ext4: fix ext4_ext_direct_IO()'s return value after converting uninit extents
[linux-2.6.git] / fs / ext4 / extents.c
1 /*
2  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3  * Written by Alex Tomas <alex@clusterfs.com>
4  *
5  * Architecture independence:
6  *   Copyright (c) 2005, Bull S.A.
7  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public Licens
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
21  */
22
23 /*
24  * Extents support for EXT4
25  *
26  * TODO:
27  *   - ext4*_error() should be used in some situations
28  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29  *   - smart tree reduction
30  */
31
32 #include <linux/module.h>
33 #include <linux/fs.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
46
47
48 /*
49  * ext_pblock:
50  * combine low and high parts of physical block number into ext4_fsblk_t
51  */
52 ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
53 {
54         ext4_fsblk_t block;
55
56         block = le32_to_cpu(ex->ee_start_lo);
57         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
58         return block;
59 }
60
61 /*
62  * idx_pblock:
63  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64  */
65 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
66 {
67         ext4_fsblk_t block;
68
69         block = le32_to_cpu(ix->ei_leaf_lo);
70         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
71         return block;
72 }
73
74 /*
75  * ext4_ext_store_pblock:
76  * stores a large physical block number into an extent struct,
77  * breaking it into parts
78  */
79 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
80 {
81         ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
82         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
83 }
84
85 /*
86  * ext4_idx_store_pblock:
87  * stores a large physical block number into an index struct,
88  * breaking it into parts
89  */
90 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
91 {
92         ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
93         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
94 }
95
96 static int ext4_ext_truncate_extend_restart(handle_t *handle,
97                                             struct inode *inode,
98                                             int needed)
99 {
100         int err;
101
102         if (!ext4_handle_valid(handle))
103                 return 0;
104         if (handle->h_buffer_credits > needed)
105                 return 0;
106         err = ext4_journal_extend(handle, needed);
107         if (err <= 0)
108                 return err;
109         err = ext4_truncate_restart_trans(handle, inode, needed);
110         /*
111          * We have dropped i_data_sem so someone might have cached again
112          * an extent we are going to truncate.
113          */
114         ext4_ext_invalidate_cache(inode);
115
116         return err;
117 }
118
119 /*
120  * could return:
121  *  - EROFS
122  *  - ENOMEM
123  */
124 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
125                                 struct ext4_ext_path *path)
126 {
127         if (path->p_bh) {
128                 /* path points to block */
129                 return ext4_journal_get_write_access(handle, path->p_bh);
130         }
131         /* path points to leaf/index in inode body */
132         /* we use in-core data, no need to protect them */
133         return 0;
134 }
135
136 /*
137  * could return:
138  *  - EROFS
139  *  - ENOMEM
140  *  - EIO
141  */
142 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
143                                 struct ext4_ext_path *path)
144 {
145         int err;
146         if (path->p_bh) {
147                 /* path points to block */
148                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
149         } else {
150                 /* path points to leaf/index in inode body */
151                 err = ext4_mark_inode_dirty(handle, inode);
152         }
153         return err;
154 }
155
156 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
157                               struct ext4_ext_path *path,
158                               ext4_lblk_t block)
159 {
160         struct ext4_inode_info *ei = EXT4_I(inode);
161         ext4_fsblk_t bg_start;
162         ext4_fsblk_t last_block;
163         ext4_grpblk_t colour;
164         ext4_group_t block_group;
165         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
166         int depth;
167
168         if (path) {
169                 struct ext4_extent *ex;
170                 depth = path->p_depth;
171
172                 /* try to predict block placement */
173                 ex = path[depth].p_ext;
174                 if (ex)
175                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
176
177                 /* it looks like index is empty;
178                  * try to find starting block from index itself */
179                 if (path[depth].p_bh)
180                         return path[depth].p_bh->b_blocknr;
181         }
182
183         /* OK. use inode's group */
184         block_group = ei->i_block_group;
185         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
186                 /*
187                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
188                  * block groups per flexgroup, reserve the first block 
189                  * group for directories and special files.  Regular 
190                  * files will start at the second block group.  This
191                  * tends to speed up directory access and improves 
192                  * fsck times.
193                  */
194                 block_group &= ~(flex_size-1);
195                 if (S_ISREG(inode->i_mode))
196                         block_group++;
197         }
198         bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
199                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
200         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
201
202         /*
203          * If we are doing delayed allocation, we don't need take
204          * colour into account.
205          */
206         if (test_opt(inode->i_sb, DELALLOC))
207                 return bg_start;
208
209         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
210                 colour = (current->pid % 16) *
211                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
212         else
213                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
214         return bg_start + colour + block;
215 }
216
217 /*
218  * Allocation for a meta data block
219  */
220 static ext4_fsblk_t
221 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
222                         struct ext4_ext_path *path,
223                         struct ext4_extent *ex, int *err)
224 {
225         ext4_fsblk_t goal, newblock;
226
227         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
228         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
229         return newblock;
230 }
231
232 static inline int ext4_ext_space_block(struct inode *inode, int check)
233 {
234         int size;
235
236         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
237                         / sizeof(struct ext4_extent);
238         if (!check) {
239 #ifdef AGGRESSIVE_TEST
240                 if (size > 6)
241                         size = 6;
242 #endif
243         }
244         return size;
245 }
246
247 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
248 {
249         int size;
250
251         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
252                         / sizeof(struct ext4_extent_idx);
253         if (!check) {
254 #ifdef AGGRESSIVE_TEST
255                 if (size > 5)
256                         size = 5;
257 #endif
258         }
259         return size;
260 }
261
262 static inline int ext4_ext_space_root(struct inode *inode, int check)
263 {
264         int size;
265
266         size = sizeof(EXT4_I(inode)->i_data);
267         size -= sizeof(struct ext4_extent_header);
268         size /= sizeof(struct ext4_extent);
269         if (!check) {
270 #ifdef AGGRESSIVE_TEST
271                 if (size > 3)
272                         size = 3;
273 #endif
274         }
275         return size;
276 }
277
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
279 {
280         int size;
281
282         size = sizeof(EXT4_I(inode)->i_data);
283         size -= sizeof(struct ext4_extent_header);
284         size /= sizeof(struct ext4_extent_idx);
285         if (!check) {
286 #ifdef AGGRESSIVE_TEST
287                 if (size > 4)
288                         size = 4;
289 #endif
290         }
291         return size;
292 }
293
294 /*
295  * Calculate the number of metadata blocks needed
296  * to allocate @blocks
297  * Worse case is one block per extent
298  */
299 int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
300 {
301         int lcap, icap, rcap, leafs, idxs, num;
302         int newextents = blocks;
303
304         rcap = ext4_ext_space_root_idx(inode, 0);
305         lcap = ext4_ext_space_block(inode, 0);
306         icap = ext4_ext_space_block_idx(inode, 0);
307
308         /* number of new leaf blocks needed */
309         num = leafs = (newextents + lcap - 1) / lcap;
310
311         /*
312          * Worse case, we need separate index block(s)
313          * to link all new leaf blocks
314          */
315         idxs = (leafs + icap - 1) / icap;
316         do {
317                 num += idxs;
318                 idxs = (idxs + icap - 1) / icap;
319         } while (idxs > rcap);
320
321         return num;
322 }
323
324 static int
325 ext4_ext_max_entries(struct inode *inode, int depth)
326 {
327         int max;
328
329         if (depth == ext_depth(inode)) {
330                 if (depth == 0)
331                         max = ext4_ext_space_root(inode, 1);
332                 else
333                         max = ext4_ext_space_root_idx(inode, 1);
334         } else {
335                 if (depth == 0)
336                         max = ext4_ext_space_block(inode, 1);
337                 else
338                         max = ext4_ext_space_block_idx(inode, 1);
339         }
340
341         return max;
342 }
343
344 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
345 {
346         ext4_fsblk_t block = ext_pblock(ext);
347         int len = ext4_ext_get_actual_len(ext);
348
349         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
350 }
351
352 static int ext4_valid_extent_idx(struct inode *inode,
353                                 struct ext4_extent_idx *ext_idx)
354 {
355         ext4_fsblk_t block = idx_pblock(ext_idx);
356
357         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
358 }
359
360 static int ext4_valid_extent_entries(struct inode *inode,
361                                 struct ext4_extent_header *eh,
362                                 int depth)
363 {
364         struct ext4_extent *ext;
365         struct ext4_extent_idx *ext_idx;
366         unsigned short entries;
367         if (eh->eh_entries == 0)
368                 return 1;
369
370         entries = le16_to_cpu(eh->eh_entries);
371
372         if (depth == 0) {
373                 /* leaf entries */
374                 ext = EXT_FIRST_EXTENT(eh);
375                 while (entries) {
376                         if (!ext4_valid_extent(inode, ext))
377                                 return 0;
378                         ext++;
379                         entries--;
380                 }
381         } else {
382                 ext_idx = EXT_FIRST_INDEX(eh);
383                 while (entries) {
384                         if (!ext4_valid_extent_idx(inode, ext_idx))
385                                 return 0;
386                         ext_idx++;
387                         entries--;
388                 }
389         }
390         return 1;
391 }
392
393 static int __ext4_ext_check(const char *function, struct inode *inode,
394                                         struct ext4_extent_header *eh,
395                                         int depth)
396 {
397         const char *error_msg;
398         int max = 0;
399
400         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
401                 error_msg = "invalid magic";
402                 goto corrupted;
403         }
404         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
405                 error_msg = "unexpected eh_depth";
406                 goto corrupted;
407         }
408         if (unlikely(eh->eh_max == 0)) {
409                 error_msg = "invalid eh_max";
410                 goto corrupted;
411         }
412         max = ext4_ext_max_entries(inode, depth);
413         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
414                 error_msg = "too large eh_max";
415                 goto corrupted;
416         }
417         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
418                 error_msg = "invalid eh_entries";
419                 goto corrupted;
420         }
421         if (!ext4_valid_extent_entries(inode, eh, depth)) {
422                 error_msg = "invalid extent entries";
423                 goto corrupted;
424         }
425         return 0;
426
427 corrupted:
428         ext4_error(inode->i_sb, function,
429                         "bad header/extent in inode #%lu: %s - magic %x, "
430                         "entries %u, max %u(%u), depth %u(%u)",
431                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
432                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
433                         max, le16_to_cpu(eh->eh_depth), depth);
434
435         return -EIO;
436 }
437
438 #define ext4_ext_check(inode, eh, depth)        \
439         __ext4_ext_check(__func__, inode, eh, depth)
440
441 int ext4_ext_check_inode(struct inode *inode)
442 {
443         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
444 }
445
446 #ifdef EXT_DEBUG
447 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
448 {
449         int k, l = path->p_depth;
450
451         ext_debug("path:");
452         for (k = 0; k <= l; k++, path++) {
453                 if (path->p_idx) {
454                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
455                             idx_pblock(path->p_idx));
456                 } else if (path->p_ext) {
457                         ext_debug("  %d:[%d]%d:%llu ",
458                                   le32_to_cpu(path->p_ext->ee_block),
459                                   ext4_ext_is_uninitialized(path->p_ext),
460                                   ext4_ext_get_actual_len(path->p_ext),
461                                   ext_pblock(path->p_ext));
462                 } else
463                         ext_debug("  []");
464         }
465         ext_debug("\n");
466 }
467
468 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
469 {
470         int depth = ext_depth(inode);
471         struct ext4_extent_header *eh;
472         struct ext4_extent *ex;
473         int i;
474
475         if (!path)
476                 return;
477
478         eh = path[depth].p_hdr;
479         ex = EXT_FIRST_EXTENT(eh);
480
481         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
482
483         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
484                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
485                           ext4_ext_is_uninitialized(ex),
486                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
487         }
488         ext_debug("\n");
489 }
490 #else
491 #define ext4_ext_show_path(inode, path)
492 #define ext4_ext_show_leaf(inode, path)
493 #endif
494
495 void ext4_ext_drop_refs(struct ext4_ext_path *path)
496 {
497         int depth = path->p_depth;
498         int i;
499
500         for (i = 0; i <= depth; i++, path++)
501                 if (path->p_bh) {
502                         brelse(path->p_bh);
503                         path->p_bh = NULL;
504                 }
505 }
506
507 /*
508  * ext4_ext_binsearch_idx:
509  * binary search for the closest index of the given block
510  * the header must be checked before calling this
511  */
512 static void
513 ext4_ext_binsearch_idx(struct inode *inode,
514                         struct ext4_ext_path *path, ext4_lblk_t block)
515 {
516         struct ext4_extent_header *eh = path->p_hdr;
517         struct ext4_extent_idx *r, *l, *m;
518
519
520         ext_debug("binsearch for %u(idx):  ", block);
521
522         l = EXT_FIRST_INDEX(eh) + 1;
523         r = EXT_LAST_INDEX(eh);
524         while (l <= r) {
525                 m = l + (r - l) / 2;
526                 if (block < le32_to_cpu(m->ei_block))
527                         r = m - 1;
528                 else
529                         l = m + 1;
530                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
531                                 m, le32_to_cpu(m->ei_block),
532                                 r, le32_to_cpu(r->ei_block));
533         }
534
535         path->p_idx = l - 1;
536         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
537                   idx_pblock(path->p_idx));
538
539 #ifdef CHECK_BINSEARCH
540         {
541                 struct ext4_extent_idx *chix, *ix;
542                 int k;
543
544                 chix = ix = EXT_FIRST_INDEX(eh);
545                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
546                   if (k != 0 &&
547                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
548                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
549                                        "first=0x%p\n", k,
550                                        ix, EXT_FIRST_INDEX(eh));
551                                 printk(KERN_DEBUG "%u <= %u\n",
552                                        le32_to_cpu(ix->ei_block),
553                                        le32_to_cpu(ix[-1].ei_block));
554                         }
555                         BUG_ON(k && le32_to_cpu(ix->ei_block)
556                                            <= le32_to_cpu(ix[-1].ei_block));
557                         if (block < le32_to_cpu(ix->ei_block))
558                                 break;
559                         chix = ix;
560                 }
561                 BUG_ON(chix != path->p_idx);
562         }
563 #endif
564
565 }
566
567 /*
568  * ext4_ext_binsearch:
569  * binary search for closest extent of the given block
570  * the header must be checked before calling this
571  */
572 static void
573 ext4_ext_binsearch(struct inode *inode,
574                 struct ext4_ext_path *path, ext4_lblk_t block)
575 {
576         struct ext4_extent_header *eh = path->p_hdr;
577         struct ext4_extent *r, *l, *m;
578
579         if (eh->eh_entries == 0) {
580                 /*
581                  * this leaf is empty:
582                  * we get such a leaf in split/add case
583                  */
584                 return;
585         }
586
587         ext_debug("binsearch for %u:  ", block);
588
589         l = EXT_FIRST_EXTENT(eh) + 1;
590         r = EXT_LAST_EXTENT(eh);
591
592         while (l <= r) {
593                 m = l + (r - l) / 2;
594                 if (block < le32_to_cpu(m->ee_block))
595                         r = m - 1;
596                 else
597                         l = m + 1;
598                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
599                                 m, le32_to_cpu(m->ee_block),
600                                 r, le32_to_cpu(r->ee_block));
601         }
602
603         path->p_ext = l - 1;
604         ext_debug("  -> %d:%llu:[%d]%d ",
605                         le32_to_cpu(path->p_ext->ee_block),
606                         ext_pblock(path->p_ext),
607                         ext4_ext_is_uninitialized(path->p_ext),
608                         ext4_ext_get_actual_len(path->p_ext));
609
610 #ifdef CHECK_BINSEARCH
611         {
612                 struct ext4_extent *chex, *ex;
613                 int k;
614
615                 chex = ex = EXT_FIRST_EXTENT(eh);
616                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
617                         BUG_ON(k && le32_to_cpu(ex->ee_block)
618                                           <= le32_to_cpu(ex[-1].ee_block));
619                         if (block < le32_to_cpu(ex->ee_block))
620                                 break;
621                         chex = ex;
622                 }
623                 BUG_ON(chex != path->p_ext);
624         }
625 #endif
626
627 }
628
629 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
630 {
631         struct ext4_extent_header *eh;
632
633         eh = ext_inode_hdr(inode);
634         eh->eh_depth = 0;
635         eh->eh_entries = 0;
636         eh->eh_magic = EXT4_EXT_MAGIC;
637         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
638         ext4_mark_inode_dirty(handle, inode);
639         ext4_ext_invalidate_cache(inode);
640         return 0;
641 }
642
643 struct ext4_ext_path *
644 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
645                                         struct ext4_ext_path *path)
646 {
647         struct ext4_extent_header *eh;
648         struct buffer_head *bh;
649         short int depth, i, ppos = 0, alloc = 0;
650
651         eh = ext_inode_hdr(inode);
652         depth = ext_depth(inode);
653
654         /* account possible depth increase */
655         if (!path) {
656                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
657                                 GFP_NOFS);
658                 if (!path)
659                         return ERR_PTR(-ENOMEM);
660                 alloc = 1;
661         }
662         path[0].p_hdr = eh;
663         path[0].p_bh = NULL;
664
665         i = depth;
666         /* walk through the tree */
667         while (i) {
668                 int need_to_validate = 0;
669
670                 ext_debug("depth %d: num %d, max %d\n",
671                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
672
673                 ext4_ext_binsearch_idx(inode, path + ppos, block);
674                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
675                 path[ppos].p_depth = i;
676                 path[ppos].p_ext = NULL;
677
678                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
679                 if (unlikely(!bh))
680                         goto err;
681                 if (!bh_uptodate_or_lock(bh)) {
682                         if (bh_submit_read(bh) < 0) {
683                                 put_bh(bh);
684                                 goto err;
685                         }
686                         /* validate the extent entries */
687                         need_to_validate = 1;
688                 }
689                 eh = ext_block_hdr(bh);
690                 ppos++;
691                 BUG_ON(ppos > depth);
692                 path[ppos].p_bh = bh;
693                 path[ppos].p_hdr = eh;
694                 i--;
695
696                 if (need_to_validate && ext4_ext_check(inode, eh, i))
697                         goto err;
698         }
699
700         path[ppos].p_depth = i;
701         path[ppos].p_ext = NULL;
702         path[ppos].p_idx = NULL;
703
704         /* find extent */
705         ext4_ext_binsearch(inode, path + ppos, block);
706         /* if not an empty leaf */
707         if (path[ppos].p_ext)
708                 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
709
710         ext4_ext_show_path(inode, path);
711
712         return path;
713
714 err:
715         ext4_ext_drop_refs(path);
716         if (alloc)
717                 kfree(path);
718         return ERR_PTR(-EIO);
719 }
720
721 /*
722  * ext4_ext_insert_index:
723  * insert new index [@logical;@ptr] into the block at @curp;
724  * check where to insert: before @curp or after @curp
725  */
726 int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
727                                 struct ext4_ext_path *curp,
728                                 int logical, ext4_fsblk_t ptr)
729 {
730         struct ext4_extent_idx *ix;
731         int len, err;
732
733         err = ext4_ext_get_access(handle, inode, curp);
734         if (err)
735                 return err;
736
737         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
738         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
739         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
740                 /* insert after */
741                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
742                         len = (len - 1) * sizeof(struct ext4_extent_idx);
743                         len = len < 0 ? 0 : len;
744                         ext_debug("insert new index %d after: %llu. "
745                                         "move %d from 0x%p to 0x%p\n",
746                                         logical, ptr, len,
747                                         (curp->p_idx + 1), (curp->p_idx + 2));
748                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
749                 }
750                 ix = curp->p_idx + 1;
751         } else {
752                 /* insert before */
753                 len = len * sizeof(struct ext4_extent_idx);
754                 len = len < 0 ? 0 : len;
755                 ext_debug("insert new index %d before: %llu. "
756                                 "move %d from 0x%p to 0x%p\n",
757                                 logical, ptr, len,
758                                 curp->p_idx, (curp->p_idx + 1));
759                 memmove(curp->p_idx + 1, curp->p_idx, len);
760                 ix = curp->p_idx;
761         }
762
763         ix->ei_block = cpu_to_le32(logical);
764         ext4_idx_store_pblock(ix, ptr);
765         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
766
767         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
768                              > le16_to_cpu(curp->p_hdr->eh_max));
769         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
770
771         err = ext4_ext_dirty(handle, inode, curp);
772         ext4_std_error(inode->i_sb, err);
773
774         return err;
775 }
776
777 /*
778  * ext4_ext_split:
779  * inserts new subtree into the path, using free index entry
780  * at depth @at:
781  * - allocates all needed blocks (new leaf and all intermediate index blocks)
782  * - makes decision where to split
783  * - moves remaining extents and index entries (right to the split point)
784  *   into the newly allocated blocks
785  * - initializes subtree
786  */
787 static int ext4_ext_split(handle_t *handle, struct inode *inode,
788                                 struct ext4_ext_path *path,
789                                 struct ext4_extent *newext, int at)
790 {
791         struct buffer_head *bh = NULL;
792         int depth = ext_depth(inode);
793         struct ext4_extent_header *neh;
794         struct ext4_extent_idx *fidx;
795         struct ext4_extent *ex;
796         int i = at, k, m, a;
797         ext4_fsblk_t newblock, oldblock;
798         __le32 border;
799         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
800         int err = 0;
801
802         /* make decision: where to split? */
803         /* FIXME: now decision is simplest: at current extent */
804
805         /* if current leaf will be split, then we should use
806          * border from split point */
807         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
808         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
809                 border = path[depth].p_ext[1].ee_block;
810                 ext_debug("leaf will be split."
811                                 " next leaf starts at %d\n",
812                                   le32_to_cpu(border));
813         } else {
814                 border = newext->ee_block;
815                 ext_debug("leaf will be added."
816                                 " next leaf starts at %d\n",
817                                 le32_to_cpu(border));
818         }
819
820         /*
821          * If error occurs, then we break processing
822          * and mark filesystem read-only. index won't
823          * be inserted and tree will be in consistent
824          * state. Next mount will repair buffers too.
825          */
826
827         /*
828          * Get array to track all allocated blocks.
829          * We need this to handle errors and free blocks
830          * upon them.
831          */
832         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
833         if (!ablocks)
834                 return -ENOMEM;
835
836         /* allocate all needed blocks */
837         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
838         for (a = 0; a < depth - at; a++) {
839                 newblock = ext4_ext_new_meta_block(handle, inode, path,
840                                                    newext, &err);
841                 if (newblock == 0)
842                         goto cleanup;
843                 ablocks[a] = newblock;
844         }
845
846         /* initialize new leaf */
847         newblock = ablocks[--a];
848         BUG_ON(newblock == 0);
849         bh = sb_getblk(inode->i_sb, newblock);
850         if (!bh) {
851                 err = -EIO;
852                 goto cleanup;
853         }
854         lock_buffer(bh);
855
856         err = ext4_journal_get_create_access(handle, bh);
857         if (err)
858                 goto cleanup;
859
860         neh = ext_block_hdr(bh);
861         neh->eh_entries = 0;
862         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
863         neh->eh_magic = EXT4_EXT_MAGIC;
864         neh->eh_depth = 0;
865         ex = EXT_FIRST_EXTENT(neh);
866
867         /* move remainder of path[depth] to the new leaf */
868         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
869         /* start copy from next extent */
870         /* TODO: we could do it by single memmove */
871         m = 0;
872         path[depth].p_ext++;
873         while (path[depth].p_ext <=
874                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
875                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
876                                 le32_to_cpu(path[depth].p_ext->ee_block),
877                                 ext_pblock(path[depth].p_ext),
878                                 ext4_ext_is_uninitialized(path[depth].p_ext),
879                                 ext4_ext_get_actual_len(path[depth].p_ext),
880                                 newblock);
881                 /*memmove(ex++, path[depth].p_ext++,
882                                 sizeof(struct ext4_extent));
883                 neh->eh_entries++;*/
884                 path[depth].p_ext++;
885                 m++;
886         }
887         if (m) {
888                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
889                 le16_add_cpu(&neh->eh_entries, m);
890         }
891
892         set_buffer_uptodate(bh);
893         unlock_buffer(bh);
894
895         err = ext4_handle_dirty_metadata(handle, inode, bh);
896         if (err)
897                 goto cleanup;
898         brelse(bh);
899         bh = NULL;
900
901         /* correct old leaf */
902         if (m) {
903                 err = ext4_ext_get_access(handle, inode, path + depth);
904                 if (err)
905                         goto cleanup;
906                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
907                 err = ext4_ext_dirty(handle, inode, path + depth);
908                 if (err)
909                         goto cleanup;
910
911         }
912
913         /* create intermediate indexes */
914         k = depth - at - 1;
915         BUG_ON(k < 0);
916         if (k)
917                 ext_debug("create %d intermediate indices\n", k);
918         /* insert new index into current index block */
919         /* current depth stored in i var */
920         i = depth - 1;
921         while (k--) {
922                 oldblock = newblock;
923                 newblock = ablocks[--a];
924                 bh = sb_getblk(inode->i_sb, newblock);
925                 if (!bh) {
926                         err = -EIO;
927                         goto cleanup;
928                 }
929                 lock_buffer(bh);
930
931                 err = ext4_journal_get_create_access(handle, bh);
932                 if (err)
933                         goto cleanup;
934
935                 neh = ext_block_hdr(bh);
936                 neh->eh_entries = cpu_to_le16(1);
937                 neh->eh_magic = EXT4_EXT_MAGIC;
938                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
939                 neh->eh_depth = cpu_to_le16(depth - i);
940                 fidx = EXT_FIRST_INDEX(neh);
941                 fidx->ei_block = border;
942                 ext4_idx_store_pblock(fidx, oldblock);
943
944                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
945                                 i, newblock, le32_to_cpu(border), oldblock);
946                 /* copy indexes */
947                 m = 0;
948                 path[i].p_idx++;
949
950                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
951                                 EXT_MAX_INDEX(path[i].p_hdr));
952                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
953                                 EXT_LAST_INDEX(path[i].p_hdr));
954                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
955                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
956                                         le32_to_cpu(path[i].p_idx->ei_block),
957                                         idx_pblock(path[i].p_idx),
958                                         newblock);
959                         /*memmove(++fidx, path[i].p_idx++,
960                                         sizeof(struct ext4_extent_idx));
961                         neh->eh_entries++;
962                         BUG_ON(neh->eh_entries > neh->eh_max);*/
963                         path[i].p_idx++;
964                         m++;
965                 }
966                 if (m) {
967                         memmove(++fidx, path[i].p_idx - m,
968                                 sizeof(struct ext4_extent_idx) * m);
969                         le16_add_cpu(&neh->eh_entries, m);
970                 }
971                 set_buffer_uptodate(bh);
972                 unlock_buffer(bh);
973
974                 err = ext4_handle_dirty_metadata(handle, inode, bh);
975                 if (err)
976                         goto cleanup;
977                 brelse(bh);
978                 bh = NULL;
979
980                 /* correct old index */
981                 if (m) {
982                         err = ext4_ext_get_access(handle, inode, path + i);
983                         if (err)
984                                 goto cleanup;
985                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
986                         err = ext4_ext_dirty(handle, inode, path + i);
987                         if (err)
988                                 goto cleanup;
989                 }
990
991                 i--;
992         }
993
994         /* insert new index */
995         err = ext4_ext_insert_index(handle, inode, path + at,
996                                     le32_to_cpu(border), newblock);
997
998 cleanup:
999         if (bh) {
1000                 if (buffer_locked(bh))
1001                         unlock_buffer(bh);
1002                 brelse(bh);
1003         }
1004
1005         if (err) {
1006                 /* free all allocated blocks in error case */
1007                 for (i = 0; i < depth; i++) {
1008                         if (!ablocks[i])
1009                                 continue;
1010                         ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
1011                 }
1012         }
1013         kfree(ablocks);
1014
1015         return err;
1016 }
1017
1018 /*
1019  * ext4_ext_grow_indepth:
1020  * implements tree growing procedure:
1021  * - allocates new block
1022  * - moves top-level data (index block or leaf) into the new block
1023  * - initializes new top-level, creating index that points to the
1024  *   just created block
1025  */
1026 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1027                                         struct ext4_ext_path *path,
1028                                         struct ext4_extent *newext)
1029 {
1030         struct ext4_ext_path *curp = path;
1031         struct ext4_extent_header *neh;
1032         struct ext4_extent_idx *fidx;
1033         struct buffer_head *bh;
1034         ext4_fsblk_t newblock;
1035         int err = 0;
1036
1037         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1038         if (newblock == 0)
1039                 return err;
1040
1041         bh = sb_getblk(inode->i_sb, newblock);
1042         if (!bh) {
1043                 err = -EIO;
1044                 ext4_std_error(inode->i_sb, err);
1045                 return err;
1046         }
1047         lock_buffer(bh);
1048
1049         err = ext4_journal_get_create_access(handle, bh);
1050         if (err) {
1051                 unlock_buffer(bh);
1052                 goto out;
1053         }
1054
1055         /* move top-level index/leaf into new block */
1056         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1057
1058         /* set size of new block */
1059         neh = ext_block_hdr(bh);
1060         /* old root could have indexes or leaves
1061          * so calculate e_max right way */
1062         if (ext_depth(inode))
1063                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1064         else
1065                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1066         neh->eh_magic = EXT4_EXT_MAGIC;
1067         set_buffer_uptodate(bh);
1068         unlock_buffer(bh);
1069
1070         err = ext4_handle_dirty_metadata(handle, inode, bh);
1071         if (err)
1072                 goto out;
1073
1074         /* create index in new top-level index: num,max,pointer */
1075         err = ext4_ext_get_access(handle, inode, curp);
1076         if (err)
1077                 goto out;
1078
1079         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1080         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1081         curp->p_hdr->eh_entries = cpu_to_le16(1);
1082         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1083
1084         if (path[0].p_hdr->eh_depth)
1085                 curp->p_idx->ei_block =
1086                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1087         else
1088                 curp->p_idx->ei_block =
1089                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1090         ext4_idx_store_pblock(curp->p_idx, newblock);
1091
1092         neh = ext_inode_hdr(inode);
1093         fidx = EXT_FIRST_INDEX(neh);
1094         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1095                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1096                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1097
1098         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1099         err = ext4_ext_dirty(handle, inode, curp);
1100 out:
1101         brelse(bh);
1102
1103         return err;
1104 }
1105
1106 /*
1107  * ext4_ext_create_new_leaf:
1108  * finds empty index and adds new leaf.
1109  * if no free index is found, then it requests in-depth growing.
1110  */
1111 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1112                                         struct ext4_ext_path *path,
1113                                         struct ext4_extent *newext)
1114 {
1115         struct ext4_ext_path *curp;
1116         int depth, i, err = 0;
1117
1118 repeat:
1119         i = depth = ext_depth(inode);
1120
1121         /* walk up to the tree and look for free index entry */
1122         curp = path + depth;
1123         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1124                 i--;
1125                 curp--;
1126         }
1127
1128         /* we use already allocated block for index block,
1129          * so subsequent data blocks should be contiguous */
1130         if (EXT_HAS_FREE_INDEX(curp)) {
1131                 /* if we found index with free entry, then use that
1132                  * entry: create all needed subtree and add new leaf */
1133                 err = ext4_ext_split(handle, inode, path, newext, i);
1134                 if (err)
1135                         goto out;
1136
1137                 /* refill path */
1138                 ext4_ext_drop_refs(path);
1139                 path = ext4_ext_find_extent(inode,
1140                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1141                                     path);
1142                 if (IS_ERR(path))
1143                         err = PTR_ERR(path);
1144         } else {
1145                 /* tree is full, time to grow in depth */
1146                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1147                 if (err)
1148                         goto out;
1149
1150                 /* refill path */
1151                 ext4_ext_drop_refs(path);
1152                 path = ext4_ext_find_extent(inode,
1153                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1154                                     path);
1155                 if (IS_ERR(path)) {
1156                         err = PTR_ERR(path);
1157                         goto out;
1158                 }
1159
1160                 /*
1161                  * only first (depth 0 -> 1) produces free space;
1162                  * in all other cases we have to split the grown tree
1163                  */
1164                 depth = ext_depth(inode);
1165                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1166                         /* now we need to split */
1167                         goto repeat;
1168                 }
1169         }
1170
1171 out:
1172         return err;
1173 }
1174
1175 /*
1176  * search the closest allocated block to the left for *logical
1177  * and returns it at @logical + it's physical address at @phys
1178  * if *logical is the smallest allocated block, the function
1179  * returns 0 at @phys
1180  * return value contains 0 (success) or error code
1181  */
1182 int
1183 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1184                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1185 {
1186         struct ext4_extent_idx *ix;
1187         struct ext4_extent *ex;
1188         int depth, ee_len;
1189
1190         BUG_ON(path == NULL);
1191         depth = path->p_depth;
1192         *phys = 0;
1193
1194         if (depth == 0 && path->p_ext == NULL)
1195                 return 0;
1196
1197         /* usually extent in the path covers blocks smaller
1198          * then *logical, but it can be that extent is the
1199          * first one in the file */
1200
1201         ex = path[depth].p_ext;
1202         ee_len = ext4_ext_get_actual_len(ex);
1203         if (*logical < le32_to_cpu(ex->ee_block)) {
1204                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1205                 while (--depth >= 0) {
1206                         ix = path[depth].p_idx;
1207                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1208                 }
1209                 return 0;
1210         }
1211
1212         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1213
1214         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1215         *phys = ext_pblock(ex) + ee_len - 1;
1216         return 0;
1217 }
1218
1219 /*
1220  * search the closest allocated block to the right for *logical
1221  * and returns it at @logical + it's physical address at @phys
1222  * if *logical is the smallest allocated block, the function
1223  * returns 0 at @phys
1224  * return value contains 0 (success) or error code
1225  */
1226 int
1227 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1228                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1229 {
1230         struct buffer_head *bh = NULL;
1231         struct ext4_extent_header *eh;
1232         struct ext4_extent_idx *ix;
1233         struct ext4_extent *ex;
1234         ext4_fsblk_t block;
1235         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1236         int ee_len;
1237
1238         BUG_ON(path == NULL);
1239         depth = path->p_depth;
1240         *phys = 0;
1241
1242         if (depth == 0 && path->p_ext == NULL)
1243                 return 0;
1244
1245         /* usually extent in the path covers blocks smaller
1246          * then *logical, but it can be that extent is the
1247          * first one in the file */
1248
1249         ex = path[depth].p_ext;
1250         ee_len = ext4_ext_get_actual_len(ex);
1251         if (*logical < le32_to_cpu(ex->ee_block)) {
1252                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1253                 while (--depth >= 0) {
1254                         ix = path[depth].p_idx;
1255                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1256                 }
1257                 *logical = le32_to_cpu(ex->ee_block);
1258                 *phys = ext_pblock(ex);
1259                 return 0;
1260         }
1261
1262         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1263
1264         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1265                 /* next allocated block in this leaf */
1266                 ex++;
1267                 *logical = le32_to_cpu(ex->ee_block);
1268                 *phys = ext_pblock(ex);
1269                 return 0;
1270         }
1271
1272         /* go up and search for index to the right */
1273         while (--depth >= 0) {
1274                 ix = path[depth].p_idx;
1275                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1276                         goto got_index;
1277         }
1278
1279         /* we've gone up to the root and found no index to the right */
1280         return 0;
1281
1282 got_index:
1283         /* we've found index to the right, let's
1284          * follow it and find the closest allocated
1285          * block to the right */
1286         ix++;
1287         block = idx_pblock(ix);
1288         while (++depth < path->p_depth) {
1289                 bh = sb_bread(inode->i_sb, block);
1290                 if (bh == NULL)
1291                         return -EIO;
1292                 eh = ext_block_hdr(bh);
1293                 /* subtract from p_depth to get proper eh_depth */
1294                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1295                         put_bh(bh);
1296                         return -EIO;
1297                 }
1298                 ix = EXT_FIRST_INDEX(eh);
1299                 block = idx_pblock(ix);
1300                 put_bh(bh);
1301         }
1302
1303         bh = sb_bread(inode->i_sb, block);
1304         if (bh == NULL)
1305                 return -EIO;
1306         eh = ext_block_hdr(bh);
1307         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1308                 put_bh(bh);
1309                 return -EIO;
1310         }
1311         ex = EXT_FIRST_EXTENT(eh);
1312         *logical = le32_to_cpu(ex->ee_block);
1313         *phys = ext_pblock(ex);
1314         put_bh(bh);
1315         return 0;
1316 }
1317
1318 /*
1319  * ext4_ext_next_allocated_block:
1320  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1321  * NOTE: it considers block number from index entry as
1322  * allocated block. Thus, index entries have to be consistent
1323  * with leaves.
1324  */
1325 static ext4_lblk_t
1326 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1327 {
1328         int depth;
1329
1330         BUG_ON(path == NULL);
1331         depth = path->p_depth;
1332
1333         if (depth == 0 && path->p_ext == NULL)
1334                 return EXT_MAX_BLOCK;
1335
1336         while (depth >= 0) {
1337                 if (depth == path->p_depth) {
1338                         /* leaf */
1339                         if (path[depth].p_ext !=
1340                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1341                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1342                 } else {
1343                         /* index */
1344                         if (path[depth].p_idx !=
1345                                         EXT_LAST_INDEX(path[depth].p_hdr))
1346                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1347                 }
1348                 depth--;
1349         }
1350
1351         return EXT_MAX_BLOCK;
1352 }
1353
1354 /*
1355  * ext4_ext_next_leaf_block:
1356  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1357  */
1358 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1359                                         struct ext4_ext_path *path)
1360 {
1361         int depth;
1362
1363         BUG_ON(path == NULL);
1364         depth = path->p_depth;
1365
1366         /* zero-tree has no leaf blocks at all */
1367         if (depth == 0)
1368                 return EXT_MAX_BLOCK;
1369
1370         /* go to index block */
1371         depth--;
1372
1373         while (depth >= 0) {
1374                 if (path[depth].p_idx !=
1375                                 EXT_LAST_INDEX(path[depth].p_hdr))
1376                         return (ext4_lblk_t)
1377                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1378                 depth--;
1379         }
1380
1381         return EXT_MAX_BLOCK;
1382 }
1383
1384 /*
1385  * ext4_ext_correct_indexes:
1386  * if leaf gets modified and modified extent is first in the leaf,
1387  * then we have to correct all indexes above.
1388  * TODO: do we need to correct tree in all cases?
1389  */
1390 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1391                                 struct ext4_ext_path *path)
1392 {
1393         struct ext4_extent_header *eh;
1394         int depth = ext_depth(inode);
1395         struct ext4_extent *ex;
1396         __le32 border;
1397         int k, err = 0;
1398
1399         eh = path[depth].p_hdr;
1400         ex = path[depth].p_ext;
1401         BUG_ON(ex == NULL);
1402         BUG_ON(eh == NULL);
1403
1404         if (depth == 0) {
1405                 /* there is no tree at all */
1406                 return 0;
1407         }
1408
1409         if (ex != EXT_FIRST_EXTENT(eh)) {
1410                 /* we correct tree if first leaf got modified only */
1411                 return 0;
1412         }
1413
1414         /*
1415          * TODO: we need correction if border is smaller than current one
1416          */
1417         k = depth - 1;
1418         border = path[depth].p_ext->ee_block;
1419         err = ext4_ext_get_access(handle, inode, path + k);
1420         if (err)
1421                 return err;
1422         path[k].p_idx->ei_block = border;
1423         err = ext4_ext_dirty(handle, inode, path + k);
1424         if (err)
1425                 return err;
1426
1427         while (k--) {
1428                 /* change all left-side indexes */
1429                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1430                         break;
1431                 err = ext4_ext_get_access(handle, inode, path + k);
1432                 if (err)
1433                         break;
1434                 path[k].p_idx->ei_block = border;
1435                 err = ext4_ext_dirty(handle, inode, path + k);
1436                 if (err)
1437                         break;
1438         }
1439
1440         return err;
1441 }
1442
1443 int
1444 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1445                                 struct ext4_extent *ex2)
1446 {
1447         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1448
1449         /*
1450          * Make sure that either both extents are uninitialized, or
1451          * both are _not_.
1452          */
1453         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1454                 return 0;
1455
1456         if (ext4_ext_is_uninitialized(ex1))
1457                 max_len = EXT_UNINIT_MAX_LEN;
1458         else
1459                 max_len = EXT_INIT_MAX_LEN;
1460
1461         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1462         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1463
1464         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1465                         le32_to_cpu(ex2->ee_block))
1466                 return 0;
1467
1468         /*
1469          * To allow future support for preallocated extents to be added
1470          * as an RO_COMPAT feature, refuse to merge to extents if
1471          * this can result in the top bit of ee_len being set.
1472          */
1473         if (ext1_ee_len + ext2_ee_len > max_len)
1474                 return 0;
1475 #ifdef AGGRESSIVE_TEST
1476         if (ext1_ee_len >= 4)
1477                 return 0;
1478 #endif
1479
1480         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1481                 return 1;
1482         return 0;
1483 }
1484
1485 /*
1486  * This function tries to merge the "ex" extent to the next extent in the tree.
1487  * It always tries to merge towards right. If you want to merge towards
1488  * left, pass "ex - 1" as argument instead of "ex".
1489  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1490  * 1 if they got merged.
1491  */
1492 int ext4_ext_try_to_merge(struct inode *inode,
1493                           struct ext4_ext_path *path,
1494                           struct ext4_extent *ex)
1495 {
1496         struct ext4_extent_header *eh;
1497         unsigned int depth, len;
1498         int merge_done = 0;
1499         int uninitialized = 0;
1500
1501         depth = ext_depth(inode);
1502         BUG_ON(path[depth].p_hdr == NULL);
1503         eh = path[depth].p_hdr;
1504
1505         while (ex < EXT_LAST_EXTENT(eh)) {
1506                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1507                         break;
1508                 /* merge with next extent! */
1509                 if (ext4_ext_is_uninitialized(ex))
1510                         uninitialized = 1;
1511                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1512                                 + ext4_ext_get_actual_len(ex + 1));
1513                 if (uninitialized)
1514                         ext4_ext_mark_uninitialized(ex);
1515
1516                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1517                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1518                                 * sizeof(struct ext4_extent);
1519                         memmove(ex + 1, ex + 2, len);
1520                 }
1521                 le16_add_cpu(&eh->eh_entries, -1);
1522                 merge_done = 1;
1523                 WARN_ON(eh->eh_entries == 0);
1524                 if (!eh->eh_entries)
1525                         ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1526                            "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1527         }
1528
1529         return merge_done;
1530 }
1531
1532 /*
1533  * check if a portion of the "newext" extent overlaps with an
1534  * existing extent.
1535  *
1536  * If there is an overlap discovered, it updates the length of the newext
1537  * such that there will be no overlap, and then returns 1.
1538  * If there is no overlap found, it returns 0.
1539  */
1540 unsigned int ext4_ext_check_overlap(struct inode *inode,
1541                                     struct ext4_extent *newext,
1542                                     struct ext4_ext_path *path)
1543 {
1544         ext4_lblk_t b1, b2;
1545         unsigned int depth, len1;
1546         unsigned int ret = 0;
1547
1548         b1 = le32_to_cpu(newext->ee_block);
1549         len1 = ext4_ext_get_actual_len(newext);
1550         depth = ext_depth(inode);
1551         if (!path[depth].p_ext)
1552                 goto out;
1553         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1554
1555         /*
1556          * get the next allocated block if the extent in the path
1557          * is before the requested block(s)
1558          */
1559         if (b2 < b1) {
1560                 b2 = ext4_ext_next_allocated_block(path);
1561                 if (b2 == EXT_MAX_BLOCK)
1562                         goto out;
1563         }
1564
1565         /* check for wrap through zero on extent logical start block*/
1566         if (b1 + len1 < b1) {
1567                 len1 = EXT_MAX_BLOCK - b1;
1568                 newext->ee_len = cpu_to_le16(len1);
1569                 ret = 1;
1570         }
1571
1572         /* check for overlap */
1573         if (b1 + len1 > b2) {
1574                 newext->ee_len = cpu_to_le16(b2 - b1);
1575                 ret = 1;
1576         }
1577 out:
1578         return ret;
1579 }
1580
1581 /*
1582  * ext4_ext_insert_extent:
1583  * tries to merge requsted extent into the existing extent or
1584  * inserts requested extent as new one into the tree,
1585  * creating new leaf in the no-space case.
1586  */
1587 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1588                                 struct ext4_ext_path *path,
1589                                 struct ext4_extent *newext, int flag)
1590 {
1591         struct ext4_extent_header *eh;
1592         struct ext4_extent *ex, *fex;
1593         struct ext4_extent *nearex; /* nearest extent */
1594         struct ext4_ext_path *npath = NULL;
1595         int depth, len, err;
1596         ext4_lblk_t next;
1597         unsigned uninitialized = 0;
1598
1599         BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1600         depth = ext_depth(inode);
1601         ex = path[depth].p_ext;
1602         BUG_ON(path[depth].p_hdr == NULL);
1603
1604         /* try to insert block into found extent and return */
1605         if (ex && (flag != EXT4_GET_BLOCKS_DIO_CREATE_EXT)
1606                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1607                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1608                                 ext4_ext_is_uninitialized(newext),
1609                                 ext4_ext_get_actual_len(newext),
1610                                 le32_to_cpu(ex->ee_block),
1611                                 ext4_ext_is_uninitialized(ex),
1612                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1613                 err = ext4_ext_get_access(handle, inode, path + depth);
1614                 if (err)
1615                         return err;
1616
1617                 /*
1618                  * ext4_can_extents_be_merged should have checked that either
1619                  * both extents are uninitialized, or both aren't. Thus we
1620                  * need to check only one of them here.
1621                  */
1622                 if (ext4_ext_is_uninitialized(ex))
1623                         uninitialized = 1;
1624                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1625                                         + ext4_ext_get_actual_len(newext));
1626                 if (uninitialized)
1627                         ext4_ext_mark_uninitialized(ex);
1628                 eh = path[depth].p_hdr;
1629                 nearex = ex;
1630                 goto merge;
1631         }
1632
1633 repeat:
1634         depth = ext_depth(inode);
1635         eh = path[depth].p_hdr;
1636         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1637                 goto has_space;
1638
1639         /* probably next leaf has space for us? */
1640         fex = EXT_LAST_EXTENT(eh);
1641         next = ext4_ext_next_leaf_block(inode, path);
1642         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1643             && next != EXT_MAX_BLOCK) {
1644                 ext_debug("next leaf block - %d\n", next);
1645                 BUG_ON(npath != NULL);
1646                 npath = ext4_ext_find_extent(inode, next, NULL);
1647                 if (IS_ERR(npath))
1648                         return PTR_ERR(npath);
1649                 BUG_ON(npath->p_depth != path->p_depth);
1650                 eh = npath[depth].p_hdr;
1651                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1652                         ext_debug("next leaf isnt full(%d)\n",
1653                                   le16_to_cpu(eh->eh_entries));
1654                         path = npath;
1655                         goto repeat;
1656                 }
1657                 ext_debug("next leaf has no free space(%d,%d)\n",
1658                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1659         }
1660
1661         /*
1662          * There is no free space in the found leaf.
1663          * We're gonna add a new leaf in the tree.
1664          */
1665         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1666         if (err)
1667                 goto cleanup;
1668         depth = ext_depth(inode);
1669         eh = path[depth].p_hdr;
1670
1671 has_space:
1672         nearex = path[depth].p_ext;
1673
1674         err = ext4_ext_get_access(handle, inode, path + depth);
1675         if (err)
1676                 goto cleanup;
1677
1678         if (!nearex) {
1679                 /* there is no extent in this leaf, create first one */
1680                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1681                                 le32_to_cpu(newext->ee_block),
1682                                 ext_pblock(newext),
1683                                 ext4_ext_is_uninitialized(newext),
1684                                 ext4_ext_get_actual_len(newext));
1685                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1686         } else if (le32_to_cpu(newext->ee_block)
1687                            > le32_to_cpu(nearex->ee_block)) {
1688 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1689                 if (nearex != EXT_LAST_EXTENT(eh)) {
1690                         len = EXT_MAX_EXTENT(eh) - nearex;
1691                         len = (len - 1) * sizeof(struct ext4_extent);
1692                         len = len < 0 ? 0 : len;
1693                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1694                                         "move %d from 0x%p to 0x%p\n",
1695                                         le32_to_cpu(newext->ee_block),
1696                                         ext_pblock(newext),
1697                                         ext4_ext_is_uninitialized(newext),
1698                                         ext4_ext_get_actual_len(newext),
1699                                         nearex, len, nearex + 1, nearex + 2);
1700                         memmove(nearex + 2, nearex + 1, len);
1701                 }
1702                 path[depth].p_ext = nearex + 1;
1703         } else {
1704                 BUG_ON(newext->ee_block == nearex->ee_block);
1705                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1706                 len = len < 0 ? 0 : len;
1707                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1708                                 "move %d from 0x%p to 0x%p\n",
1709                                 le32_to_cpu(newext->ee_block),
1710                                 ext_pblock(newext),
1711                                 ext4_ext_is_uninitialized(newext),
1712                                 ext4_ext_get_actual_len(newext),
1713                                 nearex, len, nearex + 1, nearex + 2);
1714                 memmove(nearex + 1, nearex, len);
1715                 path[depth].p_ext = nearex;
1716         }
1717
1718         le16_add_cpu(&eh->eh_entries, 1);
1719         nearex = path[depth].p_ext;
1720         nearex->ee_block = newext->ee_block;
1721         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1722         nearex->ee_len = newext->ee_len;
1723
1724 merge:
1725         /* try to merge extents to the right */
1726         if (flag != EXT4_GET_BLOCKS_DIO_CREATE_EXT)
1727                 ext4_ext_try_to_merge(inode, path, nearex);
1728
1729         /* try to merge extents to the left */
1730
1731         /* time to correct all indexes above */
1732         err = ext4_ext_correct_indexes(handle, inode, path);
1733         if (err)
1734                 goto cleanup;
1735
1736         err = ext4_ext_dirty(handle, inode, path + depth);
1737
1738 cleanup:
1739         if (npath) {
1740                 ext4_ext_drop_refs(npath);
1741                 kfree(npath);
1742         }
1743         ext4_ext_invalidate_cache(inode);
1744         return err;
1745 }
1746
1747 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1748                         ext4_lblk_t num, ext_prepare_callback func,
1749                         void *cbdata)
1750 {
1751         struct ext4_ext_path *path = NULL;
1752         struct ext4_ext_cache cbex;
1753         struct ext4_extent *ex;
1754         ext4_lblk_t next, start = 0, end = 0;
1755         ext4_lblk_t last = block + num;
1756         int depth, exists, err = 0;
1757
1758         BUG_ON(func == NULL);
1759         BUG_ON(inode == NULL);
1760
1761         while (block < last && block != EXT_MAX_BLOCK) {
1762                 num = last - block;
1763                 /* find extent for this block */
1764                 path = ext4_ext_find_extent(inode, block, path);
1765                 if (IS_ERR(path)) {
1766                         err = PTR_ERR(path);
1767                         path = NULL;
1768                         break;
1769                 }
1770
1771                 depth = ext_depth(inode);
1772                 BUG_ON(path[depth].p_hdr == NULL);
1773                 ex = path[depth].p_ext;
1774                 next = ext4_ext_next_allocated_block(path);
1775
1776                 exists = 0;
1777                 if (!ex) {
1778                         /* there is no extent yet, so try to allocate
1779                          * all requested space */
1780                         start = block;
1781                         end = block + num;
1782                 } else if (le32_to_cpu(ex->ee_block) > block) {
1783                         /* need to allocate space before found extent */
1784                         start = block;
1785                         end = le32_to_cpu(ex->ee_block);
1786                         if (block + num < end)
1787                                 end = block + num;
1788                 } else if (block >= le32_to_cpu(ex->ee_block)
1789                                         + ext4_ext_get_actual_len(ex)) {
1790                         /* need to allocate space after found extent */
1791                         start = block;
1792                         end = block + num;
1793                         if (end >= next)
1794                                 end = next;
1795                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1796                         /*
1797                          * some part of requested space is covered
1798                          * by found extent
1799                          */
1800                         start = block;
1801                         end = le32_to_cpu(ex->ee_block)
1802                                 + ext4_ext_get_actual_len(ex);
1803                         if (block + num < end)
1804                                 end = block + num;
1805                         exists = 1;
1806                 } else {
1807                         BUG();
1808                 }
1809                 BUG_ON(end <= start);
1810
1811                 if (!exists) {
1812                         cbex.ec_block = start;
1813                         cbex.ec_len = end - start;
1814                         cbex.ec_start = 0;
1815                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1816                 } else {
1817                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1818                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1819                         cbex.ec_start = ext_pblock(ex);
1820                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1821                 }
1822
1823                 BUG_ON(cbex.ec_len == 0);
1824                 err = func(inode, path, &cbex, ex, cbdata);
1825                 ext4_ext_drop_refs(path);
1826
1827                 if (err < 0)
1828                         break;
1829
1830                 if (err == EXT_REPEAT)
1831                         continue;
1832                 else if (err == EXT_BREAK) {
1833                         err = 0;
1834                         break;
1835                 }
1836
1837                 if (ext_depth(inode) != depth) {
1838                         /* depth was changed. we have to realloc path */
1839                         kfree(path);
1840                         path = NULL;
1841                 }
1842
1843                 block = cbex.ec_block + cbex.ec_len;
1844         }
1845
1846         if (path) {
1847                 ext4_ext_drop_refs(path);
1848                 kfree(path);
1849         }
1850
1851         return err;
1852 }
1853
1854 static void
1855 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1856                         __u32 len, ext4_fsblk_t start, int type)
1857 {
1858         struct ext4_ext_cache *cex;
1859         BUG_ON(len == 0);
1860         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1861         cex = &EXT4_I(inode)->i_cached_extent;
1862         cex->ec_type = type;
1863         cex->ec_block = block;
1864         cex->ec_len = len;
1865         cex->ec_start = start;
1866         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1867 }
1868
1869 /*
1870  * ext4_ext_put_gap_in_cache:
1871  * calculate boundaries of the gap that the requested block fits into
1872  * and cache this gap
1873  */
1874 static void
1875 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1876                                 ext4_lblk_t block)
1877 {
1878         int depth = ext_depth(inode);
1879         unsigned long len;
1880         ext4_lblk_t lblock;
1881         struct ext4_extent *ex;
1882
1883         ex = path[depth].p_ext;
1884         if (ex == NULL) {
1885                 /* there is no extent yet, so gap is [0;-] */
1886                 lblock = 0;
1887                 len = EXT_MAX_BLOCK;
1888                 ext_debug("cache gap(whole file):");
1889         } else if (block < le32_to_cpu(ex->ee_block)) {
1890                 lblock = block;
1891                 len = le32_to_cpu(ex->ee_block) - block;
1892                 ext_debug("cache gap(before): %u [%u:%u]",
1893                                 block,
1894                                 le32_to_cpu(ex->ee_block),
1895                                  ext4_ext_get_actual_len(ex));
1896         } else if (block >= le32_to_cpu(ex->ee_block)
1897                         + ext4_ext_get_actual_len(ex)) {
1898                 ext4_lblk_t next;
1899                 lblock = le32_to_cpu(ex->ee_block)
1900                         + ext4_ext_get_actual_len(ex);
1901
1902                 next = ext4_ext_next_allocated_block(path);
1903                 ext_debug("cache gap(after): [%u:%u] %u",
1904                                 le32_to_cpu(ex->ee_block),
1905                                 ext4_ext_get_actual_len(ex),
1906                                 block);
1907                 BUG_ON(next == lblock);
1908                 len = next - lblock;
1909         } else {
1910                 lblock = len = 0;
1911                 BUG();
1912         }
1913
1914         ext_debug(" -> %u:%lu\n", lblock, len);
1915         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1916 }
1917
1918 static int
1919 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1920                         struct ext4_extent *ex)
1921 {
1922         struct ext4_ext_cache *cex;
1923         int ret = EXT4_EXT_CACHE_NO;
1924
1925         /* 
1926          * We borrow i_block_reservation_lock to protect i_cached_extent
1927          */
1928         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1929         cex = &EXT4_I(inode)->i_cached_extent;
1930
1931         /* has cache valid data? */
1932         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1933                 goto errout;
1934
1935         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1936                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1937         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1938                 ex->ee_block = cpu_to_le32(cex->ec_block);
1939                 ext4_ext_store_pblock(ex, cex->ec_start);
1940                 ex->ee_len = cpu_to_le16(cex->ec_len);
1941                 ext_debug("%u cached by %u:%u:%llu\n",
1942                                 block,
1943                                 cex->ec_block, cex->ec_len, cex->ec_start);
1944                 ret = cex->ec_type;
1945         }
1946 errout:
1947         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1948         return ret;
1949 }
1950
1951 /*
1952  * ext4_ext_rm_idx:
1953  * removes index from the index block.
1954  * It's used in truncate case only, thus all requests are for
1955  * last index in the block only.
1956  */
1957 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1958                         struct ext4_ext_path *path)
1959 {
1960         struct buffer_head *bh;
1961         int err;
1962         ext4_fsblk_t leaf;
1963
1964         /* free index block */
1965         path--;
1966         leaf = idx_pblock(path->p_idx);
1967         BUG_ON(path->p_hdr->eh_entries == 0);
1968         err = ext4_ext_get_access(handle, inode, path);
1969         if (err)
1970                 return err;
1971         le16_add_cpu(&path->p_hdr->eh_entries, -1);
1972         err = ext4_ext_dirty(handle, inode, path);
1973         if (err)
1974                 return err;
1975         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1976         bh = sb_find_get_block(inode->i_sb, leaf);
1977         ext4_forget(handle, 1, inode, bh, leaf);
1978         ext4_free_blocks(handle, inode, leaf, 1, 1);
1979         return err;
1980 }
1981
1982 /*
1983  * ext4_ext_calc_credits_for_single_extent:
1984  * This routine returns max. credits that needed to insert an extent
1985  * to the extent tree.
1986  * When pass the actual path, the caller should calculate credits
1987  * under i_data_sem.
1988  */
1989 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1990                                                 struct ext4_ext_path *path)
1991 {
1992         if (path) {
1993                 int depth = ext_depth(inode);
1994                 int ret = 0;
1995
1996                 /* probably there is space in leaf? */
1997                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1998                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
1999
2000                         /*
2001                          *  There are some space in the leaf tree, no
2002                          *  need to account for leaf block credit
2003                          *
2004                          *  bitmaps and block group descriptor blocks
2005                          *  and other metadat blocks still need to be
2006                          *  accounted.
2007                          */
2008                         /* 1 bitmap, 1 block group descriptor */
2009                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2010                         return ret;
2011                 }
2012         }
2013
2014         return ext4_chunk_trans_blocks(inode, nrblocks);
2015 }
2016
2017 /*
2018  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2019  *
2020  * if nrblocks are fit in a single extent (chunk flag is 1), then
2021  * in the worse case, each tree level index/leaf need to be changed
2022  * if the tree split due to insert a new extent, then the old tree
2023  * index/leaf need to be updated too
2024  *
2025  * If the nrblocks are discontiguous, they could cause
2026  * the whole tree split more than once, but this is really rare.
2027  */
2028 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2029 {
2030         int index;
2031         int depth = ext_depth(inode);
2032
2033         if (chunk)
2034                 index = depth * 2;
2035         else
2036                 index = depth * 3;
2037
2038         return index;
2039 }
2040
2041 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2042                                 struct ext4_extent *ex,
2043                                 ext4_lblk_t from, ext4_lblk_t to)
2044 {
2045         struct buffer_head *bh;
2046         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2047         int i, metadata = 0;
2048
2049         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2050                 metadata = 1;
2051 #ifdef EXTENTS_STATS
2052         {
2053                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2054                 spin_lock(&sbi->s_ext_stats_lock);
2055                 sbi->s_ext_blocks += ee_len;
2056                 sbi->s_ext_extents++;
2057                 if (ee_len < sbi->s_ext_min)
2058                         sbi->s_ext_min = ee_len;
2059                 if (ee_len > sbi->s_ext_max)
2060                         sbi->s_ext_max = ee_len;
2061                 if (ext_depth(inode) > sbi->s_depth_max)
2062                         sbi->s_depth_max = ext_depth(inode);
2063                 spin_unlock(&sbi->s_ext_stats_lock);
2064         }
2065 #endif
2066         if (from >= le32_to_cpu(ex->ee_block)
2067             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2068                 /* tail removal */
2069                 ext4_lblk_t num;
2070                 ext4_fsblk_t start;
2071
2072                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2073                 start = ext_pblock(ex) + ee_len - num;
2074                 ext_debug("free last %u blocks starting %llu\n", num, start);
2075                 for (i = 0; i < num; i++) {
2076                         bh = sb_find_get_block(inode->i_sb, start + i);
2077                         ext4_forget(handle, 0, inode, bh, start + i);
2078                 }
2079                 ext4_free_blocks(handle, inode, start, num, metadata);
2080         } else if (from == le32_to_cpu(ex->ee_block)
2081                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2082                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2083                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2084         } else {
2085                 printk(KERN_INFO "strange request: removal(2) "
2086                                 "%u-%u from %u:%u\n",
2087                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2088         }
2089         return 0;
2090 }
2091
2092 static int
2093 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2094                 struct ext4_ext_path *path, ext4_lblk_t start)
2095 {
2096         int err = 0, correct_index = 0;
2097         int depth = ext_depth(inode), credits;
2098         struct ext4_extent_header *eh;
2099         ext4_lblk_t a, b, block;
2100         unsigned num;
2101         ext4_lblk_t ex_ee_block;
2102         unsigned short ex_ee_len;
2103         unsigned uninitialized = 0;
2104         struct ext4_extent *ex;
2105
2106         /* the header must be checked already in ext4_ext_remove_space() */
2107         ext_debug("truncate since %u in leaf\n", start);
2108         if (!path[depth].p_hdr)
2109                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2110         eh = path[depth].p_hdr;
2111         BUG_ON(eh == NULL);
2112
2113         /* find where to start removing */
2114         ex = EXT_LAST_EXTENT(eh);
2115
2116         ex_ee_block = le32_to_cpu(ex->ee_block);
2117         ex_ee_len = ext4_ext_get_actual_len(ex);
2118
2119         while (ex >= EXT_FIRST_EXTENT(eh) &&
2120                         ex_ee_block + ex_ee_len > start) {
2121
2122                 if (ext4_ext_is_uninitialized(ex))
2123                         uninitialized = 1;
2124                 else
2125                         uninitialized = 0;
2126
2127                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2128                          uninitialized, ex_ee_len);
2129                 path[depth].p_ext = ex;
2130
2131                 a = ex_ee_block > start ? ex_ee_block : start;
2132                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2133                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2134
2135                 ext_debug("  border %u:%u\n", a, b);
2136
2137                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2138                         block = 0;
2139                         num = 0;
2140                         BUG();
2141                 } else if (a != ex_ee_block) {
2142                         /* remove tail of the extent */
2143                         block = ex_ee_block;
2144                         num = a - block;
2145                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2146                         /* remove head of the extent */
2147                         block = a;
2148                         num = b - a;
2149                         /* there is no "make a hole" API yet */
2150                         BUG();
2151                 } else {
2152                         /* remove whole extent: excellent! */
2153                         block = ex_ee_block;
2154                         num = 0;
2155                         BUG_ON(a != ex_ee_block);
2156                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2157                 }
2158
2159                 /*
2160                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2161                  * descriptor) for each block group; assume two block
2162                  * groups plus ex_ee_len/blocks_per_block_group for
2163                  * the worst case
2164                  */
2165                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2166                 if (ex == EXT_FIRST_EXTENT(eh)) {
2167                         correct_index = 1;
2168                         credits += (ext_depth(inode)) + 1;
2169                 }
2170                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2171
2172                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2173                 if (err)
2174                         goto out;
2175
2176                 err = ext4_ext_get_access(handle, inode, path + depth);
2177                 if (err)
2178                         goto out;
2179
2180                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2181                 if (err)
2182                         goto out;
2183
2184                 if (num == 0) {
2185                         /* this extent is removed; mark slot entirely unused */
2186                         ext4_ext_store_pblock(ex, 0);
2187                         le16_add_cpu(&eh->eh_entries, -1);
2188                 }
2189
2190                 ex->ee_block = cpu_to_le32(block);
2191                 ex->ee_len = cpu_to_le16(num);
2192                 /*
2193                  * Do not mark uninitialized if all the blocks in the
2194                  * extent have been removed.
2195                  */
2196                 if (uninitialized && num)
2197                         ext4_ext_mark_uninitialized(ex);
2198
2199                 err = ext4_ext_dirty(handle, inode, path + depth);
2200                 if (err)
2201                         goto out;
2202
2203                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2204                                 ext_pblock(ex));
2205                 ex--;
2206                 ex_ee_block = le32_to_cpu(ex->ee_block);
2207                 ex_ee_len = ext4_ext_get_actual_len(ex);
2208         }
2209
2210         if (correct_index && eh->eh_entries)
2211                 err = ext4_ext_correct_indexes(handle, inode, path);
2212
2213         /* if this leaf is free, then we should
2214          * remove it from index block above */
2215         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2216                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2217
2218 out:
2219         return err;
2220 }
2221
2222 /*
2223  * ext4_ext_more_to_rm:
2224  * returns 1 if current index has to be freed (even partial)
2225  */
2226 static int
2227 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2228 {
2229         BUG_ON(path->p_idx == NULL);
2230
2231         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2232                 return 0;
2233
2234         /*
2235          * if truncate on deeper level happened, it wasn't partial,
2236          * so we have to consider current index for truncation
2237          */
2238         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2239                 return 0;
2240         return 1;
2241 }
2242
2243 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2244 {
2245         struct super_block *sb = inode->i_sb;
2246         int depth = ext_depth(inode);
2247         struct ext4_ext_path *path;
2248         handle_t *handle;
2249         int i = 0, err = 0;
2250
2251         ext_debug("truncate since %u\n", start);
2252
2253         /* probably first extent we're gonna free will be last in block */
2254         handle = ext4_journal_start(inode, depth + 1);
2255         if (IS_ERR(handle))
2256                 return PTR_ERR(handle);
2257
2258         ext4_ext_invalidate_cache(inode);
2259
2260         /*
2261          * We start scanning from right side, freeing all the blocks
2262          * after i_size and walking into the tree depth-wise.
2263          */
2264         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2265         if (path == NULL) {
2266                 ext4_journal_stop(handle);
2267                 return -ENOMEM;
2268         }
2269         path[0].p_hdr = ext_inode_hdr(inode);
2270         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2271                 err = -EIO;
2272                 goto out;
2273         }
2274         path[0].p_depth = depth;
2275
2276         while (i >= 0 && err == 0) {
2277                 if (i == depth) {
2278                         /* this is leaf block */
2279                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2280                         /* root level has p_bh == NULL, brelse() eats this */
2281                         brelse(path[i].p_bh);
2282                         path[i].p_bh = NULL;
2283                         i--;
2284                         continue;
2285                 }
2286
2287                 /* this is index block */
2288                 if (!path[i].p_hdr) {
2289                         ext_debug("initialize header\n");
2290                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2291                 }
2292
2293                 if (!path[i].p_idx) {
2294                         /* this level hasn't been touched yet */
2295                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2296                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2297                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2298                                   path[i].p_hdr,
2299                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2300                 } else {
2301                         /* we were already here, see at next index */
2302                         path[i].p_idx--;
2303                 }
2304
2305                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2306                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2307                                 path[i].p_idx);
2308                 if (ext4_ext_more_to_rm(path + i)) {
2309                         struct buffer_head *bh;
2310                         /* go to the next level */
2311                         ext_debug("move to level %d (block %llu)\n",
2312                                   i + 1, idx_pblock(path[i].p_idx));
2313                         memset(path + i + 1, 0, sizeof(*path));
2314                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2315                         if (!bh) {
2316                                 /* should we reset i_size? */
2317                                 err = -EIO;
2318                                 break;
2319                         }
2320                         if (WARN_ON(i + 1 > depth)) {
2321                                 err = -EIO;
2322                                 break;
2323                         }
2324                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2325                                                         depth - i - 1)) {
2326                                 err = -EIO;
2327                                 break;
2328                         }
2329                         path[i + 1].p_bh = bh;
2330
2331                         /* save actual number of indexes since this
2332                          * number is changed at the next iteration */
2333                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2334                         i++;
2335                 } else {
2336                         /* we finished processing this index, go up */
2337                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2338                                 /* index is empty, remove it;
2339                                  * handle must be already prepared by the
2340                                  * truncatei_leaf() */
2341                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2342                         }
2343                         /* root level has p_bh == NULL, brelse() eats this */
2344                         brelse(path[i].p_bh);
2345                         path[i].p_bh = NULL;
2346                         i--;
2347                         ext_debug("return to level %d\n", i);
2348                 }
2349         }
2350
2351         /* TODO: flexible tree reduction should be here */
2352         if (path->p_hdr->eh_entries == 0) {
2353                 /*
2354                  * truncate to zero freed all the tree,
2355                  * so we need to correct eh_depth
2356                  */
2357                 err = ext4_ext_get_access(handle, inode, path);
2358                 if (err == 0) {
2359                         ext_inode_hdr(inode)->eh_depth = 0;
2360                         ext_inode_hdr(inode)->eh_max =
2361                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2362                         err = ext4_ext_dirty(handle, inode, path);
2363                 }
2364         }
2365 out:
2366         ext4_ext_drop_refs(path);
2367         kfree(path);
2368         ext4_journal_stop(handle);
2369
2370         return err;
2371 }
2372
2373 /*
2374  * called at mount time
2375  */
2376 void ext4_ext_init(struct super_block *sb)
2377 {
2378         /*
2379          * possible initialization would be here
2380          */
2381
2382         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2383 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2384                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2385 #ifdef AGGRESSIVE_TEST
2386                 printk(", aggressive tests");
2387 #endif
2388 #ifdef CHECK_BINSEARCH
2389                 printk(", check binsearch");
2390 #endif
2391 #ifdef EXTENTS_STATS
2392                 printk(", stats");
2393 #endif
2394                 printk("\n");
2395 #endif
2396 #ifdef EXTENTS_STATS
2397                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2398                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2399                 EXT4_SB(sb)->s_ext_max = 0;
2400 #endif
2401         }
2402 }
2403
2404 /*
2405  * called at umount time
2406  */
2407 void ext4_ext_release(struct super_block *sb)
2408 {
2409         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2410                 return;
2411
2412 #ifdef EXTENTS_STATS
2413         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2414                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2415                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2416                         sbi->s_ext_blocks, sbi->s_ext_extents,
2417                         sbi->s_ext_blocks / sbi->s_ext_extents);
2418                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2419                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2420         }
2421 #endif
2422 }
2423
2424 static void bi_complete(struct bio *bio, int error)
2425 {
2426         complete((struct completion *)bio->bi_private);
2427 }
2428
2429 /* FIXME!! we need to try to merge to left or right after zero-out  */
2430 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2431 {
2432         int ret = -EIO;
2433         struct bio *bio;
2434         int blkbits, blocksize;
2435         sector_t ee_pblock;
2436         struct completion event;
2437         unsigned int ee_len, len, done, offset;
2438
2439
2440         blkbits   = inode->i_blkbits;
2441         blocksize = inode->i_sb->s_blocksize;
2442         ee_len    = ext4_ext_get_actual_len(ex);
2443         ee_pblock = ext_pblock(ex);
2444
2445         /* convert ee_pblock to 512 byte sectors */
2446         ee_pblock = ee_pblock << (blkbits - 9);
2447
2448         while (ee_len > 0) {
2449
2450                 if (ee_len > BIO_MAX_PAGES)
2451                         len = BIO_MAX_PAGES;
2452                 else
2453                         len = ee_len;
2454
2455                 bio = bio_alloc(GFP_NOIO, len);
2456                 bio->bi_sector = ee_pblock;
2457                 bio->bi_bdev   = inode->i_sb->s_bdev;
2458
2459                 done = 0;
2460                 offset = 0;
2461                 while (done < len) {
2462                         ret = bio_add_page(bio, ZERO_PAGE(0),
2463                                                         blocksize, offset);
2464                         if (ret != blocksize) {
2465                                 /*
2466                                  * We can't add any more pages because of
2467                                  * hardware limitations.  Start a new bio.
2468                                  */
2469                                 break;
2470                         }
2471                         done++;
2472                         offset += blocksize;
2473                         if (offset >= PAGE_CACHE_SIZE)
2474                                 offset = 0;
2475                 }
2476
2477                 init_completion(&event);
2478                 bio->bi_private = &event;
2479                 bio->bi_end_io = bi_complete;
2480                 submit_bio(WRITE, bio);
2481                 wait_for_completion(&event);
2482
2483                 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2484                         ret = 0;
2485                 else {
2486                         ret = -EIO;
2487                         break;
2488                 }
2489                 bio_put(bio);
2490                 ee_len    -= done;
2491                 ee_pblock += done  << (blkbits - 9);
2492         }
2493         return ret;
2494 }
2495
2496 #define EXT4_EXT_ZERO_LEN 7
2497 /*
2498  * This function is called by ext4_ext_get_blocks() if someone tries to write
2499  * to an uninitialized extent. It may result in splitting the uninitialized
2500  * extent into multiple extents (upto three - one initialized and two
2501  * uninitialized).
2502  * There are three possibilities:
2503  *   a> There is no split required: Entire extent should be initialized
2504  *   b> Splits in two extents: Write is happening at either end of the extent
2505  *   c> Splits in three extents: Somone is writing in middle of the extent
2506  */
2507 static int ext4_ext_convert_to_initialized(handle_t *handle,
2508                                                 struct inode *inode,
2509                                                 struct ext4_ext_path *path,
2510                                                 ext4_lblk_t iblock,
2511                                                 unsigned int max_blocks)
2512 {
2513         struct ext4_extent *ex, newex, orig_ex;
2514         struct ext4_extent *ex1 = NULL;
2515         struct ext4_extent *ex2 = NULL;
2516         struct ext4_extent *ex3 = NULL;
2517         struct ext4_extent_header *eh;
2518         ext4_lblk_t ee_block;
2519         unsigned int allocated, ee_len, depth;
2520         ext4_fsblk_t newblock;
2521         int err = 0;
2522         int ret = 0;
2523
2524         depth = ext_depth(inode);
2525         eh = path[depth].p_hdr;
2526         ex = path[depth].p_ext;
2527         ee_block = le32_to_cpu(ex->ee_block);
2528         ee_len = ext4_ext_get_actual_len(ex);
2529         allocated = ee_len - (iblock - ee_block);
2530         newblock = iblock - ee_block + ext_pblock(ex);
2531         ex2 = ex;
2532         orig_ex.ee_block = ex->ee_block;
2533         orig_ex.ee_len   = cpu_to_le16(ee_len);
2534         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2535
2536         err = ext4_ext_get_access(handle, inode, path + depth);
2537         if (err)
2538                 goto out;
2539         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2540         if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2541                 err =  ext4_ext_zeroout(inode, &orig_ex);
2542                 if (err)
2543                         goto fix_extent_len;
2544                 /* update the extent length and mark as initialized */
2545                 ex->ee_block = orig_ex.ee_block;
2546                 ex->ee_len   = orig_ex.ee_len;
2547                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2548                 ext4_ext_dirty(handle, inode, path + depth);
2549                 /* zeroed the full extent */
2550                 return allocated;
2551         }
2552
2553         /* ex1: ee_block to iblock - 1 : uninitialized */
2554         if (iblock > ee_block) {
2555                 ex1 = ex;
2556                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2557                 ext4_ext_mark_uninitialized(ex1);
2558                 ex2 = &newex;
2559         }
2560         /*
2561          * for sanity, update the length of the ex2 extent before
2562          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2563          * overlap of blocks.
2564          */
2565         if (!ex1 && allocated > max_blocks)
2566                 ex2->ee_len = cpu_to_le16(max_blocks);
2567         /* ex3: to ee_block + ee_len : uninitialised */
2568         if (allocated > max_blocks) {
2569                 unsigned int newdepth;
2570                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2571                 if (allocated <= EXT4_EXT_ZERO_LEN) {
2572                         /*
2573                          * iblock == ee_block is handled by the zerouout
2574                          * at the beginning.
2575                          * Mark first half uninitialized.
2576                          * Mark second half initialized and zero out the
2577                          * initialized extent
2578                          */
2579                         ex->ee_block = orig_ex.ee_block;
2580                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2581                         ext4_ext_mark_uninitialized(ex);
2582                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2583                         ext4_ext_dirty(handle, inode, path + depth);
2584
2585                         ex3 = &newex;
2586                         ex3->ee_block = cpu_to_le32(iblock);
2587                         ext4_ext_store_pblock(ex3, newblock);
2588                         ex3->ee_len = cpu_to_le16(allocated);
2589                         err = ext4_ext_insert_extent(handle, inode, path,
2590                                                         ex3, 0);
2591                         if (err == -ENOSPC) {
2592                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2593                                 if (err)
2594                                         goto fix_extent_len;
2595                                 ex->ee_block = orig_ex.ee_block;
2596                                 ex->ee_len   = orig_ex.ee_len;
2597                                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2598                                 ext4_ext_dirty(handle, inode, path + depth);
2599                                 /* blocks available from iblock */
2600                                 return allocated;
2601
2602                         } else if (err)
2603                                 goto fix_extent_len;
2604
2605                         /*
2606                          * We need to zero out the second half because
2607                          * an fallocate request can update file size and
2608                          * converting the second half to initialized extent
2609                          * implies that we can leak some junk data to user
2610                          * space.
2611                          */
2612                         err =  ext4_ext_zeroout(inode, ex3);
2613                         if (err) {
2614                                 /*
2615                                  * We should actually mark the
2616                                  * second half as uninit and return error
2617                                  * Insert would have changed the extent
2618                                  */
2619                                 depth = ext_depth(inode);
2620                                 ext4_ext_drop_refs(path);
2621                                 path = ext4_ext_find_extent(inode,
2622                                                                 iblock, path);
2623                                 if (IS_ERR(path)) {
2624                                         err = PTR_ERR(path);
2625                                         return err;
2626                                 }
2627                                 /* get the second half extent details */
2628                                 ex = path[depth].p_ext;
2629                                 err = ext4_ext_get_access(handle, inode,
2630                                                                 path + depth);
2631                                 if (err)
2632                                         return err;
2633                                 ext4_ext_mark_uninitialized(ex);
2634                                 ext4_ext_dirty(handle, inode, path + depth);
2635                                 return err;
2636                         }
2637
2638                         /* zeroed the second half */
2639                         return allocated;
2640                 }
2641                 ex3 = &newex;
2642                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2643                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2644                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2645                 ext4_ext_mark_uninitialized(ex3);
2646                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2647                 if (err == -ENOSPC) {
2648                         err =  ext4_ext_zeroout(inode, &orig_ex);
2649                         if (err)
2650                                 goto fix_extent_len;
2651                         /* update the extent length and mark as initialized */
2652                         ex->ee_block = orig_ex.ee_block;
2653                         ex->ee_len   = orig_ex.ee_len;
2654                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2655                         ext4_ext_dirty(handle, inode, path + depth);
2656                         /* zeroed the full extent */
2657                         /* blocks available from iblock */
2658                         return allocated;
2659
2660                 } else if (err)
2661                         goto fix_extent_len;
2662                 /*
2663                  * The depth, and hence eh & ex might change
2664                  * as part of the insert above.
2665                  */
2666                 newdepth = ext_depth(inode);
2667                 /*
2668                  * update the extent length after successful insert of the
2669                  * split extent
2670                  */
2671                 orig_ex.ee_len = cpu_to_le16(ee_len -
2672                                                 ext4_ext_get_actual_len(ex3));
2673                 depth = newdepth;
2674                 ext4_ext_drop_refs(path);
2675                 path = ext4_ext_find_extent(inode, iblock, path);
2676                 if (IS_ERR(path)) {
2677                         err = PTR_ERR(path);
2678                         goto out;
2679                 }
2680                 eh = path[depth].p_hdr;
2681                 ex = path[depth].p_ext;
2682                 if (ex2 != &newex)
2683                         ex2 = ex;
2684
2685                 err = ext4_ext_get_access(handle, inode, path + depth);
2686                 if (err)
2687                         goto out;
2688
2689                 allocated = max_blocks;
2690
2691                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2692                  * to insert a extent in the middle zerout directly
2693                  * otherwise give the extent a chance to merge to left
2694                  */
2695                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2696                                                         iblock != ee_block) {
2697                         err =  ext4_ext_zeroout(inode, &orig_ex);
2698                         if (err)
2699                                 goto fix_extent_len;
2700                         /* update the extent length and mark as initialized */
2701                         ex->ee_block = orig_ex.ee_block;
2702                         ex->ee_len   = orig_ex.ee_len;
2703                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2704                         ext4_ext_dirty(handle, inode, path + depth);
2705                         /* zero out the first half */
2706                         /* blocks available from iblock */
2707                         return allocated;
2708                 }
2709         }
2710         /*
2711          * If there was a change of depth as part of the
2712          * insertion of ex3 above, we need to update the length
2713          * of the ex1 extent again here
2714          */
2715         if (ex1 && ex1 != ex) {
2716                 ex1 = ex;
2717                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2718                 ext4_ext_mark_uninitialized(ex1);
2719                 ex2 = &newex;
2720         }
2721         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2722         ex2->ee_block = cpu_to_le32(iblock);
2723         ext4_ext_store_pblock(ex2, newblock);
2724         ex2->ee_len = cpu_to_le16(allocated);
2725         if (ex2 != ex)
2726                 goto insert;
2727         /*
2728          * New (initialized) extent starts from the first block
2729          * in the current extent. i.e., ex2 == ex
2730          * We have to see if it can be merged with the extent
2731          * on the left.
2732          */
2733         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2734                 /*
2735                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2736                  * since it merges towards right _only_.
2737                  */
2738                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2739                 if (ret) {
2740                         err = ext4_ext_correct_indexes(handle, inode, path);
2741                         if (err)
2742                                 goto out;
2743                         depth = ext_depth(inode);
2744                         ex2--;
2745                 }
2746         }
2747         /*
2748          * Try to Merge towards right. This might be required
2749          * only when the whole extent is being written to.
2750          * i.e. ex2 == ex and ex3 == NULL.
2751          */
2752         if (!ex3) {
2753                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2754                 if (ret) {
2755                         err = ext4_ext_correct_indexes(handle, inode, path);
2756                         if (err)
2757                                 goto out;
2758                 }
2759         }
2760         /* Mark modified extent as dirty */
2761         err = ext4_ext_dirty(handle, inode, path + depth);
2762         goto out;
2763 insert:
2764         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2765         if (err == -ENOSPC) {
2766                 err =  ext4_ext_zeroout(inode, &orig_ex);
2767                 if (err)
2768                         goto fix_extent_len;
2769                 /* update the extent length and mark as initialized */
2770                 ex->ee_block = orig_ex.ee_block;
2771                 ex->ee_len   = orig_ex.ee_len;
2772                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2773                 ext4_ext_dirty(handle, inode, path + depth);
2774                 /* zero out the first half */
2775                 return allocated;
2776         } else if (err)
2777                 goto fix_extent_len;
2778 out:
2779         ext4_ext_show_leaf(inode, path);
2780         return err ? err : allocated;
2781
2782 fix_extent_len:
2783         ex->ee_block = orig_ex.ee_block;
2784         ex->ee_len   = orig_ex.ee_len;
2785         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2786         ext4_ext_mark_uninitialized(ex);
2787         ext4_ext_dirty(handle, inode, path + depth);
2788         return err;
2789 }
2790
2791 /*
2792  * This function is called by ext4_ext_get_blocks() from
2793  * ext4_get_blocks_dio_write() when DIO to write
2794  * to an uninitialized extent.
2795  *
2796  * Writing to an uninitized extent may result in splitting the uninitialized
2797  * extent into multiple /intialized unintialized extents (up to three)
2798  * There are three possibilities:
2799  *   a> There is no split required: Entire extent should be uninitialized
2800  *   b> Splits in two extents: Write is happening at either end of the extent
2801  *   c> Splits in three extents: Somone is writing in middle of the extent
2802  *
2803  * One of more index blocks maybe needed if the extent tree grow after
2804  * the unintialized extent split. To prevent ENOSPC occur at the IO
2805  * complete, we need to split the uninitialized extent before DIO submit
2806  * the IO. The uninitilized extent called at this time will be split
2807  * into three uninitialized extent(at most). After IO complete, the part
2808  * being filled will be convert to initialized by the end_io callback function
2809  * via ext4_convert_unwritten_extents().
2810  */
2811 static int ext4_split_unwritten_extents(handle_t *handle,
2812                                         struct inode *inode,
2813                                         struct ext4_ext_path *path,
2814                                         ext4_lblk_t iblock,
2815                                         unsigned int max_blocks,
2816                                         int flags)
2817 {
2818         struct ext4_extent *ex, newex, orig_ex;
2819         struct ext4_extent *ex1 = NULL;
2820         struct ext4_extent *ex2 = NULL;
2821         struct ext4_extent *ex3 = NULL;
2822         struct ext4_extent_header *eh;
2823         ext4_lblk_t ee_block;
2824         unsigned int allocated, ee_len, depth;
2825         ext4_fsblk_t newblock;
2826         int err = 0;
2827         int ret = 0;
2828
2829         ext_debug("ext4_split_unwritten_extents: inode %lu,"
2830                   "iblock %llu, max_blocks %u\n", inode->i_ino,
2831                   (unsigned long long)iblock, max_blocks);
2832         depth = ext_depth(inode);
2833         eh = path[depth].p_hdr;
2834         ex = path[depth].p_ext;
2835         ee_block = le32_to_cpu(ex->ee_block);
2836         ee_len = ext4_ext_get_actual_len(ex);
2837         allocated = ee_len - (iblock - ee_block);
2838         newblock = iblock - ee_block + ext_pblock(ex);
2839         ex2 = ex;
2840         orig_ex.ee_block = ex->ee_block;
2841         orig_ex.ee_len   = cpu_to_le16(ee_len);
2842         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2843
2844         /*
2845          * if the entire unintialized extent length less than
2846          * the size of extent to write, there is no need to split
2847          * uninitialized extent
2848          */
2849         if (allocated <= max_blocks)
2850                 return ret;
2851
2852         err = ext4_ext_get_access(handle, inode, path + depth);
2853         if (err)
2854                 goto out;
2855         /* ex1: ee_block to iblock - 1 : uninitialized */
2856         if (iblock > ee_block) {
2857                 ex1 = ex;
2858                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2859                 ext4_ext_mark_uninitialized(ex1);
2860                 ex2 = &newex;
2861         }
2862         /*
2863          * for sanity, update the length of the ex2 extent before
2864          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2865          * overlap of blocks.
2866          */
2867         if (!ex1 && allocated > max_blocks)
2868                 ex2->ee_len = cpu_to_le16(max_blocks);
2869         /* ex3: to ee_block + ee_len : uninitialised */
2870         if (allocated > max_blocks) {
2871                 unsigned int newdepth;
2872                 ex3 = &newex;
2873                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2874                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2875                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2876                 ext4_ext_mark_uninitialized(ex3);
2877                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2878                 if (err == -ENOSPC) {
2879                         err =  ext4_ext_zeroout(inode, &orig_ex);
2880                         if (err)
2881                                 goto fix_extent_len;
2882                         /* update the extent length and mark as initialized */
2883                         ex->ee_block = orig_ex.ee_block;
2884                         ex->ee_len   = orig_ex.ee_len;
2885                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2886                         ext4_ext_dirty(handle, inode, path + depth);
2887                         /* zeroed the full extent */
2888                         /* blocks available from iblock */
2889                         return allocated;
2890
2891                 } else if (err)
2892                         goto fix_extent_len;
2893                 /*
2894                  * The depth, and hence eh & ex might change
2895                  * as part of the insert above.
2896                  */
2897                 newdepth = ext_depth(inode);
2898                 /*
2899                  * update the extent length after successful insert of the
2900                  * split extent
2901                  */
2902                 orig_ex.ee_len = cpu_to_le16(ee_len -
2903                                                 ext4_ext_get_actual_len(ex3));
2904                 depth = newdepth;
2905                 ext4_ext_drop_refs(path);
2906                 path = ext4_ext_find_extent(inode, iblock, path);
2907                 if (IS_ERR(path)) {
2908                         err = PTR_ERR(path);
2909                         goto out;
2910                 }
2911                 eh = path[depth].p_hdr;
2912                 ex = path[depth].p_ext;
2913                 if (ex2 != &newex)
2914                         ex2 = ex;
2915
2916                 err = ext4_ext_get_access(handle, inode, path + depth);
2917                 if (err)
2918                         goto out;
2919
2920                 allocated = max_blocks;
2921         }
2922         /*
2923          * If there was a change of depth as part of the
2924          * insertion of ex3 above, we need to update the length
2925          * of the ex1 extent again here
2926          */
2927         if (ex1 && ex1 != ex) {
2928                 ex1 = ex;
2929                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2930                 ext4_ext_mark_uninitialized(ex1);
2931                 ex2 = &newex;
2932         }
2933         /*
2934          * ex2: iblock to iblock + maxblocks-1 : to be direct IO written,
2935          * uninitialised still.
2936          */
2937         ex2->ee_block = cpu_to_le32(iblock);
2938         ext4_ext_store_pblock(ex2, newblock);
2939         ex2->ee_len = cpu_to_le16(allocated);
2940         ext4_ext_mark_uninitialized(ex2);
2941         if (ex2 != ex)
2942                 goto insert;
2943         /* Mark modified extent as dirty */
2944         err = ext4_ext_dirty(handle, inode, path + depth);
2945         ext_debug("out here\n");
2946         goto out;
2947 insert:
2948         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2949         if (err == -ENOSPC) {
2950                 err =  ext4_ext_zeroout(inode, &orig_ex);
2951                 if (err)
2952                         goto fix_extent_len;
2953                 /* update the extent length and mark as initialized */
2954                 ex->ee_block = orig_ex.ee_block;
2955                 ex->ee_len   = orig_ex.ee_len;
2956                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2957                 ext4_ext_dirty(handle, inode, path + depth);
2958                 /* zero out the first half */
2959                 return allocated;
2960         } else if (err)
2961                 goto fix_extent_len;
2962 out:
2963         ext4_ext_show_leaf(inode, path);
2964         return err ? err : allocated;
2965
2966 fix_extent_len:
2967         ex->ee_block = orig_ex.ee_block;
2968         ex->ee_len   = orig_ex.ee_len;
2969         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2970         ext4_ext_mark_uninitialized(ex);
2971         ext4_ext_dirty(handle, inode, path + depth);
2972         return err;
2973 }
2974 static int ext4_convert_unwritten_extents_dio(handle_t *handle,
2975                                               struct inode *inode,
2976                                               struct ext4_ext_path *path)
2977 {
2978         struct ext4_extent *ex;
2979         struct ext4_extent_header *eh;
2980         int depth;
2981         int err = 0;
2982         int ret = 0;
2983
2984         depth = ext_depth(inode);
2985         eh = path[depth].p_hdr;
2986         ex = path[depth].p_ext;
2987
2988         err = ext4_ext_get_access(handle, inode, path + depth);
2989         if (err)
2990                 goto out;
2991         /* first mark the extent as initialized */
2992         ext4_ext_mark_initialized(ex);
2993
2994         /*
2995          * We have to see if it can be merged with the extent
2996          * on the left.
2997          */
2998         if (ex > EXT_FIRST_EXTENT(eh)) {
2999                 /*
3000                  * To merge left, pass "ex - 1" to try_to_merge(),
3001                  * since it merges towards right _only_.
3002                  */
3003                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3004                 if (ret) {
3005                         err = ext4_ext_correct_indexes(handle, inode, path);
3006                         if (err)
3007                                 goto out;
3008                         depth = ext_depth(inode);
3009                         ex--;
3010                 }
3011         }
3012         /*
3013          * Try to Merge towards right.
3014          */
3015         ret = ext4_ext_try_to_merge(inode, path, ex);
3016         if (ret) {
3017                 err = ext4_ext_correct_indexes(handle, inode, path);
3018                 if (err)
3019                         goto out;
3020                 depth = ext_depth(inode);
3021         }
3022         /* Mark modified extent as dirty */
3023         err = ext4_ext_dirty(handle, inode, path + depth);
3024 out:
3025         ext4_ext_show_leaf(inode, path);
3026         return err;
3027 }
3028
3029 static int
3030 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3031                         ext4_lblk_t iblock, unsigned int max_blocks,
3032                         struct ext4_ext_path *path, int flags,
3033                         unsigned int allocated, struct buffer_head *bh_result,
3034                         ext4_fsblk_t newblock)
3035 {
3036         int ret = 0;
3037         int err = 0;
3038         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3039
3040         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3041                   "block %llu, max_blocks %u, flags %d, allocated %u",
3042                   inode->i_ino, (unsigned long long)iblock, max_blocks,
3043                   flags, allocated);
3044         ext4_ext_show_leaf(inode, path);
3045
3046         /* DIO get_block() before submit the IO, split the extent */
3047         if (flags == EXT4_GET_BLOCKS_DIO_CREATE_EXT) {
3048                 ret = ext4_split_unwritten_extents(handle,
3049                                                 inode, path, iblock,
3050                                                 max_blocks, flags);
3051                 /* flag the io_end struct that we need convert when IO done */
3052                 if (io)
3053                         io->flag = DIO_AIO_UNWRITTEN;
3054                 goto out;
3055         }
3056         /* DIO end_io complete, convert the filled extent to written */
3057         if (flags == EXT4_GET_BLOCKS_DIO_CONVERT_EXT) {
3058                 ret = ext4_convert_unwritten_extents_dio(handle, inode,
3059                                                         path);
3060                 goto out2;
3061         }
3062         /* buffered IO case */
3063         /*
3064          * repeat fallocate creation request
3065          * we already have an unwritten extent
3066          */
3067         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3068                 goto map_out;
3069
3070         /* buffered READ or buffered write_begin() lookup */
3071         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3072                 /*
3073                  * We have blocks reserved already.  We
3074                  * return allocated blocks so that delalloc
3075                  * won't do block reservation for us.  But
3076                  * the buffer head will be unmapped so that
3077                  * a read from the block returns 0s.
3078                  */
3079                 set_buffer_unwritten(bh_result);
3080                 goto out1;
3081         }
3082
3083         /* buffered write, writepage time, convert*/
3084         ret = ext4_ext_convert_to_initialized(handle, inode,
3085                                                 path, iblock,
3086                                                 max_blocks);
3087 out:
3088         if (ret <= 0) {
3089                 err = ret;
3090                 goto out2;
3091         } else
3092                 allocated = ret;
3093         set_buffer_new(bh_result);
3094 map_out:
3095         set_buffer_mapped(bh_result);
3096 out1:
3097         if (allocated > max_blocks)
3098                 allocated = max_blocks;
3099         ext4_ext_show_leaf(inode, path);
3100         bh_result->b_bdev = inode->i_sb->s_bdev;
3101         bh_result->b_blocknr = newblock;
3102 out2:
3103         if (path) {
3104                 ext4_ext_drop_refs(path);
3105                 kfree(path);
3106         }
3107         return err ? err : allocated;
3108 }
3109 /*
3110  * Block allocation/map/preallocation routine for extents based files
3111  *
3112  *
3113  * Need to be called with
3114  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3115  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3116  *
3117  * return > 0, number of of blocks already mapped/allocated
3118  *          if create == 0 and these are pre-allocated blocks
3119  *              buffer head is unmapped
3120  *          otherwise blocks are mapped
3121  *
3122  * return = 0, if plain look up failed (blocks have not been allocated)
3123  *          buffer head is unmapped
3124  *
3125  * return < 0, error case.
3126  */
3127 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3128                         ext4_lblk_t iblock,
3129                         unsigned int max_blocks, struct buffer_head *bh_result,
3130                         int flags)
3131 {
3132         struct ext4_ext_path *path = NULL;
3133         struct ext4_extent_header *eh;
3134         struct ext4_extent newex, *ex;
3135         ext4_fsblk_t newblock;
3136         int err = 0, depth, ret, cache_type;
3137         unsigned int allocated = 0;
3138         struct ext4_allocation_request ar;
3139         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3140
3141         __clear_bit(BH_New, &bh_result->b_state);
3142         ext_debug("blocks %u/%u requested for inode %lu\n",
3143                         iblock, max_blocks, inode->i_ino);
3144
3145         /* check in cache */
3146         cache_type = ext4_ext_in_cache(inode, iblock, &newex);
3147         if (cache_type) {
3148                 if (cache_type == EXT4_EXT_CACHE_GAP) {
3149                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3150                                 /*
3151                                  * block isn't allocated yet and
3152                                  * user doesn't want to allocate it
3153                                  */
3154                                 goto out2;
3155                         }
3156                         /* we should allocate requested block */
3157                 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3158                         /* block is already allocated */
3159                         newblock = iblock
3160                                    - le32_to_cpu(newex.ee_block)
3161                                    + ext_pblock(&newex);
3162                         /* number of remaining blocks in the extent */
3163                         allocated = ext4_ext_get_actual_len(&newex) -
3164                                         (iblock - le32_to_cpu(newex.ee_block));
3165                         goto out;
3166                 } else {
3167                         BUG();
3168                 }
3169         }
3170
3171         /* find extent for this block */
3172         path = ext4_ext_find_extent(inode, iblock, NULL);
3173         if (IS_ERR(path)) {
3174                 err = PTR_ERR(path);
3175                 path = NULL;
3176                 goto out2;
3177         }
3178
3179         depth = ext_depth(inode);
3180
3181         /*
3182          * consistent leaf must not be empty;
3183          * this situation is possible, though, _during_ tree modification;
3184          * this is why assert can't be put in ext4_ext_find_extent()
3185          */
3186         BUG_ON(path[depth].p_ext == NULL && depth != 0);
3187         eh = path[depth].p_hdr;
3188
3189         ex = path[depth].p_ext;
3190         if (ex) {
3191                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3192                 ext4_fsblk_t ee_start = ext_pblock(ex);
3193                 unsigned short ee_len;
3194
3195                 /*
3196                  * Uninitialized extents are treated as holes, except that
3197                  * we split out initialized portions during a write.
3198                  */
3199                 ee_len = ext4_ext_get_actual_len(ex);
3200                 /* if found extent covers block, simply return it */
3201                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
3202                         newblock = iblock - ee_block + ee_start;
3203                         /* number of remaining blocks in the extent */
3204                         allocated = ee_len - (iblock - ee_block);
3205                         ext_debug("%u fit into %u:%d -> %llu\n", iblock,
3206                                         ee_block, ee_len, newblock);
3207
3208                         /* Do not put uninitialized extent in the cache */
3209                         if (!ext4_ext_is_uninitialized(ex)) {
3210                                 ext4_ext_put_in_cache(inode, ee_block,
3211                                                         ee_len, ee_start,
3212                                                         EXT4_EXT_CACHE_EXTENT);
3213                                 goto out;
3214                         }
3215                         ret = ext4_ext_handle_uninitialized_extents(handle,
3216                                         inode, iblock, max_blocks, path,
3217                                         flags, allocated, bh_result, newblock);
3218                         return ret;
3219                 }
3220         }
3221
3222         /*
3223          * requested block isn't allocated yet;
3224          * we couldn't try to create block if create flag is zero
3225          */
3226         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3227                 /*
3228                  * put just found gap into cache to speed up
3229                  * subsequent requests
3230                  */
3231                 ext4_ext_put_gap_in_cache(inode, path, iblock);
3232                 goto out2;
3233         }
3234         /*
3235          * Okay, we need to do block allocation.
3236          */
3237
3238         /* find neighbour allocated blocks */
3239         ar.lleft = iblock;
3240         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3241         if (err)
3242                 goto out2;
3243         ar.lright = iblock;
3244         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3245         if (err)
3246                 goto out2;
3247
3248         /*
3249          * See if request is beyond maximum number of blocks we can have in
3250          * a single extent. For an initialized extent this limit is
3251          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3252          * EXT_UNINIT_MAX_LEN.
3253          */
3254         if (max_blocks > EXT_INIT_MAX_LEN &&
3255             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3256                 max_blocks = EXT_INIT_MAX_LEN;
3257         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
3258                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3259                 max_blocks = EXT_UNINIT_MAX_LEN;
3260
3261         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
3262         newex.ee_block = cpu_to_le32(iblock);
3263         newex.ee_len = cpu_to_le16(max_blocks);
3264         err = ext4_ext_check_overlap(inode, &newex, path);
3265         if (err)
3266                 allocated = ext4_ext_get_actual_len(&newex);
3267         else
3268                 allocated = max_blocks;
3269
3270         /* allocate new block */
3271         ar.inode = inode;
3272         ar.goal = ext4_ext_find_goal(inode, path, iblock);
3273         ar.logical = iblock;
3274         ar.len = allocated;
3275         if (S_ISREG(inode->i_mode))
3276                 ar.flags = EXT4_MB_HINT_DATA;
3277         else
3278                 /* disable in-core preallocation for non-regular files */
3279                 ar.flags = 0;
3280         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3281         if (!newblock)
3282                 goto out2;
3283         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3284                   ar.goal, newblock, allocated);
3285
3286         /* try to insert new extent into found leaf and return */
3287         ext4_ext_store_pblock(&newex, newblock);
3288         newex.ee_len = cpu_to_le16(ar.len);
3289         /* Mark uninitialized */
3290         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3291                 ext4_ext_mark_uninitialized(&newex);
3292                 /*
3293                  * io_end structure was created for every async
3294                  * direct IO write to the middle of the file.
3295                  * To avoid unecessary convertion for every aio dio rewrite
3296                  * to the mid of file, here we flag the IO that is really
3297                  * need the convertion.
3298                  *
3299                  */
3300                 if (io && flags == EXT4_GET_BLOCKS_DIO_CREATE_EXT)
3301                         io->flag = DIO_AIO_UNWRITTEN;
3302         }
3303         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3304         if (err) {
3305                 /* free data blocks we just allocated */
3306                 /* not a good idea to call discard here directly,
3307                  * but otherwise we'd need to call it every free() */
3308                 ext4_discard_preallocations(inode);
3309                 ext4_free_blocks(handle, inode, ext_pblock(&newex),
3310                                         ext4_ext_get_actual_len(&newex), 0);
3311                 goto out2;
3312         }
3313
3314         /* previous routine could use block we allocated */
3315         newblock = ext_pblock(&newex);
3316         allocated = ext4_ext_get_actual_len(&newex);
3317         set_buffer_new(bh_result);
3318
3319         /* Cache only when it is _not_ an uninitialized extent */
3320         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
3321                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
3322                                                 EXT4_EXT_CACHE_EXTENT);
3323 out:
3324         if (allocated > max_blocks)
3325                 allocated = max_blocks;
3326         ext4_ext_show_leaf(inode, path);
3327         set_buffer_mapped(bh_result);
3328         bh_result->b_bdev = inode->i_sb->s_bdev;
3329         bh_result->b_blocknr = newblock;
3330 out2:
3331         if (path) {
3332                 ext4_ext_drop_refs(path);
3333                 kfree(path);
3334         }
3335         return err ? err : allocated;
3336 }
3337
3338 void ext4_ext_truncate(struct inode *inode)
3339 {
3340         struct address_space *mapping = inode->i_mapping;
3341         struct super_block *sb = inode->i_sb;
3342         ext4_lblk_t last_block;
3343         handle_t *handle;
3344         int err = 0;
3345
3346         /*
3347          * probably first extent we're gonna free will be last in block
3348          */
3349         err = ext4_writepage_trans_blocks(inode);
3350         handle = ext4_journal_start(inode, err);
3351         if (IS_ERR(handle))
3352                 return;
3353
3354         if (inode->i_size & (sb->s_blocksize - 1))
3355                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3356
3357         if (ext4_orphan_add(handle, inode))
3358                 goto out_stop;
3359
3360         down_write(&EXT4_I(inode)->i_data_sem);
3361         ext4_ext_invalidate_cache(inode);
3362
3363         ext4_discard_preallocations(inode);
3364
3365         /*
3366          * TODO: optimization is possible here.
3367          * Probably we need not scan at all,
3368          * because page truncation is enough.
3369          */
3370
3371         /* we have to know where to truncate from in crash case */
3372         EXT4_I(inode)->i_disksize = inode->i_size;
3373         ext4_mark_inode_dirty(handle, inode);
3374
3375         last_block = (inode->i_size + sb->s_blocksize - 1)
3376                         >> EXT4_BLOCK_SIZE_BITS(sb);
3377         err = ext4_ext_remove_space(inode, last_block);
3378
3379         /* In a multi-transaction truncate, we only make the final
3380          * transaction synchronous.
3381          */
3382         if (IS_SYNC(inode))
3383                 ext4_handle_sync(handle);
3384
3385 out_stop:
3386         up_write(&EXT4_I(inode)->i_data_sem);
3387         /*
3388          * If this was a simple ftruncate() and the file will remain alive,
3389          * then we need to clear up the orphan record which we created above.
3390          * However, if this was a real unlink then we were called by
3391          * ext4_delete_inode(), and we allow that function to clean up the
3392          * orphan info for us.
3393          */
3394         if (inode->i_nlink)
3395                 ext4_orphan_del(handle, inode);
3396
3397         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3398         ext4_mark_inode_dirty(handle, inode);
3399         ext4_journal_stop(handle);
3400 }
3401
3402 static void ext4_falloc_update_inode(struct inode *inode,
3403                                 int mode, loff_t new_size, int update_ctime)
3404 {
3405         struct timespec now;
3406
3407         if (update_ctime) {
3408                 now = current_fs_time(inode->i_sb);
3409                 if (!timespec_equal(&inode->i_ctime, &now))
3410                         inode->i_ctime = now;
3411         }
3412         /*
3413          * Update only when preallocation was requested beyond
3414          * the file size.
3415          */
3416         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3417                 if (new_size > i_size_read(inode))
3418                         i_size_write(inode, new_size);
3419                 if (new_size > EXT4_I(inode)->i_disksize)
3420                         ext4_update_i_disksize(inode, new_size);
3421         }
3422
3423 }
3424
3425 /*
3426  * preallocate space for a file. This implements ext4's fallocate inode
3427  * operation, which gets called from sys_fallocate system call.
3428  * For block-mapped files, posix_fallocate should fall back to the method
3429  * of writing zeroes to the required new blocks (the same behavior which is
3430  * expected for file systems which do not support fallocate() system call).
3431  */
3432 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3433 {
3434         handle_t *handle;
3435         ext4_lblk_t block;
3436         loff_t new_size;
3437         unsigned int max_blocks;
3438         int ret = 0;
3439         int ret2 = 0;
3440         int retries = 0;
3441         struct buffer_head map_bh;
3442         unsigned int credits, blkbits = inode->i_blkbits;
3443
3444         /*
3445          * currently supporting (pre)allocate mode for extent-based
3446          * files _only_
3447          */
3448         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3449                 return -EOPNOTSUPP;
3450
3451         /* preallocation to directories is currently not supported */
3452         if (S_ISDIR(inode->i_mode))
3453                 return -ENODEV;
3454
3455         block = offset >> blkbits;
3456         /*
3457          * We can't just convert len to max_blocks because
3458          * If blocksize = 4096 offset = 3072 and len = 2048
3459          */
3460         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3461                                                         - block;
3462         /*
3463          * credits to insert 1 extent into extent tree
3464          */
3465         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3466         mutex_lock(&inode->i_mutex);
3467 retry:
3468         while (ret >= 0 && ret < max_blocks) {
3469                 block = block + ret;
3470                 max_blocks = max_blocks - ret;
3471                 handle = ext4_journal_start(inode, credits);
3472                 if (IS_ERR(handle)) {
3473                         ret = PTR_ERR(handle);
3474                         break;
3475                 }
3476                 map_bh.b_state = 0;
3477                 ret = ext4_get_blocks(handle, inode, block,
3478                                       max_blocks, &map_bh,
3479                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3480                 if (ret <= 0) {
3481 #ifdef EXT4FS_DEBUG
3482                         WARN_ON(ret <= 0);
3483                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3484                                     "returned error inode#%lu, block=%u, "
3485                                     "max_blocks=%u", __func__,
3486                                     inode->i_ino, block, max_blocks);
3487 #endif
3488                         ext4_mark_inode_dirty(handle, inode);
3489                         ret2 = ext4_journal_stop(handle);
3490                         break;
3491                 }
3492                 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3493                                                 blkbits) >> blkbits))
3494                         new_size = offset + len;
3495                 else
3496                         new_size = (block + ret) << blkbits;
3497
3498                 ext4_falloc_update_inode(inode, mode, new_size,
3499                                                 buffer_new(&map_bh));
3500                 ext4_mark_inode_dirty(handle, inode);
3501                 ret2 = ext4_journal_stop(handle);
3502                 if (ret2)
3503                         break;
3504         }
3505         if (ret == -ENOSPC &&
3506                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3507                 ret = 0;
3508                 goto retry;
3509         }
3510         mutex_unlock(&inode->i_mutex);
3511         return ret > 0 ? ret2 : ret;
3512 }
3513
3514 /*
3515  * This function convert a range of blocks to written extents
3516  * The caller of this function will pass the start offset and the size.
3517  * all unwritten extents within this range will be converted to
3518  * written extents.
3519  *
3520  * This function is called from the direct IO end io call back
3521  * function, to convert the fallocated extents after IO is completed.
3522  * Returns 0 on success.
3523  */
3524 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3525                                     loff_t len)
3526 {
3527         handle_t *handle;
3528         ext4_lblk_t block;
3529         unsigned int max_blocks;
3530         int ret = 0;
3531         int ret2 = 0;
3532         struct buffer_head map_bh;
3533         unsigned int credits, blkbits = inode->i_blkbits;
3534
3535         block = offset >> blkbits;
3536         /*
3537          * We can't just convert len to max_blocks because
3538          * If blocksize = 4096 offset = 3072 and len = 2048
3539          */
3540         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3541                                                         - block;
3542         /*
3543          * credits to insert 1 extent into extent tree
3544          */
3545         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3546         while (ret >= 0 && ret < max_blocks) {
3547                 block = block + ret;
3548                 max_blocks = max_blocks - ret;
3549                 handle = ext4_journal_start(inode, credits);
3550                 if (IS_ERR(handle)) {
3551                         ret = PTR_ERR(handle);
3552                         break;
3553                 }
3554                 map_bh.b_state = 0;
3555                 ret = ext4_get_blocks(handle, inode, block,
3556                                       max_blocks, &map_bh,
3557                                       EXT4_GET_BLOCKS_DIO_CONVERT_EXT);
3558                 if (ret <= 0) {
3559                         WARN_ON(ret <= 0);
3560                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3561                                     "returned error inode#%lu, block=%u, "
3562                                     "max_blocks=%u", __func__,
3563                                     inode->i_ino, block, max_blocks);
3564                 }
3565                 ext4_mark_inode_dirty(handle, inode);
3566                 ret2 = ext4_journal_stop(handle);
3567                 if (ret <= 0 || ret2 )
3568                         break;
3569         }
3570         return ret > 0 ? ret2 : ret;
3571 }
3572 /*
3573  * Callback function called for each extent to gather FIEMAP information.
3574  */
3575 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3576                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3577                        void *data)
3578 {
3579         struct fiemap_extent_info *fieinfo = data;
3580         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3581         __u64   logical;
3582         __u64   physical;
3583         __u64   length;
3584         __u32   flags = 0;
3585         int     error;
3586
3587         logical =  (__u64)newex->ec_block << blksize_bits;
3588
3589         if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3590                 pgoff_t offset;
3591                 struct page *page;
3592                 struct buffer_head *bh = NULL;
3593
3594                 offset = logical >> PAGE_SHIFT;
3595                 page = find_get_page(inode->i_mapping, offset);
3596                 if (!page || !page_has_buffers(page))
3597                         return EXT_CONTINUE;
3598
3599                 bh = page_buffers(page);
3600
3601                 if (!bh)
3602                         return EXT_CONTINUE;
3603
3604                 if (buffer_delay(bh)) {
3605                         flags |= FIEMAP_EXTENT_DELALLOC;
3606                         page_cache_release(page);
3607                 } else {
3608                         page_cache_release(page);
3609                         return EXT_CONTINUE;
3610                 }
3611         }
3612
3613         physical = (__u64)newex->ec_start << blksize_bits;
3614         length =   (__u64)newex->ec_len << blksize_bits;
3615
3616         if (ex && ext4_ext_is_uninitialized(ex))
3617                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3618
3619         /*
3620          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3621          *
3622          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3623          * this also indicates no more allocated blocks.
3624          *
3625          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3626          */
3627         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3628             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3629                 loff_t size = i_size_read(inode);
3630                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3631
3632                 flags |= FIEMAP_EXTENT_LAST;
3633                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3634                     logical+length > size)
3635                         length = (size - logical + bs - 1) & ~(bs-1);
3636         }
3637
3638         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3639                                         length, flags);
3640         if (error < 0)
3641                 return error;
3642         if (error == 1)
3643                 return EXT_BREAK;
3644
3645         return EXT_CONTINUE;
3646 }
3647
3648 /* fiemap flags we can handle specified here */
3649 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3650
3651 static int ext4_xattr_fiemap(struct inode *inode,
3652                                 struct fiemap_extent_info *fieinfo)
3653 {
3654         __u64 physical = 0;
3655         __u64 length;
3656         __u32 flags = FIEMAP_EXTENT_LAST;
3657         int blockbits = inode->i_sb->s_blocksize_bits;
3658         int error = 0;
3659
3660         /* in-inode? */
3661         if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3662                 struct ext4_iloc iloc;
3663                 int offset;     /* offset of xattr in inode */
3664
3665                 error = ext4_get_inode_loc(inode, &iloc);
3666                 if (error)
3667                         return error;
3668                 physical = iloc.bh->b_blocknr << blockbits;
3669                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3670                                 EXT4_I(inode)->i_extra_isize;
3671                 physical += offset;
3672                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3673                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3674         } else { /* external block */
3675                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3676                 length = inode->i_sb->s_blocksize;
3677         }
3678
3679         if (physical)
3680                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3681                                                 length, flags);
3682         return (error < 0 ? error : 0);
3683 }
3684
3685 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3686                 __u64 start, __u64 len)
3687 {
3688         ext4_lblk_t start_blk;
3689         ext4_lblk_t len_blks;
3690         int error = 0;
3691
3692         /* fallback to generic here if not in extents fmt */
3693         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3694                 return generic_block_fiemap(inode, fieinfo, start, len,
3695                         ext4_get_block);
3696
3697         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3698                 return -EBADR;
3699
3700         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3701                 error = ext4_xattr_fiemap(inode, fieinfo);
3702         } else {
3703                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3704                 len_blks = len >> inode->i_sb->s_blocksize_bits;
3705
3706                 /*
3707                  * Walk the extent tree gathering extent information.
3708                  * ext4_ext_fiemap_cb will push extents back to user.
3709                  */
3710                 down_read(&EXT4_I(inode)->i_data_sem);
3711                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3712                                           ext4_ext_fiemap_cb, fieinfo);
3713                 up_read(&EXT4_I(inode)->i_data_sem);
3714         }
3715
3716         return error;
3717 }
3718