]> Pileus Git - ~andy/linux/blob - fs/ext4/inline.c
bae987549dc367736eaf0beeedb29e8e514e46e6
[~andy/linux] / fs / ext4 / inline.c
1 /*
2  * Copyright (c) 2012 Taobao.
3  * Written by Tao Ma <boyu.mt@taobao.com>
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of version 2.1 of the GNU Lesser General Public License
7  * as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  */
14 #include "ext4_jbd2.h"
15 #include "ext4.h"
16 #include "xattr.h"
17 #include "truncate.h"
18 #include <linux/fiemap.h>
19
20 #define EXT4_XATTR_SYSTEM_DATA  "data"
21 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
22 #define EXT4_INLINE_DOTDOT_OFFSET       2
23 #define EXT4_INLINE_DOTDOT_SIZE         4
24
25 int ext4_get_inline_size(struct inode *inode)
26 {
27         if (EXT4_I(inode)->i_inline_off)
28                 return EXT4_I(inode)->i_inline_size;
29
30         return 0;
31 }
32
33 static int get_max_inline_xattr_value_size(struct inode *inode,
34                                            struct ext4_iloc *iloc)
35 {
36         struct ext4_xattr_ibody_header *header;
37         struct ext4_xattr_entry *entry;
38         struct ext4_inode *raw_inode;
39         int free, min_offs;
40
41         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42                         EXT4_GOOD_OLD_INODE_SIZE -
43                         EXT4_I(inode)->i_extra_isize -
44                         sizeof(struct ext4_xattr_ibody_header);
45
46         /*
47          * We need to subtract another sizeof(__u32) since an in-inode xattr
48          * needs an empty 4 bytes to indicate the gap between the xattr entry
49          * and the name/value pair.
50          */
51         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52                 return EXT4_XATTR_SIZE(min_offs -
53                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54                         EXT4_XATTR_ROUND - sizeof(__u32));
55
56         raw_inode = ext4_raw_inode(iloc);
57         header = IHDR(inode, raw_inode);
58         entry = IFIRST(header);
59
60         /* Compute min_offs. */
61         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62                 if (!entry->e_value_block && entry->e_value_size) {
63                         size_t offs = le16_to_cpu(entry->e_value_offs);
64                         if (offs < min_offs)
65                                 min_offs = offs;
66                 }
67         }
68         free = min_offs -
69                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71         if (EXT4_I(inode)->i_inline_off) {
72                 entry = (struct ext4_xattr_entry *)
73                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76                 goto out;
77         }
78
79         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81         if (free > EXT4_XATTR_ROUND)
82                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83         else
84                 free = 0;
85
86 out:
87         return free;
88 }
89
90 /*
91  * Get the maximum size we now can store in an inode.
92  * If we can't find the space for a xattr entry, don't use the space
93  * of the extents since we have no space to indicate the inline data.
94  */
95 int ext4_get_max_inline_size(struct inode *inode)
96 {
97         int error, max_inline_size;
98         struct ext4_iloc iloc;
99
100         if (EXT4_I(inode)->i_extra_isize == 0)
101                 return 0;
102
103         error = ext4_get_inode_loc(inode, &iloc);
104         if (error) {
105                 ext4_error_inode(inode, __func__, __LINE__, 0,
106                                  "can't get inode location %lu",
107                                  inode->i_ino);
108                 return 0;
109         }
110
111         down_read(&EXT4_I(inode)->xattr_sem);
112         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113         up_read(&EXT4_I(inode)->xattr_sem);
114
115         brelse(iloc.bh);
116
117         if (!max_inline_size)
118                 return 0;
119
120         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121 }
122
123 int ext4_has_inline_data(struct inode *inode)
124 {
125         return ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) &&
126                EXT4_I(inode)->i_inline_off;
127 }
128
129 /*
130  * this function does not take xattr_sem, which is OK because it is
131  * currently only used in a code path coming form ext4_iget, before
132  * the new inode has been unlocked
133  */
134 int ext4_find_inline_data_nolock(struct inode *inode)
135 {
136         struct ext4_xattr_ibody_find is = {
137                 .s = { .not_found = -ENODATA, },
138         };
139         struct ext4_xattr_info i = {
140                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
141                 .name = EXT4_XATTR_SYSTEM_DATA,
142         };
143         int error;
144
145         if (EXT4_I(inode)->i_extra_isize == 0)
146                 return 0;
147
148         error = ext4_get_inode_loc(inode, &is.iloc);
149         if (error)
150                 return error;
151
152         error = ext4_xattr_ibody_find(inode, &i, &is);
153         if (error)
154                 goto out;
155
156         if (!is.s.not_found) {
157                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
158                                         (void *)ext4_raw_inode(&is.iloc));
159                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
160                                 le32_to_cpu(is.s.here->e_value_size);
161                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
162         }
163 out:
164         brelse(is.iloc.bh);
165         return error;
166 }
167
168 static int ext4_read_inline_data(struct inode *inode, void *buffer,
169                                  unsigned int len,
170                                  struct ext4_iloc *iloc)
171 {
172         struct ext4_xattr_entry *entry;
173         struct ext4_xattr_ibody_header *header;
174         int cp_len = 0;
175         struct ext4_inode *raw_inode;
176
177         if (!len)
178                 return 0;
179
180         BUG_ON(len > EXT4_I(inode)->i_inline_size);
181
182         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
183                         len : EXT4_MIN_INLINE_DATA_SIZE;
184
185         raw_inode = ext4_raw_inode(iloc);
186         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
187
188         len -= cp_len;
189         buffer += cp_len;
190
191         if (!len)
192                 goto out;
193
194         header = IHDR(inode, raw_inode);
195         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
196                                             EXT4_I(inode)->i_inline_off);
197         len = min_t(unsigned int, len,
198                     (unsigned int)le32_to_cpu(entry->e_value_size));
199
200         memcpy(buffer,
201                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
202         cp_len += len;
203
204 out:
205         return cp_len;
206 }
207
208 /*
209  * write the buffer to the inline inode.
210  * If 'create' is set, we don't need to do the extra copy in the xattr
211  * value since it is already handled by ext4_xattr_ibody_inline_set.
212  * That saves us one memcpy.
213  */
214 void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
215                             void *buffer, loff_t pos, unsigned int len)
216 {
217         struct ext4_xattr_entry *entry;
218         struct ext4_xattr_ibody_header *header;
219         struct ext4_inode *raw_inode;
220         int cp_len = 0;
221
222         BUG_ON(!EXT4_I(inode)->i_inline_off);
223         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
224
225         raw_inode = ext4_raw_inode(iloc);
226         buffer += pos;
227
228         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
229                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
230                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
231                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
232
233                 len -= cp_len;
234                 buffer += cp_len;
235                 pos += cp_len;
236         }
237
238         if (!len)
239                 return;
240
241         pos -= EXT4_MIN_INLINE_DATA_SIZE;
242         header = IHDR(inode, raw_inode);
243         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
244                                             EXT4_I(inode)->i_inline_off);
245
246         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
247                buffer, len);
248 }
249
250 static int ext4_create_inline_data(handle_t *handle,
251                                    struct inode *inode, unsigned len)
252 {
253         int error;
254         void *value = NULL;
255         struct ext4_xattr_ibody_find is = {
256                 .s = { .not_found = -ENODATA, },
257         };
258         struct ext4_xattr_info i = {
259                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
260                 .name = EXT4_XATTR_SYSTEM_DATA,
261         };
262
263         error = ext4_get_inode_loc(inode, &is.iloc);
264         if (error)
265                 return error;
266
267         error = ext4_journal_get_write_access(handle, is.iloc.bh);
268         if (error)
269                 goto out;
270
271         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272                 value = EXT4_ZERO_XATTR_VALUE;
273                 len -= EXT4_MIN_INLINE_DATA_SIZE;
274         } else {
275                 value = "";
276                 len = 0;
277         }
278
279         /* Insert the the xttr entry. */
280         i.value = value;
281         i.value_len = len;
282
283         error = ext4_xattr_ibody_find(inode, &i, &is);
284         if (error)
285                 goto out;
286
287         BUG_ON(!is.s.not_found);
288
289         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290         if (error) {
291                 if (error == -ENOSPC)
292                         ext4_clear_inode_state(inode,
293                                                EXT4_STATE_MAY_INLINE_DATA);
294                 goto out;
295         }
296
297         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298                 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301                                       (void *)ext4_raw_inode(&is.iloc));
302         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305         get_bh(is.iloc.bh);
306         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
307
308 out:
309         brelse(is.iloc.bh);
310         return error;
311 }
312
313 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
314                                    unsigned int len)
315 {
316         int error;
317         void *value = NULL;
318         struct ext4_xattr_ibody_find is = {
319                 .s = { .not_found = -ENODATA, },
320         };
321         struct ext4_xattr_info i = {
322                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
323                 .name = EXT4_XATTR_SYSTEM_DATA,
324         };
325
326         /* If the old space is ok, write the data directly. */
327         if (len <= EXT4_I(inode)->i_inline_size)
328                 return 0;
329
330         error = ext4_get_inode_loc(inode, &is.iloc);
331         if (error)
332                 return error;
333
334         error = ext4_xattr_ibody_find(inode, &i, &is);
335         if (error)
336                 goto out;
337
338         BUG_ON(is.s.not_found);
339
340         len -= EXT4_MIN_INLINE_DATA_SIZE;
341         value = kzalloc(len, GFP_NOFS);
342         if (!value)
343                 goto out;
344
345         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
346                                      value, len);
347         if (error == -ENODATA)
348                 goto out;
349
350         error = ext4_journal_get_write_access(handle, is.iloc.bh);
351         if (error)
352                 goto out;
353
354         /* Update the xttr entry. */
355         i.value = value;
356         i.value_len = len;
357
358         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
359         if (error)
360                 goto out;
361
362         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
363                                       (void *)ext4_raw_inode(&is.iloc));
364         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
365                                 le32_to_cpu(is.s.here->e_value_size);
366         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
367         get_bh(is.iloc.bh);
368         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
369
370 out:
371         kfree(value);
372         brelse(is.iloc.bh);
373         return error;
374 }
375
376 int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
377                              unsigned int len)
378 {
379         int ret, size;
380         struct ext4_inode_info *ei = EXT4_I(inode);
381
382         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
383                 return -ENOSPC;
384
385         size = ext4_get_max_inline_size(inode);
386         if (size < len)
387                 return -ENOSPC;
388
389         down_write(&EXT4_I(inode)->xattr_sem);
390
391         if (ei->i_inline_off)
392                 ret = ext4_update_inline_data(handle, inode, len);
393         else
394                 ret = ext4_create_inline_data(handle, inode, len);
395
396         up_write(&EXT4_I(inode)->xattr_sem);
397
398         return ret;
399 }
400
401 static int ext4_destroy_inline_data_nolock(handle_t *handle,
402                                            struct inode *inode)
403 {
404         struct ext4_inode_info *ei = EXT4_I(inode);
405         struct ext4_xattr_ibody_find is = {
406                 .s = { .not_found = 0, },
407         };
408         struct ext4_xattr_info i = {
409                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
410                 .name = EXT4_XATTR_SYSTEM_DATA,
411                 .value = NULL,
412                 .value_len = 0,
413         };
414         int error;
415
416         if (!ei->i_inline_off)
417                 return 0;
418
419         error = ext4_get_inode_loc(inode, &is.iloc);
420         if (error)
421                 return error;
422
423         error = ext4_xattr_ibody_find(inode, &i, &is);
424         if (error)
425                 goto out;
426
427         error = ext4_journal_get_write_access(handle, is.iloc.bh);
428         if (error)
429                 goto out;
430
431         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
432         if (error)
433                 goto out;
434
435         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
436                 0, EXT4_MIN_INLINE_DATA_SIZE);
437
438         if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
439                                       EXT4_FEATURE_INCOMPAT_EXTENTS)) {
440                 if (S_ISDIR(inode->i_mode) ||
441                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
442                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
443                         ext4_ext_tree_init(handle, inode);
444                 }
445         }
446         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
447
448         get_bh(is.iloc.bh);
449         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
450
451         EXT4_I(inode)->i_inline_off = 0;
452         EXT4_I(inode)->i_inline_size = 0;
453         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
454 out:
455         brelse(is.iloc.bh);
456         if (error == -ENODATA)
457                 error = 0;
458         return error;
459 }
460
461 static int ext4_read_inline_page(struct inode *inode, struct page *page)
462 {
463         void *kaddr;
464         int ret = 0;
465         size_t len;
466         struct ext4_iloc iloc;
467
468         BUG_ON(!PageLocked(page));
469         BUG_ON(!ext4_has_inline_data(inode));
470         BUG_ON(page->index);
471
472         if (!EXT4_I(inode)->i_inline_off) {
473                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
474                              inode->i_ino);
475                 goto out;
476         }
477
478         ret = ext4_get_inode_loc(inode, &iloc);
479         if (ret)
480                 goto out;
481
482         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
483         kaddr = kmap_atomic(page);
484         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
485         flush_dcache_page(page);
486         kunmap_atomic(kaddr);
487         zero_user_segment(page, len, PAGE_CACHE_SIZE);
488         SetPageUptodate(page);
489         brelse(iloc.bh);
490
491 out:
492         return ret;
493 }
494
495 int ext4_readpage_inline(struct inode *inode, struct page *page)
496 {
497         int ret = 0;
498
499         down_read(&EXT4_I(inode)->xattr_sem);
500         if (!ext4_has_inline_data(inode)) {
501                 up_read(&EXT4_I(inode)->xattr_sem);
502                 return -EAGAIN;
503         }
504
505         /*
506          * Current inline data can only exist in the 1st page,
507          * So for all the other pages, just set them uptodate.
508          */
509         if (!page->index)
510                 ret = ext4_read_inline_page(inode, page);
511         else if (!PageUptodate(page)) {
512                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
513                 SetPageUptodate(page);
514         }
515
516         up_read(&EXT4_I(inode)->xattr_sem);
517
518         unlock_page(page);
519         return ret >= 0 ? 0 : ret;
520 }
521
522 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
523                                               struct inode *inode,
524                                               unsigned flags)
525 {
526         int ret, needed_blocks;
527         handle_t *handle = NULL;
528         int retries = 0, sem_held = 0;
529         struct page *page = NULL;
530         unsigned from, to;
531         struct ext4_iloc iloc;
532
533         if (!ext4_has_inline_data(inode)) {
534                 /*
535                  * clear the flag so that no new write
536                  * will trap here again.
537                  */
538                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
539                 return 0;
540         }
541
542         needed_blocks = ext4_writepage_trans_blocks(inode);
543
544         ret = ext4_get_inode_loc(inode, &iloc);
545         if (ret)
546                 return ret;
547
548 retry:
549         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
550         if (IS_ERR(handle)) {
551                 ret = PTR_ERR(handle);
552                 handle = NULL;
553                 goto out;
554         }
555
556         /* We cannot recurse into the filesystem as the transaction is already
557          * started */
558         flags |= AOP_FLAG_NOFS;
559
560         page = grab_cache_page_write_begin(mapping, 0, flags);
561         if (!page) {
562                 ret = -ENOMEM;
563                 goto out;
564         }
565
566         down_write(&EXT4_I(inode)->xattr_sem);
567         sem_held = 1;
568         /* If some one has already done this for us, just exit. */
569         if (!ext4_has_inline_data(inode)) {
570                 ret = 0;
571                 goto out;
572         }
573
574         from = 0;
575         to = ext4_get_inline_size(inode);
576         if (!PageUptodate(page)) {
577                 ret = ext4_read_inline_page(inode, page);
578                 if (ret < 0)
579                         goto out;
580         }
581
582         ret = ext4_destroy_inline_data_nolock(handle, inode);
583         if (ret)
584                 goto out;
585
586         if (ext4_should_dioread_nolock(inode))
587                 ret = __block_write_begin(page, from, to, ext4_get_block_write);
588         else
589                 ret = __block_write_begin(page, from, to, ext4_get_block);
590
591         if (!ret && ext4_should_journal_data(inode)) {
592                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
593                                              from, to, NULL,
594                                              do_journal_get_write_access);
595         }
596
597         if (ret) {
598                 unlock_page(page);
599                 page_cache_release(page);
600                 ext4_orphan_add(handle, inode);
601                 up_write(&EXT4_I(inode)->xattr_sem);
602                 sem_held = 0;
603                 ext4_journal_stop(handle);
604                 handle = NULL;
605                 ext4_truncate_failed_write(inode);
606                 /*
607                  * If truncate failed early the inode might
608                  * still be on the orphan list; we need to
609                  * make sure the inode is removed from the
610                  * orphan list in that case.
611                  */
612                 if (inode->i_nlink)
613                         ext4_orphan_del(NULL, inode);
614         }
615
616         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617                 goto retry;
618
619         block_commit_write(page, from, to);
620 out:
621         if (page) {
622                 unlock_page(page);
623                 page_cache_release(page);
624         }
625         if (sem_held)
626                 up_write(&EXT4_I(inode)->xattr_sem);
627         if (handle)
628                 ext4_journal_stop(handle);
629         brelse(iloc.bh);
630         return ret;
631 }
632
633 /*
634  * Try to write data in the inode.
635  * If the inode has inline data, check whether the new write can be
636  * in the inode also. If not, create the page the handle, move the data
637  * to the page make it update and let the later codes create extent for it.
638  */
639 int ext4_try_to_write_inline_data(struct address_space *mapping,
640                                   struct inode *inode,
641                                   loff_t pos, unsigned len,
642                                   unsigned flags,
643                                   struct page **pagep)
644 {
645         int ret;
646         handle_t *handle;
647         struct page *page;
648         struct ext4_iloc iloc;
649
650         if (pos + len > ext4_get_max_inline_size(inode))
651                 goto convert;
652
653         ret = ext4_get_inode_loc(inode, &iloc);
654         if (ret)
655                 return ret;
656
657         /*
658          * The possible write could happen in the inode,
659          * so try to reserve the space in inode first.
660          */
661         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
662         if (IS_ERR(handle)) {
663                 ret = PTR_ERR(handle);
664                 handle = NULL;
665                 goto out;
666         }
667
668         ret = ext4_prepare_inline_data(handle, inode, pos + len);
669         if (ret && ret != -ENOSPC)
670                 goto out;
671
672         /* We don't have space in inline inode, so convert it to extent. */
673         if (ret == -ENOSPC) {
674                 ext4_journal_stop(handle);
675                 brelse(iloc.bh);
676                 goto convert;
677         }
678
679         flags |= AOP_FLAG_NOFS;
680
681         page = grab_cache_page_write_begin(mapping, 0, flags);
682         if (!page) {
683                 ret = -ENOMEM;
684                 goto out;
685         }
686
687         *pagep = page;
688         down_read(&EXT4_I(inode)->xattr_sem);
689         if (!ext4_has_inline_data(inode)) {
690                 ret = 0;
691                 unlock_page(page);
692                 page_cache_release(page);
693                 goto out_up_read;
694         }
695
696         if (!PageUptodate(page)) {
697                 ret = ext4_read_inline_page(inode, page);
698                 if (ret < 0)
699                         goto out_up_read;
700         }
701
702         ret = 1;
703         handle = NULL;
704 out_up_read:
705         up_read(&EXT4_I(inode)->xattr_sem);
706 out:
707         if (handle)
708                 ext4_journal_stop(handle);
709         brelse(iloc.bh);
710         return ret;
711 convert:
712         return ext4_convert_inline_data_to_extent(mapping,
713                                                   inode, flags);
714 }
715
716 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
717                                unsigned copied, struct page *page)
718 {
719         int ret;
720         void *kaddr;
721         struct ext4_iloc iloc;
722
723         if (unlikely(copied < len)) {
724                 if (!PageUptodate(page)) {
725                         copied = 0;
726                         goto out;
727                 }
728         }
729
730         ret = ext4_get_inode_loc(inode, &iloc);
731         if (ret) {
732                 ext4_std_error(inode->i_sb, ret);
733                 copied = 0;
734                 goto out;
735         }
736
737         down_write(&EXT4_I(inode)->xattr_sem);
738         BUG_ON(!ext4_has_inline_data(inode));
739
740         kaddr = kmap_atomic(page);
741         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
742         kunmap_atomic(kaddr);
743         SetPageUptodate(page);
744         /* clear page dirty so that writepages wouldn't work for us. */
745         ClearPageDirty(page);
746
747         up_write(&EXT4_I(inode)->xattr_sem);
748         brelse(iloc.bh);
749 out:
750         return copied;
751 }
752
753 struct buffer_head *
754 ext4_journalled_write_inline_data(struct inode *inode,
755                                   unsigned len,
756                                   struct page *page)
757 {
758         int ret;
759         void *kaddr;
760         struct ext4_iloc iloc;
761
762         ret = ext4_get_inode_loc(inode, &iloc);
763         if (ret) {
764                 ext4_std_error(inode->i_sb, ret);
765                 return NULL;
766         }
767
768         down_write(&EXT4_I(inode)->xattr_sem);
769         kaddr = kmap_atomic(page);
770         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
771         kunmap_atomic(kaddr);
772         up_write(&EXT4_I(inode)->xattr_sem);
773
774         return iloc.bh;
775 }
776
777 /*
778  * Try to make the page cache and handle ready for the inline data case.
779  * We can call this function in 2 cases:
780  * 1. The inode is created and the first write exceeds inline size. We can
781  *    clear the inode state safely.
782  * 2. The inode has inline data, then we need to read the data, make it
783  *    update and dirty so that ext4_da_writepages can handle it. We don't
784  *    need to start the journal since the file's metatdata isn't changed now.
785  */
786 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
787                                                  struct inode *inode,
788                                                  unsigned flags,
789                                                  void **fsdata)
790 {
791         int ret = 0, inline_size;
792         struct page *page;
793
794         page = grab_cache_page_write_begin(mapping, 0, flags);
795         if (!page)
796                 return -ENOMEM;
797
798         down_read(&EXT4_I(inode)->xattr_sem);
799         if (!ext4_has_inline_data(inode)) {
800                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
801                 goto out;
802         }
803
804         inline_size = ext4_get_inline_size(inode);
805
806         if (!PageUptodate(page)) {
807                 ret = ext4_read_inline_page(inode, page);
808                 if (ret < 0)
809                         goto out;
810         }
811
812         ret = __block_write_begin(page, 0, inline_size,
813                                   ext4_da_get_block_prep);
814         if (ret) {
815                 ext4_truncate_failed_write(inode);
816                 goto out;
817         }
818
819         SetPageDirty(page);
820         SetPageUptodate(page);
821         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
822         *fsdata = (void *)CONVERT_INLINE_DATA;
823
824 out:
825         up_read(&EXT4_I(inode)->xattr_sem);
826         if (page) {
827                 unlock_page(page);
828                 page_cache_release(page);
829         }
830         return ret;
831 }
832
833 /*
834  * Prepare the write for the inline data.
835  * If the the data can be written into the inode, we just read
836  * the page and make it uptodate, and start the journal.
837  * Otherwise read the page, makes it dirty so that it can be
838  * handle in writepages(the i_disksize update is left to the
839  * normal ext4_da_write_end).
840  */
841 int ext4_da_write_inline_data_begin(struct address_space *mapping,
842                                     struct inode *inode,
843                                     loff_t pos, unsigned len,
844                                     unsigned flags,
845                                     struct page **pagep,
846                                     void **fsdata)
847 {
848         int ret, inline_size;
849         handle_t *handle;
850         struct page *page;
851         struct ext4_iloc iloc;
852
853         ret = ext4_get_inode_loc(inode, &iloc);
854         if (ret)
855                 return ret;
856
857         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
858         if (IS_ERR(handle)) {
859                 ret = PTR_ERR(handle);
860                 handle = NULL;
861                 goto out;
862         }
863
864         inline_size = ext4_get_max_inline_size(inode);
865
866         ret = -ENOSPC;
867         if (inline_size >= pos + len) {
868                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
869                 if (ret && ret != -ENOSPC)
870                         goto out;
871         }
872
873         if (ret == -ENOSPC) {
874                 ret = ext4_da_convert_inline_data_to_extent(mapping,
875                                                             inode,
876                                                             flags,
877                                                             fsdata);
878                 goto out;
879         }
880
881         /*
882          * We cannot recurse into the filesystem as the transaction
883          * is already started.
884          */
885         flags |= AOP_FLAG_NOFS;
886
887         page = grab_cache_page_write_begin(mapping, 0, flags);
888         if (!page) {
889                 ret = -ENOMEM;
890                 goto out;
891         }
892
893         down_read(&EXT4_I(inode)->xattr_sem);
894         if (!ext4_has_inline_data(inode)) {
895                 ret = 0;
896                 goto out_release_page;
897         }
898
899         if (!PageUptodate(page)) {
900                 ret = ext4_read_inline_page(inode, page);
901                 if (ret < 0)
902                         goto out_release_page;
903         }
904
905         up_read(&EXT4_I(inode)->xattr_sem);
906         *pagep = page;
907         handle = NULL;
908         brelse(iloc.bh);
909         return 1;
910 out_release_page:
911         up_read(&EXT4_I(inode)->xattr_sem);
912         unlock_page(page);
913         page_cache_release(page);
914 out:
915         if (handle)
916                 ext4_journal_stop(handle);
917         brelse(iloc.bh);
918         return ret;
919 }
920
921 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
922                                   unsigned len, unsigned copied,
923                                   struct page *page)
924 {
925         int i_size_changed = 0;
926
927         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
928
929         /*
930          * No need to use i_size_read() here, the i_size
931          * cannot change under us because we hold i_mutex.
932          *
933          * But it's important to update i_size while still holding page lock:
934          * page writeout could otherwise come in and zero beyond i_size.
935          */
936         if (pos+copied > inode->i_size) {
937                 i_size_write(inode, pos+copied);
938                 i_size_changed = 1;
939         }
940         unlock_page(page);
941         page_cache_release(page);
942
943         /*
944          * Don't mark the inode dirty under page lock. First, it unnecessarily
945          * makes the holding time of page lock longer. Second, it forces lock
946          * ordering of page lock and transaction start for journaling
947          * filesystems.
948          */
949         if (i_size_changed)
950                 mark_inode_dirty(inode);
951
952         return copied;
953 }
954
955 #ifdef INLINE_DIR_DEBUG
956 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
957                           void *inline_start, int inline_size)
958 {
959         int offset;
960         unsigned short de_len;
961         struct ext4_dir_entry_2 *de = inline_start;
962         void *dlimit = inline_start + inline_size;
963
964         trace_printk("inode %lu\n", dir->i_ino);
965         offset = 0;
966         while ((void *)de < dlimit) {
967                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
968                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
969                              offset, de_len, de->name_len, de->name,
970                              de->name_len, le32_to_cpu(de->inode));
971                 if (ext4_check_dir_entry(dir, NULL, de, bh,
972                                          inline_start, inline_size, offset))
973                         BUG();
974
975                 offset += de_len;
976                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
977         }
978 }
979 #else
980 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
981 #endif
982
983 /*
984  * Add a new entry into a inline dir.
985  * It will return -ENOSPC if no space is available, and -EIO
986  * and -EEXIST if directory entry already exists.
987  */
988 static int ext4_add_dirent_to_inline(handle_t *handle,
989                                      struct dentry *dentry,
990                                      struct inode *inode,
991                                      struct ext4_iloc *iloc,
992                                      void *inline_start, int inline_size)
993 {
994         struct inode    *dir = dentry->d_parent->d_inode;
995         const char      *name = dentry->d_name.name;
996         int             namelen = dentry->d_name.len;
997         int             err;
998         struct ext4_dir_entry_2 *de;
999
1000         err = ext4_find_dest_de(dir, inode, iloc->bh,
1001                                 inline_start, inline_size,
1002                                 name, namelen, &de);
1003         if (err)
1004                 return err;
1005
1006         err = ext4_journal_get_write_access(handle, iloc->bh);
1007         if (err)
1008                 return err;
1009         ext4_insert_dentry(inode, de, inline_size, name, namelen);
1010
1011         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1012
1013         /*
1014          * XXX shouldn't update any times until successful
1015          * completion of syscall, but too many callers depend
1016          * on this.
1017          *
1018          * XXX similarly, too many callers depend on
1019          * ext4_new_inode() setting the times, but error
1020          * recovery deletes the inode, so the worst that can
1021          * happen is that the times are slightly out of date
1022          * and/or different from the directory change time.
1023          */
1024         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1025         ext4_update_dx_flag(dir);
1026         dir->i_version++;
1027         ext4_mark_inode_dirty(handle, dir);
1028         return 1;
1029 }
1030
1031 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1032                                        struct ext4_iloc *iloc)
1033 {
1034         struct ext4_xattr_entry *entry;
1035         struct ext4_xattr_ibody_header *header;
1036
1037         BUG_ON(!EXT4_I(inode)->i_inline_off);
1038
1039         header = IHDR(inode, ext4_raw_inode(iloc));
1040         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1041                                             EXT4_I(inode)->i_inline_off);
1042
1043         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1044 }
1045
1046 /* Set the final de to cover the whole block. */
1047 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1048 {
1049         struct ext4_dir_entry_2 *de, *prev_de;
1050         void *limit;
1051         int de_len;
1052
1053         de = (struct ext4_dir_entry_2 *)de_buf;
1054         if (old_size) {
1055                 limit = de_buf + old_size;
1056                 do {
1057                         prev_de = de;
1058                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1059                         de_buf += de_len;
1060                         de = (struct ext4_dir_entry_2 *)de_buf;
1061                 } while (de_buf < limit);
1062
1063                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1064                                                         old_size, new_size);
1065         } else {
1066                 /* this is just created, so create an empty entry. */
1067                 de->inode = 0;
1068                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1069         }
1070 }
1071
1072 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1073                                   struct ext4_iloc *iloc)
1074 {
1075         int ret;
1076         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1077         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1078
1079         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1080                 return -ENOSPC;
1081
1082         ret = ext4_update_inline_data(handle, dir,
1083                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1084         if (ret)
1085                 return ret;
1086
1087         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1088                              EXT4_I(dir)->i_inline_size -
1089                                                 EXT4_MIN_INLINE_DATA_SIZE);
1090         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1091         return 0;
1092 }
1093
1094 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1095                                      struct ext4_iloc *iloc,
1096                                      void *buf, int inline_size)
1097 {
1098         ext4_create_inline_data(handle, inode, inline_size);
1099         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1100         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1101 }
1102
1103 static int ext4_finish_convert_inline_dir(handle_t *handle,
1104                                           struct inode *inode,
1105                                           struct buffer_head *dir_block,
1106                                           void *buf,
1107                                           int inline_size)
1108 {
1109         int err, csum_size = 0, header_size = 0;
1110         struct ext4_dir_entry_2 *de;
1111         struct ext4_dir_entry_tail *t;
1112         void *target = dir_block->b_data;
1113
1114         /*
1115          * First create "." and ".." and then copy the dir information
1116          * back to the block.
1117          */
1118         de = (struct ext4_dir_entry_2 *)target;
1119         de = ext4_init_dot_dotdot(inode, de,
1120                 inode->i_sb->s_blocksize, csum_size,
1121                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1122         header_size = (void *)de - target;
1123
1124         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1125                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1126
1127         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1128                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1129                 csum_size = sizeof(struct ext4_dir_entry_tail);
1130
1131         inode->i_size = inode->i_sb->s_blocksize;
1132         i_size_write(inode, inode->i_sb->s_blocksize);
1133         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1134         ext4_update_final_de(dir_block->b_data,
1135                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1136                         inode->i_sb->s_blocksize - csum_size);
1137
1138         if (csum_size) {
1139                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1140                                      inode->i_sb->s_blocksize);
1141                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1142         }
1143         set_buffer_uptodate(dir_block);
1144         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1145         if (err)
1146                 goto out;
1147         set_buffer_verified(dir_block);
1148 out:
1149         return err;
1150 }
1151
1152 static int ext4_convert_inline_data_nolock(handle_t *handle,
1153                                            struct inode *inode,
1154                                            struct ext4_iloc *iloc)
1155 {
1156         int error;
1157         void *buf = NULL;
1158         struct buffer_head *data_bh = NULL;
1159         struct ext4_map_blocks map;
1160         int inline_size;
1161
1162         inline_size = ext4_get_inline_size(inode);
1163         buf = kmalloc(inline_size, GFP_NOFS);
1164         if (!buf) {
1165                 error = -ENOMEM;
1166                 goto out;
1167         }
1168
1169         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1170         if (error < 0)
1171                 goto out;
1172
1173         error = ext4_destroy_inline_data_nolock(handle, inode);
1174         if (error)
1175                 goto out;
1176
1177         map.m_lblk = 0;
1178         map.m_len = 1;
1179         map.m_flags = 0;
1180         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1181         if (error < 0)
1182                 goto out_restore;
1183         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1184                 error = -EIO;
1185                 goto out_restore;
1186         }
1187
1188         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1189         if (!data_bh) {
1190                 error = -ENOMEM;
1191                 goto out_restore;
1192         }
1193
1194         lock_buffer(data_bh);
1195         error = ext4_journal_get_create_access(handle, data_bh);
1196         if (error) {
1197                 unlock_buffer(data_bh);
1198                 error = -EIO;
1199                 goto out_restore;
1200         }
1201         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1202
1203         if (!S_ISDIR(inode->i_mode)) {
1204                 memcpy(data_bh->b_data, buf, inline_size);
1205                 set_buffer_uptodate(data_bh);
1206                 error = ext4_handle_dirty_metadata(handle,
1207                                                    inode, data_bh);
1208         } else {
1209                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1210                                                        buf, inline_size);
1211         }
1212
1213         unlock_buffer(data_bh);
1214 out_restore:
1215         if (error)
1216                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1217
1218 out:
1219         brelse(data_bh);
1220         kfree(buf);
1221         return error;
1222 }
1223
1224 /*
1225  * Try to add the new entry to the inline data.
1226  * If succeeds, return 0. If not, extended the inline dir and copied data to
1227  * the new created block.
1228  */
1229 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1230                               struct inode *inode)
1231 {
1232         int ret, inline_size;
1233         void *inline_start;
1234         struct ext4_iloc iloc;
1235         struct inode *dir = dentry->d_parent->d_inode;
1236
1237         ret = ext4_get_inode_loc(dir, &iloc);
1238         if (ret)
1239                 return ret;
1240
1241         down_write(&EXT4_I(dir)->xattr_sem);
1242         if (!ext4_has_inline_data(dir))
1243                 goto out;
1244
1245         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1246                                                  EXT4_INLINE_DOTDOT_SIZE;
1247         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1248
1249         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1250                                         inline_start, inline_size);
1251         if (ret != -ENOSPC)
1252                 goto out;
1253
1254         /* check whether it can be inserted to inline xattr space. */
1255         inline_size = EXT4_I(dir)->i_inline_size -
1256                         EXT4_MIN_INLINE_DATA_SIZE;
1257         if (!inline_size) {
1258                 /* Try to use the xattr space.*/
1259                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1260                 if (ret && ret != -ENOSPC)
1261                         goto out;
1262
1263                 inline_size = EXT4_I(dir)->i_inline_size -
1264                                 EXT4_MIN_INLINE_DATA_SIZE;
1265         }
1266
1267         if (inline_size) {
1268                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1269
1270                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1271                                                 inline_start, inline_size);
1272
1273                 if (ret != -ENOSPC)
1274                         goto out;
1275         }
1276
1277         /*
1278          * The inline space is filled up, so create a new block for it.
1279          * As the extent tree will be created, we have to save the inline
1280          * dir first.
1281          */
1282         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1283
1284 out:
1285         ext4_mark_inode_dirty(handle, dir);
1286         up_write(&EXT4_I(dir)->xattr_sem);
1287         brelse(iloc.bh);
1288         return ret;
1289 }
1290
1291 /*
1292  * This function fills a red-black tree with information from an
1293  * inlined dir.  It returns the number directory entries loaded
1294  * into the tree.  If there is an error it is returned in err.
1295  */
1296 int htree_inlinedir_to_tree(struct file *dir_file,
1297                             struct inode *dir, ext4_lblk_t block,
1298                             struct dx_hash_info *hinfo,
1299                             __u32 start_hash, __u32 start_minor_hash,
1300                             int *has_inline_data)
1301 {
1302         int err = 0, count = 0;
1303         unsigned int parent_ino;
1304         int pos;
1305         struct ext4_dir_entry_2 *de;
1306         struct inode *inode = file_inode(dir_file);
1307         int ret, inline_size = 0;
1308         struct ext4_iloc iloc;
1309         void *dir_buf = NULL;
1310         struct ext4_dir_entry_2 fake;
1311
1312         ret = ext4_get_inode_loc(inode, &iloc);
1313         if (ret)
1314                 return ret;
1315
1316         down_read(&EXT4_I(inode)->xattr_sem);
1317         if (!ext4_has_inline_data(inode)) {
1318                 up_read(&EXT4_I(inode)->xattr_sem);
1319                 *has_inline_data = 0;
1320                 goto out;
1321         }
1322
1323         inline_size = ext4_get_inline_size(inode);
1324         dir_buf = kmalloc(inline_size, GFP_NOFS);
1325         if (!dir_buf) {
1326                 ret = -ENOMEM;
1327                 up_read(&EXT4_I(inode)->xattr_sem);
1328                 goto out;
1329         }
1330
1331         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1332         up_read(&EXT4_I(inode)->xattr_sem);
1333         if (ret < 0)
1334                 goto out;
1335
1336         pos = 0;
1337         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1338         while (pos < inline_size) {
1339                 /*
1340                  * As inlined dir doesn't store any information about '.' and
1341                  * only the inode number of '..' is stored, we have to handle
1342                  * them differently.
1343                  */
1344                 if (pos == 0) {
1345                         fake.inode = cpu_to_le32(inode->i_ino);
1346                         fake.name_len = 1;
1347                         strcpy(fake.name, ".");
1348                         fake.rec_len = ext4_rec_len_to_disk(
1349                                                 EXT4_DIR_REC_LEN(fake.name_len),
1350                                                 inline_size);
1351                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1352                         de = &fake;
1353                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1354                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1355                         fake.inode = cpu_to_le32(parent_ino);
1356                         fake.name_len = 2;
1357                         strcpy(fake.name, "..");
1358                         fake.rec_len = ext4_rec_len_to_disk(
1359                                                 EXT4_DIR_REC_LEN(fake.name_len),
1360                                                 inline_size);
1361                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1362                         de = &fake;
1363                         pos = EXT4_INLINE_DOTDOT_SIZE;
1364                 } else {
1365                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1366                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1367                         if (ext4_check_dir_entry(inode, dir_file, de,
1368                                          iloc.bh, dir_buf,
1369                                          inline_size, pos)) {
1370                                 ret = count;
1371                                 goto out;
1372                         }
1373                 }
1374
1375                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1376                 if ((hinfo->hash < start_hash) ||
1377                     ((hinfo->hash == start_hash) &&
1378                      (hinfo->minor_hash < start_minor_hash)))
1379                         continue;
1380                 if (de->inode == 0)
1381                         continue;
1382                 err = ext4_htree_store_dirent(dir_file,
1383                                    hinfo->hash, hinfo->minor_hash, de);
1384                 if (err) {
1385                         count = err;
1386                         goto out;
1387                 }
1388                 count++;
1389         }
1390         ret = count;
1391 out:
1392         kfree(dir_buf);
1393         brelse(iloc.bh);
1394         return ret;
1395 }
1396
1397 /*
1398  * So this function is called when the volume is mkfsed with
1399  * dir_index disabled. In order to keep f_pos persistent
1400  * after we convert from an inlined dir to a blocked based,
1401  * we just pretend that we are a normal dir and return the
1402  * offset as if '.' and '..' really take place.
1403  *
1404  */
1405 int ext4_read_inline_dir(struct file *file,
1406                          struct dir_context *ctx,
1407                          int *has_inline_data)
1408 {
1409         unsigned int offset, parent_ino;
1410         int i;
1411         struct ext4_dir_entry_2 *de;
1412         struct super_block *sb;
1413         struct inode *inode = file_inode(file);
1414         int ret, inline_size = 0;
1415         struct ext4_iloc iloc;
1416         void *dir_buf = NULL;
1417         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1418
1419         ret = ext4_get_inode_loc(inode, &iloc);
1420         if (ret)
1421                 return ret;
1422
1423         down_read(&EXT4_I(inode)->xattr_sem);
1424         if (!ext4_has_inline_data(inode)) {
1425                 up_read(&EXT4_I(inode)->xattr_sem);
1426                 *has_inline_data = 0;
1427                 goto out;
1428         }
1429
1430         inline_size = ext4_get_inline_size(inode);
1431         dir_buf = kmalloc(inline_size, GFP_NOFS);
1432         if (!dir_buf) {
1433                 ret = -ENOMEM;
1434                 up_read(&EXT4_I(inode)->xattr_sem);
1435                 goto out;
1436         }
1437
1438         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1439         up_read(&EXT4_I(inode)->xattr_sem);
1440         if (ret < 0)
1441                 goto out;
1442
1443         ret = 0;
1444         sb = inode->i_sb;
1445         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1446         offset = ctx->pos;
1447
1448         /*
1449          * dotdot_offset and dotdot_size is the real offset and
1450          * size for ".." and "." if the dir is block based while
1451          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1452          * So we will use extra_offset and extra_size to indicate them
1453          * during the inline dir iteration.
1454          */
1455         dotdot_offset = EXT4_DIR_REC_LEN(1);
1456         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1457         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1458         extra_size = extra_offset + inline_size;
1459
1460         /*
1461          * If the version has changed since the last call to
1462          * readdir(2), then we might be pointing to an invalid
1463          * dirent right now.  Scan from the start of the inline
1464          * dir to make sure.
1465          */
1466         if (file->f_version != inode->i_version) {
1467                 for (i = 0; i < extra_size && i < offset;) {
1468                         /*
1469                          * "." is with offset 0 and
1470                          * ".." is dotdot_offset.
1471                          */
1472                         if (!i) {
1473                                 i = dotdot_offset;
1474                                 continue;
1475                         } else if (i == dotdot_offset) {
1476                                 i = dotdot_size;
1477                                 continue;
1478                         }
1479                         /* for other entry, the real offset in
1480                          * the buf has to be tuned accordingly.
1481                          */
1482                         de = (struct ext4_dir_entry_2 *)
1483                                 (dir_buf + i - extra_offset);
1484                         /* It's too expensive to do a full
1485                          * dirent test each time round this
1486                          * loop, but we do have to test at
1487                          * least that it is non-zero.  A
1488                          * failure will be detected in the
1489                          * dirent test below. */
1490                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1491                                 < EXT4_DIR_REC_LEN(1))
1492                                 break;
1493                         i += ext4_rec_len_from_disk(de->rec_len,
1494                                                     extra_size);
1495                 }
1496                 offset = i;
1497                 ctx->pos = offset;
1498                 file->f_version = inode->i_version;
1499         }
1500
1501         while (ctx->pos < extra_size) {
1502                 if (ctx->pos == 0) {
1503                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1504                                 goto out;
1505                         ctx->pos = dotdot_offset;
1506                         continue;
1507                 }
1508
1509                 if (ctx->pos == dotdot_offset) {
1510                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1511                                 goto out;
1512                         ctx->pos = dotdot_size;
1513                         continue;
1514                 }
1515
1516                 de = (struct ext4_dir_entry_2 *)
1517                         (dir_buf + ctx->pos - extra_offset);
1518                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1519                                          extra_size, ctx->pos))
1520                         goto out;
1521                 if (le32_to_cpu(de->inode)) {
1522                         if (!dir_emit(ctx, de->name, de->name_len,
1523                                       le32_to_cpu(de->inode),
1524                                       get_dtype(sb, de->file_type)))
1525                                 goto out;
1526                 }
1527                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1528         }
1529 out:
1530         kfree(dir_buf);
1531         brelse(iloc.bh);
1532         return ret;
1533 }
1534
1535 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1536                                         struct ext4_dir_entry_2 **parent_de,
1537                                         int *retval)
1538 {
1539         struct ext4_iloc iloc;
1540
1541         *retval = ext4_get_inode_loc(inode, &iloc);
1542         if (*retval)
1543                 return NULL;
1544
1545         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1546
1547         return iloc.bh;
1548 }
1549
1550 /*
1551  * Try to create the inline data for the new dir.
1552  * If it succeeds, return 0, otherwise return the error.
1553  * In case of ENOSPC, the caller should create the normal disk layout dir.
1554  */
1555 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1556                                struct inode *inode)
1557 {
1558         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1559         struct ext4_iloc iloc;
1560         struct ext4_dir_entry_2 *de;
1561
1562         ret = ext4_get_inode_loc(inode, &iloc);
1563         if (ret)
1564                 return ret;
1565
1566         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1567         if (ret)
1568                 goto out;
1569
1570         /*
1571          * For inline dir, we only save the inode information for the ".."
1572          * and create a fake dentry to cover the left space.
1573          */
1574         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1575         de->inode = cpu_to_le32(parent->i_ino);
1576         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1577         de->inode = 0;
1578         de->rec_len = ext4_rec_len_to_disk(
1579                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1580                                 inline_size);
1581         set_nlink(inode, 2);
1582         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1583 out:
1584         brelse(iloc.bh);
1585         return ret;
1586 }
1587
1588 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1589                                         const struct qstr *d_name,
1590                                         struct ext4_dir_entry_2 **res_dir,
1591                                         int *has_inline_data)
1592 {
1593         int ret;
1594         struct ext4_iloc iloc;
1595         void *inline_start;
1596         int inline_size;
1597
1598         if (ext4_get_inode_loc(dir, &iloc))
1599                 return NULL;
1600
1601         down_read(&EXT4_I(dir)->xattr_sem);
1602         if (!ext4_has_inline_data(dir)) {
1603                 *has_inline_data = 0;
1604                 goto out;
1605         }
1606
1607         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1608                                                 EXT4_INLINE_DOTDOT_SIZE;
1609         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1610         ret = search_dir(iloc.bh, inline_start, inline_size,
1611                          dir, d_name, 0, res_dir);
1612         if (ret == 1)
1613                 goto out_find;
1614         if (ret < 0)
1615                 goto out;
1616
1617         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1618                 goto out;
1619
1620         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1621         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1622
1623         ret = search_dir(iloc.bh, inline_start, inline_size,
1624                          dir, d_name, 0, res_dir);
1625         if (ret == 1)
1626                 goto out_find;
1627
1628 out:
1629         brelse(iloc.bh);
1630         iloc.bh = NULL;
1631 out_find:
1632         up_read(&EXT4_I(dir)->xattr_sem);
1633         return iloc.bh;
1634 }
1635
1636 int ext4_delete_inline_entry(handle_t *handle,
1637                              struct inode *dir,
1638                              struct ext4_dir_entry_2 *de_del,
1639                              struct buffer_head *bh,
1640                              int *has_inline_data)
1641 {
1642         int err, inline_size;
1643         struct ext4_iloc iloc;
1644         void *inline_start;
1645
1646         err = ext4_get_inode_loc(dir, &iloc);
1647         if (err)
1648                 return err;
1649
1650         down_write(&EXT4_I(dir)->xattr_sem);
1651         if (!ext4_has_inline_data(dir)) {
1652                 *has_inline_data = 0;
1653                 goto out;
1654         }
1655
1656         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1657                 EXT4_MIN_INLINE_DATA_SIZE) {
1658                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1659                                         EXT4_INLINE_DOTDOT_SIZE;
1660                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1661                                 EXT4_INLINE_DOTDOT_SIZE;
1662         } else {
1663                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1664                 inline_size = ext4_get_inline_size(dir) -
1665                                 EXT4_MIN_INLINE_DATA_SIZE;
1666         }
1667
1668         err = ext4_journal_get_write_access(handle, bh);
1669         if (err)
1670                 goto out;
1671
1672         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1673                                         inline_start, inline_size, 0);
1674         if (err)
1675                 goto out;
1676
1677         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1678         err = ext4_mark_inode_dirty(handle, dir);
1679         if (unlikely(err))
1680                 goto out;
1681
1682         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1683 out:
1684         up_write(&EXT4_I(dir)->xattr_sem);
1685         brelse(iloc.bh);
1686         if (err != -ENOENT)
1687                 ext4_std_error(dir->i_sb, err);
1688         return err;
1689 }
1690
1691 /*
1692  * Get the inline dentry at offset.
1693  */
1694 static inline struct ext4_dir_entry_2 *
1695 ext4_get_inline_entry(struct inode *inode,
1696                       struct ext4_iloc *iloc,
1697                       unsigned int offset,
1698                       void **inline_start,
1699                       int *inline_size)
1700 {
1701         void *inline_pos;
1702
1703         BUG_ON(offset > ext4_get_inline_size(inode));
1704
1705         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1706                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1707                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1708         } else {
1709                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1710                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1711                 *inline_size = ext4_get_inline_size(inode) -
1712                                 EXT4_MIN_INLINE_DATA_SIZE;
1713         }
1714
1715         if (inline_start)
1716                 *inline_start = inline_pos;
1717         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1718 }
1719
1720 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1721 {
1722         int err, inline_size;
1723         struct ext4_iloc iloc;
1724         void *inline_pos;
1725         unsigned int offset;
1726         struct ext4_dir_entry_2 *de;
1727         int ret = 1;
1728
1729         err = ext4_get_inode_loc(dir, &iloc);
1730         if (err) {
1731                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1732                                  err, dir->i_ino);
1733                 return 1;
1734         }
1735
1736         down_read(&EXT4_I(dir)->xattr_sem);
1737         if (!ext4_has_inline_data(dir)) {
1738                 *has_inline_data = 0;
1739                 goto out;
1740         }
1741
1742         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1743         if (!le32_to_cpu(de->inode)) {
1744                 ext4_warning(dir->i_sb,
1745                              "bad inline directory (dir #%lu) - no `..'",
1746                              dir->i_ino);
1747                 ret = 1;
1748                 goto out;
1749         }
1750
1751         offset = EXT4_INLINE_DOTDOT_SIZE;
1752         while (offset < dir->i_size) {
1753                 de = ext4_get_inline_entry(dir, &iloc, offset,
1754                                            &inline_pos, &inline_size);
1755                 if (ext4_check_dir_entry(dir, NULL, de,
1756                                          iloc.bh, inline_pos,
1757                                          inline_size, offset)) {
1758                         ext4_warning(dir->i_sb,
1759                                      "bad inline directory (dir #%lu) - "
1760                                      "inode %u, rec_len %u, name_len %d"
1761                                      "inline size %d\n",
1762                                      dir->i_ino, le32_to_cpu(de->inode),
1763                                      le16_to_cpu(de->rec_len), de->name_len,
1764                                      inline_size);
1765                         ret = 1;
1766                         goto out;
1767                 }
1768                 if (le32_to_cpu(de->inode)) {
1769                         ret = 0;
1770                         goto out;
1771                 }
1772                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1773         }
1774
1775 out:
1776         up_read(&EXT4_I(dir)->xattr_sem);
1777         brelse(iloc.bh);
1778         return ret;
1779 }
1780
1781 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1782 {
1783         int ret;
1784
1785         down_write(&EXT4_I(inode)->xattr_sem);
1786         ret = ext4_destroy_inline_data_nolock(handle, inode);
1787         up_write(&EXT4_I(inode)->xattr_sem);
1788
1789         return ret;
1790 }
1791
1792 int ext4_inline_data_fiemap(struct inode *inode,
1793                             struct fiemap_extent_info *fieinfo,
1794                             int *has_inline)
1795 {
1796         __u64 physical = 0;
1797         __u64 length;
1798         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1799         int error = 0;
1800         struct ext4_iloc iloc;
1801
1802         down_read(&EXT4_I(inode)->xattr_sem);
1803         if (!ext4_has_inline_data(inode)) {
1804                 *has_inline = 0;
1805                 goto out;
1806         }
1807
1808         error = ext4_get_inode_loc(inode, &iloc);
1809         if (error)
1810                 goto out;
1811
1812         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1813         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1814         physical += offsetof(struct ext4_inode, i_block);
1815         length = i_size_read(inode);
1816
1817         if (physical)
1818                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1819                                                 length, flags);
1820         brelse(iloc.bh);
1821 out:
1822         up_read(&EXT4_I(inode)->xattr_sem);
1823         return (error < 0 ? error : 0);
1824 }
1825
1826 /*
1827  * Called during xattr set, and if we can sparse space 'needed',
1828  * just create the extent tree evict the data to the outer block.
1829  *
1830  * We use jbd2 instead of page cache to move data to the 1st block
1831  * so that the whole transaction can be committed as a whole and
1832  * the data isn't lost because of the delayed page cache write.
1833  */
1834 int ext4_try_to_evict_inline_data(handle_t *handle,
1835                                   struct inode *inode,
1836                                   int needed)
1837 {
1838         int error;
1839         struct ext4_xattr_entry *entry;
1840         struct ext4_xattr_ibody_header *header;
1841         struct ext4_inode *raw_inode;
1842         struct ext4_iloc iloc;
1843
1844         error = ext4_get_inode_loc(inode, &iloc);
1845         if (error)
1846                 return error;
1847
1848         raw_inode = ext4_raw_inode(&iloc);
1849         header = IHDR(inode, raw_inode);
1850         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1851                                             EXT4_I(inode)->i_inline_off);
1852         if (EXT4_XATTR_LEN(entry->e_name_len) +
1853             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1854                 error = -ENOSPC;
1855                 goto out;
1856         }
1857
1858         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1859 out:
1860         brelse(iloc.bh);
1861         return error;
1862 }
1863
1864 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1865 {
1866         handle_t *handle;
1867         int inline_size, value_len, needed_blocks;
1868         size_t i_size;
1869         void *value = NULL;
1870         struct ext4_xattr_ibody_find is = {
1871                 .s = { .not_found = -ENODATA, },
1872         };
1873         struct ext4_xattr_info i = {
1874                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1875                 .name = EXT4_XATTR_SYSTEM_DATA,
1876         };
1877
1878
1879         needed_blocks = ext4_writepage_trans_blocks(inode);
1880         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1881         if (IS_ERR(handle))
1882                 return;
1883
1884         down_write(&EXT4_I(inode)->xattr_sem);
1885         if (!ext4_has_inline_data(inode)) {
1886                 *has_inline = 0;
1887                 ext4_journal_stop(handle);
1888                 return;
1889         }
1890
1891         if (ext4_orphan_add(handle, inode))
1892                 goto out;
1893
1894         if (ext4_get_inode_loc(inode, &is.iloc))
1895                 goto out;
1896
1897         down_write(&EXT4_I(inode)->i_data_sem);
1898         i_size = inode->i_size;
1899         inline_size = ext4_get_inline_size(inode);
1900         EXT4_I(inode)->i_disksize = i_size;
1901
1902         if (i_size < inline_size) {
1903                 /* Clear the content in the xattr space. */
1904                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1905                         if (ext4_xattr_ibody_find(inode, &i, &is))
1906                                 goto out_error;
1907
1908                         BUG_ON(is.s.not_found);
1909
1910                         value_len = le32_to_cpu(is.s.here->e_value_size);
1911                         value = kmalloc(value_len, GFP_NOFS);
1912                         if (!value)
1913                                 goto out_error;
1914
1915                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1916                                                 value, value_len))
1917                                 goto out_error;
1918
1919                         i.value = value;
1920                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1921                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1922                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1923                                 goto out_error;
1924                 }
1925
1926                 /* Clear the content within i_blocks. */
1927                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE)
1928                         memset(ext4_raw_inode(&is.iloc)->i_block + i_size, 0,
1929                                         EXT4_MIN_INLINE_DATA_SIZE - i_size);
1930
1931                 EXT4_I(inode)->i_inline_size = i_size <
1932                                         EXT4_MIN_INLINE_DATA_SIZE ?
1933                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1934         }
1935
1936 out_error:
1937         up_write(&EXT4_I(inode)->i_data_sem);
1938 out:
1939         brelse(is.iloc.bh);
1940         up_write(&EXT4_I(inode)->xattr_sem);
1941         kfree(value);
1942         if (inode->i_nlink)
1943                 ext4_orphan_del(handle, inode);
1944
1945         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1946         ext4_mark_inode_dirty(handle, inode);
1947         if (IS_SYNC(inode))
1948                 ext4_handle_sync(handle);
1949
1950         ext4_journal_stop(handle);
1951         return;
1952 }
1953
1954 int ext4_convert_inline_data(struct inode *inode)
1955 {
1956         int error, needed_blocks;
1957         handle_t *handle;
1958         struct ext4_iloc iloc;
1959
1960         if (!ext4_has_inline_data(inode)) {
1961                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1962                 return 0;
1963         }
1964
1965         needed_blocks = ext4_writepage_trans_blocks(inode);
1966
1967         iloc.bh = NULL;
1968         error = ext4_get_inode_loc(inode, &iloc);
1969         if (error)
1970                 return error;
1971
1972         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1973         if (IS_ERR(handle)) {
1974                 error = PTR_ERR(handle);
1975                 goto out_free;
1976         }
1977
1978         down_write(&EXT4_I(inode)->xattr_sem);
1979         if (!ext4_has_inline_data(inode)) {
1980                 up_write(&EXT4_I(inode)->xattr_sem);
1981                 goto out;
1982         }
1983
1984         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1985         up_write(&EXT4_I(inode)->xattr_sem);
1986 out:
1987         ext4_journal_stop(handle);
1988 out_free:
1989         brelse(iloc.bh);
1990         return error;
1991 }