NULL)) {
2172                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2173                 return -EIO;
2174         }
2175         /* find where to start removing */
2176         ex = EXT_LAST_EXTENT(eh);
2177
2178         ex_ee_block = le32_to_cpu(ex->ee_block);
2179         ex_ee_len = ext4_ext_get_actual_len(ex);
2180
2181         while (ex >= EXT_FIRST_EXTENT(eh) &&
2182                         ex_ee_block + ex_ee_len > start) {
2183
2184                 if (ext4_ext_is_uninitialized(ex))
2185                         uninitialized = 1;
2186                 else
2187                         uninitialized = 0;
2188
2189                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2190                          uninitialized, ex_ee_len);
2191                 path[depth].p_ext = ex;
2192
2193                 a = ex_ee_block > start ? ex_ee_block : start;
2194                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2195                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2196
2197                 ext_debug("  border %u:%u\n", a, b);
2198
2199                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2200                         block = 0;
2201                         num = 0;
2202                         BUG();
2203                 } else if (a != ex_ee_block) {
2204                         /* remove tail of the extent */
2205                         block = ex_ee_block;
2206                         num = a - block;
2207                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2208                         /* remove head of the extent */
2209                         block = a;
2210                         num = b - a;
2211                         /* there is no "make a hole" API yet */
2212                         BUG();
2213                 } else {
2214                         /* remove whole extent: excellent! */
2215                         block = ex_ee_block;
2216                         num = 0;
2217                         BUG_ON(a != ex_ee_block);
2218                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2219                 }
2220
2221                 /*
2222                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2223                  * descriptor) for each block group; assume two block
2224                  * groups plus ex_ee_len/blocks_per_block_group for
2225                  * the worst case
2226                  */
2227                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2228                 if (ex == EXT_FIRST_EXTENT(eh)) {
2229                         correct_index = 1;
2230                         credits += (ext_depth(inode)) + 1;
2231                 }
2232                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2233
2234                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2235                 if (err)
2236                         goto out;
2237
2238                 err = ext4_ext_get_access(handle, inode, path + depth);
2239                 if (err)
2240                         goto out;
2241
2242                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2243                 if (err)
2244                         goto out;
2245
2246                 if (num == 0) {
2247                         /* this extent is removed; mark slot entirely unused */
2248                         ext4_ext_store_pblock(ex, 0);
2249                         le16_add_cpu(&eh->eh_entries, -1);
2250                 }
2251
2252                 ex->ee_block = cpu_to_le32(block);
2253                 ex->ee_len = cpu_to_le16(num);
2254                 /*
2255                  * Do not mark uninitialized if all the blocks in the
2256                  * extent have been removed.
2257                  */
2258                 if (uninitialized && num)
2259                         ext4_ext_mark_uninitialized(ex);
2260
2261                 err = ext4_ext_dirty(handle, inode, path + depth);
2262                 if (err)
2263                         goto out;
2264
2265                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2266                                 ext4_ext_pblock(ex));
2267                 ex--;
2268                 ex_ee_block = le32_to_cpu(ex->ee_block);
2269                 ex_ee_len = ext4_ext_get_actual_len(ex);
2270         }
2271
2272         if (correct_index && eh->eh_entries)
2273                 err = ext4_ext_correct_indexes(handle, inode, path);
2274
2275         /* if this leaf is free, then we should
2276          * remove it from index block above */
2277         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2278                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2279
2280 out:
2281         return err;
2282 }
2283
2284 /*
2285  * ext4_ext_more_to_rm:
2286  * returns 1 if current index has to be freed (even partial)
2287  */
2288 static int
2289 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2290 {
2291         BUG_ON(path->p_idx == NULL);
2292
2293         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2294                 return 0;
2295
2296         /*
2297          * if truncate on deeper level happened, it wasn't partial,
2298          * so we have to consider current index for truncation
2299          */
2300         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2301                 return 0;
2302         return 1;
2303 }
2304
2305 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2306 {
2307         struct super_block *sb = inode->i_sb;
2308         int depth = ext_depth(inode);
2309         struct ext4_ext_path *path;
2310         handle_t *handle;
2311         int i, err;
2312
2313         ext_debug("truncate since %u\n", start);
2314
2315         /* probably first extent we're gonna free will be last in block */
2316         handle = ext4_journal_start(inode, depth + 1);
2317         if (IS_ERR(handle))
2318                 return PTR_ERR(handle);
2319
2320 again:
2321         ext4_ext_invalidate_cache(inode);
2322
2323         /*
2324          * We start scanning from right side, freeing all the blocks
2325          * after i_size and walking into the tree depth-wise.
2326          */
2327         depth = ext_depth(inode);
2328         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2329         if (path == NULL) {
2330                 ext4_journal_stop(handle);
2331                 return -ENOMEM;
2332         }
2333         path[0].p_depth = depth;
2334         path[0].p_hdr = ext_inode_hdr(inode);
2335         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2336                 err = -EIO;
2337                 goto out;
2338         }
2339         i = err = 0;
2340
2341         while (i >= 0 && err == 0) {
2342                 if (i == depth) {
2343                         /* this is leaf block */
2344                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2345                         /* root level has p_bh == NULL, brelse() eats this */
2346                         brelse(path[i].p_bh);
2347                         path[i].p_bh = NULL;
2348                         i--;
2349                         continue;
2350                 }
2351
2352                 /* this is index block */
2353                 if (!path[i].p_hdr) {
2354                         ext_debug("initialize header\n");
2355                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2356                 }
2357
2358                 if (!path[i].p_idx) {
2359                         /* this level hasn't been touched yet */
2360                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2361                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2362                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2363                                   path[i].p_hdr,
2364                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2365                 } else {
2366                         /* we were already here, see at next index */
2367                         path[i].p_idx--;
2368                 }
2369
2370                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2371                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2372                                 path[i].p_idx);
2373                 if (ext4_ext_more_to_rm(path + i)) {
2374                         struct buffer_head *bh;
2375                         /* go to the next level */
2376                         ext_debug("move to level %d (block %llu)\n",
2377                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2378                         memset(path + i + 1, 0, sizeof(*path));
2379                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2380                         if (!bh) {
2381                                 /* should we reset i_size? */
2382                                 err = -EIO;
2383                                 break;
2384                         }
2385                         if (WARN_ON(i + 1 > depth)) {
2386                                 err = -EIO;
2387                                 break;
2388                         }
2389                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2390                                                         depth - i - 1)) {
2391                                 err = -EIO;
2392                                 break;
2393                         }
2394                         path[i + 1].p_bh = bh;
2395
2396                         /* save actual number of indexes since this
2397                          * number is changed at the next iteration */
2398                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2399                         i++;
2400                 } else {
2401                         /* we finished processing this index, go up */
2402                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2403                                 /* index is empty, remove it;
2404                                  * handle must be already prepared by the
2405                                  * truncatei_leaf() */
2406                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2407                         }
2408                         /* root level has p_bh == NULL, brelse() eats this */
2409                         brelse(path[i].p_bh);
2410                         path[i].p_bh = NULL;
2411                         i--;
2412                         ext_debug("return to level %d\n", i);
2413                 }
2414         }
2415
2416         /* TODO: flexible tree reduction should be here */
2417         if (path->p_hdr->eh_entries == 0) {
2418                 /*
2419                  * truncate to zero freed all the tree,
2420                  * so we need to correct eh_depth
2421                  */
2422                 err = ext4_ext_get_access(handle, inode, path);
2423                 if (err == 0) {
2424                         ext_inode_hdr(inode)->eh_depth = 0;
2425                         ext_inode_hdr(inode)->eh_max =
2426                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2427                         err = ext4_ext_dirty(handle, inode, path);
2428                 }
2429         }
2430 out:
2431         ext4_ext_drop_refs(path);
2432         kfree(path);
2433         if (err == -EAGAIN)
2434                 goto again;
2435         ext4_journal_stop(handle);
2436
2437         return err;
2438 }
2439
2440 /*
2441  * called at mount time
2442  */
2443 void ext4_ext_init(struct super_block *sb)
2444 {
2445         /*
2446          * possible initialization would be here
2447          */
2448
2449         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2450 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2451                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2452 #ifdef AGGRESSIVE_TEST
2453                 printk(", aggressive tests");
2454 #endif
2455 #ifdef CHECK_BINSEARCH
2456                 printk(", check binsearch");
2457 #endif
2458 #ifdef EXTENTS_STATS
2459                 printk(", stats");
2460 #endif
2461                 printk("\n");
2462 #endif
2463 #ifdef EXTENTS_STATS
2464                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2465                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2466                 EXT4_SB(sb)->s_ext_max = 0;
2467 #endif
2468         }
2469 }
2470
2471 /*
2472  * called at umount time
2473  */
2474 void ext4_ext_release(struct super_block *sb)
2475 {
2476         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2477                 return;
2478
2479 #ifdef EXTENTS_STATS
2480         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2481                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2482                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2483                         sbi->s_ext_blocks, sbi->s_ext_extents,
2484                         sbi->s_ext_blocks / sbi->s_ext_extents);
2485                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2486                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2487         }
2488 #endif
2489 }
2490
2491 /* FIXME!! we need to try to merge to left or right after zero-out  */
2492 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2493 {
2494         ext4_fsblk_t ee_pblock;
2495         unsigned int ee_len;
2496         int ret;
2497
2498         ee_len    = ext4_ext_get_actual_len(ex);
2499         ee_pblock = ext4_ext_pblock(ex);
2500
2501         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2502         if (ret > 0)
2503                 ret = 0;
2504
2505         return ret;
2506 }
2507
2508 #define EXT4_EXT_ZERO_LEN 7
2509 /*
2510  * This function is called by ext4_ext_map_blocks() if someone tries to write
2511  * to an uninitialized extent. It may result in splitting the uninitialized
2512  * extent into multiple extents (upto three - one initialized and two
2513  * uninitialized).
2514  * There are three possibilities:
2515  *   a> There is no split required: Entire extent should be initialized
2516  *   b> Splits in two extents: Write is happening at either end of the extent
2517  *   c> Splits in three extents: Somone is writing in middle of the extent
2518  */
2519 static int ext4_ext_convert_to_initialized(handle_t *handle,
2520                                            struct inode *inode,
2521                                            struct ext4_map_blocks *map,
2522                                            struct ext4_ext_path *path)
2523 {
2524         struct ext4_extent *ex, newex, orig_ex;
2525         struct ext4_extent *ex1 = NULL;
2526         struct ext4_extent *ex2 = NULL;
2527         struct ext4_extent *ex3 = NULL;
2528         struct ext4_extent_header *eh;
2529         ext4_lblk_t ee_block, eof_block;
2530         unsigned int allocated, ee_len, depth;
2531         ext4_fsblk_t newblock;
2532         int err = 0;
2533         int ret = 0;
2534         int may_zeroout;
2535
2536         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2537                 "block %llu, max_blocks %u\n", inode->i_ino,
2538                 (unsigned long long)map->m_lblk, map->m_len);
2539
2540         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2541                 inode->i_sb->s_blocksize_bits;
2542         if (eof_block < map->m_lblk + map->m_len)
2543                 eof_block = map->m_lblk + map->m_len;
2544
2545         depth = ext_depth(inode);
2546         eh = path[depth].p_hdr;
2547         ex = path[depth].p_ext;
2548         ee_block = le32_to_cpu(ex->ee_block);
2549         ee_len = ext4_ext_get_actual_len(ex);
2550         allocated = ee_len - (map->m_lblk - ee_block);
2551         newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2552
2553         ex2 = ex;
2554         orig_ex.ee_block = ex->ee_block;
2555         orig_ex.ee_len   = cpu_to_le16(ee_len);
2556         ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2557
2558         /*
2559          * It is safe to convert extent to initialized via explicit
2560          * zeroout only if extent is fully insde i_size or new_size.
2561          */
2562         may_zeroout = ee_block + ee_len <= eof_block;
2563
2564         err = ext4_ext_get_access(handle, inode, path + depth);
2565         if (err)
2566                 goto out;
2567         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2568         if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2569                 err =  ext4_ext_zeroout(inode, &orig_ex);
2570                 if (err)
2571                         goto fix_extent_len;
2572                 /* update the extent length and mark as initialized */
2573                 ex->ee_block = orig_ex.ee_block;
2574                 ex->ee_len   = orig_ex.ee_len;
2575                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2576                 ext4_ext_dirty(handle, inode, path + depth);
2577                 /* zeroed the full extent */
2578                 return allocated;
2579         }
2580
2581         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2582         if (map->m_lblk > ee_block) {
2583                 ex1 = ex;
2584                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2585                 ext4_ext_mark_uninitialized(ex1);
2586                 ex2 = &newex;
2587         }
2588         /*
2589          * for sanity, update the length of the ex2 extent before
2590          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2591          * overlap of blocks.
2592          */
2593         if (!ex1 && allocated > map->m_len)
2594                 ex2->ee_len = cpu_to_le16(map->m_len);
2595         /* ex3: to ee_block + ee_len : uninitialised */
2596         if (allocated > map->m_len) {
2597                 unsigned int newdepth;
2598                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2599                 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2600                         /*
2601                          * map->m_lblk == ee_block is handled by the zerouout
2602                          * at the beginning.
2603                          * Mark first half uninitialized.
2604                          * Mark second half initialized and zero out the
2605                          * initialized extent
2606                          */
2607                         ex->ee_block = orig_ex.ee_block;
2608                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2609                         ext4_ext_mark_uninitialized(ex);
2610                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2611                         ext4_ext_dirty(handle, inode, path + depth);
2612
2613                         ex3 = &newex;
2614                         ex3->ee_block = cpu_to_le32(map->m_lblk);
2615                         ext4_ext_store_pblock(ex3, newblock);
2616                         ex3->ee_len = cpu_to_le16(allocated);
2617                         err = ext4_ext_insert_extent(handle, inode, path,
2618                                                         ex3, 0);
2619                         if (err == -ENOSPC) {
2620                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2621                                 if (err)
2622                                         goto fix_extent_len;
2623                                 ex->ee_block = orig_ex.ee_block;
2624                                 ex->ee_len   = orig_ex.ee_len;
2625                                 ext4_ext_store_pblock(ex,
2626                                         ext4_ext_pblock(&orig_ex));
2627                                 ext4_ext_dirty(handle, inode, path + depth);
2628                                 /* blocks available from map->m_lblk */
2629                                 return allocated;
2630
2631                         } else if (err)
2632                                 goto fix_extent_len;
2633
2634                         /*
2635                          * We need to zero out the second half because
2636                          * an fallocate request can update file size and
2637                          * converting the second half to initialized extent
2638                          * implies that we can leak some junk data to user
2639                          * space.
2640                          */
2641                         err =  ext4_ext_zeroout(inode, ex3);
2642                         if (err) {
2643                                 /*
2644                                  * We should actually mark the
2645                                  * second half as uninit and return error
2646                                  * Insert would have changed the extent
2647                                  */
2648                                 depth = ext_depth(inode);
2649                                 ext4_ext_drop_refs(path);
2650                                 path = ext4_ext_find_extent(inode, map->m_lblk,
2651                                                             path);
2652                                 if (IS_ERR(path)) {
2653                                         err = PTR_ERR(path);
2654                                         return err;
2655                                 }
2656                                 /* get the second half extent details */
2657                                 ex = path[depth].p_ext;
2658                                 err = ext4_ext_get_access(handle, inode,
2659                                                                 path + depth);
2660                                 if (err)
2661                                         return err;
2662                                 ext4_ext_mark_uninitialized(ex);
2663                                 ext4_ext_dirty(handle, inode, path + depth);
2664                                 return err;
2665                         }
2666
2667                         /* zeroed the second half */
2668                         return allocated;
2669                 }
2670                 ex3 = &newex;
2671                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2672                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2673                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2674                 ext4_ext_mark_uninitialized(ex3);
2675                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2676                 if (err == -ENOSPC && may_zeroout) {
2677                         err =  ext4_ext_zeroout(inode, &orig_ex);
2678                         if (err)
2679                                 goto fix_extent_len;
2680                         /* update the extent length and mark as initialized */
2681                         ex->ee_block = orig_ex.ee_block;
2682                         ex->ee_len   = orig_ex.ee_len;
2683                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2684                         ext4_ext_dirty(handle, inode, path + depth);
2685                         /* zeroed the full extent */
2686                         /* blocks available from map->m_lblk */
2687                         return allocated;
2688
2689                 } else if (err)
2690                         goto fix_extent_len;
2691                 /*
2692                  * The depth, and hence eh & ex might change
2693                  * as part of the insert above.
2694                  */
2695                 newdepth = ext_depth(inode);
2696                 /*
2697                  * update the extent length after successful insert of the
2698                  * split extent
2699                  */
2700                 ee_len -= ext4_ext_get_actual_len(ex3);
2701                 orig_ex.ee_len = cpu_to_le16(ee_len);
2702                 may_zeroout = ee_block + ee_len <= eof_block;
2703
2704                 depth = newdepth;
2705                 ext4_ext_drop_refs(path);
2706                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2707                 if (IS_ERR(path)) {
2708                         err = PTR_ERR(path);
2709                         goto out;
2710                 }
2711                 eh = path[depth].p_hdr;
2712                 ex = path[depth].p_ext;
2713                 if (ex2 != &newex)
2714                         ex2 = ex;
2715
2716                 err = ext4_ext_get_access(handle, inode, path + depth);
2717                 if (err)
2718                         goto out;
2719
2720                 allocated = map->m_len;
2721
2722                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2723                  * to insert a extent in the middle zerout directly
2724                  * otherwise give the extent a chance to merge to left
2725                  */
2726                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2727                         map->m_lblk != ee_block && may_zeroout) {
2728                         err =  ext4_ext_zeroout(inode, &orig_ex);
2729                         if (err)
2730                                 goto fix_extent_len;
2731                         /* update the extent length and mark as initialized */
2732                         ex->ee_block = orig_ex.ee_block;
2733                         ex->ee_len   = orig_ex.ee_len;
2734                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2735                         ext4_ext_dirty(handle, inode, path + depth);
2736                         /* zero out the first half */
2737                         /* blocks available from map->m_lblk */
2738                         return allocated;
2739                 }
2740         }
2741         /*
2742          * If there was a change of depth as part of the
2743          * insertion of ex3 above, we need to update the length
2744          * of the ex1 extent again here
2745          */
2746         if (ex1 && ex1 != ex) {
2747                 ex1 = ex;
2748                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2749                 ext4_ext_mark_uninitialized(ex1);
2750                 ex2 = &newex;
2751         }
2752         /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2753         ex2->ee_block = cpu_to_le32(map->m_lblk);
2754         ext4_ext_store_pblock(ex2, newblock);
2755         ex2->ee_len = cpu_to_le16(allocated);
2756         if (ex2 != ex)
2757                 goto insert;
2758         /*
2759          * New (initialized) extent starts from the first block
2760          * in the current extent. i.e., ex2 == ex
2761          * We have to see if it can be merged with the extent
2762          * on the left.
2763          */
2764         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2765                 /*
2766                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2767                  * since it merges towards right _only_.
2768                  */
2769                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2770                 if (ret) {
2771                         err = ext4_ext_correct_indexes(handle, inode, path);
2772                         if (err)
2773                                 goto out;
2774                         depth = ext_depth(inode);
2775                         ex2--;
2776                 }
2777         }
2778         /*
2779          * Try to Merge towards right. This might be required
2780          * only when the whole extent is being written to.
2781          * i.e. ex2 == ex and ex3 == NULL.
2782          */
2783         if (!ex3) {
2784                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2785                 if (ret) {
2786                         err = ext4_ext_correct_indexes(handle, inode, path);
2787                         if (err)
2788                                 goto out;
2789                 }
2790         }
2791         /* Mark modified extent as dirty */
2792         err = ext4_ext_dirty(handle, inode, path + depth);
2793         goto out;
2794 insert:
2795         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2796         if (err == -ENOSPC && may_zeroout) {
2797                 err =  ext4_ext_zeroout(inode, &orig_ex);
2798                 if (err)
2799                         goto fix_extent_len;
2800                 /* update the extent length and mark as initialized */
2801                 ex->ee_block = orig_ex.ee_block;
2802                 ex->ee_len   = orig_ex.ee_len;
2803                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2804                 ext4_ext_dirty(handle, inode, path + depth);
2805                 /* zero out the first half */
2806                 return allocated;
2807         } else if (err)
2808                 goto fix_extent_len;
2809 out:
2810         ext4_ext_show_leaf(inode, path);
2811         return err ? err : allocated;
2812
2813 fix_extent_len:
2814         ex->ee_block = orig_ex.ee_block;
2815         ex->ee_len   = orig_ex.ee_len;
2816         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2817         ext4_ext_mark_uninitialized(ex);
2818         ext4_ext_dirty(handle, inode, path + depth);
2819         return err;
2820 }
2821
2822 /*
2823  * This function is called by ext4_ext_map_blocks() from
2824  * ext4_get_blocks_dio_write() when DIO to write
2825  * to an uninitialized extent.
2826  *
2827  * Writing to an uninitized extent may result in splitting the uninitialized
2828  * extent into multiple /intialized unintialized extents (up to three)
2829  * There are three possibilities:
2830  *   a> There is no split required: Entire extent should be uninitialized
2831  *   b> Splits in two extents: Write is happening at either end of the extent
2832  *   c> Splits in three extents: Somone is writing in middle of the extent
2833  *
2834  * One of more index blocks maybe needed if the extent tree grow after
2835  * the unintialized extent split. To prevent ENOSPC occur at the IO
2836  * complete, we need to split the uninitialized extent before DIO submit
2837  * the IO. The uninitialized extent called at this time will be split
2838  * into three uninitialized extent(at most). After IO complete, the part
2839  * being filled will be convert to initialized by the end_io callback function
2840  * via ext4_convert_unwritten_extents().
2841  *
2842  * Returns the size of uninitialized extent to be written on success.
2843  */
2844 static int ext4_split_unwritten_extents(handle_t *handle,
2845                                         struct inode *inode,
2846                                         struct ext4_map_blocks *map,
2847                                         struct ext4_ext_path *path,
2848                                         int flags)
2849 {
2850         struct ext4_extent *ex, newex, orig_ex;
2851         struct ext4_extent *ex1 = NULL;
2852         struct ext4_extent *ex2 = NULL;
2853         struct ext4_extent *ex3 = NULL;
2854         ext4_lblk_t ee_block, eof_block;
2855         unsigned int allocated, ee_len, depth;
2856         ext4_fsblk_t newblock;
2857         int err = 0;
2858         int may_zeroout;
2859
2860         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2861                 "block %llu, max_blocks %u\n", inode->i_ino,
2862                 (unsigned long long)map->m_lblk, map->m_len);
2863
2864         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2865                 inode->i_sb->s_blocksize_bits;
2866         if (eof_block < map->m_lblk + map->m_len)
2867                 eof_block = map->m_lblk + map->m_len;
2868
2869         depth = ext_depth(inode);
2870         ex = path[depth].p_ext;
2871         ee_block = le32_to_cpu(ex->ee_block);
2872         ee_len = ext4_ext_get_actual_len(ex);
2873         allocated = ee_len - (map->m_lblk - ee_block);
2874         newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2875
2876         ex2 = ex;
2877         orig_ex.ee_block = ex->ee_block;
2878         orig_ex.ee_len   = cpu_to_le16(ee_len);
2879         ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2880
2881         /*
2882          * It is safe to convert extent to initialized via explicit
2883          * zeroout only if extent is fully insde i_size or new_size.
2884          */
2885         may_zeroout = ee_block + ee_len <= eof_block;
2886
2887         /*
2888          * If the uninitialized extent begins at the same logical
2889          * block where the write begins, and the write completely
2890          * covers the extent, then we don't need to split it.
2891          */
2892         if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2893                 return allocated;
2894
2895         err = ext4_ext_get_access(handle, inode, path + depth);
2896         if (err)
2897                 goto out;
2898         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2899         if (map->m_lblk > ee_block) {
2900                 ex1 = ex;
2901                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2902                 ext4_ext_mark_uninitialized(ex1);
2903                 ex2 = &newex;
2904         }
2905         /*
2906          * for sanity, update the length of the ex2 extent before
2907          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2908          * overlap of blocks.
2909          */
2910         if (!ex1 && allocated > map->m_len)
2911                 ex2->ee_len = cpu_to_le16(map->m_len);
2912         /* ex3: to ee_block + ee_len : uninitialised */
2913         if (allocated > map->m_len) {
2914                 unsigned int newdepth;
2915                 ex3 = &newex;
2916                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2917                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2918                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2919                 ext4_ext_mark_uninitialized(ex3);
2920                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2921                 if (err == -ENOSPC && may_zeroout) {
2922                         err =  ext4_ext_zeroout(inode, &orig_ex);
2923                         if (err)
2924                                 goto fix_extent_len;
2925                         /* update the extent length and mark as initialized */
2926                         ex->ee_block = orig_ex.ee_block;
2927                         ex->ee_len   = orig_ex.ee_len;
2928                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2929                         ext4_ext_dirty(handle, inode, path + depth);
2930                         /* zeroed the full extent */
2931                         /* blocks available from map->m_lblk */
2932                         return allocated;
2933
2934                 } else if (err)
2935                         goto fix_extent_len;
2936                 /*
2937                  * The depth, and hence eh & ex might change
2938                  * as part of the insert above.
2939                  */
2940                 newdepth = ext_depth(inode);
2941                 /*
2942                  * update the extent length after successful insert of the
2943                  * split extent
2944                  */
2945                 ee_len -= ext4_ext_get_actual_len(ex3);
2946                 orig_ex.ee_len = cpu_to_le16(ee_len);
2947                 may_zeroout = ee_block + ee_len <= eof_block;
2948
2949                 depth = newdepth;
2950                 ext4_ext_drop_refs(path);
2951                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2952                 if (IS_ERR(path)) {
2953                         err = PTR_ERR(path);
2954                         goto out;
2955                 }
2956                 ex = path[depth].p_ext;
2957                 if (ex2 != &newex)
2958                         ex2 = ex;
2959
2960                 err = ext4_ext_get_access(handle, inode, path + depth);
2961                 if (err)
2962                         goto out;
2963
2964                 allocated = map->m_len;
2965         }
2966         /*
2967          * If there was a change of depth as part of the
2968          * insertion of ex3 above, we need to update the length
2969          * of the ex1 extent again here
2970          */
2971         if (ex1 && ex1 != ex) {
2972                 ex1 = ex;
2973                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2974                 ext4_ext_mark_uninitialized(ex1);
2975                 ex2 = &newex;
2976         }
2977         /*
2978          * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
2979          * using direct I/O, uninitialised still.
2980          */
2981         ex2->ee_block = cpu_to_le32(map->m_lblk);
2982         ext4_ext_store_pblock(ex2, newblock);
2983         ex2->ee_len = cpu_to_le16(allocated);
2984         ext4_ext_mark_uninitialized(ex2);
2985         if (ex2 != ex)
2986                 goto insert;
2987         /* Mark modified extent as dirty */
2988         err = ext4_ext_dirty(handle, inode, path + depth);
2989         ext_debug("out here\n");
2990         goto out;
2991 insert:
2992         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2993         if (err == -ENOSPC && may_zeroout) {
2994                 err =  ext4_ext_zeroout(inode, &orig_ex);
2995                 if (err)
2996                         goto fix_extent_len;
2997                 /* update the extent length and mark as initialized */
2998                 ex->ee_block = orig_ex.ee_block;
2999                 ex->ee_len   = orig_ex.ee_len;
3000                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3001                 ext4_ext_dirty(handle, inode, path + depth);
3002                 /* zero out the first half */
3003                 return allocated;
3004         } else if (err)
3005                 goto fix_extent_len;
3006 out:
3007         ext4_ext_show_leaf(inode, path);
3008         return err ? err : allocated;
3009
3010 fix_extent_len:
3011         ex->ee_block = orig_ex.ee_block;
3012         ex->ee_len   = orig_ex.ee_len;
3013         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3014         ext4_ext_mark_uninitialized(ex);
3015         ext4_ext_dirty(handle, inode, path + depth);
3016         return err;
3017 }
3018 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3019                                               struct inode *inode,
3020                                               struct ext4_ext_path *path)
3021 {
3022         struct ext4_extent *ex;
3023         struct ext4_extent_header *eh;
3024         int depth;
3025         int err = 0;
3026         int ret = 0;
3027
3028         depth = ext_depth(inode);
3029         eh = path[depth].p_hdr;
3030         ex = path[depth].p_ext;
3031
3032         err = ext4_ext_get_access(handle, inode, path + depth);
3033         if (err)
3034                 goto out;
3035         /* first mark the extent as initialized */
3036         ext4_ext_mark_initialized(ex);
3037
3038         /*
3039          * We have to see if it can be merged with the extent
3040          * on the left.
3041          */
3042         if (ex > EXT_FIRST_EXTENT(eh)) {
3043                 /*
3044                  * To merge left, pass "ex - 1" to try_to_merge(),
3045                  * since it merges towards right _only_.
3046                  */
3047                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3048                 if (ret) {
3049                         err = ext4_ext_correct_indexes(handle, inode, path);
3050                         if (err)
3051                                 goto out;
3052                         depth = ext_depth(inode);
3053                         ex--;
3054                 }
3055         }
3056         /*
3057          * Try to Merge towards right.
3058          */
3059         ret = ext4_ext_try_to_merge(inode, path, ex);
3060         if (ret) {
3061                 err = ext4_ext_correct_indexes(handle, inode, path);
3062                 if (err)
3063                         goto out;
3064                 depth = ext_depth(inode);
3065         }
3066         /* Mark modified extent as dirty */
3067         err = ext4_ext_dirty(handle, inode, path + depth);
3068 out:
3069         ext4_ext_show_leaf(inode, path);
3070         return err;
3071 }
3072
3073 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3074                         sector_t block, int count)
3075 {
3076         int i;
3077         for (i = 0; i < count; i++)
3078                 unmap_underlying_metadata(bdev, block + i);
3079 }
3080
3081 /*
3082  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3083  */
3084 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3085                               struct ext4_map_blocks *map,
3086                               struct ext4_ext_path *path,
3087                               unsigned int len)
3088 {
3089         int i, depth;
3090         struct ext4_extent_header *eh;
3091         struct ext4_extent *ex, *last_ex;
3092
3093         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3094                 return 0;
3095
3096         depth = ext_depth(inode);
3097         eh = path[depth].p_hdr;
3098         ex = path[depth].p_ext;
3099
3100         if (unlikely(!eh->eh_entries)) {
3101                 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3102                                  "EOFBLOCKS_FL set");
3103                 return -EIO;
3104         }
3105         last_ex = EXT_LAST_EXTENT(eh);
3106         /*
3107          * We should clear the EOFBLOCKS_FL flag if we are writing the
3108          * last block in the last extent in the file.  We test this by
3109          * first checking to see if the caller to
3110          * ext4_ext_get_blocks() was interested in the last block (or
3111          * a block beyond the last block) in the current extent.  If
3112          * this turns out to be false, we can bail out from this
3113          * function immediately.
3114          */
3115         if (map->m_lblk + len < le32_to_cpu(last_ex->ee_block) +
3116             ext4_ext_get_actual_len(last_ex))
3117                 return 0;
3118         /*
3119          * If the caller does appear to be planning to write at or
3120          * beyond the end of the current extent, we then test to see
3121          * if the current extent is the last extent in the file, by
3122          * checking to make sure it was reached via the rightmost node
3123          * at each level of the tree.
3124          */
3125         for (i = depth-1; i >= 0; i--)
3126                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3127                         return 0;
3128         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3129         return ext4_mark_inode_dirty(handle, inode);
3130 }
3131
3132 static int
3133 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3134                         struct ext4_map_blocks *map,
3135                         struct ext4_ext_path *path, int flags,
3136                         unsigned int allocated, ext4_fsblk_t newblock)
3137 {
3138         int ret = 0;
3139         int err = 0;
3140         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3141
3142         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3143                   "block %llu, max_blocks %u, flags %d, allocated %u",
3144                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3145                   flags, allocated);
3146         ext4_ext_show_leaf(inode, path);
3147
3148         /* get_block() before submit the IO, split the extent */
3149         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3150                 ret = ext4_split_unwritten_extents(handle, inode, map,
3151                                                    path, flags);
3152                 /*
3153                  * Flag the inode(non aio case) or end_io struct (aio case)
3154                  * that this IO needs to convertion to written when IO is
3155                  * completed
3156                  */
3157                 if (io)
3158                         io->flag = EXT4_IO_END_UNWRITTEN;
3159                 else
3160                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3161                 if (ext4_should_dioread_nolock(inode))
3162                         map->m_flags |= EXT4_MAP_UNINIT;
3163                 goto out;
3164         }
3165         /* IO end_io complete, convert the filled extent to written */
3166         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3167                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3168                                                         path);
3169                 if (ret >= 0) {
3170                         ext4_update_inode_fsync_trans(handle, inode, 1);
3171                         err = check_eofblocks_fl(handle, inode, map, path,
3172                                                  map->m_len);
3173                 } else
3174                         err = ret;
3175                 goto out2;
3176         }
3177         /* buffered IO case */
3178         /*
3179          * repeat fallocate creation request
3180          * we already have an unwritten extent
3181          */
3182         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3183                 goto map_out;
3184
3185         /* buffered READ or buffered write_begin() lookup */
3186         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3187                 /*
3188                  * We have blocks reserved already.  We
3189                  * return allocated blocks so that delalloc
3190                  * won't do block reservation for us.  But
3191                  * the buffer head will be unmapped so that
3192                  * a read from the block returns 0s.
3193                  */
3194                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3195                 goto out1;
3196         }
3197
3198         /* buffered write, writepage time, convert*/
3199         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3200         if (ret >= 0) {
3201                 ext4_update_inode_fsync_trans(handle, inode, 1);
3202                 err = check_eofblocks_fl(handle, inode, map, path, map->m_len);
3203                 if (err < 0)
3204                         goto out2;
3205         }
3206
3207 out:
3208         if (ret <= 0) {
3209                 err = ret;
3210                 goto out2;
3211         } else
3212                 allocated = ret;
3213         map->m_flags |= EXT4_MAP_NEW;
3214         /*
3215          * if we allocated more blocks than requested
3216          * we need to make sure we unmap the extra block
3217          * allocated. The actual needed block will get
3218          * unmapped later when we find the buffer_head marked
3219          * new.
3220          */
3221         if (allocated > map->m_len) {
3222                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3223                                         newblock + map->m_len,
3224                                         allocated - map->m_len);
3225                 allocated = map->m_len;
3226         }
3227
3228         /*
3229          * If we have done fallocate with the offset that is already
3230          * delayed allocated, we would have block reservation
3231          * and quota reservation done in the delayed write path.
3232          * But fallocate would have already updated quota and block
3233          * count for this offset. So cancel these reservation
3234          */
3235         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3236                 ext4_da_update_reserve_space(inode, allocated, 0);
3237
3238 map_out:
3239         map->m_flags |= EXT4_MAP_MAPPED;
3240 out1:
3241         if (allocated > map->m_len)
3242                 allocated = map->m_len;
3243         ext4_ext_show_leaf(inode, path);
3244         map->m_pblk = newblock;
3245         map->m_len = allocated;
3246 out2:
3247         if (path) {
3248                 ext4_ext_drop_refs(path);
3249                 kfree(path);
3250         }
3251         return err ? err : allocated;
3252 }
3253
3254 /*
3255  * Block allocation/map/preallocation routine for extents based files
3256  *
3257  *
3258  * Need to be called with
3259  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3260  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3261  *
3262  * return > 0, number of of blocks already mapped/allocated
3263  *          if create == 0 and these are pre-allocated blocks
3264  *              buffer head is unmapped
3265  *          otherwise blocks are mapped
3266  *
3267  * return = 0, if plain look up failed (blocks have not been allocated)
3268  *          buffer head is unmapped
3269  *
3270  * return < 0, error case.
3271  */
3272 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3273                         struct ext4_map_blocks *map, int flags)
3274 {
3275         struct ext4_ext_path *path = NULL;
3276         struct ext4_extent_header *eh;
3277         struct ext4_extent newex, *ex;
3278         ext4_fsblk_t newblock;
3279         int err = 0, depth, ret, cache_type;
3280         unsigned int allocated = 0;
3281         struct ext4_allocation_request ar;
3282         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3283
3284         ext_debug("blocks %u/%u requested for inode %lu\n",
3285                   map->m_lblk, map->m_len, inode->i_ino);
3286
3287         /* check in cache */
3288         cache_type = ext4_ext_in_cache(inode, map->m_lblk, &newex);
3289         if (cache_type) {
3290                 if (cache_type == EXT4_EXT_CACHE_GAP) {
3291                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3292                                 /*
3293                                  * block isn't allocated yet and
3294                                  * user doesn't want to allocate it
3295                                  */
3296                                 goto out2;
3297                         }
3298                         /* we should allocate requested block */
3299                 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3300                         /* block is already allocated */
3301                         newblock = map->m_lblk
3302                                    - le32_to_cpu(newex.ee_block)
3303                                    + ext4_ext_pblock(&newex);
3304                         /* number of remaining blocks in the extent */
3305                         allocated = ext4_ext_get_actual_len(&newex) -
3306                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3307                         goto out;
3308                 } else {
3309                         BUG();
3310                 }
3311         }
3312
3313         /* find extent for this block */
3314         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3315         if (IS_ERR(path)) {
3316                 err = PTR_ERR(path);
3317                 path = NULL;
3318                 goto out2;
3319         }
3320
3321         depth = ext_depth(inode);
3322
3323         /*
3324          * consistent leaf must not be empty;
3325          * this situation is possible, though, _during_ tree modification;
3326          * this is why assert can't be put in ext4_ext_find_extent()
3327          */
3328         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3329                 EXT4_ERROR_INODE(inode, "bad extent address "
3330                                  "lblock: %lu, depth: %d pblock %lld",
3331                                  (unsigned long) map->m_lblk, depth,
3332                                  path[depth].p_block);
3333                 err = -EIO;
3334                 goto out2;
3335         }
3336         eh = path[depth].p_hdr;
3337
3338         ex = path[depth].p_ext;
3339         if (ex) {
3340                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3341                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3342                 unsigned short ee_len;
3343
3344                 /*
3345                  * Uninitialized extents are treated as holes, except that
3346                  * we split out initialized portions during a write.
3347                  */
3348                 ee_len = ext4_ext_get_actual_len(ex);
3349                 /* if found extent covers block, simply return it */
3350                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3351                         newblock = map->m_lblk - ee_block + ee_start;
3352                         /* number of remaining blocks in the extent */
3353                         allocated = ee_len - (map->m_lblk - ee_block);
3354                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3355                                   ee_block, ee_len, newblock);
3356
3357                         /* Do not put uninitialized extent in the cache */
3358                         if (!ext4_ext_is_uninitialized(ex)) {
3359                                 ext4_ext_put_in_cache(inode, ee_block,
3360                                                         ee_len, ee_start,
3361                                                         EXT4_EXT_CACHE_EXTENT);
3362                                 goto out;
3363                         }
3364                         ret = ext4_ext_handle_uninitialized_extents(handle,
3365                                         inode, map, path, flags, allocated,
3366                                         newblock);
3367                         return ret;
3368                 }
3369         }
3370
3371         /*
3372          * requested block isn't allocated yet;
3373          * we couldn't try to create block if create flag is zero
3374          */
3375         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3376                 /*
3377                  * put just found gap into cache to speed up
3378                  * subsequent requests
3379                  */
3380                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3381                 goto out2;
3382         }
3383         /*
3384          * Okay, we need to do block allocation.
3385          */
3386
3387         /* find neighbour allocated blocks */
3388         ar.lleft = map->m_lblk;
3389         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3390         if (err)
3391                 goto out2;
3392         ar.lright = map->m_lblk;
3393         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3394         if (err)
3395                 goto out2;
3396
3397         /*
3398          * See if request is beyond maximum number of blocks we can have in
3399          * a single extent. For an initialized extent this limit is
3400          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3401          * EXT_UNINIT_MAX_LEN.
3402          */
3403         if (map->m_len > EXT_INIT_MAX_LEN &&
3404             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3405                 map->m_len = EXT_INIT_MAX_LEN;
3406         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3407                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3408                 map->m_len = EXT_UNINIT_MAX_LEN;
3409
3410         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3411         newex.ee_block = cpu_to_le32(map->m_lblk);
3412         newex.ee_len = cpu_to_le16(map->m_len);
3413         err = ext4_ext_check_overlap(inode, &newex, path);
3414         if (err)
3415                 allocated = ext4_ext_get_actual_len(&newex);
3416         else
3417                 allocated = map->m_len;
3418
3419         /* allocate new block */
3420         ar.inode = inode;
3421         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3422         ar.logical = map->m_lblk;
3423         ar.len = allocated;
3424         if (S_ISREG(inode->i_mode))
3425                 ar.flags = EXT4_MB_HINT_DATA;
3426         else
3427                 /* disable in-core preallocation for non-regular files */
3428                 ar.flags = 0;
3429         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3430         if (!newblock)
3431                 goto out2;
3432         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3433                   ar.goal, newblock, allocated);
3434
3435         /* try to insert new extent into found leaf and return */
3436         ext4_ext_store_pblock(&newex, newblock);
3437         newex.ee_len = cpu_to_le16(ar.len);
3438         /* Mark uninitialized */
3439         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3440                 ext4_ext_mark_uninitialized(&newex);
3441                 /*
3442                  * io_end structure was created for every IO write to an
3443                  * uninitialized extent. To avoid unecessary conversion,
3444                  * here we flag the IO that really needs the conversion.
3445                  * For non asycn direct IO case, flag the inode state
3446                  * that we need to perform convertion when IO is done.
3447                  */
3448                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3449                         if (io)
3450                                 io->flag = EXT4_IO_END_UNWRITTEN;
3451                         else
3452                                 ext4_set_inode_state(inode,
3453                                                      EXT4_STATE_DIO_UNWRITTEN);
3454                 }
3455                 if (ext4_should_dioread_nolock(inode))
3456                         map->m_flags |= EXT4_MAP_UNINIT;
3457         }
3458
3459         err = check_eofblocks_fl(handle, inode, map, path, ar.len);
3460         if (err)
3461                 goto out2;
3462
3463         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3464         if (err) {
3465                 /* free data blocks we just allocated */
3466                 /* not a good idea to call discard here directly,
3467                  * but otherwise we'd need to call it every free() */
3468                 ext4_discard_preallocations(inode);
3469                 ext4_free_blocks(handle, inode, 0, ext4_ext_pblock(&newex),
3470                                  ext4_ext_get_actual_len(&newex), 0);
3471                 goto out2;
3472         }
3473
3474         /* previous routine could use block we allocated */
3475         newblock = ext4_ext_pblock(&newex);
3476         allocated = ext4_ext_get_actual_len(&newex);
3477         if (allocated > map->m_len)
3478                 allocated = map->m_len;
3479         map->m_flags |= EXT4_MAP_NEW;
3480
3481         /*
3482          * Update reserved blocks/metadata blocks after successful
3483          * block allocation which had been deferred till now.
3484          */
3485         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3486                 ext4_da_update_reserve_space(inode, allocated, 1);
3487
3488         /*
3489          * Cache the extent and update transaction to commit on fdatasync only
3490          * when it is _not_ an uninitialized extent.
3491          */
3492         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3493                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock,
3494                                                 EXT4_EXT_CACHE_EXTENT);
3495                 ext4_update_inode_fsync_trans(handle, inode, 1);
3496         } else
3497                 ext4_update_inode_fsync_trans(handle, inode, 0);
3498 out:
3499         if (allocated > map->m_len)
3500                 allocated = map->m_len;
3501         ext4_ext_show_leaf(inode, path);
3502         map->m_flags |= EXT4_MAP_MAPPED;
3503         map->m_pblk = newblock;
3504         map->m_len = allocated;
3505 out2:
3506         if (path) {
3507                 ext4_ext_drop_refs(path);
3508                 kfree(path);
3509         }
3510         return err ? err : allocated;
3511 }
3512
3513 void ext4_ext_truncate(struct inode *inode)
3514 {
3515         struct address_space *mapping = inode->i_mapping;
3516         struct super_block *sb = inode->i_sb;
3517         ext4_lblk_t last_block;
3518         handle_t *handle;
3519         int err = 0;
3520
3521         /*
3522          * probably first extent we're gonna free will be last in block
3523          */
3524         err = ext4_writepage_trans_blocks(inode);
3525         handle = ext4_journal_start(inode, err);
3526         if (IS_ERR(handle))
3527                 return;
3528
3529         if (inode->i_size & (sb->s_blocksize - 1))
3530                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3531
3532         if (ext4_orphan_add(handle, inode))
3533                 goto out_stop;
3534
3535         down_write(&EXT4_I(inode)->i_data_sem);
3536         ext4_ext_invalidate_cache(inode);
3537
3538         ext4_discard_preallocations(inode);
3539
3540         /*
3541          * TODO: optimization is possible here.
3542          * Probably we need not scan at all,
3543          * because page truncation is enough.
3544          */
3545
3546         /* we have to know where to truncate from in crash case */
3547         EXT4_I(inode)->i_disksize = inode->i_size;
3548         ext4_mark_inode_dirty(handle, inode);
3549
3550         last_block = (inode->i_size + sb->s_blocksize - 1)
3551                         >> EXT4_BLOCK_SIZE_BITS(sb);
3552         err = ext4_ext_remove_space(inode, last_block);
3553
3554         /* In a multi-transaction truncate, we only make the final
3555          * transaction synchronous.
3556          */
3557         if (IS_SYNC(inode))
3558                 ext4_handle_sync(handle);
3559
3560 out_stop:
3561         up_write(&EXT4_I(inode)->i_data_sem);
3562         /*
3563          * If this was a simple ftruncate() and the file will remain alive,
3564          * then we need to clear up the orphan record which we created above.
3565          * However, if this was a real unlink then we were called by
3566          * ext4_delete_inode(), and we allow that function to clean up the
3567          * orphan info for us.
3568          */
3569         if (inode->i_nlink)
3570                 ext4_orphan_del(handle, inode);
3571
3572         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3573         ext4_mark_inode_dirty(handle, inode);
3574         ext4_journal_stop(handle);
3575 }
3576
3577 static void ext4_falloc_update_inode(struct inode *inode,
3578                                 int mode, loff_t new_size, int update_ctime)
3579 {
3580         struct timespec now;
3581
3582         if (update_ctime) {
3583                 now = current_fs_time(inode->i_sb);
3584                 if (!timespec_equal(&inode->i_ctime, &now))
3585                         inode->i_ctime = now;
3586         }
3587         /*
3588          * Update only when preallocation was requested beyond
3589          * the file size.
3590          */
3591         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3592                 if (new_size > i_size_read(inode))
3593                         i_size_write(inode, new_size);
3594                 if (new_size > EXT4_I(inode)->i_disksize)
3595                         ext4_update_i_disksize(inode, new_size);
3596         } else {
3597                 /*
3598                  * Mark that we allocate beyond EOF so the subsequent truncate
3599                  * can proceed even if the new size is the same as i_size.
3600                  */
3601                 if (new_size > i_size_read(inode))
3602                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3603         }
3604
3605 }
3606
3607 /*
3608  * preallocate space for a file. This implements ext4's fallocate inode
3609  * operation, which gets called from sys_fallocate system call.
3610  * For block-mapped files, posix_fallocate should fall back to the method
3611  * of writing zeroes to the required new blocks (the same behavior which is
3612  * expected for file systems which do not support fallocate() system call).
3613  */
3614 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3615 {
3616         handle_t *handle;
3617         loff_t new_size;
3618         unsigned int max_blocks;
3619         int ret = 0;
3620         int ret2 = 0;
3621         int retries = 0;
3622         struct ext4_map_blocks map;
3623         unsigned int credits, blkbits = inode->i_blkbits;
3624
3625         /*
3626          * currently supporting (pre)allocate mode for extent-based
3627          * files _only_
3628          */
3629         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3630                 return -EOPNOTSUPP;
3631
3632         /* preallocation to directories is currently not supported */
3633         if (S_ISDIR(inode->i_mode))
3634                 return -ENODEV;
3635
3636         map.m_lblk = offset >> blkbits;
3637         /*
3638          * We can't just convert len to max_blocks because
3639          * If blocksize = 4096 offset = 3072 and len = 2048
3640          */
3641         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3642                 - map.m_lblk;
3643         /*
3644          * credits to insert 1 extent into extent tree
3645          */
3646         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3647         mutex_lock(&inode->i_mutex);
3648         ret = inode_newsize_ok(inode, (len + offset));
3649         if (ret) {
3650                 mutex_unlock(&inode->i_mutex);
3651                 return ret;
3652         }
3653 retry:
3654         while (ret >= 0 && ret < max_blocks) {
3655                 map.m_lblk = map.m_lblk + ret;
3656                 map.m_len = max_blocks = max_blocks - ret;
3657                 handle = ext4_journal_start(inode, credits);
3658                 if (IS_ERR(handle)) {
3659                         ret = PTR_ERR(handle);
3660                         break;
3661                 }
3662                 ret = ext4_map_blocks(handle, inode, &map,
3663                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3664                 if (ret <= 0) {
3665 #ifdef EXT4FS_DEBUG
3666                         WARN_ON(ret <= 0);
3667                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3668                                     "returned error inode#%lu, block=%u, "
3669                                     "max_blocks=%u", __func__,
3670                                     inode->i_ino, map.m_lblk, max_blocks);
3671 #endif
3672                         ext4_mark_inode_dirty(handle, inode);
3673                         ret2 = ext4_journal_stop(handle);
3674                         break;
3675                 }
3676                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3677                                                 blkbits) >> blkbits))
3678                         new_size = offset + len;
3679                 else
3680                         new_size = (map.m_lblk + ret) << blkbits;
3681
3682                 ext4_falloc_update_inode(inode, mode, new_size,
3683                                          (map.m_flags & EXT4_MAP_NEW));
3684                 ext4_mark_inode_dirty(handle, inode);
3685                 ret2 = ext4_journal_stop(handle);
3686                 if (ret2)
3687                         break;
3688         }
3689         if (ret == -ENOSPC &&
3690                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3691                 ret = 0;
3692                 goto retry;
3693         }
3694         mutex_unlock(&inode->i_mutex);
3695         return ret > 0 ? ret2 : ret;
3696 }
3697
3698 /*
3699  * This function convert a range of blocks to written extents
3700  * The caller of this function will pass the start offset and the size.
3701  * all unwritten extents within this range will be converted to
3702  * written extents.
3703  *
3704  * This function is called from the direct IO end io call back
3705  * function, to convert the fallocated extents after IO is completed.
3706  * Returns 0 on success.
3707  */
3708 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3709                                     ssize_t len)
3710 {
3711         handle_t *handle;
3712         unsigned int max_blocks;
3713         int ret = 0;
3714         int ret2 = 0;
3715         struct ext4_map_blocks map;
3716         unsigned int credits, blkbits = inode->i_blkbits;
3717
3718         map.m_lblk = offset >> blkbits;
3719         /*
3720          * We can't just convert len to max_blocks because
3721          * If blocksize = 4096 offset = 3072 and len = 2048
3722          */
3723         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3724                       map.m_lblk);
3725         /*
3726          * credits to insert 1 extent into extent tree
3727          */
3728         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3729         while (ret >= 0 && ret < max_blocks) {
3730                 map.m_lblk += ret;
3731                 map.m_len = (max_blocks -= ret);
3732                 handle = ext4_journal_start(inode, credits);
3733                 if (IS_ERR(handle)) {
3734                         ret = PTR_ERR(handle);
3735                         break;
3736                 }
3737                 ret = ext4_map_blocks(handle, inode, &map,
3738                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3739                 if (ret <= 0) {
3740                         WARN_ON(ret <= 0);
3741                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3742                                     "returned error inode#%lu, block=%u, "
3743                                     "max_blocks=%u", __func__,
3744                                     inode->i_ino, map.m_lblk, map.m_len);
3745                 }
3746                 ext4_mark_inode_dirty(handle, inode);
3747                 ret2 = ext4_journal_stop(handle);
3748                 if (ret <= 0 || ret2 )
3749                         break;
3750         }
3751         return ret > 0 ? ret2 : ret;
3752 }
3753 /*
3754  * Callback function called for each extent to gather FIEMAP information.
3755  */
3756 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3757                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3758                        void *data)
3759 {
3760         struct fiemap_extent_info *fieinfo = data;
3761         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3762         __u64   logical;
3763         __u64   physical;
3764         __u64   length;
3765         __u32   flags = 0;
3766         int     error;
3767
3768         logical =  (__u64)newex->ec_block << blksize_bits;
3769
3770         if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3771                 pgoff_t offset;
3772                 struct page *page;
3773                 struct buffer_head *bh = NULL;
3774
3775                 offset = logical >> PAGE_SHIFT;
3776                 page = find_get_page(inode->i_mapping, offset);
3777                 if (!page || !page_has_buffers(page))
3778                         return EXT_CONTINUE;
3779
3780                 bh = page_buffers(page);
3781
3782                 if (!bh)
3783                         return EXT_CONTINUE;
3784
3785                 if (buffer_delay(bh)) {
3786                         flags |= FIEMAP_EXTENT_DELALLOC;
3787                         page_cache_release(page);
3788                 } else {
3789                         page_cache_release(page);
3790                         return EXT_CONTINUE;
3791                 }
3792         }
3793
3794         physical = (__u64)newex->ec_start << blksize_bits;
3795         length =   (__u64)newex->ec_len << blksize_bits;
3796
3797         if (ex && ext4_ext_is_uninitialized(ex))
3798                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3799
3800         /*
3801          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3802          *
3803          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3804          * this also indicates no more allocated blocks.
3805          *
3806          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3807          */
3808         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3809             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3810                 loff_t size = i_size_read(inode);
3811                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3812
3813                 flags |= FIEMAP_EXTENT_LAST;
3814                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3815                     logical+length > size)
3816                         length = (size - logical + bs - 1) & ~(bs-1);
3817         }
3818
3819         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3820                                         length, flags);
3821         if (error < 0)
3822                 return error;
3823         if (error == 1)
3824                 return EXT_BREAK;
3825
3826         return EXT_CONTINUE;
3827 }
3828
3829 /* fiemap flags we can handle specified here */
3830 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3831
3832 static int ext4_xattr_fiemap(struct inode *inode,
3833                                 struct fiemap_extent_info *fieinfo)
3834 {
3835         __u64 physical = 0;
3836         __u64 length;
3837         __u32 flags = FIEMAP_EXTENT_LAST;
3838         int blockbits = inode->i_sb->s_blocksize_bits;
3839         int error = 0;
3840
3841         /* in-inode? */
3842         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3843                 struct ext4_iloc iloc;
3844                 int offset;     /* offset of xattr in inode */
3845
3846                 error = ext4_get_inode_loc(inode, &iloc);
3847                 if (error)
3848                         return error;
3849                 physical = iloc.bh->b_blocknr << blockbits;
3850                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3851                                 EXT4_I(inode)->i_extra_isize;
3852                 physical += offset;
3853                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3854                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3855                 brelse(iloc.bh);
3856         } else { /* external block */
3857                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3858                 length = inode->i_sb->s_blocksize;
3859         }
3860
3861         if (physical)
3862                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3863                                                 length, flags);
3864         return (error < 0 ? error : 0);
3865 }
3866
3867 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3868                 __u64 start, __u64 len)
3869 {
3870         ext4_lblk_t start_blk;
3871         int error = 0;
3872
3873         /* fallback to generic here if not in extents fmt */
3874         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3875                 return generic_block_fiemap(inode, fieinfo, start, len,
3876                         ext4_get_block);
3877
3878         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3879                 return -EBADR;
3880
3881         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3882                 error = ext4_xattr_fiemap(inode, fieinfo);
3883         } else {
3884                 ext4_lblk_t len_blks;
3885                 __u64 last_blk;
3886
3887                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3888                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3889                 if (last_blk >= EXT_MAX_BLOCK)
3890                         last_blk = EXT_MAX_BLOCK-1;
3891                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3892
3893                 /*
3894                  * Walk the extent tree gathering extent information.
3895                  * ext4_ext_fiemap_cb will push extents back to user.
3896                  */
3897                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3898                                           ext4_ext_fiemap_cb, fieinfo);
3899         }
3900
3901         return error;
3902 }
3903