ext4: move __func__ into a macro for ext4_warning, ext4_error
[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, sector_t lblock)
300 {
301         struct ext4_inode_info *ei = EXT4_I(inode);
302         int idxs, num = 0;
303
304         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
305                 / sizeof(struct ext4_extent_idx));
306
307         /*
308          * If the new delayed allocation block is contiguous with the
309          * previous da block, it can share index blocks with the
310          * previous block, so we only need to allocate a new index
311          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
312          * an additional index block, and at ldxs**3 blocks, yet
313          * another index blocks.
314          */
315         if (ei->i_da_metadata_calc_len &&
316             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
317                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
318                         num++;
319                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
320                         num++;
321                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
322                         num++;
323                         ei->i_da_metadata_calc_len = 0;
324                 } else
325                         ei->i_da_metadata_calc_len++;
326                 ei->i_da_metadata_calc_last_lblock++;
327                 return num;
328         }
329
330         /*
331          * In the worst case we need a new set of index blocks at
332          * every level of the inode's extent tree.
333          */
334         ei->i_da_metadata_calc_len = 1;
335         ei->i_da_metadata_calc_last_lblock = lblock;
336         return ext_depth(inode) + 1;
337 }
338
339 static int
340 ext4_ext_max_entries(struct inode *inode, int depth)
341 {
342         int max;
343
344         if (depth == ext_depth(inode)) {
345                 if (depth == 0)
346                         max = ext4_ext_space_root(inode, 1);
347                 else
348                         max = ext4_ext_space_root_idx(inode, 1);
349         } else {
350                 if (depth == 0)
351                         max = ext4_ext_space_block(inode, 1);
352                 else
353                         max = ext4_ext_space_block_idx(inode, 1);
354         }
355
356         return max;
357 }
358
359 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
360 {
361         ext4_fsblk_t block = ext_pblock(ext);
362         int len = ext4_ext_get_actual_len(ext);
363
364         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
365 }
366
367 static int ext4_valid_extent_idx(struct inode *inode,
368                                 struct ext4_extent_idx *ext_idx)
369 {
370         ext4_fsblk_t block = idx_pblock(ext_idx);
371
372         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
373 }
374
375 static int ext4_valid_extent_entries(struct inode *inode,
376                                 struct ext4_extent_header *eh,
377                                 int depth)
378 {
379         struct ext4_extent *ext;
380         struct ext4_extent_idx *ext_idx;
381         unsigned short entries;
382         if (eh->eh_entries == 0)
383                 return 1;
384
385         entries = le16_to_cpu(eh->eh_entries);
386
387         if (depth == 0) {
388                 /* leaf entries */
389                 ext = EXT_FIRST_EXTENT(eh);
390                 while (entries) {
391                         if (!ext4_valid_extent(inode, ext))
392                                 return 0;
393                         ext++;
394                         entries--;
395                 }
396         } else {
397                 ext_idx = EXT_FIRST_INDEX(eh);
398                 while (entries) {
399                         if (!ext4_valid_extent_idx(inode, ext_idx))
400                                 return 0;
401                         ext_idx++;
402                         entries--;
403                 }
404         }
405         return 1;
406 }
407
408 static int __ext4_ext_check(const char *function, struct inode *inode,
409                                         struct ext4_extent_header *eh,
410                                         int depth)
411 {
412         const char *error_msg;
413         int max = 0;
414
415         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
416                 error_msg = "invalid magic";
417                 goto corrupted;
418         }
419         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
420                 error_msg = "unexpected eh_depth";
421                 goto corrupted;
422         }
423         if (unlikely(eh->eh_max == 0)) {
424                 error_msg = "invalid eh_max";
425                 goto corrupted;
426         }
427         max = ext4_ext_max_entries(inode, depth);
428         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
429                 error_msg = "too large eh_max";
430                 goto corrupted;
431         }
432         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
433                 error_msg = "invalid eh_entries";
434                 goto corrupted;
435         }
436         if (!ext4_valid_extent_entries(inode, eh, depth)) {
437                 error_msg = "invalid extent entries";
438                 goto corrupted;
439         }
440         return 0;
441
442 corrupted:
443         __ext4_error(inode->i_sb, function,
444                         "bad header/extent in inode #%lu: %s - magic %x, "
445                         "entries %u, max %u(%u), depth %u(%u)",
446                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
447                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
448                         max, le16_to_cpu(eh->eh_depth), depth);
449
450         return -EIO;
451 }
452
453 #define ext4_ext_check(inode, eh, depth)        \
454         __ext4_ext_check(__func__, inode, eh, depth)
455
456 int ext4_ext_check_inode(struct inode *inode)
457 {
458         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
459 }
460
461 #ifdef EXT_DEBUG
462 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
463 {
464         int k, l = path->p_depth;
465
466         ext_debug("path:");
467         for (k = 0; k <= l; k++, path++) {
468                 if (path->p_idx) {
469                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
470                             idx_pblock(path->p_idx));
471                 } else if (path->p_ext) {
472                         ext_debug("  %d:[%d]%d:%llu ",
473                                   le32_to_cpu(path->p_ext->ee_block),
474                                   ext4_ext_is_uninitialized(path->p_ext),
475                                   ext4_ext_get_actual_len(path->p_ext),
476                                   ext_pblock(path->p_ext));
477                 } else
478                         ext_debug("  []");
479         }
480         ext_debug("\n");
481 }
482
483 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
484 {
485         int depth = ext_depth(inode);
486         struct ext4_extent_header *eh;
487         struct ext4_extent *ex;
488         int i;
489
490         if (!path)
491                 return;
492
493         eh = path[depth].p_hdr;
494         ex = EXT_FIRST_EXTENT(eh);
495
496         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
497
498         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
499                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
500                           ext4_ext_is_uninitialized(ex),
501                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
502         }
503         ext_debug("\n");
504 }
505 #else
506 #define ext4_ext_show_path(inode, path)
507 #define ext4_ext_show_leaf(inode, path)
508 #endif
509
510 void ext4_ext_drop_refs(struct ext4_ext_path *path)
511 {
512         int depth = path->p_depth;
513         int i;
514
515         for (i = 0; i <= depth; i++, path++)
516                 if (path->p_bh) {
517                         brelse(path->p_bh);
518                         path->p_bh = NULL;
519                 }
520 }
521
522 /*
523  * ext4_ext_binsearch_idx:
524  * binary search for the closest index of the given block
525  * the header must be checked before calling this
526  */
527 static void
528 ext4_ext_binsearch_idx(struct inode *inode,
529                         struct ext4_ext_path *path, ext4_lblk_t block)
530 {
531         struct ext4_extent_header *eh = path->p_hdr;
532         struct ext4_extent_idx *r, *l, *m;
533
534
535         ext_debug("binsearch for %u(idx):  ", block);
536
537         l = EXT_FIRST_INDEX(eh) + 1;
538         r = EXT_LAST_INDEX(eh);
539         while (l <= r) {
540                 m = l + (r - l) / 2;
541                 if (block < le32_to_cpu(m->ei_block))
542                         r = m - 1;
543                 else
544                         l = m + 1;
545                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
546                                 m, le32_to_cpu(m->ei_block),
547                                 r, le32_to_cpu(r->ei_block));
548         }
549
550         path->p_idx = l - 1;
551         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
552                   idx_pblock(path->p_idx));
553
554 #ifdef CHECK_BINSEARCH
555         {
556                 struct ext4_extent_idx *chix, *ix;
557                 int k;
558
559                 chix = ix = EXT_FIRST_INDEX(eh);
560                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
561                   if (k != 0 &&
562                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
563                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
564                                        "first=0x%p\n", k,
565                                        ix, EXT_FIRST_INDEX(eh));
566                                 printk(KERN_DEBUG "%u <= %u\n",
567                                        le32_to_cpu(ix->ei_block),
568                                        le32_to_cpu(ix[-1].ei_block));
569                         }
570                         BUG_ON(k && le32_to_cpu(ix->ei_block)
571                                            <= le32_to_cpu(ix[-1].ei_block));
572                         if (block < le32_to_cpu(ix->ei_block))
573                                 break;
574                         chix = ix;
575                 }
576                 BUG_ON(chix != path->p_idx);
577         }
578 #endif
579
580 }
581
582 /*
583  * ext4_ext_binsearch:
584  * binary search for closest extent of the given block
585  * the header must be checked before calling this
586  */
587 static void
588 ext4_ext_binsearch(struct inode *inode,
589                 struct ext4_ext_path *path, ext4_lblk_t block)
590 {
591         struct ext4_extent_header *eh = path->p_hdr;
592         struct ext4_extent *r, *l, *m;
593
594         if (eh->eh_entries == 0) {
595                 /*
596                  * this leaf is empty:
597                  * we get such a leaf in split/add case
598                  */
599                 return;
600         }
601
602         ext_debug("binsearch for %u:  ", block);
603
604         l = EXT_FIRST_EXTENT(eh) + 1;
605         r = EXT_LAST_EXTENT(eh);
606
607         while (l <= r) {
608                 m = l + (r - l) / 2;
609                 if (block < le32_to_cpu(m->ee_block))
610                         r = m - 1;
611                 else
612                         l = m + 1;
613                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
614                                 m, le32_to_cpu(m->ee_block),
615                                 r, le32_to_cpu(r->ee_block));
616         }
617
618         path->p_ext = l - 1;
619         ext_debug("  -> %d:%llu:[%d]%d ",
620                         le32_to_cpu(path->p_ext->ee_block),
621                         ext_pblock(path->p_ext),
622                         ext4_ext_is_uninitialized(path->p_ext),
623                         ext4_ext_get_actual_len(path->p_ext));
624
625 #ifdef CHECK_BINSEARCH
626         {
627                 struct ext4_extent *chex, *ex;
628                 int k;
629
630                 chex = ex = EXT_FIRST_EXTENT(eh);
631                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
632                         BUG_ON(k && le32_to_cpu(ex->ee_block)
633                                           <= le32_to_cpu(ex[-1].ee_block));
634                         if (block < le32_to_cpu(ex->ee_block))
635                                 break;
636                         chex = ex;
637                 }
638                 BUG_ON(chex != path->p_ext);
639         }
640 #endif
641
642 }
643
644 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
645 {
646         struct ext4_extent_header *eh;
647
648         eh = ext_inode_hdr(inode);
649         eh->eh_depth = 0;
650         eh->eh_entries = 0;
651         eh->eh_magic = EXT4_EXT_MAGIC;
652         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
653         ext4_mark_inode_dirty(handle, inode);
654         ext4_ext_invalidate_cache(inode);
655         return 0;
656 }
657
658 struct ext4_ext_path *
659 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
660                                         struct ext4_ext_path *path)
661 {
662         struct ext4_extent_header *eh;
663         struct buffer_head *bh;
664         short int depth, i, ppos = 0, alloc = 0;
665
666         eh = ext_inode_hdr(inode);
667         depth = ext_depth(inode);
668
669         /* account possible depth increase */
670         if (!path) {
671                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
672                                 GFP_NOFS);
673                 if (!path)
674                         return ERR_PTR(-ENOMEM);
675                 alloc = 1;
676         }
677         path[0].p_hdr = eh;
678         path[0].p_bh = NULL;
679
680         i = depth;
681         /* walk through the tree */
682         while (i) {
683                 int need_to_validate = 0;
684
685                 ext_debug("depth %d: num %d, max %d\n",
686                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
687
688                 ext4_ext_binsearch_idx(inode, path + ppos, block);
689                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
690                 path[ppos].p_depth = i;
691                 path[ppos].p_ext = NULL;
692
693                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
694                 if (unlikely(!bh))
695                         goto err;
696                 if (!bh_uptodate_or_lock(bh)) {
697                         if (bh_submit_read(bh) < 0) {
698                                 put_bh(bh);
699                                 goto err;
700                         }
701                         /* validate the extent entries */
702                         need_to_validate = 1;
703                 }
704                 eh = ext_block_hdr(bh);
705                 ppos++;
706                 BUG_ON(ppos > depth);
707                 path[ppos].p_bh = bh;
708                 path[ppos].p_hdr = eh;
709                 i--;
710
711                 if (need_to_validate && ext4_ext_check(inode, eh, i))
712                         goto err;
713         }
714
715         path[ppos].p_depth = i;
716         path[ppos].p_ext = NULL;
717         path[ppos].p_idx = NULL;
718
719         /* find extent */
720         ext4_ext_binsearch(inode, path + ppos, block);
721         /* if not an empty leaf */
722         if (path[ppos].p_ext)
723                 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
724
725         ext4_ext_show_path(inode, path);
726
727         return path;
728
729 err:
730         ext4_ext_drop_refs(path);
731         if (alloc)
732                 kfree(path);
733         return ERR_PTR(-EIO);
734 }
735
736 /*
737  * ext4_ext_insert_index:
738  * insert new index [@logical;@ptr] into the block at @curp;
739  * check where to insert: before @curp or after @curp
740  */
741 int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
742                                 struct ext4_ext_path *curp,
743                                 int logical, ext4_fsblk_t ptr)
744 {
745         struct ext4_extent_idx *ix;
746         int len, err;
747
748         err = ext4_ext_get_access(handle, inode, curp);
749         if (err)
750                 return err;
751
752         BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
753         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
754         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
755                 /* insert after */
756                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
757                         len = (len - 1) * sizeof(struct ext4_extent_idx);
758                         len = len < 0 ? 0 : len;
759                         ext_debug("insert new index %d after: %llu. "
760                                         "move %d from 0x%p to 0x%p\n",
761                                         logical, ptr, len,
762                                         (curp->p_idx + 1), (curp->p_idx + 2));
763                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
764                 }
765                 ix = curp->p_idx + 1;
766         } else {
767                 /* insert before */
768                 len = len * sizeof(struct ext4_extent_idx);
769                 len = len < 0 ? 0 : len;
770                 ext_debug("insert new index %d before: %llu. "
771                                 "move %d from 0x%p to 0x%p\n",
772                                 logical, ptr, len,
773                                 curp->p_idx, (curp->p_idx + 1));
774                 memmove(curp->p_idx + 1, curp->p_idx, len);
775                 ix = curp->p_idx;
776         }
777
778         ix->ei_block = cpu_to_le32(logical);
779         ext4_idx_store_pblock(ix, ptr);
780         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
781
782         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
783                              > le16_to_cpu(curp->p_hdr->eh_max));
784         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
785
786         err = ext4_ext_dirty(handle, inode, curp);
787         ext4_std_error(inode->i_sb, err);
788
789         return err;
790 }
791
792 /*
793  * ext4_ext_split:
794  * inserts new subtree into the path, using free index entry
795  * at depth @at:
796  * - allocates all needed blocks (new leaf and all intermediate index blocks)
797  * - makes decision where to split
798  * - moves remaining extents and index entries (right to the split point)
799  *   into the newly allocated blocks
800  * - initializes subtree
801  */
802 static int ext4_ext_split(handle_t *handle, struct inode *inode,
803                                 struct ext4_ext_path *path,
804                                 struct ext4_extent *newext, int at)
805 {
806         struct buffer_head *bh = NULL;
807         int depth = ext_depth(inode);
808         struct ext4_extent_header *neh;
809         struct ext4_extent_idx *fidx;
810         struct ext4_extent *ex;
811         int i = at, k, m, a;
812         ext4_fsblk_t newblock, oldblock;
813         __le32 border;
814         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
815         int err = 0;
816
817         /* make decision: where to split? */
818         /* FIXME: now decision is simplest: at current extent */
819
820         /* if current leaf will be split, then we should use
821          * border from split point */
822         BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
823         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
824                 border = path[depth].p_ext[1].ee_block;
825                 ext_debug("leaf will be split."
826                                 " next leaf starts at %d\n",
827                                   le32_to_cpu(border));
828         } else {
829                 border = newext->ee_block;
830                 ext_debug("leaf will be added."
831                                 " next leaf starts at %d\n",
832                                 le32_to_cpu(border));
833         }
834
835         /*
836          * If error occurs, then we break processing
837          * and mark filesystem read-only. index won't
838          * be inserted and tree will be in consistent
839          * state. Next mount will repair buffers too.
840          */
841
842         /*
843          * Get array to track all allocated blocks.
844          * We need this to handle errors and free blocks
845          * upon them.
846          */
847         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
848         if (!ablocks)
849                 return -ENOMEM;
850
851         /* allocate all needed blocks */
852         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
853         for (a = 0; a < depth - at; a++) {
854                 newblock = ext4_ext_new_meta_block(handle, inode, path,
855                                                    newext, &err);
856                 if (newblock == 0)
857                         goto cleanup;
858                 ablocks[a] = newblock;
859         }
860
861         /* initialize new leaf */
862         newblock = ablocks[--a];
863         BUG_ON(newblock == 0);
864         bh = sb_getblk(inode->i_sb, newblock);
865         if (!bh) {
866                 err = -EIO;
867                 goto cleanup;
868         }
869         lock_buffer(bh);
870
871         err = ext4_journal_get_create_access(handle, bh);
872         if (err)
873                 goto cleanup;
874
875         neh = ext_block_hdr(bh);
876         neh->eh_entries = 0;
877         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
878         neh->eh_magic = EXT4_EXT_MAGIC;
879         neh->eh_depth = 0;
880         ex = EXT_FIRST_EXTENT(neh);
881
882         /* move remainder of path[depth] to the new leaf */
883         BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
884         /* start copy from next extent */
885         /* TODO: we could do it by single memmove */
886         m = 0;
887         path[depth].p_ext++;
888         while (path[depth].p_ext <=
889                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
890                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
891                                 le32_to_cpu(path[depth].p_ext->ee_block),
892                                 ext_pblock(path[depth].p_ext),
893                                 ext4_ext_is_uninitialized(path[depth].p_ext),
894                                 ext4_ext_get_actual_len(path[depth].p_ext),
895                                 newblock);
896                 /*memmove(ex++, path[depth].p_ext++,
897                                 sizeof(struct ext4_extent));
898                 neh->eh_entries++;*/
899                 path[depth].p_ext++;
900                 m++;
901         }
902         if (m) {
903                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
904                 le16_add_cpu(&neh->eh_entries, m);
905         }
906
907         set_buffer_uptodate(bh);
908         unlock_buffer(bh);
909
910         err = ext4_handle_dirty_metadata(handle, inode, bh);
911         if (err)
912                 goto cleanup;
913         brelse(bh);
914         bh = NULL;
915
916         /* correct old leaf */
917         if (m) {
918                 err = ext4_ext_get_access(handle, inode, path + depth);
919                 if (err)
920                         goto cleanup;
921                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
922                 err = ext4_ext_dirty(handle, inode, path + depth);
923                 if (err)
924                         goto cleanup;
925
926         }
927
928         /* create intermediate indexes */
929         k = depth - at - 1;
930         BUG_ON(k < 0);
931         if (k)
932                 ext_debug("create %d intermediate indices\n", k);
933         /* insert new index into current index block */
934         /* current depth stored in i var */
935         i = depth - 1;
936         while (k--) {
937                 oldblock = newblock;
938                 newblock = ablocks[--a];
939                 bh = sb_getblk(inode->i_sb, newblock);
940                 if (!bh) {
941                         err = -EIO;
942                         goto cleanup;
943                 }
944                 lock_buffer(bh);
945
946                 err = ext4_journal_get_create_access(handle, bh);
947                 if (err)
948                         goto cleanup;
949
950                 neh = ext_block_hdr(bh);
951                 neh->eh_entries = cpu_to_le16(1);
952                 neh->eh_magic = EXT4_EXT_MAGIC;
953                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
954                 neh->eh_depth = cpu_to_le16(depth - i);
955                 fidx = EXT_FIRST_INDEX(neh);
956                 fidx->ei_block = border;
957                 ext4_idx_store_pblock(fidx, oldblock);
958
959                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
960                                 i, newblock, le32_to_cpu(border), oldblock);
961                 /* copy indexes */
962                 m = 0;
963                 path[i].p_idx++;
964
965                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
966                                 EXT_MAX_INDEX(path[i].p_hdr));
967                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
968                                 EXT_LAST_INDEX(path[i].p_hdr));
969                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
970                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
971                                         le32_to_cpu(path[i].p_idx->ei_block),
972                                         idx_pblock(path[i].p_idx),
973                                         newblock);
974                         /*memmove(++fidx, path[i].p_idx++,
975                                         sizeof(struct ext4_extent_idx));
976                         neh->eh_entries++;
977                         BUG_ON(neh->eh_entries > neh->eh_max);*/
978                         path[i].p_idx++;
979                         m++;
980                 }
981                 if (m) {
982                         memmove(++fidx, path[i].p_idx - m,
983                                 sizeof(struct ext4_extent_idx) * m);
984                         le16_add_cpu(&neh->eh_entries, m);
985                 }
986                 set_buffer_uptodate(bh);
987                 unlock_buffer(bh);
988
989                 err = ext4_handle_dirty_metadata(handle, inode, bh);
990                 if (err)
991                         goto cleanup;
992                 brelse(bh);
993                 bh = NULL;
994
995                 /* correct old index */
996                 if (m) {
997                         err = ext4_ext_get_access(handle, inode, path + i);
998                         if (err)
999                                 goto cleanup;
1000                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1001                         err = ext4_ext_dirty(handle, inode, path + i);
1002                         if (err)
1003                                 goto cleanup;
1004                 }
1005
1006                 i--;
1007         }
1008
1009         /* insert new index */
1010         err = ext4_ext_insert_index(handle, inode, path + at,
1011                                     le32_to_cpu(border), newblock);
1012
1013 cleanup:
1014         if (bh) {
1015                 if (buffer_locked(bh))
1016                         unlock_buffer(bh);
1017                 brelse(bh);
1018         }
1019
1020         if (err) {
1021                 /* free all allocated blocks in error case */
1022                 for (i = 0; i < depth; i++) {
1023                         if (!ablocks[i])
1024                                 continue;
1025                         ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1026                                          EXT4_FREE_BLOCKS_METADATA);
1027                 }
1028         }
1029         kfree(ablocks);
1030
1031         return err;
1032 }
1033
1034 /*
1035  * ext4_ext_grow_indepth:
1036  * implements tree growing procedure:
1037  * - allocates new block
1038  * - moves top-level data (index block or leaf) into the new block
1039  * - initializes new top-level, creating index that points to the
1040  *   just created block
1041  */
1042 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1043                                         struct ext4_ext_path *path,
1044                                         struct ext4_extent *newext)
1045 {
1046         struct ext4_ext_path *curp = path;
1047         struct ext4_extent_header *neh;
1048         struct ext4_extent_idx *fidx;
1049         struct buffer_head *bh;
1050         ext4_fsblk_t newblock;
1051         int err = 0;
1052
1053         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1054         if (newblock == 0)
1055                 return err;
1056
1057         bh = sb_getblk(inode->i_sb, newblock);
1058         if (!bh) {
1059                 err = -EIO;
1060                 ext4_std_error(inode->i_sb, err);
1061                 return err;
1062         }
1063         lock_buffer(bh);
1064
1065         err = ext4_journal_get_create_access(handle, bh);
1066         if (err) {
1067                 unlock_buffer(bh);
1068                 goto out;
1069         }
1070
1071         /* move top-level index/leaf into new block */
1072         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1073
1074         /* set size of new block */
1075         neh = ext_block_hdr(bh);
1076         /* old root could have indexes or leaves
1077          * so calculate e_max right way */
1078         if (ext_depth(inode))
1079                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1080         else
1081                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1082         neh->eh_magic = EXT4_EXT_MAGIC;
1083         set_buffer_uptodate(bh);
1084         unlock_buffer(bh);
1085
1086         err = ext4_handle_dirty_metadata(handle, inode, bh);
1087         if (err)
1088                 goto out;
1089
1090         /* create index in new top-level index: num,max,pointer */
1091         err = ext4_ext_get_access(handle, inode, curp);
1092         if (err)
1093                 goto out;
1094
1095         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1096         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1097         curp->p_hdr->eh_entries = cpu_to_le16(1);
1098         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1099
1100         if (path[0].p_hdr->eh_depth)
1101                 curp->p_idx->ei_block =
1102                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1103         else
1104                 curp->p_idx->ei_block =
1105                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1106         ext4_idx_store_pblock(curp->p_idx, newblock);
1107
1108         neh = ext_inode_hdr(inode);
1109         fidx = EXT_FIRST_INDEX(neh);
1110         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1111                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1112                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1113
1114         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1115         err = ext4_ext_dirty(handle, inode, curp);
1116 out:
1117         brelse(bh);
1118
1119         return err;
1120 }
1121
1122 /*
1123  * ext4_ext_create_new_leaf:
1124  * finds empty index and adds new leaf.
1125  * if no free index is found, then it requests in-depth growing.
1126  */
1127 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1128                                         struct ext4_ext_path *path,
1129                                         struct ext4_extent *newext)
1130 {
1131         struct ext4_ext_path *curp;
1132         int depth, i, err = 0;
1133
1134 repeat:
1135         i = depth = ext_depth(inode);
1136
1137         /* walk up to the tree and look for free index entry */
1138         curp = path + depth;
1139         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1140                 i--;
1141                 curp--;
1142         }
1143
1144         /* we use already allocated block for index block,
1145          * so subsequent data blocks should be contiguous */
1146         if (EXT_HAS_FREE_INDEX(curp)) {
1147                 /* if we found index with free entry, then use that
1148                  * entry: create all needed subtree and add new leaf */
1149                 err = ext4_ext_split(handle, inode, path, newext, i);
1150                 if (err)
1151                         goto out;
1152
1153                 /* refill path */
1154                 ext4_ext_drop_refs(path);
1155                 path = ext4_ext_find_extent(inode,
1156                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1157                                     path);
1158                 if (IS_ERR(path))
1159                         err = PTR_ERR(path);
1160         } else {
1161                 /* tree is full, time to grow in depth */
1162                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1163                 if (err)
1164                         goto out;
1165
1166                 /* refill path */
1167                 ext4_ext_drop_refs(path);
1168                 path = ext4_ext_find_extent(inode,
1169                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1170                                     path);
1171                 if (IS_ERR(path)) {
1172                         err = PTR_ERR(path);
1173                         goto out;
1174                 }
1175
1176                 /*
1177                  * only first (depth 0 -> 1) produces free space;
1178                  * in all other cases we have to split the grown tree
1179                  */
1180                 depth = ext_depth(inode);
1181                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1182                         /* now we need to split */
1183                         goto repeat;
1184                 }
1185         }
1186
1187 out:
1188         return err;
1189 }
1190
1191 /*
1192  * search the closest allocated block to the left for *logical
1193  * and returns it at @logical + it's physical address at @phys
1194  * if *logical is the smallest allocated block, the function
1195  * returns 0 at @phys
1196  * return value contains 0 (success) or error code
1197  */
1198 int
1199 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1200                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1201 {
1202         struct ext4_extent_idx *ix;
1203         struct ext4_extent *ex;
1204         int depth, ee_len;
1205
1206         BUG_ON(path == NULL);
1207         depth = path->p_depth;
1208         *phys = 0;
1209
1210         if (depth == 0 && path->p_ext == NULL)
1211                 return 0;
1212
1213         /* usually extent in the path covers blocks smaller
1214          * then *logical, but it can be that extent is the
1215          * first one in the file */
1216
1217         ex = path[depth].p_ext;
1218         ee_len = ext4_ext_get_actual_len(ex);
1219         if (*logical < le32_to_cpu(ex->ee_block)) {
1220                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1221                 while (--depth >= 0) {
1222                         ix = path[depth].p_idx;
1223                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1224                 }
1225                 return 0;
1226         }
1227
1228         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1229
1230         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1231         *phys = ext_pblock(ex) + ee_len - 1;
1232         return 0;
1233 }
1234
1235 /*
1236  * search the closest allocated block to the right for *logical
1237  * and returns it at @logical + it's physical address at @phys
1238  * if *logical is the smallest allocated block, the function
1239  * returns 0 at @phys
1240  * return value contains 0 (success) or error code
1241  */
1242 int
1243 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1244                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1245 {
1246         struct buffer_head *bh = NULL;
1247         struct ext4_extent_header *eh;
1248         struct ext4_extent_idx *ix;
1249         struct ext4_extent *ex;
1250         ext4_fsblk_t block;
1251         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1252         int ee_len;
1253
1254         BUG_ON(path == NULL);
1255         depth = path->p_depth;
1256         *phys = 0;
1257
1258         if (depth == 0 && path->p_ext == NULL)
1259                 return 0;
1260
1261         /* usually extent in the path covers blocks smaller
1262          * then *logical, but it can be that extent is the
1263          * first one in the file */
1264
1265         ex = path[depth].p_ext;
1266         ee_len = ext4_ext_get_actual_len(ex);
1267         if (*logical < le32_to_cpu(ex->ee_block)) {
1268                 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1269                 while (--depth >= 0) {
1270                         ix = path[depth].p_idx;
1271                         BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1272                 }
1273                 *logical = le32_to_cpu(ex->ee_block);
1274                 *phys = ext_pblock(ex);
1275                 return 0;
1276         }
1277
1278         BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1279
1280         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1281                 /* next allocated block in this leaf */
1282                 ex++;
1283                 *logical = le32_to_cpu(ex->ee_block);
1284                 *phys = ext_pblock(ex);
1285                 return 0;
1286         }
1287
1288         /* go up and search for index to the right */
1289         while (--depth >= 0) {
1290                 ix = path[depth].p_idx;
1291                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1292                         goto got_index;
1293         }
1294
1295         /* we've gone up to the root and found no index to the right */
1296         return 0;
1297
1298 got_index:
1299         /* we've found index to the right, let's
1300          * follow it and find the closest allocated
1301          * block to the right */
1302         ix++;
1303         block = idx_pblock(ix);
1304         while (++depth < path->p_depth) {
1305                 bh = sb_bread(inode->i_sb, block);
1306                 if (bh == NULL)
1307                         return -EIO;
1308                 eh = ext_block_hdr(bh);
1309                 /* subtract from p_depth to get proper eh_depth */
1310                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1311                         put_bh(bh);
1312                         return -EIO;
1313                 }
1314                 ix = EXT_FIRST_INDEX(eh);
1315                 block = idx_pblock(ix);
1316                 put_bh(bh);
1317         }
1318
1319         bh = sb_bread(inode->i_sb, block);
1320         if (bh == NULL)
1321                 return -EIO;
1322         eh = ext_block_hdr(bh);
1323         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1324                 put_bh(bh);
1325                 return -EIO;
1326         }
1327         ex = EXT_FIRST_EXTENT(eh);
1328         *logical = le32_to_cpu(ex->ee_block);
1329         *phys = ext_pblock(ex);
1330         put_bh(bh);
1331         return 0;
1332 }
1333
1334 /*
1335  * ext4_ext_next_allocated_block:
1336  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1337  * NOTE: it considers block number from index entry as
1338  * allocated block. Thus, index entries have to be consistent
1339  * with leaves.
1340  */
1341 static ext4_lblk_t
1342 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1343 {
1344         int depth;
1345
1346         BUG_ON(path == NULL);
1347         depth = path->p_depth;
1348
1349         if (depth == 0 && path->p_ext == NULL)
1350                 return EXT_MAX_BLOCK;
1351
1352         while (depth >= 0) {
1353                 if (depth == path->p_depth) {
1354                         /* leaf */
1355                         if (path[depth].p_ext !=
1356                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1357                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1358                 } else {
1359                         /* index */
1360                         if (path[depth].p_idx !=
1361                                         EXT_LAST_INDEX(path[depth].p_hdr))
1362                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1363                 }
1364                 depth--;
1365         }
1366
1367         return EXT_MAX_BLOCK;
1368 }
1369
1370 /*
1371  * ext4_ext_next_leaf_block:
1372  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1373  */
1374 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1375                                         struct ext4_ext_path *path)
1376 {
1377         int depth;
1378
1379         BUG_ON(path == NULL);
1380         depth = path->p_depth;
1381
1382         /* zero-tree has no leaf blocks at all */
1383         if (depth == 0)
1384                 return EXT_MAX_BLOCK;
1385
1386         /* go to index block */
1387         depth--;
1388
1389         while (depth >= 0) {
1390                 if (path[depth].p_idx !=
1391                                 EXT_LAST_INDEX(path[depth].p_hdr))
1392                         return (ext4_lblk_t)
1393                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1394                 depth--;
1395         }
1396
1397         return EXT_MAX_BLOCK;
1398 }
1399
1400 /*
1401  * ext4_ext_correct_indexes:
1402  * if leaf gets modified and modified extent is first in the leaf,
1403  * then we have to correct all indexes above.
1404  * TODO: do we need to correct tree in all cases?
1405  */
1406 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1407                                 struct ext4_ext_path *path)
1408 {
1409         struct ext4_extent_header *eh;
1410         int depth = ext_depth(inode);
1411         struct ext4_extent *ex;
1412         __le32 border;
1413         int k, err = 0;
1414
1415         eh = path[depth].p_hdr;
1416         ex = path[depth].p_ext;
1417         BUG_ON(ex == NULL);
1418         BUG_ON(eh == NULL);
1419
1420         if (depth == 0) {
1421                 /* there is no tree at all */
1422                 return 0;
1423         }
1424
1425         if (ex != EXT_FIRST_EXTENT(eh)) {
1426                 /* we correct tree if first leaf got modified only */
1427                 return 0;
1428         }
1429
1430         /*
1431          * TODO: we need correction if border is smaller than current one
1432          */
1433         k = depth - 1;
1434         border = path[depth].p_ext->ee_block;
1435         err = ext4_ext_get_access(handle, inode, path + k);
1436         if (err)
1437                 return err;
1438         path[k].p_idx->ei_block = border;
1439         err = ext4_ext_dirty(handle, inode, path + k);
1440         if (err)
1441                 return err;
1442
1443         while (k--) {
1444                 /* change all left-side indexes */
1445                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1446                         break;
1447                 err = ext4_ext_get_access(handle, inode, path + k);
1448                 if (err)
1449                         break;
1450                 path[k].p_idx->ei_block = border;
1451                 err = ext4_ext_dirty(handle, inode, path + k);
1452                 if (err)
1453                         break;
1454         }
1455
1456         return err;
1457 }
1458
1459 int
1460 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1461                                 struct ext4_extent *ex2)
1462 {
1463         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1464
1465         /*
1466          * Make sure that either both extents are uninitialized, or
1467          * both are _not_.
1468          */
1469         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1470                 return 0;
1471
1472         if (ext4_ext_is_uninitialized(ex1))
1473                 max_len = EXT_UNINIT_MAX_LEN;
1474         else
1475                 max_len = EXT_INIT_MAX_LEN;
1476
1477         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1478         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1479
1480         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1481                         le32_to_cpu(ex2->ee_block))
1482                 return 0;
1483
1484         /*
1485          * To allow future support for preallocated extents to be added
1486          * as an RO_COMPAT feature, refuse to merge to extents if
1487          * this can result in the top bit of ee_len being set.
1488          */
1489         if (ext1_ee_len + ext2_ee_len > max_len)
1490                 return 0;
1491 #ifdef AGGRESSIVE_TEST
1492         if (ext1_ee_len >= 4)
1493                 return 0;
1494 #endif
1495
1496         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1497                 return 1;
1498         return 0;
1499 }
1500
1501 /*
1502  * This function tries to merge the "ex" extent to the next extent in the tree.
1503  * It always tries to merge towards right. If you want to merge towards
1504  * left, pass "ex - 1" as argument instead of "ex".
1505  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1506  * 1 if they got merged.
1507  */
1508 int ext4_ext_try_to_merge(struct inode *inode,
1509                           struct ext4_ext_path *path,
1510                           struct ext4_extent *ex)
1511 {
1512         struct ext4_extent_header *eh;
1513         unsigned int depth, len;
1514         int merge_done = 0;
1515         int uninitialized = 0;
1516
1517         depth = ext_depth(inode);
1518         BUG_ON(path[depth].p_hdr == NULL);
1519         eh = path[depth].p_hdr;
1520
1521         while (ex < EXT_LAST_EXTENT(eh)) {
1522                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1523                         break;
1524                 /* merge with next extent! */
1525                 if (ext4_ext_is_uninitialized(ex))
1526                         uninitialized = 1;
1527                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1528                                 + ext4_ext_get_actual_len(ex + 1));
1529                 if (uninitialized)
1530                         ext4_ext_mark_uninitialized(ex);
1531
1532                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1533                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1534                                 * sizeof(struct ext4_extent);
1535                         memmove(ex + 1, ex + 2, len);
1536                 }
1537                 le16_add_cpu(&eh->eh_entries, -1);
1538                 merge_done = 1;
1539                 WARN_ON(eh->eh_entries == 0);
1540                 if (!eh->eh_entries)
1541                         ext4_error(inode->i_sb,
1542                                    "inode#%lu, eh->eh_entries = 0!",
1543                                    inode->i_ino);
1544         }
1545
1546         return merge_done;
1547 }
1548
1549 /*
1550  * check if a portion of the "newext" extent overlaps with an
1551  * existing extent.
1552  *
1553  * If there is an overlap discovered, it updates the length of the newext
1554  * such that there will be no overlap, and then returns 1.
1555  * If there is no overlap found, it returns 0.
1556  */
1557 unsigned int ext4_ext_check_overlap(struct inode *inode,
1558                                     struct ext4_extent *newext,
1559                                     struct ext4_ext_path *path)
1560 {
1561         ext4_lblk_t b1, b2;
1562         unsigned int depth, len1;
1563         unsigned int ret = 0;
1564
1565         b1 = le32_to_cpu(newext->ee_block);
1566         len1 = ext4_ext_get_actual_len(newext);
1567         depth = ext_depth(inode);
1568         if (!path[depth].p_ext)
1569                 goto out;
1570         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1571
1572         /*
1573          * get the next allocated block if the extent in the path
1574          * is before the requested block(s)
1575          */
1576         if (b2 < b1) {
1577                 b2 = ext4_ext_next_allocated_block(path);
1578                 if (b2 == EXT_MAX_BLOCK)
1579                         goto out;
1580         }
1581
1582         /* check for wrap through zero on extent logical start block*/
1583         if (b1 + len1 < b1) {
1584                 len1 = EXT_MAX_BLOCK - b1;
1585                 newext->ee_len = cpu_to_le16(len1);
1586                 ret = 1;
1587         }
1588
1589         /* check for overlap */
1590         if (b1 + len1 > b2) {
1591                 newext->ee_len = cpu_to_le16(b2 - b1);
1592                 ret = 1;
1593         }
1594 out:
1595         return ret;
1596 }
1597
1598 /*
1599  * ext4_ext_insert_extent:
1600  * tries to merge requsted extent into the existing extent or
1601  * inserts requested extent as new one into the tree,
1602  * creating new leaf in the no-space case.
1603  */
1604 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1605                                 struct ext4_ext_path *path,
1606                                 struct ext4_extent *newext, int flag)
1607 {
1608         struct ext4_extent_header *eh;
1609         struct ext4_extent *ex, *fex;
1610         struct ext4_extent *nearex; /* nearest extent */
1611         struct ext4_ext_path *npath = NULL;
1612         int depth, len, err;
1613         ext4_lblk_t next;
1614         unsigned uninitialized = 0;
1615
1616         BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1617         depth = ext_depth(inode);
1618         ex = path[depth].p_ext;
1619         BUG_ON(path[depth].p_hdr == NULL);
1620
1621         /* try to insert block into found extent and return */
1622         if (ex && (flag != EXT4_GET_BLOCKS_DIO_CREATE_EXT)
1623                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1624                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1625                                 ext4_ext_is_uninitialized(newext),
1626                                 ext4_ext_get_actual_len(newext),
1627                                 le32_to_cpu(ex->ee_block),
1628                                 ext4_ext_is_uninitialized(ex),
1629                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1630                 err = ext4_ext_get_access(handle, inode, path + depth);
1631                 if (err)
1632                         return err;
1633
1634                 /*
1635                  * ext4_can_extents_be_merged should have checked that either
1636                  * both extents are uninitialized, or both aren't. Thus we
1637                  * need to check only one of them here.
1638                  */
1639                 if (ext4_ext_is_uninitialized(ex))
1640                         uninitialized = 1;
1641                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1642                                         + ext4_ext_get_actual_len(newext));
1643                 if (uninitialized)
1644                         ext4_ext_mark_uninitialized(ex);
1645                 eh = path[depth].p_hdr;
1646                 nearex = ex;
1647                 goto merge;
1648         }
1649
1650 repeat:
1651         depth = ext_depth(inode);
1652         eh = path[depth].p_hdr;
1653         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1654                 goto has_space;
1655
1656         /* probably next leaf has space for us? */
1657         fex = EXT_LAST_EXTENT(eh);
1658         next = ext4_ext_next_leaf_block(inode, path);
1659         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1660             && next != EXT_MAX_BLOCK) {
1661                 ext_debug("next leaf block - %d\n", next);
1662                 BUG_ON(npath != NULL);
1663                 npath = ext4_ext_find_extent(inode, next, NULL);
1664                 if (IS_ERR(npath))
1665                         return PTR_ERR(npath);
1666                 BUG_ON(npath->p_depth != path->p_depth);
1667                 eh = npath[depth].p_hdr;
1668                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1669                         ext_debug("next leaf isnt full(%d)\n",
1670                                   le16_to_cpu(eh->eh_entries));
1671                         path = npath;
1672                         goto repeat;
1673                 }
1674                 ext_debug("next leaf has no free space(%d,%d)\n",
1675                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1676         }
1677
1678         /*
1679          * There is no free space in the found leaf.
1680          * We're gonna add a new leaf in the tree.
1681          */
1682         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1683         if (err)
1684                 goto cleanup;
1685         depth = ext_depth(inode);
1686         eh = path[depth].p_hdr;
1687
1688 has_space:
1689         nearex = path[depth].p_ext;
1690
1691         err = ext4_ext_get_access(handle, inode, path + depth);
1692         if (err)
1693                 goto cleanup;
1694
1695         if (!nearex) {
1696                 /* there is no extent in this leaf, create first one */
1697                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1698                                 le32_to_cpu(newext->ee_block),
1699                                 ext_pblock(newext),
1700                                 ext4_ext_is_uninitialized(newext),
1701                                 ext4_ext_get_actual_len(newext));
1702                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1703         } else if (le32_to_cpu(newext->ee_block)
1704                            > le32_to_cpu(nearex->ee_block)) {
1705 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1706                 if (nearex != EXT_LAST_EXTENT(eh)) {
1707                         len = EXT_MAX_EXTENT(eh) - nearex;
1708                         len = (len - 1) * sizeof(struct ext4_extent);
1709                         len = len < 0 ? 0 : len;
1710                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1711                                         "move %d from 0x%p to 0x%p\n",
1712                                         le32_to_cpu(newext->ee_block),
1713                                         ext_pblock(newext),
1714                                         ext4_ext_is_uninitialized(newext),
1715                                         ext4_ext_get_actual_len(newext),
1716                                         nearex, len, nearex + 1, nearex + 2);
1717                         memmove(nearex + 2, nearex + 1, len);
1718                 }
1719                 path[depth].p_ext = nearex + 1;
1720         } else {
1721                 BUG_ON(newext->ee_block == nearex->ee_block);
1722                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1723                 len = len < 0 ? 0 : len;
1724                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1725                                 "move %d from 0x%p to 0x%p\n",
1726                                 le32_to_cpu(newext->ee_block),
1727                                 ext_pblock(newext),
1728                                 ext4_ext_is_uninitialized(newext),
1729                                 ext4_ext_get_actual_len(newext),
1730                                 nearex, len, nearex + 1, nearex + 2);
1731                 memmove(nearex + 1, nearex, len);
1732                 path[depth].p_ext = nearex;
1733         }
1734
1735         le16_add_cpu(&eh->eh_entries, 1);
1736         nearex = path[depth].p_ext;
1737         nearex->ee_block = newext->ee_block;
1738         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1739         nearex->ee_len = newext->ee_len;
1740
1741 merge:
1742         /* try to merge extents to the right */
1743         if (flag != EXT4_GET_BLOCKS_DIO_CREATE_EXT)
1744                 ext4_ext_try_to_merge(inode, path, nearex);
1745
1746         /* try to merge extents to the left */
1747
1748         /* time to correct all indexes above */
1749         err = ext4_ext_correct_indexes(handle, inode, path);
1750         if (err)
1751                 goto cleanup;
1752
1753         err = ext4_ext_dirty(handle, inode, path + depth);
1754
1755 cleanup:
1756         if (npath) {
1757                 ext4_ext_drop_refs(npath);
1758                 kfree(npath);
1759         }
1760         ext4_ext_invalidate_cache(inode);
1761         return err;
1762 }
1763
1764 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1765                         ext4_lblk_t num, ext_prepare_callback func,
1766                         void *cbdata)
1767 {
1768         struct ext4_ext_path *path = NULL;
1769         struct ext4_ext_cache cbex;
1770         struct ext4_extent *ex;
1771         ext4_lblk_t next, start = 0, end = 0;
1772         ext4_lblk_t last = block + num;
1773         int depth, exists, err = 0;
1774
1775         BUG_ON(func == NULL);
1776         BUG_ON(inode == NULL);
1777
1778         while (block < last && block != EXT_MAX_BLOCK) {
1779                 num = last - block;
1780                 /* find extent for this block */
1781                 down_read(&EXT4_I(inode)->i_data_sem);
1782                 path = ext4_ext_find_extent(inode, block, path);
1783                 up_read(&EXT4_I(inode)->i_data_sem);
1784                 if (IS_ERR(path)) {
1785                         err = PTR_ERR(path);
1786                         path = NULL;
1787                         break;
1788                 }
1789
1790                 depth = ext_depth(inode);
1791                 BUG_ON(path[depth].p_hdr == NULL);
1792                 ex = path[depth].p_ext;
1793                 next = ext4_ext_next_allocated_block(path);
1794
1795                 exists = 0;
1796                 if (!ex) {
1797                         /* there is no extent yet, so try to allocate
1798                          * all requested space */
1799                         start = block;
1800                         end = block + num;
1801                 } else if (le32_to_cpu(ex->ee_block) > block) {
1802                         /* need to allocate space before found extent */
1803                         start = block;
1804                         end = le32_to_cpu(ex->ee_block);
1805                         if (block + num < end)
1806                                 end = block + num;
1807                 } else if (block >= le32_to_cpu(ex->ee_block)
1808                                         + ext4_ext_get_actual_len(ex)) {
1809                         /* need to allocate space after found extent */
1810                         start = block;
1811                         end = block + num;
1812                         if (end >= next)
1813                                 end = next;
1814                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1815                         /*
1816                          * some part of requested space is covered
1817                          * by found extent
1818                          */
1819                         start = block;
1820                         end = le32_to_cpu(ex->ee_block)
1821                                 + ext4_ext_get_actual_len(ex);
1822                         if (block + num < end)
1823                                 end = block + num;
1824                         exists = 1;
1825                 } else {
1826                         BUG();
1827                 }
1828                 BUG_ON(end <= start);
1829
1830                 if (!exists) {
1831                         cbex.ec_block = start;
1832                         cbex.ec_len = end - start;
1833                         cbex.ec_start = 0;
1834                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1835                 } else {
1836                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1837                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1838                         cbex.ec_start = ext_pblock(ex);
1839                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1840                 }
1841
1842                 BUG_ON(cbex.ec_len == 0);
1843                 err = func(inode, path, &cbex, ex, cbdata);
1844                 ext4_ext_drop_refs(path);
1845
1846                 if (err < 0)
1847                         break;
1848
1849                 if (err == EXT_REPEAT)
1850                         continue;
1851                 else if (err == EXT_BREAK) {
1852                         err = 0;
1853                         break;
1854                 }
1855
1856                 if (ext_depth(inode) != depth) {
1857                         /* depth was changed. we have to realloc path */
1858                         kfree(path);
1859                         path = NULL;
1860                 }
1861
1862                 block = cbex.ec_block + cbex.ec_len;
1863         }
1864
1865         if (path) {
1866                 ext4_ext_drop_refs(path);
1867                 kfree(path);
1868         }
1869
1870         return err;
1871 }
1872
1873 static void
1874 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1875                         __u32 len, ext4_fsblk_t start, int type)
1876 {
1877         struct ext4_ext_cache *cex;
1878         BUG_ON(len == 0);
1879         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1880         cex = &EXT4_I(inode)->i_cached_extent;
1881         cex->ec_type = type;
1882         cex->ec_block = block;
1883         cex->ec_len = len;
1884         cex->ec_start = start;
1885         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1886 }
1887
1888 /*
1889  * ext4_ext_put_gap_in_cache:
1890  * calculate boundaries of the gap that the requested block fits into
1891  * and cache this gap
1892  */
1893 static void
1894 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1895                                 ext4_lblk_t block)
1896 {
1897         int depth = ext_depth(inode);
1898         unsigned long len;
1899         ext4_lblk_t lblock;
1900         struct ext4_extent *ex;
1901
1902         ex = path[depth].p_ext;
1903         if (ex == NULL) {
1904                 /* there is no extent yet, so gap is [0;-] */
1905                 lblock = 0;
1906                 len = EXT_MAX_BLOCK;
1907                 ext_debug("cache gap(whole file):");
1908         } else if (block < le32_to_cpu(ex->ee_block)) {
1909                 lblock = block;
1910                 len = le32_to_cpu(ex->ee_block) - block;
1911                 ext_debug("cache gap(before): %u [%u:%u]",
1912                                 block,
1913                                 le32_to_cpu(ex->ee_block),
1914                                  ext4_ext_get_actual_len(ex));
1915         } else if (block >= le32_to_cpu(ex->ee_block)
1916                         + ext4_ext_get_actual_len(ex)) {
1917                 ext4_lblk_t next;
1918                 lblock = le32_to_cpu(ex->ee_block)
1919                         + ext4_ext_get_actual_len(ex);
1920
1921                 next = ext4_ext_next_allocated_block(path);
1922                 ext_debug("cache gap(after): [%u:%u] %u",
1923                                 le32_to_cpu(ex->ee_block),
1924                                 ext4_ext_get_actual_len(ex),
1925                                 block);
1926                 BUG_ON(next == lblock);
1927                 len = next - lblock;
1928         } else {
1929                 lblock = len = 0;
1930                 BUG();
1931         }
1932
1933         ext_debug(" -> %u:%lu\n", lblock, len);
1934         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1935 }
1936
1937 static int
1938 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1939                         struct ext4_extent *ex)
1940 {
1941         struct ext4_ext_cache *cex;
1942         int ret = EXT4_EXT_CACHE_NO;
1943
1944         /* 
1945          * We borrow i_block_reservation_lock to protect i_cached_extent
1946          */
1947         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1948         cex = &EXT4_I(inode)->i_cached_extent;
1949
1950         /* has cache valid data? */
1951         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1952                 goto errout;
1953
1954         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1955                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1956         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1957                 ex->ee_block = cpu_to_le32(cex->ec_block);
1958                 ext4_ext_store_pblock(ex, cex->ec_start);
1959                 ex->ee_len = cpu_to_le16(cex->ec_len);
1960                 ext_debug("%u cached by %u:%u:%llu\n",
1961                                 block,
1962                                 cex->ec_block, cex->ec_len, cex->ec_start);
1963                 ret = cex->ec_type;
1964         }
1965 errout:
1966         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1967         return ret;
1968 }
1969
1970 /*
1971  * ext4_ext_rm_idx:
1972  * removes index from the index block.
1973  * It's used in truncate case only, thus all requests are for
1974  * last index in the block only.
1975  */
1976 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1977                         struct ext4_ext_path *path)
1978 {
1979         int err;
1980         ext4_fsblk_t leaf;
1981
1982         /* free index block */
1983         path--;
1984         leaf = idx_pblock(path->p_idx);
1985         BUG_ON(path->p_hdr->eh_entries == 0);
1986         err = ext4_ext_get_access(handle, inode, path);
1987         if (err)
1988                 return err;
1989         le16_add_cpu(&path->p_hdr->eh_entries, -1);
1990         err = ext4_ext_dirty(handle, inode, path);
1991         if (err)
1992                 return err;
1993         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1994         ext4_free_blocks(handle, inode, 0, leaf, 1,
1995                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1996         return err;
1997 }
1998
1999 /*
2000  * ext4_ext_calc_credits_for_single_extent:
2001  * This routine returns max. credits that needed to insert an extent
2002  * to the extent tree.
2003  * When pass the actual path, the caller should calculate credits
2004  * under i_data_sem.
2005  */
2006 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2007                                                 struct ext4_ext_path *path)
2008 {
2009         if (path) {
2010                 int depth = ext_depth(inode);
2011                 int ret = 0;
2012
2013                 /* probably there is space in leaf? */
2014                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2015                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2016
2017                         /*
2018                          *  There are some space in the leaf tree, no
2019                          *  need to account for leaf block credit
2020                          *
2021                          *  bitmaps and block group descriptor blocks
2022                          *  and other metadat blocks still need to be
2023                          *  accounted.
2024                          */
2025                         /* 1 bitmap, 1 block group descriptor */
2026                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2027                         return ret;
2028                 }
2029         }
2030
2031         return ext4_chunk_trans_blocks(inode, nrblocks);
2032 }
2033
2034 /*
2035  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2036  *
2037  * if nrblocks are fit in a single extent (chunk flag is 1), then
2038  * in the worse case, each tree level index/leaf need to be changed
2039  * if the tree split due to insert a new extent, then the old tree
2040  * index/leaf need to be updated too
2041  *
2042  * If the nrblocks are discontiguous, they could cause
2043  * the whole tree split more than once, but this is really rare.
2044  */
2045 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2046 {
2047         int index;
2048         int depth = ext_depth(inode);
2049
2050         if (chunk)
2051                 index = depth * 2;
2052         else
2053                 index = depth * 3;
2054
2055         return index;
2056 }
2057
2058 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2059                                 struct ext4_extent *ex,
2060                                 ext4_lblk_t from, ext4_lblk_t to)
2061 {
2062         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2063         int flags = EXT4_FREE_BLOCKS_FORGET;
2064
2065         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2066                 flags |= EXT4_FREE_BLOCKS_METADATA;
2067 #ifdef EXTENTS_STATS
2068         {
2069                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2070                 spin_lock(&sbi->s_ext_stats_lock);
2071                 sbi->s_ext_blocks += ee_len;
2072                 sbi->s_ext_extents++;
2073                 if (ee_len < sbi->s_ext_min)
2074                         sbi->s_ext_min = ee_len;
2075                 if (ee_len > sbi->s_ext_max)
2076                         sbi->s_ext_max = ee_len;
2077                 if (ext_depth(inode) > sbi->s_depth_max)
2078                         sbi->s_depth_max = ext_depth(inode);
2079                 spin_unlock(&sbi->s_ext_stats_lock);
2080         }
2081 #endif
2082         if (from >= le32_to_cpu(ex->ee_block)
2083             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2084                 /* tail removal */
2085                 ext4_lblk_t num;
2086                 ext4_fsblk_t start;
2087
2088                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2089                 start = ext_pblock(ex) + ee_len - num;
2090                 ext_debug("free last %u blocks starting %llu\n", num, start);
2091                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2092         } else if (from == le32_to_cpu(ex->ee_block)
2093                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2094                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2095                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2096         } else {
2097                 printk(KERN_INFO "strange request: removal(2) "
2098                                 "%u-%u from %u:%u\n",
2099                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2100         }
2101         return 0;
2102 }
2103
2104 static int
2105 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2106                 struct ext4_ext_path *path, ext4_lblk_t start)
2107 {
2108         int err = 0, correct_index = 0;
2109         int depth = ext_depth(inode), credits;
2110         struct ext4_extent_header *eh;
2111         ext4_lblk_t a, b, block;
2112         unsigned num;
2113         ext4_lblk_t ex_ee_block;
2114         unsigned short ex_ee_len;
2115         unsigned uninitialized = 0;
2116         struct ext4_extent *ex;
2117
2118         /* the header must be checked already in ext4_ext_remove_space() */
2119         ext_debug("truncate since %u in leaf\n", start);
2120         if (!path[depth].p_hdr)
2121                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2122         eh = path[depth].p_hdr;
2123         BUG_ON(eh == NULL);
2124
2125         /* find where to start removing */
2126         ex = EXT_LAST_EXTENT(eh);
2127
2128         ex_ee_block = le32_to_cpu(ex->ee_block);
2129         ex_ee_len = ext4_ext_get_actual_len(ex);
2130
2131         while (ex >= EXT_FIRST_EXTENT(eh) &&
2132                         ex_ee_block + ex_ee_len > start) {
2133
2134                 if (ext4_ext_is_uninitialized(ex))
2135                         uninitialized = 1;
2136                 else
2137                         uninitialized = 0;
2138
2139                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2140                          uninitialized, ex_ee_len);
2141                 path[depth].p_ext = ex;
2142
2143                 a = ex_ee_block > start ? ex_ee_block : start;
2144                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2145                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2146
2147                 ext_debug("  border %u:%u\n", a, b);
2148
2149                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2150                         block = 0;
2151                         num = 0;
2152                         BUG();
2153                 } else if (a != ex_ee_block) {
2154                         /* remove tail of the extent */
2155                         block = ex_ee_block;
2156                         num = a - block;
2157                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2158                         /* remove head of the extent */
2159                         block = a;
2160                         num = b - a;
2161                         /* there is no "make a hole" API yet */
2162                         BUG();
2163                 } else {
2164                         /* remove whole extent: excellent! */
2165                         block = ex_ee_block;
2166                         num = 0;
2167                         BUG_ON(a != ex_ee_block);
2168                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2169                 }
2170
2171                 /*
2172                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2173                  * descriptor) for each block group; assume two block
2174                  * groups plus ex_ee_len/blocks_per_block_group for
2175                  * the worst case
2176                  */
2177                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2178                 if (ex == EXT_FIRST_EXTENT(eh)) {
2179                         correct_index = 1;
2180                         credits += (ext_depth(inode)) + 1;
2181                 }
2182                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2183
2184                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2185                 if (err)
2186                         goto out;
2187
2188                 err = ext4_ext_get_access(handle, inode, path + depth);
2189                 if (err)
2190                         goto out;
2191
2192                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2193                 if (err)
2194                         goto out;
2195
2196                 if (num == 0) {
2197                         /* this extent is removed; mark slot entirely unused */
2198                         ext4_ext_store_pblock(ex, 0);
2199                         le16_add_cpu(&eh->eh_entries, -1);
2200                 }
2201
2202                 ex->ee_block = cpu_to_le32(block);
2203                 ex->ee_len = cpu_to_le16(num);
2204                 /*
2205                  * Do not mark uninitialized if all the blocks in the
2206                  * extent have been removed.
2207                  */
2208                 if (uninitialized && num)
2209                         ext4_ext_mark_uninitialized(ex);
2210
2211                 err = ext4_ext_dirty(handle, inode, path + depth);
2212                 if (err)
2213                         goto out;
2214
2215                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2216                                 ext_pblock(ex));
2217                 ex--;
2218                 ex_ee_block = le32_to_cpu(ex->ee_block);
2219                 ex_ee_len = ext4_ext_get_actual_len(ex);
2220         }
2221
2222         if (correct_index && eh->eh_entries)
2223                 err = ext4_ext_correct_indexes(handle, inode, path);
2224
2225         /* if this leaf is free, then we should
2226          * remove it from index block above */
2227         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2228                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2229
2230 out:
2231         return err;
2232 }
2233
2234 /*
2235  * ext4_ext_more_to_rm:
2236  * returns 1 if current index has to be freed (even partial)
2237  */
2238 static int
2239 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2240 {
2241         BUG_ON(path->p_idx == NULL);
2242
2243         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2244                 return 0;
2245
2246         /*
2247          * if truncate on deeper level happened, it wasn't partial,
2248          * so we have to consider current index for truncation
2249          */
2250         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2251                 return 0;
2252         return 1;
2253 }
2254
2255 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2256 {
2257         struct super_block *sb = inode->i_sb;
2258         int depth = ext_depth(inode);
2259         struct ext4_ext_path *path;
2260         handle_t *handle;
2261         int i = 0, err = 0;
2262
2263         ext_debug("truncate since %u\n", start);
2264
2265         /* probably first extent we're gonna free will be last in block */
2266         handle = ext4_journal_start(inode, depth + 1);
2267         if (IS_ERR(handle))
2268                 return PTR_ERR(handle);
2269
2270         ext4_ext_invalidate_cache(inode);
2271
2272         /*
2273          * We start scanning from right side, freeing all the blocks
2274          * after i_size and walking into the tree depth-wise.
2275          */
2276         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2277         if (path == NULL) {
2278                 ext4_journal_stop(handle);
2279                 return -ENOMEM;
2280         }
2281         path[0].p_hdr = ext_inode_hdr(inode);
2282         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2283                 err = -EIO;
2284                 goto out;
2285         }
2286         path[0].p_depth = depth;
2287
2288         while (i >= 0 && err == 0) {
2289                 if (i == depth) {
2290                         /* this is leaf block */
2291                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2292                         /* root level has p_bh == NULL, brelse() eats this */
2293                         brelse(path[i].p_bh);
2294                         path[i].p_bh = NULL;
2295                         i--;
2296                         continue;
2297                 }
2298
2299                 /* this is index block */
2300                 if (!path[i].p_hdr) {
2301                         ext_debug("initialize header\n");
2302                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2303                 }
2304
2305                 if (!path[i].p_idx) {
2306                         /* this level hasn't been touched yet */
2307                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2308                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2309                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2310                                   path[i].p_hdr,
2311                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2312                 } else {
2313                         /* we were already here, see at next index */
2314                         path[i].p_idx--;
2315                 }
2316
2317                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2318                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2319                                 path[i].p_idx);
2320                 if (ext4_ext_more_to_rm(path + i)) {
2321                         struct buffer_head *bh;
2322                         /* go to the next level */
2323                         ext_debug("move to level %d (block %llu)\n",
2324                                   i + 1, idx_pblock(path[i].p_idx));
2325                         memset(path + i + 1, 0, sizeof(*path));
2326                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2327                         if (!bh) {
2328                                 /* should we reset i_size? */
2329                                 err = -EIO;
2330                                 break;
2331                         }
2332                         if (WARN_ON(i + 1 > depth)) {
2333                                 err = -EIO;
2334                                 break;
2335                         }
2336                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2337                                                         depth - i - 1)) {
2338                                 err = -EIO;
2339                                 break;
2340                         }
2341                         path[i + 1].p_bh = bh;
2342
2343                         /* save actual number of indexes since this
2344                          * number is changed at the next iteration */
2345                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2346                         i++;
2347                 } else {
2348                         /* we finished processing this index, go up */
2349                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2350                                 /* index is empty, remove it;
2351                                  * handle must be already prepared by the
2352                                  * truncatei_leaf() */
2353                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2354                         }
2355                         /* root level has p_bh == NULL, brelse() eats this */
2356                         brelse(path[i].p_bh);
2357                         path[i].p_bh = NULL;
2358                         i--;
2359                         ext_debug("return to level %d\n", i);
2360                 }
2361         }
2362
2363         /* TODO: flexible tree reduction should be here */
2364         if (path->p_hdr->eh_entries == 0) {
2365                 /*
2366                  * truncate to zero freed all the tree,
2367                  * so we need to correct eh_depth
2368                  */
2369                 err = ext4_ext_get_access(handle, inode, path);
2370                 if (err == 0) {
2371                         ext_inode_hdr(inode)->eh_depth = 0;
2372                         ext_inode_hdr(inode)->eh_max =
2373                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2374                         err = ext4_ext_dirty(handle, inode, path);
2375                 }
2376         }
2377 out:
2378         ext4_ext_drop_refs(path);
2379         kfree(path);
2380         ext4_journal_stop(handle);
2381
2382         return err;
2383 }
2384
2385 /*
2386  * called at mount time
2387  */
2388 void ext4_ext_init(struct super_block *sb)
2389 {
2390         /*
2391          * possible initialization would be here
2392          */
2393
2394         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2395 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2396                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2397 #ifdef AGGRESSIVE_TEST
2398                 printk(", aggressive tests");
2399 #endif
2400 #ifdef CHECK_BINSEARCH
2401                 printk(", check binsearch");
2402 #endif
2403 #ifdef EXTENTS_STATS
2404                 printk(", stats");
2405 #endif
2406                 printk("\n");
2407 #endif
2408 #ifdef EXTENTS_STATS
2409                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2410                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2411                 EXT4_SB(sb)->s_ext_max = 0;
2412 #endif
2413         }
2414 }
2415
2416 /*
2417  * called at umount time
2418  */
2419 void ext4_ext_release(struct super_block *sb)
2420 {
2421         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2422                 return;
2423
2424 #ifdef EXTENTS_STATS
2425         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2426                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2427                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2428                         sbi->s_ext_blocks, sbi->s_ext_extents,
2429                         sbi->s_ext_blocks / sbi->s_ext_extents);
2430                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2431                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2432         }
2433 #endif
2434 }
2435
2436 static void bi_complete(struct bio *bio, int error)
2437 {
2438         complete((struct completion *)bio->bi_private);
2439 }
2440
2441 /* FIXME!! we need to try to merge to left or right after zero-out  */
2442 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2443 {
2444         int ret = -EIO;
2445         struct bio *bio;
2446         int blkbits, blocksize;
2447         sector_t ee_pblock;
2448         struct completion event;
2449         unsigned int ee_len, len, done, offset;
2450
2451
2452         blkbits   = inode->i_blkbits;
2453         blocksize = inode->i_sb->s_blocksize;
2454         ee_len    = ext4_ext_get_actual_len(ex);
2455         ee_pblock = ext_pblock(ex);
2456
2457         /* convert ee_pblock to 512 byte sectors */
2458         ee_pblock = ee_pblock << (blkbits - 9);
2459
2460         while (ee_len > 0) {
2461
2462                 if (ee_len > BIO_MAX_PAGES)
2463                         len = BIO_MAX_PAGES;
2464                 else
2465                         len = ee_len;
2466
2467                 bio = bio_alloc(GFP_NOIO, len);
2468                 bio->bi_sector = ee_pblock;
2469                 bio->bi_bdev   = inode->i_sb->s_bdev;
2470
2471                 done = 0;
2472                 offset = 0;
2473                 while (done < len) {
2474                         ret = bio_add_page(bio, ZERO_PAGE(0),
2475                                                         blocksize, offset);
2476                         if (ret != blocksize) {
2477                                 /*
2478                                  * We can't add any more pages because of
2479                                  * hardware limitations.  Start a new bio.
2480                                  */
2481                                 break;
2482                         }
2483                         done++;
2484                         offset += blocksize;
2485                         if (offset >= PAGE_CACHE_SIZE)
2486                                 offset = 0;
2487                 }
2488
2489                 init_completion(&event);
2490                 bio->bi_private = &event;
2491                 bio->bi_end_io = bi_complete;
2492                 submit_bio(WRITE, bio);
2493                 wait_for_completion(&event);
2494
2495                 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2496                         ret = 0;
2497                 else {
2498                         ret = -EIO;
2499                         break;
2500                 }
2501                 bio_put(bio);
2502                 ee_len    -= done;
2503                 ee_pblock += done  << (blkbits - 9);
2504         }
2505         return ret;
2506 }
2507
2508 #define EXT4_EXT_ZERO_LEN 7
2509 /*
2510  * This function is called by ext4_ext_get_blocks() if someone tries to write
2511  * to an uninitialized extent. It may result in splitting the uninitialized
2512  * extent into multiple extents (upto three - one initialized and two
2513  * uninitialized).
2514  * There are three possibilities:
2515  *   a> There is no split required: Entire extent should be initialized
2516  *   b> Splits in two extents: Write is happening at either end of the extent
2517  *   c> Splits in three extents: Somone is writing in middle of the extent
2518  */
2519 static int ext4_ext_convert_to_initialized(handle_t *handle,
2520                                                 struct inode *inode,
2521                                                 struct ext4_ext_path *path,
2522                                                 ext4_lblk_t iblock,
2523                                                 unsigned int max_blocks)
2524 {
2525         struct ext4_extent *ex, newex, orig_ex;
2526         struct ext4_extent *ex1 = NULL;
2527         struct ext4_extent *ex2 = NULL;
2528         struct ext4_extent *ex3 = NULL;
2529         struct ext4_extent_header *eh;
2530         ext4_lblk_t ee_block;
2531         unsigned int allocated, ee_len, depth;
2532         ext4_fsblk_t newblock;
2533         int err = 0;
2534         int ret = 0;
2535
2536         depth = ext_depth(inode);
2537         eh = path[depth].p_hdr;
2538         ex = path[depth].p_ext;
2539         ee_block = le32_to_cpu(ex->ee_block);
2540         ee_len = ext4_ext_get_actual_len(ex);
2541         allocated = ee_len - (iblock - ee_block);
2542         newblock = iblock - ee_block + ext_pblock(ex);
2543         ex2 = ex;
2544         orig_ex.ee_block = ex->ee_block;
2545         orig_ex.ee_len   = cpu_to_le16(ee_len);
2546         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2547
2548         err = ext4_ext_get_access(handle, inode, path + depth);
2549         if (err)
2550                 goto out;
2551         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2552         if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2553                 err =  ext4_ext_zeroout(inode, &orig_ex);
2554                 if (err)
2555                         goto fix_extent_len;
2556                 /* update the extent length and mark as initialized */
2557                 ex->ee_block = orig_ex.ee_block;
2558                 ex->ee_len   = orig_ex.ee_len;
2559                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2560                 ext4_ext_dirty(handle, inode, path + depth);
2561                 /* zeroed the full extent */
2562                 return allocated;
2563         }
2564
2565         /* ex1: ee_block to iblock - 1 : uninitialized */
2566         if (iblock > ee_block) {
2567                 ex1 = ex;
2568                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2569                 ext4_ext_mark_uninitialized(ex1);
2570                 ex2 = &newex;
2571         }
2572         /*
2573          * for sanity, update the length of the ex2 extent before
2574          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2575          * overlap of blocks.
2576          */
2577         if (!ex1 && allocated > max_blocks)
2578                 ex2->ee_len = cpu_to_le16(max_blocks);
2579         /* ex3: to ee_block + ee_len : uninitialised */
2580         if (allocated > max_blocks) {
2581                 unsigned int newdepth;
2582                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2583                 if (allocated <= EXT4_EXT_ZERO_LEN) {
2584                         /*
2585                          * iblock == ee_block is handled by the zerouout
2586                          * at the beginning.
2587                          * Mark first half uninitialized.
2588                          * Mark second half initialized and zero out the
2589                          * initialized extent
2590                          */
2591                         ex->ee_block = orig_ex.ee_block;
2592                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2593                         ext4_ext_mark_uninitialized(ex);
2594                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2595                         ext4_ext_dirty(handle, inode, path + depth);
2596
2597                         ex3 = &newex;
2598                         ex3->ee_block = cpu_to_le32(iblock);
2599                         ext4_ext_store_pblock(ex3, newblock);
2600                         ex3->ee_len = cpu_to_le16(allocated);
2601                         err = ext4_ext_insert_extent(handle, inode, path,
2602                                                         ex3, 0);
2603                         if (err == -ENOSPC) {
2604                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2605                                 if (err)
2606                                         goto fix_extent_len;
2607                                 ex->ee_block = orig_ex.ee_block;
2608                                 ex->ee_len   = orig_ex.ee_len;
2609                                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2610                                 ext4_ext_dirty(handle, inode, path + depth);
2611                                 /* blocks available from iblock */
2612                                 return allocated;
2613
2614                         } else if (err)
2615                                 goto fix_extent_len;
2616
2617                         /*
2618                          * We need to zero out the second half because
2619                          * an fallocate request can update file size and
2620                          * converting the second half to initialized extent
2621                          * implies that we can leak some junk data to user
2622                          * space.
2623                          */
2624                         err =  ext4_ext_zeroout(inode, ex3);
2625                         if (err) {
2626                                 /*
2627                                  * We should actually mark the
2628                                  * second half as uninit and return error
2629                                  * Insert would have changed the extent
2630                                  */
2631                                 depth = ext_depth(inode);
2632                                 ext4_ext_drop_refs(path);
2633                                 path = ext4_ext_find_extent(inode,
2634                                                                 iblock, path);
2635                                 if (IS_ERR(path)) {
2636                                         err = PTR_ERR(path);
2637                                         return err;
2638                                 }
2639                                 /* get the second half extent details */
2640                                 ex = path[depth].p_ext;
2641                                 err = ext4_ext_get_access(handle, inode,
2642                                                                 path + depth);
2643                                 if (err)
2644                                         return err;
2645                                 ext4_ext_mark_uninitialized(ex);
2646                                 ext4_ext_dirty(handle, inode, path + depth);
2647                                 return err;
2648                         }
2649
2650                         /* zeroed the second half */
2651                         return allocated;
2652                 }
2653                 ex3 = &newex;
2654                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2655                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2656                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2657                 ext4_ext_mark_uninitialized(ex3);
2658                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2659                 if (err == -ENOSPC) {
2660                         err =  ext4_ext_zeroout(inode, &orig_ex);
2661                         if (err)
2662                                 goto fix_extent_len;
2663                         /* update the extent length and mark as initialized */
2664                         ex->ee_block = orig_ex.ee_block;
2665                         ex->ee_len   = orig_ex.ee_len;
2666                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2667                         ext4_ext_dirty(handle, inode, path + depth);
2668                         /* zeroed the full extent */
2669                         /* blocks available from iblock */
2670                         return allocated;
2671
2672                 } else if (err)
2673                         goto fix_extent_len;
2674                 /*
2675                  * The depth, and hence eh & ex might change
2676                  * as part of the insert above.
2677                  */
2678                 newdepth = ext_depth(inode);
2679                 /*
2680                  * update the extent length after successful insert of the
2681                  * split extent
2682                  */
2683                 orig_ex.ee_len = cpu_to_le16(ee_len -
2684                                                 ext4_ext_get_actual_len(ex3));
2685                 depth = newdepth;
2686                 ext4_ext_drop_refs(path);
2687                 path = ext4_ext_find_extent(inode, iblock, path);
2688                 if (IS_ERR(path)) {
2689                         err = PTR_ERR(path);
2690                         goto out;
2691                 }
2692                 eh = path[depth].p_hdr;
2693                 ex = path[depth].p_ext;
2694                 if (ex2 != &newex)
2695                         ex2 = ex;
2696
2697                 err = ext4_ext_get_access(handle, inode, path + depth);
2698                 if (err)
2699                         goto out;
2700
2701                 allocated = max_blocks;
2702
2703                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2704                  * to insert a extent in the middle zerout directly
2705                  * otherwise give the extent a chance to merge to left
2706                  */
2707                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2708                                                         iblock != ee_block) {
2709                         err =  ext4_ext_zeroout(inode, &orig_ex);
2710                         if (err)
2711                                 goto fix_extent_len;
2712                         /* update the extent length and mark as initialized */
2713                         ex->ee_block = orig_ex.ee_block;
2714                         ex->ee_len   = orig_ex.ee_len;
2715                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2716                         ext4_ext_dirty(handle, inode, path + depth);
2717                         /* zero out the first half */
2718                         /* blocks available from iblock */
2719                         return allocated;
2720                 }
2721         }
2722         /*
2723          * If there was a change of depth as part of the
2724          * insertion of ex3 above, we need to update the length
2725          * of the ex1 extent again here
2726          */
2727         if (ex1 && ex1 != ex) {
2728                 ex1 = ex;
2729                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2730                 ext4_ext_mark_uninitialized(ex1);
2731                 ex2 = &newex;
2732         }
2733         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2734         ex2->ee_block = cpu_to_le32(iblock);
2735         ext4_ext_store_pblock(ex2, newblock);
2736         ex2->ee_len = cpu_to_le16(allocated);
2737         if (ex2 != ex)
2738                 goto insert;
2739         /*
2740          * New (initialized) extent starts from the first block
2741          * in the current extent. i.e., ex2 == ex
2742          * We have to see if it can be merged with the extent
2743          * on the left.
2744          */
2745         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2746                 /*
2747                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2748                  * since it merges towards right _only_.
2749                  */
2750                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2751                 if (ret) {
2752                         err = ext4_ext_correct_indexes(handle, inode, path);
2753                         if (err)
2754                                 goto out;
2755                         depth = ext_depth(inode);
2756                         ex2--;
2757                 }
2758         }
2759         /*
2760          * Try to Merge towards right. This might be required
2761          * only when the whole extent is being written to.
2762          * i.e. ex2 == ex and ex3 == NULL.
2763          */
2764         if (!ex3) {
2765                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2766                 if (ret) {
2767                         err = ext4_ext_correct_indexes(handle, inode, path);
2768                         if (err)
2769                                 goto out;
2770                 }
2771         }
2772         /* Mark modified extent as dirty */
2773         err = ext4_ext_dirty(handle, inode, path + depth);
2774         goto out;
2775 insert:
2776         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2777         if (err == -ENOSPC) {
2778                 err =  ext4_ext_zeroout(inode, &orig_ex);
2779                 if (err)
2780                         goto fix_extent_len;
2781                 /* update the extent length and mark as initialized */
2782                 ex->ee_block = orig_ex.ee_block;
2783                 ex->ee_len   = orig_ex.ee_len;
2784                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2785                 ext4_ext_dirty(handle, inode, path + depth);
2786                 /* zero out the first half */
2787                 return allocated;
2788         } else if (err)
2789                 goto fix_extent_len;
2790 out:
2791         ext4_ext_show_leaf(inode, path);
2792         return err ? err : allocated;
2793
2794 fix_extent_len:
2795         ex->ee_block = orig_ex.ee_block;
2796         ex->ee_len   = orig_ex.ee_len;
2797         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2798         ext4_ext_mark_uninitialized(ex);
2799         ext4_ext_dirty(handle, inode, path + depth);
2800         return err;
2801 }
2802
2803 /*
2804  * This function is called by ext4_ext_get_blocks() from
2805  * ext4_get_blocks_dio_write() when DIO to write
2806  * to an uninitialized extent.
2807  *
2808  * Writing to an uninitized extent may result in splitting the uninitialized
2809  * extent into multiple /intialized unintialized extents (up to three)
2810  * There are three possibilities:
2811  *   a> There is no split required: Entire extent should be uninitialized
2812  *   b> Splits in two extents: Write is happening at either end of the extent
2813  *   c> Splits in three extents: Somone is writing in middle of the extent
2814  *
2815  * One of more index blocks maybe needed if the extent tree grow after
2816  * the unintialized extent split. To prevent ENOSPC occur at the IO
2817  * complete, we need to split the uninitialized extent before DIO submit
2818  * the IO. The uninitilized extent called at this time will be split
2819  * into three uninitialized extent(at most). After IO complete, the part
2820  * being filled will be convert to initialized by the end_io callback function
2821  * via ext4_convert_unwritten_extents().
2822  *
2823  * Returns the size of uninitialized extent to be written on success.
2824  */
2825 static int ext4_split_unwritten_extents(handle_t *handle,
2826                                         struct inode *inode,
2827                                         struct ext4_ext_path *path,
2828                                         ext4_lblk_t iblock,
2829                                         unsigned int max_blocks,
2830                                         int flags)
2831 {
2832         struct ext4_extent *ex, newex, orig_ex;
2833         struct ext4_extent *ex1 = NULL;
2834         struct ext4_extent *ex2 = NULL;
2835         struct ext4_extent *ex3 = NULL;
2836         struct ext4_extent_header *eh;
2837         ext4_lblk_t ee_block;
2838         unsigned int allocated, ee_len, depth;
2839         ext4_fsblk_t newblock;
2840         int err = 0;
2841
2842         ext_debug("ext4_split_unwritten_extents: inode %lu,"
2843                   "iblock %llu, max_blocks %u\n", inode->i_ino,
2844                   (unsigned long long)iblock, max_blocks);
2845         depth = ext_depth(inode);
2846         eh = path[depth].p_hdr;
2847         ex = path[depth].p_ext;
2848         ee_block = le32_to_cpu(ex->ee_block);
2849         ee_len = ext4_ext_get_actual_len(ex);
2850         allocated = ee_len - (iblock - ee_block);
2851         newblock = iblock - ee_block + ext_pblock(ex);
2852         ex2 = ex;
2853         orig_ex.ee_block = ex->ee_block;
2854         orig_ex.ee_len   = cpu_to_le16(ee_len);
2855         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2856
2857         /*
2858          * If the uninitialized extent begins at the same logical
2859          * block where the write begins, and the write completely
2860          * covers the extent, then we don't need to split it.
2861          */
2862         if ((iblock == ee_block) && (allocated <= max_blocks))
2863                 return allocated;
2864
2865         err = ext4_ext_get_access(handle, inode, path + depth);
2866         if (err)
2867                 goto out;
2868         /* ex1: ee_block to iblock - 1 : uninitialized */
2869         if (iblock > ee_block) {
2870                 ex1 = ex;
2871                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2872                 ext4_ext_mark_uninitialized(ex1);
2873                 ex2 = &newex;
2874         }
2875         /*
2876          * for sanity, update the length of the ex2 extent before
2877          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2878          * overlap of blocks.
2879          */
2880         if (!ex1 && allocated > max_blocks)
2881                 ex2->ee_len = cpu_to_le16(max_blocks);
2882         /* ex3: to ee_block + ee_len : uninitialised */
2883         if (allocated > max_blocks) {
2884                 unsigned int newdepth;
2885                 ex3 = &newex;
2886                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2887                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2888                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2889                 ext4_ext_mark_uninitialized(ex3);
2890                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2891                 if (err == -ENOSPC) {
2892                         err =  ext4_ext_zeroout(inode, &orig_ex);
2893                         if (err)
2894                                 goto fix_extent_len;
2895                         /* update the extent length and mark as initialized */
2896                         ex->ee_block = orig_ex.ee_block;
2897                         ex->ee_len   = orig_ex.ee_len;
2898                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2899                         ext4_ext_dirty(handle, inode, path + depth);
2900                         /* zeroed the full extent */
2901                         /* blocks available from iblock */
2902                         return allocated;
2903
2904                 } else if (err)
2905                         goto fix_extent_len;
2906                 /*
2907                  * The depth, and hence eh & ex might change
2908                  * as part of the insert above.
2909                  */
2910                 newdepth = ext_depth(inode);
2911                 /*
2912                  * update the extent length after successful insert of the
2913                  * split extent
2914                  */
2915                 orig_ex.ee_len = cpu_to_le16(ee_len -
2916                                                 ext4_ext_get_actual_len(ex3));
2917                 depth = newdepth;
2918                 ext4_ext_drop_refs(path);
2919                 path = ext4_ext_find_extent(inode, iblock, path);
2920                 if (IS_ERR(path)) {
2921                         err = PTR_ERR(path);
2922                         goto out;
2923                 }
2924                 eh = path[depth].p_hdr;
2925                 ex = path[depth].p_ext;
2926                 if (ex2 != &newex)
2927                         ex2 = ex;
2928
2929                 err = ext4_ext_get_access(handle, inode, path + depth);
2930                 if (err)
2931                         goto out;
2932
2933                 allocated = max_blocks;
2934         }
2935         /*
2936          * If there was a change of depth as part of the
2937          * insertion of ex3 above, we need to update the length
2938          * of the ex1 extent again here
2939          */
2940         if (ex1 && ex1 != ex) {
2941                 ex1 = ex;
2942                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2943                 ext4_ext_mark_uninitialized(ex1);
2944                 ex2 = &newex;
2945         }
2946         /*
2947          * ex2: iblock to iblock + maxblocks-1 : to be direct IO written,
2948          * uninitialised still.
2949          */
2950         ex2->ee_block = cpu_to_le32(iblock);
2951         ext4_ext_store_pblock(ex2, newblock);
2952         ex2->ee_len = cpu_to_le16(allocated);
2953         ext4_ext_mark_uninitialized(ex2);
2954         if (ex2 != ex)
2955                 goto insert;
2956         /* Mark modified extent as dirty */
2957         err = ext4_ext_dirty(handle, inode, path + depth);
2958         ext_debug("out here\n");
2959         goto out;
2960 insert:
2961         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2962         if (err == -ENOSPC) {
2963                 err =  ext4_ext_zeroout(inode, &orig_ex);
2964                 if (err)
2965                         goto fix_extent_len;
2966                 /* update the extent length and mark as initialized */
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_dirty(handle, inode, path + depth);
2971                 /* zero out the first half */
2972                 return allocated;
2973         } else if (err)
2974                 goto fix_extent_len;
2975 out:
2976         ext4_ext_show_leaf(inode, path);
2977         return err ? err : allocated;
2978
2979 fix_extent_len:
2980         ex->ee_block = orig_ex.ee_block;
2981         ex->ee_len   = orig_ex.ee_len;
2982         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2983         ext4_ext_mark_uninitialized(ex);
2984         ext4_ext_dirty(handle, inode, path + depth);
2985         return err;
2986 }
2987 static int ext4_convert_unwritten_extents_dio(handle_t *handle,
2988                                               struct inode *inode,
2989                                               struct ext4_ext_path *path)
2990 {
2991         struct ext4_extent *ex;
2992         struct ext4_extent_header *eh;
2993         int depth;
2994         int err = 0;
2995         int ret = 0;
2996
2997         depth = ext_depth(inode);
2998         eh = path[depth].p_hdr;
2999         ex = path[depth].p_ext;
3000
3001         err = ext4_ext_get_access(handle, inode, path + depth);
3002         if (err)
3003                 goto out;
3004         /* first mark the extent as initialized */
3005         ext4_ext_mark_initialized(ex);
3006
3007         /*
3008          * We have to see if it can be merged with the extent
3009          * on the left.
3010          */
3011         if (ex > EXT_FIRST_EXTENT(eh)) {
3012                 /*
3013                  * To merge left, pass "ex - 1" to try_to_merge(),
3014                  * since it merges towards right _only_.
3015                  */
3016                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3017                 if (ret) {
3018                         err = ext4_ext_correct_indexes(handle, inode, path);
3019                         if (err)
3020                                 goto out;
3021                         depth = ext_depth(inode);
3022                         ex--;
3023                 }
3024         }
3025         /*
3026          * Try to Merge towards right.
3027          */
3028         ret = ext4_ext_try_to_merge(inode, path, ex);
3029         if (ret) {
3030                 err = ext4_ext_correct_indexes(handle, inode, path);
3031                 if (err)
3032                         goto out;
3033                 depth = ext_depth(inode);
3034         }
3035         /* Mark modified extent as dirty */
3036         err = ext4_ext_dirty(handle, inode, path + depth);
3037 out:
3038         ext4_ext_show_leaf(inode, path);
3039         return err;
3040 }
3041
3042 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3043                         sector_t block, int count)
3044 {
3045         int i;
3046         for (i = 0; i < count; i++)
3047                 unmap_underlying_metadata(bdev, block + i);
3048 }
3049
3050 static int
3051 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3052                         ext4_lblk_t iblock, unsigned int max_blocks,
3053                         struct ext4_ext_path *path, int flags,
3054                         unsigned int allocated, struct buffer_head *bh_result,
3055                         ext4_fsblk_t newblock)
3056 {
3057         int ret = 0;
3058         int err = 0;
3059         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3060
3061         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3062                   "block %llu, max_blocks %u, flags %d, allocated %u",
3063                   inode->i_ino, (unsigned long long)iblock, max_blocks,
3064                   flags, allocated);
3065         ext4_ext_show_leaf(inode, path);
3066
3067         /* DIO get_block() before submit the IO, split the extent */
3068         if (flags == EXT4_GET_BLOCKS_DIO_CREATE_EXT) {
3069                 ret = ext4_split_unwritten_extents(handle,
3070                                                 inode, path, iblock,
3071                                                 max_blocks, flags);
3072                 /*
3073                  * Flag the inode(non aio case) or end_io struct (aio case)
3074                  * that this IO needs to convertion to written when IO is
3075                  * completed
3076                  */
3077                 if (io)
3078                         io->flag = DIO_AIO_UNWRITTEN;
3079                 else
3080                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3081                 goto out;
3082         }
3083         /* async DIO end_io complete, convert the filled extent to written */
3084         if (flags == EXT4_GET_BLOCKS_DIO_CONVERT_EXT) {
3085                 ret = ext4_convert_unwritten_extents_dio(handle, inode,
3086                                                         path);
3087                 if (ret >= 0)
3088                         ext4_update_inode_fsync_trans(handle, inode, 1);
3089                 goto out2;
3090         }
3091         /* buffered IO case */
3092         /*
3093          * repeat fallocate creation request
3094          * we already have an unwritten extent
3095          */
3096         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3097                 goto map_out;
3098
3099         /* buffered READ or buffered write_begin() lookup */
3100         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3101                 /*
3102                  * We have blocks reserved already.  We
3103                  * return allocated blocks so that delalloc
3104                  * won't do block reservation for us.  But
3105                  * the buffer head will be unmapped so that
3106                  * a read from the block returns 0s.
3107                  */
3108                 set_buffer_unwritten(bh_result);
3109                 goto out1;
3110         }
3111
3112         /* buffered write, writepage time, convert*/
3113         ret = ext4_ext_convert_to_initialized(handle, inode,
3114                                                 path, iblock,
3115                                                 max_blocks);
3116         if (ret >= 0)
3117                 ext4_update_inode_fsync_trans(handle, inode, 1);
3118 out:
3119         if (ret <= 0) {
3120                 err = ret;
3121                 goto out2;
3122         } else
3123                 allocated = ret;
3124         set_buffer_new(bh_result);
3125         /*
3126          * if we allocated more blocks than requested
3127          * we need to make sure we unmap the extra block
3128          * allocated. The actual needed block will get
3129          * unmapped later when we find the buffer_head marked
3130          * new.
3131          */
3132         if (allocated > max_blocks) {
3133                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3134                                         newblock + max_blocks,
3135                                         allocated - max_blocks);
3136                 allocated = max_blocks;
3137         }
3138
3139         /*
3140          * If we have done fallocate with the offset that is already
3141          * delayed allocated, we would have block reservation
3142          * and quota reservation done in the delayed write path.
3143          * But fallocate would have already updated quota and block
3144          * count for this offset. So cancel these reservation
3145          */
3146         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3147                 ext4_da_update_reserve_space(inode, allocated, 0);
3148
3149 map_out:
3150         set_buffer_mapped(bh_result);
3151 out1:
3152         if (allocated > max_blocks)
3153                 allocated = max_blocks;
3154         ext4_ext_show_leaf(inode, path);
3155         bh_result->b_bdev = inode->i_sb->s_bdev;
3156         bh_result->b_blocknr = newblock;
3157 out2:
3158         if (path) {
3159                 ext4_ext_drop_refs(path);
3160                 kfree(path);
3161         }
3162         return err ? err : allocated;
3163 }
3164 /*
3165  * Block allocation/map/preallocation routine for extents based files
3166  *
3167  *
3168  * Need to be called with
3169  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3170  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3171  *
3172  * return > 0, number of of blocks already mapped/allocated
3173  *          if create == 0 and these are pre-allocated blocks
3174  *              buffer head is unmapped
3175  *          otherwise blocks are mapped
3176  *
3177  * return = 0, if plain look up failed (blocks have not been allocated)
3178  *          buffer head is unmapped
3179  *
3180  * return < 0, error case.
3181  */
3182 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3183                         ext4_lblk_t iblock,
3184                         unsigned int max_blocks, struct buffer_head *bh_result,
3185                         int flags)
3186 {
3187         struct ext4_ext_path *path = NULL;
3188         struct ext4_extent_header *eh;
3189         struct ext4_extent newex, *ex;
3190         ext4_fsblk_t newblock;
3191         int err = 0, depth, ret, cache_type;
3192         unsigned int allocated = 0;
3193         struct ext4_allocation_request ar;
3194         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3195
3196         __clear_bit(BH_New, &bh_result->b_state);
3197         ext_debug("blocks %u/%u requested for inode %lu\n",
3198                         iblock, max_blocks, inode->i_ino);
3199
3200         /* check in cache */
3201         cache_type = ext4_ext_in_cache(inode, iblock, &newex);
3202         if (cache_type) {
3203                 if (cache_type == EXT4_EXT_CACHE_GAP) {
3204                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3205                                 /*
3206                                  * block isn't allocated yet and
3207                                  * user doesn't want to allocate it
3208                                  */
3209                                 goto out2;
3210                         }
3211                         /* we should allocate requested block */
3212                 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3213                         /* block is already allocated */
3214                         newblock = iblock
3215                                    - le32_to_cpu(newex.ee_block)
3216                                    + ext_pblock(&newex);
3217                         /* number of remaining blocks in the extent */
3218                         allocated = ext4_ext_get_actual_len(&newex) -
3219                                         (iblock - le32_to_cpu(newex.ee_block));
3220                         goto out;
3221                 } else {
3222                         BUG();
3223                 }
3224         }
3225
3226         /* find extent for this block */
3227         path = ext4_ext_find_extent(inode, iblock, NULL);
3228         if (IS_ERR(path)) {
3229                 err = PTR_ERR(path);
3230                 path = NULL;
3231                 goto out2;
3232         }
3233
3234         depth = ext_depth(inode);
3235
3236         /*
3237          * consistent leaf must not be empty;
3238          * this situation is possible, though, _during_ tree modification;
3239          * this is why assert can't be put in ext4_ext_find_extent()
3240          */
3241         if (path[depth].p_ext == NULL && depth != 0) {
3242                 ext4_error(inode->i_sb, "bad extent address "
3243                            "inode: %lu, iblock: %d, depth: %d",
3244                            inode->i_ino, iblock, depth);
3245                 err = -EIO;
3246                 goto out2;
3247         }
3248         eh = path[depth].p_hdr;
3249
3250         ex = path[depth].p_ext;
3251         if (ex) {
3252                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3253                 ext4_fsblk_t ee_start = ext_pblock(ex);
3254                 unsigned short ee_len;
3255
3256                 /*
3257                  * Uninitialized extents are treated as holes, except that
3258                  * we split out initialized portions during a write.
3259                  */
3260                 ee_len = ext4_ext_get_actual_len(ex);
3261                 /* if found extent covers block, simply return it */
3262                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
3263                         newblock = iblock - ee_block + ee_start;
3264                         /* number of remaining blocks in the extent */
3265                         allocated = ee_len - (iblock - ee_block);
3266                         ext_debug("%u fit into %u:%d -> %llu\n", iblock,
3267                                         ee_block, ee_len, newblock);
3268
3269                         /* Do not put uninitialized extent in the cache */
3270                         if (!ext4_ext_is_uninitialized(ex)) {
3271                                 ext4_ext_put_in_cache(inode, ee_block,
3272                                                         ee_len, ee_start,
3273                                                         EXT4_EXT_CACHE_EXTENT);
3274                                 goto out;
3275                         }
3276                         ret = ext4_ext_handle_uninitialized_extents(handle,
3277                                         inode, iblock, max_blocks, path,
3278                                         flags, allocated, bh_result, newblock);
3279                         return ret;
3280                 }
3281         }
3282
3283         /*
3284          * requested block isn't allocated yet;
3285          * we couldn't try to create block if create flag is zero
3286          */
3287         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3288                 /*
3289                  * put just found gap into cache to speed up
3290                  * subsequent requests
3291                  */
3292                 ext4_ext_put_gap_in_cache(inode, path, iblock);
3293                 goto out2;
3294         }
3295         /*
3296          * Okay, we need to do block allocation.
3297          */
3298
3299         /* find neighbour allocated blocks */
3300         ar.lleft = iblock;
3301         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3302         if (err)
3303                 goto out2;
3304         ar.lright = iblock;
3305         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3306         if (err)
3307                 goto out2;
3308
3309         /*
3310          * See if request is beyond maximum number of blocks we can have in
3311          * a single extent. For an initialized extent this limit is
3312          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3313          * EXT_UNINIT_MAX_LEN.
3314          */
3315         if (max_blocks > EXT_INIT_MAX_LEN &&
3316             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3317                 max_blocks = EXT_INIT_MAX_LEN;
3318         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
3319                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3320                 max_blocks = EXT_UNINIT_MAX_LEN;
3321
3322         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
3323         newex.ee_block = cpu_to_le32(iblock);
3324         newex.ee_len = cpu_to_le16(max_blocks);
3325         err = ext4_ext_check_overlap(inode, &newex, path);
3326         if (err)
3327                 allocated = ext4_ext_get_actual_len(&newex);
3328         else
3329                 allocated = max_blocks;
3330
3331         /* allocate new block */
3332         ar.inode = inode;
3333         ar.goal = ext4_ext_find_goal(inode, path, iblock);
3334         ar.logical = iblock;
3335         ar.len = allocated;
3336         if (S_ISREG(inode->i_mode))
3337                 ar.flags = EXT4_MB_HINT_DATA;
3338         else
3339                 /* disable in-core preallocation for non-regular files */
3340                 ar.flags = 0;
3341         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3342         if (!newblock)
3343                 goto out2;
3344         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3345                   ar.goal, newblock, allocated);
3346
3347         /* try to insert new extent into found leaf and return */
3348         ext4_ext_store_pblock(&newex, newblock);
3349         newex.ee_len = cpu_to_le16(ar.len);
3350         /* Mark uninitialized */
3351         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3352                 ext4_ext_mark_uninitialized(&newex);
3353                 /*
3354                  * io_end structure was created for every async
3355                  * direct IO write to the middle of the file.
3356                  * To avoid unecessary convertion for every aio dio rewrite
3357                  * to the mid of file, here we flag the IO that is really
3358                  * need the convertion.
3359                  * For non asycn direct IO case, flag the inode state
3360                  * that we need to perform convertion when IO is done.
3361                  */
3362                 if (flags == EXT4_GET_BLOCKS_DIO_CREATE_EXT) {
3363                         if (io)
3364                                 io->flag = DIO_AIO_UNWRITTEN;
3365                         else
3366                                 ext4_set_inode_state(inode,
3367                                                      EXT4_STATE_DIO_UNWRITTEN);
3368                 }
3369         }
3370         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3371         if (err) {
3372                 /* free data blocks we just allocated */
3373                 /* not a good idea to call discard here directly,
3374                  * but otherwise we'd need to call it every free() */
3375                 ext4_discard_preallocations(inode);
3376                 ext4_free_blocks(handle, inode, 0, ext_pblock(&newex),
3377                                  ext4_ext_get_actual_len(&newex), 0);
3378                 goto out2;
3379         }
3380
3381         /* previous routine could use block we allocated */
3382         newblock = ext_pblock(&newex);
3383         allocated = ext4_ext_get_actual_len(&newex);
3384         if (allocated > max_blocks)
3385                 allocated = max_blocks;
3386         set_buffer_new(bh_result);
3387
3388         /*
3389          * Update reserved blocks/metadata blocks after successful
3390          * block allocation which had been deferred till now.
3391          */
3392         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3393                 ext4_da_update_reserve_space(inode, allocated, 1);
3394
3395         /*
3396          * Cache the extent and update transaction to commit on fdatasync only
3397          * when it is _not_ an uninitialized extent.
3398          */
3399         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3400                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
3401                                                 EXT4_EXT_CACHE_EXTENT);
3402                 ext4_update_inode_fsync_trans(handle, inode, 1);
3403         } else
3404                 ext4_update_inode_fsync_trans(handle, inode, 0);
3405 out:
3406         if (allocated > max_blocks)
3407                 allocated = max_blocks;
3408         ext4_ext_show_leaf(inode, path);
3409         set_buffer_mapped(bh_result);
3410         bh_result->b_bdev = inode->i_sb->s_bdev;
3411         bh_result->b_blocknr = newblock;
3412 out2:
3413         if (path) {
3414                 ext4_ext_drop_refs(path);
3415                 kfree(path);
3416         }
3417         return err ? err : allocated;
3418 }
3419
3420 void ext4_ext_truncate(struct inode *inode)
3421 {
3422         struct address_space *mapping = inode->i_mapping;
3423         struct super_block *sb = inode->i_sb;
3424         ext4_lblk_t last_block;
3425         handle_t *handle;
3426         int err = 0;
3427
3428         /*
3429          * probably first extent we're gonna free will be last in block
3430          */
3431         err = ext4_writepage_trans_blocks(inode);
3432         handle = ext4_journal_start(inode, err);
3433         if (IS_ERR(handle))
3434                 return;
3435
3436         if (inode->i_size & (sb->s_blocksize - 1))
3437                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3438
3439         if (ext4_orphan_add(handle, inode))
3440                 goto out_stop;
3441
3442         down_write(&EXT4_I(inode)->i_data_sem);
3443         ext4_ext_invalidate_cache(inode);
3444
3445         ext4_discard_preallocations(inode);
3446
3447         /*
3448          * TODO: optimization is possible here.
3449          * Probably we need not scan at all,
3450          * because page truncation is enough.
3451          */
3452
3453         /* we have to know where to truncate from in crash case */
3454         EXT4_I(inode)->i_disksize = inode->i_size;
3455         ext4_mark_inode_dirty(handle, inode);
3456
3457         last_block = (inode->i_size + sb->s_blocksize - 1)
3458                         >> EXT4_BLOCK_SIZE_BITS(sb);
3459         err = ext4_ext_remove_space(inode, last_block);
3460
3461         /* In a multi-transaction truncate, we only make the final
3462          * transaction synchronous.
3463          */
3464         if (IS_SYNC(inode))
3465                 ext4_handle_sync(handle);
3466
3467 out_stop:
3468         up_write(&EXT4_I(inode)->i_data_sem);
3469         /*
3470          * If this was a simple ftruncate() and the file will remain alive,
3471          * then we need to clear up the orphan record which we created above.
3472          * However, if this was a real unlink then we were called by
3473          * ext4_delete_inode(), and we allow that function to clean up the
3474          * orphan info for us.
3475          */
3476         if (inode->i_nlink)
3477                 ext4_orphan_del(handle, inode);
3478
3479         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3480         ext4_mark_inode_dirty(handle, inode);
3481         ext4_journal_stop(handle);
3482 }
3483
3484 static void ext4_falloc_update_inode(struct inode *inode,
3485                                 int mode, loff_t new_size, int update_ctime)
3486 {
3487         struct timespec now;
3488
3489         if (update_ctime) {
3490                 now = current_fs_time(inode->i_sb);
3491                 if (!timespec_equal(&inode->i_ctime, &now))
3492                         inode->i_ctime = now;
3493         }
3494         /*
3495          * Update only when preallocation was requested beyond
3496          * the file size.
3497          */
3498         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3499                 if (new_size > i_size_read(inode))
3500                         i_size_write(inode, new_size);
3501                 if (new_size > EXT4_I(inode)->i_disksize)
3502                         ext4_update_i_disksize(inode, new_size);
3503         }
3504
3505 }
3506
3507 /*
3508  * preallocate space for a file. This implements ext4's fallocate inode
3509  * operation, which gets called from sys_fallocate system call.
3510  * For block-mapped files, posix_fallocate should fall back to the method
3511  * of writing zeroes to the required new blocks (the same behavior which is
3512  * expected for file systems which do not support fallocate() system call).
3513  */
3514 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3515 {
3516         handle_t *handle;
3517         ext4_lblk_t block;
3518         loff_t new_size;
3519         unsigned int max_blocks;
3520         int ret = 0;
3521         int ret2 = 0;
3522         int retries = 0;
3523         struct buffer_head map_bh;
3524         unsigned int credits, blkbits = inode->i_blkbits;
3525
3526         /*
3527          * currently supporting (pre)allocate mode for extent-based
3528          * files _only_
3529          */
3530         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3531                 return -EOPNOTSUPP;
3532
3533         /* preallocation to directories is currently not supported */
3534         if (S_ISDIR(inode->i_mode))
3535                 return -ENODEV;
3536
3537         block = offset >> blkbits;
3538         /*
3539          * We can't just convert len to max_blocks because
3540          * If blocksize = 4096 offset = 3072 and len = 2048
3541          */
3542         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3543                                                         - block;
3544         /*
3545          * credits to insert 1 extent into extent tree
3546          */
3547         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3548         mutex_lock(&inode->i_mutex);
3549 retry:
3550         while (ret >= 0 && ret < max_blocks) {
3551                 block = block + ret;
3552                 max_blocks = max_blocks - ret;
3553                 handle = ext4_journal_start(inode, credits);
3554                 if (IS_ERR(handle)) {
3555                         ret = PTR_ERR(handle);
3556                         break;
3557                 }
3558                 map_bh.b_state = 0;
3559                 ret = ext4_get_blocks(handle, inode, block,
3560                                       max_blocks, &map_bh,
3561                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3562                 if (ret <= 0) {
3563 #ifdef EXT4FS_DEBUG
3564                         WARN_ON(ret <= 0);
3565                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3566                                     "returned error inode#%lu, block=%u, "
3567                                     "max_blocks=%u", __func__,
3568                                     inode->i_ino, block, max_blocks);
3569 #endif
3570                         ext4_mark_inode_dirty(handle, inode);
3571                         ret2 = ext4_journal_stop(handle);
3572                         break;
3573                 }
3574                 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3575                                                 blkbits) >> blkbits))
3576                         new_size = offset + len;
3577                 else
3578                         new_size = (block + ret) << blkbits;
3579
3580                 ext4_falloc_update_inode(inode, mode, new_size,
3581                                                 buffer_new(&map_bh));
3582                 ext4_mark_inode_dirty(handle, inode);
3583                 ret2 = ext4_journal_stop(handle);
3584                 if (ret2)
3585                         break;
3586         }
3587         if (ret == -ENOSPC &&
3588                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3589                 ret = 0;
3590                 goto retry;
3591         }
3592         mutex_unlock(&inode->i_mutex);
3593         return ret > 0 ? ret2 : ret;
3594 }
3595
3596 /*
3597  * This function convert a range of blocks to written extents
3598  * The caller of this function will pass the start offset and the size.
3599  * all unwritten extents within this range will be converted to
3600  * written extents.
3601  *
3602  * This function is called from the direct IO end io call back
3603  * function, to convert the fallocated extents after IO is completed.
3604  * Returns 0 on success.
3605  */
3606 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3607                                     ssize_t len)
3608 {
3609         handle_t *handle;
3610         ext4_lblk_t block;
3611         unsigned int max_blocks;
3612         int ret = 0;
3613         int ret2 = 0;
3614         struct buffer_head map_bh;
3615         unsigned int credits, blkbits = inode->i_blkbits;
3616
3617         block = offset >> blkbits;
3618         /*
3619          * We can't just convert len to max_blocks because
3620          * If blocksize = 4096 offset = 3072 and len = 2048
3621          */
3622         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3623                                                         - block;
3624         /*
3625          * credits to insert 1 extent into extent tree
3626          */
3627         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3628         while (ret >= 0 && ret < max_blocks) {
3629                 block = block + ret;
3630                 max_blocks = max_blocks - ret;
3631                 handle = ext4_journal_start(inode, credits);
3632                 if (IS_ERR(handle)) {
3633                         ret = PTR_ERR(handle);
3634                         break;
3635                 }
3636                 map_bh.b_state = 0;
3637                 ret = ext4_get_blocks(handle, inode, block,
3638                                       max_blocks, &map_bh,
3639                                       EXT4_GET_BLOCKS_DIO_CONVERT_EXT);
3640                 if (ret <= 0) {
3641                         WARN_ON(ret <= 0);
3642                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3643                                     "returned error inode#%lu, block=%u, "
3644                                     "max_blocks=%u", __func__,
3645                                     inode->i_ino, block, max_blocks);
3646                 }
3647                 ext4_mark_inode_dirty(handle, inode);
3648                 ret2 = ext4_journal_stop(handle);
3649                 if (ret <= 0 || ret2 )
3650                         break;
3651         }
3652         return ret > 0 ? ret2 : ret;
3653 }
3654 /*
3655  * Callback function called for each extent to gather FIEMAP information.
3656  */
3657 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3658                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3659                        void *data)
3660 {
3661         struct fiemap_extent_info *fieinfo = data;
3662         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3663         __u64   logical;
3664         __u64   physical;
3665         __u64   length;
3666         __u32   flags = 0;
3667         int     error;
3668
3669         logical =  (__u64)newex->ec_block << blksize_bits;
3670
3671         if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3672                 pgoff_t offset;
3673                 struct page *page;
3674                 struct buffer_head *bh = NULL;
3675
3676                 offset = logical >> PAGE_SHIFT;
3677                 page = find_get_page(inode->i_mapping, offset);
3678                 if (!page || !page_has_buffers(page))
3679                         return EXT_CONTINUE;
3680
3681                 bh = page_buffers(page);
3682
3683                 if (!bh)
3684                         return EXT_CONTINUE;
3685
3686                 if (buffer_delay(bh)) {
3687                         flags |= FIEMAP_EXTENT_DELALLOC;
3688                         page_cache_release(page);
3689                 } else {
3690                         page_cache_release(page);
3691                         return EXT_CONTINUE;
3692                 }
3693         }
3694
3695         physical = (__u64)newex->ec_start << blksize_bits;
3696         length =   (__u64)newex->ec_len << blksize_bits;
3697
3698         if (ex && ext4_ext_is_uninitialized(ex))
3699                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3700
3701         /*
3702          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3703          *
3704          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3705          * this also indicates no more allocated blocks.
3706          *
3707          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3708          */
3709         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3710             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3711                 loff_t size = i_size_read(inode);
3712                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3713
3714                 flags |= FIEMAP_EXTENT_LAST;
3715                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3716                     logical+length > size)
3717                         length = (size - logical + bs - 1) & ~(bs-1);
3718         }
3719
3720         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3721                                         length, flags);
3722         if (error < 0)
3723                 return error;
3724         if (error == 1)
3725                 return EXT_BREAK;
3726
3727         return EXT_CONTINUE;
3728 }
3729
3730 /* fiemap flags we can handle specified here */
3731 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3732
3733 static int ext4_xattr_fiemap(struct inode *inode,
3734                                 struct fiemap_extent_info *fieinfo)
3735 {
3736         __u64 physical = 0;
3737         __u64 length;
3738         __u32 flags = FIEMAP_EXTENT_LAST;
3739         int blockbits = inode->i_sb->s_blocksize_bits;
3740         int error = 0;
3741
3742         /* in-inode? */
3743         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3744                 struct ext4_iloc iloc;
3745                 int offset;     /* offset of xattr in inode */
3746
3747                 error = ext4_get_inode_loc(inode, &iloc);
3748                 if (error)
3749                         return error;
3750                 physical = iloc.bh->b_blocknr << blockbits;
3751                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3752                                 EXT4_I(inode)->i_extra_isize;
3753                 physical += offset;
3754                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3755                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3756         } else { /* external block */
3757                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3758                 length = inode->i_sb->s_blocksize;
3759         }
3760
3761         if (physical)
3762                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3763                                                 length, flags);
3764         return (error < 0 ? error : 0);
3765 }
3766
3767 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3768                 __u64 start, __u64 len)
3769 {
3770         ext4_lblk_t start_blk;
3771         ext4_lblk_t len_blks;
3772         int error = 0;
3773
3774         /* fallback to generic here if not in extents fmt */
3775         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3776                 return generic_block_fiemap(inode, fieinfo, start, len,
3777                         ext4_get_block);
3778
3779         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3780                 return -EBADR;
3781
3782         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3783                 error = ext4_xattr_fiemap(inode, fieinfo);
3784         } else {
3785                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3786                 len_blks = len >> inode->i_sb->s_blocksize_bits;
3787
3788                 /*
3789                  * Walk the extent tree gathering extent information.
3790                  * ext4_ext_fiemap_cb will push extents back to user.
3791                  */
3792                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3793                                           ext4_ext_fiemap_cb, fieinfo);
3794         }
3795
3796         return error;
3797 }
3798