GNU Linux-libre 4.9.294-gnu1
[releases.git] / 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
15 #include <linux/fiemap.h>
16
17 #include "ext4_jbd2.h"
18 #include "ext4.h"
19 #include "xattr.h"
20 #include "truncate.h"
21
22 #define EXT4_XATTR_SYSTEM_DATA  "data"
23 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
24 #define EXT4_INLINE_DOTDOT_OFFSET       2
25 #define EXT4_INLINE_DOTDOT_SIZE         4
26
27 static int ext4_get_inline_size(struct inode *inode)
28 {
29         if (EXT4_I(inode)->i_inline_off)
30                 return EXT4_I(inode)->i_inline_size;
31
32         return 0;
33 }
34
35 static int get_max_inline_xattr_value_size(struct inode *inode,
36                                            struct ext4_iloc *iloc)
37 {
38         struct ext4_xattr_ibody_header *header;
39         struct ext4_xattr_entry *entry;
40         struct ext4_inode *raw_inode;
41         int free, min_offs;
42
43         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44                         EXT4_GOOD_OLD_INODE_SIZE -
45                         EXT4_I(inode)->i_extra_isize -
46                         sizeof(struct ext4_xattr_ibody_header);
47
48         /*
49          * We need to subtract another sizeof(__u32) since an in-inode xattr
50          * needs an empty 4 bytes to indicate the gap between the xattr entry
51          * and the name/value pair.
52          */
53         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54                 return EXT4_XATTR_SIZE(min_offs -
55                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56                         EXT4_XATTR_ROUND - sizeof(__u32));
57
58         raw_inode = ext4_raw_inode(iloc);
59         header = IHDR(inode, raw_inode);
60         entry = IFIRST(header);
61
62         /* Compute min_offs. */
63         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64                 if (!entry->e_value_block && entry->e_value_size) {
65                         size_t offs = le16_to_cpu(entry->e_value_offs);
66                         if (offs < min_offs)
67                                 min_offs = offs;
68                 }
69         }
70         free = min_offs -
71                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73         if (EXT4_I(inode)->i_inline_off) {
74                 entry = (struct ext4_xattr_entry *)
75                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78                 goto out;
79         }
80
81         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83         if (free > EXT4_XATTR_ROUND)
84                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85         else
86                 free = 0;
87
88 out:
89         return free;
90 }
91
92 /*
93  * Get the maximum size we now can store in an inode.
94  * If we can't find the space for a xattr entry, don't use the space
95  * of the extents since we have no space to indicate the inline data.
96  */
97 int ext4_get_max_inline_size(struct inode *inode)
98 {
99         int error, max_inline_size;
100         struct ext4_iloc iloc;
101
102         if (EXT4_I(inode)->i_extra_isize == 0)
103                 return 0;
104
105         error = ext4_get_inode_loc(inode, &iloc);
106         if (error) {
107                 ext4_error_inode(inode, __func__, __LINE__, 0,
108                                  "can't get inode location %lu",
109                                  inode->i_ino);
110                 return 0;
111         }
112
113         down_read(&EXT4_I(inode)->xattr_sem);
114         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
115         up_read(&EXT4_I(inode)->xattr_sem);
116
117         brelse(iloc.bh);
118
119         if (!max_inline_size)
120                 return 0;
121
122         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
123 }
124
125 /*
126  * this function does not take xattr_sem, which is OK because it is
127  * currently only used in a code path coming form ext4_iget, before
128  * the new inode has been unlocked
129  */
130 int ext4_find_inline_data_nolock(struct inode *inode)
131 {
132         struct ext4_xattr_ibody_find is = {
133                 .s = { .not_found = -ENODATA, },
134         };
135         struct ext4_xattr_info i = {
136                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
137                 .name = EXT4_XATTR_SYSTEM_DATA,
138         };
139         int error;
140
141         if (EXT4_I(inode)->i_extra_isize == 0)
142                 return 0;
143
144         error = ext4_get_inode_loc(inode, &is.iloc);
145         if (error)
146                 return error;
147
148         error = ext4_xattr_ibody_find(inode, &i, &is);
149         if (error)
150                 goto out;
151
152         if (!is.s.not_found) {
153                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154                                         (void *)ext4_raw_inode(&is.iloc));
155                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156                                 le32_to_cpu(is.s.here->e_value_size);
157                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158         }
159 out:
160         brelse(is.iloc.bh);
161         return error;
162 }
163
164 static int ext4_read_inline_data(struct inode *inode, void *buffer,
165                                  unsigned int len,
166                                  struct ext4_iloc *iloc)
167 {
168         struct ext4_xattr_entry *entry;
169         struct ext4_xattr_ibody_header *header;
170         int cp_len = 0;
171         struct ext4_inode *raw_inode;
172
173         if (!len)
174                 return 0;
175
176         BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179                         len : EXT4_MIN_INLINE_DATA_SIZE;
180
181         raw_inode = ext4_raw_inode(iloc);
182         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184         len -= cp_len;
185         buffer += cp_len;
186
187         if (!len)
188                 goto out;
189
190         header = IHDR(inode, raw_inode);
191         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192                                             EXT4_I(inode)->i_inline_off);
193         len = min_t(unsigned int, len,
194                     (unsigned int)le32_to_cpu(entry->e_value_size));
195
196         memcpy(buffer,
197                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198         cp_len += len;
199
200 out:
201         return cp_len;
202 }
203
204 /*
205  * write the buffer to the inline inode.
206  * If 'create' is set, we don't need to do the extra copy in the xattr
207  * value since it is already handled by ext4_xattr_ibody_inline_set.
208  * That saves us one memcpy.
209  */
210 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211                                    void *buffer, loff_t pos, unsigned int len)
212 {
213         struct ext4_xattr_entry *entry;
214         struct ext4_xattr_ibody_header *header;
215         struct ext4_inode *raw_inode;
216         int cp_len = 0;
217
218         BUG_ON(!EXT4_I(inode)->i_inline_off);
219         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
220
221         raw_inode = ext4_raw_inode(iloc);
222         buffer += pos;
223
224         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
225                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
226                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
227                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
228
229                 len -= cp_len;
230                 buffer += cp_len;
231                 pos += cp_len;
232         }
233
234         if (!len)
235                 return;
236
237         pos -= EXT4_MIN_INLINE_DATA_SIZE;
238         header = IHDR(inode, raw_inode);
239         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
240                                             EXT4_I(inode)->i_inline_off);
241
242         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
243                buffer, len);
244 }
245
246 static int ext4_create_inline_data(handle_t *handle,
247                                    struct inode *inode, unsigned len)
248 {
249         int error;
250         void *value = NULL;
251         struct ext4_xattr_ibody_find is = {
252                 .s = { .not_found = -ENODATA, },
253         };
254         struct ext4_xattr_info i = {
255                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
256                 .name = EXT4_XATTR_SYSTEM_DATA,
257         };
258
259         error = ext4_get_inode_loc(inode, &is.iloc);
260         if (error)
261                 return error;
262
263         BUFFER_TRACE(is.iloc.bh, "get_write_access");
264         error = ext4_journal_get_write_access(handle, is.iloc.bh);
265         if (error)
266                 goto out;
267
268         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
269                 value = EXT4_ZERO_XATTR_VALUE;
270                 len -= EXT4_MIN_INLINE_DATA_SIZE;
271         } else {
272                 value = "";
273                 len = 0;
274         }
275
276         /* Insert the the xttr entry. */
277         i.value = value;
278         i.value_len = len;
279
280         error = ext4_xattr_ibody_find(inode, &i, &is);
281         if (error)
282                 goto out;
283
284         BUG_ON(!is.s.not_found);
285
286         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
287         if (error) {
288                 if (error == -ENOSPC)
289                         ext4_clear_inode_state(inode,
290                                                EXT4_STATE_MAY_INLINE_DATA);
291                 goto out;
292         }
293
294         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
295                 0, EXT4_MIN_INLINE_DATA_SIZE);
296
297         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
298                                       (void *)ext4_raw_inode(&is.iloc));
299         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
300         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
301         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
302         get_bh(is.iloc.bh);
303         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
304
305 out:
306         brelse(is.iloc.bh);
307         return error;
308 }
309
310 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
311                                    unsigned int len)
312 {
313         int error;
314         void *value = NULL;
315         struct ext4_xattr_ibody_find is = {
316                 .s = { .not_found = -ENODATA, },
317         };
318         struct ext4_xattr_info i = {
319                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
320                 .name = EXT4_XATTR_SYSTEM_DATA,
321         };
322
323         /* If the old space is ok, write the data directly. */
324         if (len <= EXT4_I(inode)->i_inline_size)
325                 return 0;
326
327         error = ext4_get_inode_loc(inode, &is.iloc);
328         if (error)
329                 return error;
330
331         error = ext4_xattr_ibody_find(inode, &i, &is);
332         if (error)
333                 goto out;
334
335         BUG_ON(is.s.not_found);
336
337         len -= EXT4_MIN_INLINE_DATA_SIZE;
338         value = kzalloc(len, GFP_NOFS);
339         if (!value) {
340                 error = -ENOMEM;
341                 goto out;
342         }
343
344         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
345                                      value, len);
346         if (error == -ENODATA)
347                 goto out;
348
349         BUFFER_TRACE(is.iloc.bh, "get_write_access");
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 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
377                                     unsigned int len)
378 {
379         int ret, size, no_expand;
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         ext4_write_lock_xattr(inode, &no_expand);
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         ext4_write_unlock_xattr(inode, &no_expand);
397         return ret;
398 }
399
400 static int ext4_destroy_inline_data_nolock(handle_t *handle,
401                                            struct inode *inode)
402 {
403         struct ext4_inode_info *ei = EXT4_I(inode);
404         struct ext4_xattr_ibody_find is = {
405                 .s = { .not_found = 0, },
406         };
407         struct ext4_xattr_info i = {
408                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
409                 .name = EXT4_XATTR_SYSTEM_DATA,
410                 .value = NULL,
411                 .value_len = 0,
412         };
413         int error;
414
415         if (!ei->i_inline_off)
416                 return 0;
417
418         error = ext4_get_inode_loc(inode, &is.iloc);
419         if (error)
420                 return error;
421
422         error = ext4_xattr_ibody_find(inode, &i, &is);
423         if (error)
424                 goto out;
425
426         BUFFER_TRACE(is.iloc.bh, "get_write_access");
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         memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
438
439         if (ext4_has_feature_extents(inode->i_sb)) {
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_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_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, no_expand;
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         ext4_write_lock_xattr(inode, &no_expand);
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,
588                                           ext4_get_block_unwritten);
589         } else
590                 ret = __block_write_begin(page, from, to, ext4_get_block);
591
592         if (!ret && ext4_should_journal_data(inode)) {
593                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
594                                              from, to, NULL,
595                                              do_journal_get_write_access);
596         }
597
598         if (ret) {
599                 unlock_page(page);
600                 put_page(page);
601                 page = NULL;
602                 ext4_orphan_add(handle, inode);
603                 ext4_write_unlock_xattr(inode, &no_expand);
604                 sem_held = 0;
605                 ext4_journal_stop(handle);
606                 handle = NULL;
607                 ext4_truncate_failed_write(inode);
608                 /*
609                  * If truncate failed early the inode might
610                  * still be on the orphan list; we need to
611                  * make sure the inode is removed from the
612                  * orphan list in that case.
613                  */
614                 if (inode->i_nlink)
615                         ext4_orphan_del(NULL, inode);
616         }
617
618         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
619                 goto retry;
620
621         if (page)
622                 block_commit_write(page, from, to);
623 out:
624         if (page) {
625                 unlock_page(page);
626                 put_page(page);
627         }
628         if (sem_held)
629                 ext4_write_unlock_xattr(inode, &no_expand);
630         if (handle)
631                 ext4_journal_stop(handle);
632         brelse(iloc.bh);
633         return ret;
634 }
635
636 /*
637  * Try to write data in the inode.
638  * If the inode has inline data, check whether the new write can be
639  * in the inode also. If not, create the page the handle, move the data
640  * to the page make it update and let the later codes create extent for it.
641  */
642 int ext4_try_to_write_inline_data(struct address_space *mapping,
643                                   struct inode *inode,
644                                   loff_t pos, unsigned len,
645                                   unsigned flags,
646                                   struct page **pagep)
647 {
648         int ret;
649         handle_t *handle;
650         struct page *page;
651         struct ext4_iloc iloc;
652
653         if (pos + len > ext4_get_max_inline_size(inode))
654                 goto convert;
655
656         ret = ext4_get_inode_loc(inode, &iloc);
657         if (ret)
658                 return ret;
659
660         /*
661          * The possible write could happen in the inode,
662          * so try to reserve the space in inode first.
663          */
664         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
665         if (IS_ERR(handle)) {
666                 ret = PTR_ERR(handle);
667                 handle = NULL;
668                 goto out;
669         }
670
671         ret = ext4_prepare_inline_data(handle, inode, pos + len);
672         if (ret && ret != -ENOSPC)
673                 goto out;
674
675         /* We don't have space in inline inode, so convert it to extent. */
676         if (ret == -ENOSPC) {
677                 ext4_journal_stop(handle);
678                 brelse(iloc.bh);
679                 goto convert;
680         }
681
682         ret = ext4_journal_get_write_access(handle, iloc.bh);
683         if (ret)
684                 goto out;
685
686         flags |= AOP_FLAG_NOFS;
687
688         page = grab_cache_page_write_begin(mapping, 0, flags);
689         if (!page) {
690                 ret = -ENOMEM;
691                 goto out;
692         }
693
694         *pagep = page;
695         down_read(&EXT4_I(inode)->xattr_sem);
696         if (!ext4_has_inline_data(inode)) {
697                 ret = 0;
698                 unlock_page(page);
699                 put_page(page);
700                 goto out_up_read;
701         }
702
703         if (!PageUptodate(page)) {
704                 ret = ext4_read_inline_page(inode, page);
705                 if (ret < 0) {
706                         unlock_page(page);
707                         put_page(page);
708                         goto out_up_read;
709                 }
710         }
711
712         ret = 1;
713         handle = NULL;
714 out_up_read:
715         up_read(&EXT4_I(inode)->xattr_sem);
716 out:
717         if (handle && (ret != 1))
718                 ext4_journal_stop(handle);
719         brelse(iloc.bh);
720         return ret;
721 convert:
722         return ext4_convert_inline_data_to_extent(mapping,
723                                                   inode, flags);
724 }
725
726 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
727                                unsigned copied, struct page *page)
728 {
729         int ret, no_expand;
730         void *kaddr;
731         struct ext4_iloc iloc;
732
733         if (unlikely(copied < len)) {
734                 if (!PageUptodate(page)) {
735                         copied = 0;
736                         goto out;
737                 }
738         }
739
740         ret = ext4_get_inode_loc(inode, &iloc);
741         if (ret) {
742                 ext4_std_error(inode->i_sb, ret);
743                 copied = 0;
744                 goto out;
745         }
746
747         ext4_write_lock_xattr(inode, &no_expand);
748         BUG_ON(!ext4_has_inline_data(inode));
749
750         /*
751          * ei->i_inline_off may have changed since ext4_write_begin()
752          * called ext4_try_to_write_inline_data()
753          */
754         (void) ext4_find_inline_data_nolock(inode);
755
756         kaddr = kmap_atomic(page);
757         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
758         kunmap_atomic(kaddr);
759         SetPageUptodate(page);
760         /* clear page dirty so that writepages wouldn't work for us. */
761         ClearPageDirty(page);
762
763         ext4_write_unlock_xattr(inode, &no_expand);
764         brelse(iloc.bh);
765         mark_inode_dirty(inode);
766 out:
767         return copied;
768 }
769
770 struct buffer_head *
771 ext4_journalled_write_inline_data(struct inode *inode,
772                                   unsigned len,
773                                   struct page *page)
774 {
775         int ret, no_expand;
776         void *kaddr;
777         struct ext4_iloc iloc;
778
779         ret = ext4_get_inode_loc(inode, &iloc);
780         if (ret) {
781                 ext4_std_error(inode->i_sb, ret);
782                 return NULL;
783         }
784
785         ext4_write_lock_xattr(inode, &no_expand);
786         kaddr = kmap_atomic(page);
787         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
788         kunmap_atomic(kaddr);
789         ext4_write_unlock_xattr(inode, &no_expand);
790
791         return iloc.bh;
792 }
793
794 /*
795  * Try to make the page cache and handle ready for the inline data case.
796  * We can call this function in 2 cases:
797  * 1. The inode is created and the first write exceeds inline size. We can
798  *    clear the inode state safely.
799  * 2. The inode has inline data, then we need to read the data, make it
800  *    update and dirty so that ext4_da_writepages can handle it. We don't
801  *    need to start the journal since the file's metatdata isn't changed now.
802  */
803 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
804                                                  struct inode *inode,
805                                                  unsigned flags,
806                                                  void **fsdata)
807 {
808         int ret = 0, inline_size;
809         struct page *page;
810
811         page = grab_cache_page_write_begin(mapping, 0, flags);
812         if (!page)
813                 return -ENOMEM;
814
815         down_read(&EXT4_I(inode)->xattr_sem);
816         if (!ext4_has_inline_data(inode)) {
817                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
818                 goto out;
819         }
820
821         inline_size = ext4_get_inline_size(inode);
822
823         if (!PageUptodate(page)) {
824                 ret = ext4_read_inline_page(inode, page);
825                 if (ret < 0)
826                         goto out;
827         }
828
829         ret = __block_write_begin(page, 0, inline_size,
830                                   ext4_da_get_block_prep);
831         if (ret) {
832                 up_read(&EXT4_I(inode)->xattr_sem);
833                 unlock_page(page);
834                 put_page(page);
835                 ext4_truncate_failed_write(inode);
836                 return ret;
837         }
838
839         SetPageDirty(page);
840         SetPageUptodate(page);
841         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
842         *fsdata = (void *)CONVERT_INLINE_DATA;
843
844 out:
845         up_read(&EXT4_I(inode)->xattr_sem);
846         if (page) {
847                 unlock_page(page);
848                 put_page(page);
849         }
850         return ret;
851 }
852
853 /*
854  * Prepare the write for the inline data.
855  * If the the data can be written into the inode, we just read
856  * the page and make it uptodate, and start the journal.
857  * Otherwise read the page, makes it dirty so that it can be
858  * handle in writepages(the i_disksize update is left to the
859  * normal ext4_da_write_end).
860  */
861 int ext4_da_write_inline_data_begin(struct address_space *mapping,
862                                     struct inode *inode,
863                                     loff_t pos, unsigned len,
864                                     unsigned flags,
865                                     struct page **pagep,
866                                     void **fsdata)
867 {
868         int ret, inline_size;
869         handle_t *handle;
870         struct page *page;
871         struct ext4_iloc iloc;
872         int retries = 0;
873
874         ret = ext4_get_inode_loc(inode, &iloc);
875         if (ret)
876                 return ret;
877
878 retry_journal:
879         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
880         if (IS_ERR(handle)) {
881                 ret = PTR_ERR(handle);
882                 goto out;
883         }
884
885         inline_size = ext4_get_max_inline_size(inode);
886
887         ret = -ENOSPC;
888         if (inline_size >= pos + len) {
889                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
890                 if (ret && ret != -ENOSPC)
891                         goto out_journal;
892         }
893
894         /*
895          * We cannot recurse into the filesystem as the transaction
896          * is already started.
897          */
898         flags |= AOP_FLAG_NOFS;
899
900         if (ret == -ENOSPC) {
901                 ext4_journal_stop(handle);
902                 ret = ext4_da_convert_inline_data_to_extent(mapping,
903                                                             inode,
904                                                             flags,
905                                                             fsdata);
906                 if (ret == -ENOSPC &&
907                     ext4_should_retry_alloc(inode->i_sb, &retries))
908                         goto retry_journal;
909                 goto out;
910         }
911
912         page = grab_cache_page_write_begin(mapping, 0, flags);
913         if (!page) {
914                 ret = -ENOMEM;
915                 goto out_journal;
916         }
917
918         down_read(&EXT4_I(inode)->xattr_sem);
919         if (!ext4_has_inline_data(inode)) {
920                 ret = 0;
921                 goto out_release_page;
922         }
923
924         if (!PageUptodate(page)) {
925                 ret = ext4_read_inline_page(inode, page);
926                 if (ret < 0)
927                         goto out_release_page;
928         }
929         ret = ext4_journal_get_write_access(handle, iloc.bh);
930         if (ret)
931                 goto out_release_page;
932
933         up_read(&EXT4_I(inode)->xattr_sem);
934         *pagep = page;
935         brelse(iloc.bh);
936         return 1;
937 out_release_page:
938         up_read(&EXT4_I(inode)->xattr_sem);
939         unlock_page(page);
940         put_page(page);
941 out_journal:
942         ext4_journal_stop(handle);
943 out:
944         brelse(iloc.bh);
945         return ret;
946 }
947
948 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
949                                   unsigned len, unsigned copied,
950                                   struct page *page)
951 {
952         int ret;
953
954         ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
955         if (ret < 0) {
956                 unlock_page(page);
957                 put_page(page);
958                 return ret;
959         }
960         copied = ret;
961
962         /*
963          * No need to use i_size_read() here, the i_size
964          * cannot change under us because we hold i_mutex.
965          *
966          * But it's important to update i_size while still holding page lock:
967          * page writeout could otherwise come in and zero beyond i_size.
968          */
969         if (pos+copied > inode->i_size)
970                 i_size_write(inode, pos+copied);
971         unlock_page(page);
972         put_page(page);
973
974         /*
975          * Don't mark the inode dirty under page lock. First, it unnecessarily
976          * makes the holding time of page lock longer. Second, it forces lock
977          * ordering of page lock and transaction start for journaling
978          * filesystems.
979          */
980         mark_inode_dirty(inode);
981
982         return copied;
983 }
984
985 #ifdef INLINE_DIR_DEBUG
986 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
987                           void *inline_start, int inline_size)
988 {
989         int offset;
990         unsigned short de_len;
991         struct ext4_dir_entry_2 *de = inline_start;
992         void *dlimit = inline_start + inline_size;
993
994         trace_printk("inode %lu\n", dir->i_ino);
995         offset = 0;
996         while ((void *)de < dlimit) {
997                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
998                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
999                              offset, de_len, de->name_len, de->name,
1000                              de->name_len, le32_to_cpu(de->inode));
1001                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1002                                          inline_start, inline_size, offset))
1003                         BUG();
1004
1005                 offset += de_len;
1006                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1007         }
1008 }
1009 #else
1010 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1011 #endif
1012
1013 /*
1014  * Add a new entry into a inline dir.
1015  * It will return -ENOSPC if no space is available, and -EIO
1016  * and -EEXIST if directory entry already exists.
1017  */
1018 static int ext4_add_dirent_to_inline(handle_t *handle,
1019                                      struct ext4_filename *fname,
1020                                      struct inode *dir,
1021                                      struct inode *inode,
1022                                      struct ext4_iloc *iloc,
1023                                      void *inline_start, int inline_size)
1024 {
1025         int             err;
1026         struct ext4_dir_entry_2 *de;
1027
1028         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1029                                 inline_size, fname, &de);
1030         if (err)
1031                 return err;
1032
1033         BUFFER_TRACE(iloc->bh, "get_write_access");
1034         err = ext4_journal_get_write_access(handle, iloc->bh);
1035         if (err)
1036                 return err;
1037         ext4_insert_dentry(dir, inode, de, inline_size, fname);
1038
1039         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1040
1041         /*
1042          * XXX shouldn't update any times until successful
1043          * completion of syscall, but too many callers depend
1044          * on this.
1045          *
1046          * XXX similarly, too many callers depend on
1047          * ext4_new_inode() setting the times, but error
1048          * recovery deletes the inode, so the worst that can
1049          * happen is that the times are slightly out of date
1050          * and/or different from the directory change time.
1051          */
1052         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1053         ext4_update_dx_flag(dir);
1054         dir->i_version++;
1055         ext4_mark_inode_dirty(handle, dir);
1056         return 1;
1057 }
1058
1059 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1060                                        struct ext4_iloc *iloc)
1061 {
1062         struct ext4_xattr_entry *entry;
1063         struct ext4_xattr_ibody_header *header;
1064
1065         BUG_ON(!EXT4_I(inode)->i_inline_off);
1066
1067         header = IHDR(inode, ext4_raw_inode(iloc));
1068         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1069                                             EXT4_I(inode)->i_inline_off);
1070
1071         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1072 }
1073
1074 /* Set the final de to cover the whole block. */
1075 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1076 {
1077         struct ext4_dir_entry_2 *de, *prev_de;
1078         void *limit;
1079         int de_len;
1080
1081         de = (struct ext4_dir_entry_2 *)de_buf;
1082         if (old_size) {
1083                 limit = de_buf + old_size;
1084                 do {
1085                         prev_de = de;
1086                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1087                         de_buf += de_len;
1088                         de = (struct ext4_dir_entry_2 *)de_buf;
1089                 } while (de_buf < limit);
1090
1091                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1092                                                         old_size, new_size);
1093         } else {
1094                 /* this is just created, so create an empty entry. */
1095                 de->inode = 0;
1096                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1097         }
1098 }
1099
1100 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1101                                   struct ext4_iloc *iloc)
1102 {
1103         int ret;
1104         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1105         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1106
1107         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1108                 return -ENOSPC;
1109
1110         ret = ext4_update_inline_data(handle, dir,
1111                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1112         if (ret)
1113                 return ret;
1114
1115         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1116                              EXT4_I(dir)->i_inline_size -
1117                                                 EXT4_MIN_INLINE_DATA_SIZE);
1118         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1119         return 0;
1120 }
1121
1122 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1123                                      struct ext4_iloc *iloc,
1124                                      void *buf, int inline_size)
1125 {
1126         ext4_create_inline_data(handle, inode, inline_size);
1127         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1128         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1129 }
1130
1131 static int ext4_finish_convert_inline_dir(handle_t *handle,
1132                                           struct inode *inode,
1133                                           struct buffer_head *dir_block,
1134                                           void *buf,
1135                                           int inline_size)
1136 {
1137         int err, csum_size = 0, header_size = 0;
1138         struct ext4_dir_entry_2 *de;
1139         struct ext4_dir_entry_tail *t;
1140         void *target = dir_block->b_data;
1141
1142         /*
1143          * First create "." and ".." and then copy the dir information
1144          * back to the block.
1145          */
1146         de = (struct ext4_dir_entry_2 *)target;
1147         de = ext4_init_dot_dotdot(inode, de,
1148                 inode->i_sb->s_blocksize, csum_size,
1149                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1150         header_size = (void *)de - target;
1151
1152         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1153                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1154
1155         if (ext4_has_metadata_csum(inode->i_sb))
1156                 csum_size = sizeof(struct ext4_dir_entry_tail);
1157
1158         inode->i_size = inode->i_sb->s_blocksize;
1159         i_size_write(inode, inode->i_sb->s_blocksize);
1160         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1161         ext4_update_final_de(dir_block->b_data,
1162                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1163                         inode->i_sb->s_blocksize - csum_size);
1164
1165         if (csum_size) {
1166                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1167                                      inode->i_sb->s_blocksize);
1168                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1169         }
1170         set_buffer_uptodate(dir_block);
1171         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1172         if (err)
1173                 return err;
1174         set_buffer_verified(dir_block);
1175         return ext4_mark_inode_dirty(handle, inode);
1176 }
1177
1178 static int ext4_convert_inline_data_nolock(handle_t *handle,
1179                                            struct inode *inode,
1180                                            struct ext4_iloc *iloc)
1181 {
1182         int error;
1183         void *buf = NULL;
1184         struct buffer_head *data_bh = NULL;
1185         struct ext4_map_blocks map;
1186         int inline_size;
1187
1188         inline_size = ext4_get_inline_size(inode);
1189         buf = kmalloc(inline_size, GFP_NOFS);
1190         if (!buf) {
1191                 error = -ENOMEM;
1192                 goto out;
1193         }
1194
1195         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1196         if (error < 0)
1197                 goto out;
1198
1199         /*
1200          * Make sure the inline directory entries pass checks before we try to
1201          * convert them, so that we avoid touching stuff that needs fsck.
1202          */
1203         if (S_ISDIR(inode->i_mode)) {
1204                 error = ext4_check_all_de(inode, iloc->bh,
1205                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1206                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1207                 if (error)
1208                         goto out;
1209         }
1210
1211         error = ext4_destroy_inline_data_nolock(handle, inode);
1212         if (error)
1213                 goto out;
1214
1215         map.m_lblk = 0;
1216         map.m_len = 1;
1217         map.m_flags = 0;
1218         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1219         if (error < 0)
1220                 goto out_restore;
1221         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1222                 error = -EIO;
1223                 goto out_restore;
1224         }
1225
1226         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1227         if (!data_bh) {
1228                 error = -ENOMEM;
1229                 goto out_restore;
1230         }
1231
1232         lock_buffer(data_bh);
1233         error = ext4_journal_get_create_access(handle, data_bh);
1234         if (error) {
1235                 unlock_buffer(data_bh);
1236                 error = -EIO;
1237                 goto out_restore;
1238         }
1239         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1240
1241         if (!S_ISDIR(inode->i_mode)) {
1242                 memcpy(data_bh->b_data, buf, inline_size);
1243                 set_buffer_uptodate(data_bh);
1244                 error = ext4_handle_dirty_metadata(handle,
1245                                                    inode, data_bh);
1246         } else {
1247                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1248                                                        buf, inline_size);
1249         }
1250
1251         unlock_buffer(data_bh);
1252 out_restore:
1253         if (error)
1254                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1255
1256 out:
1257         brelse(data_bh);
1258         kfree(buf);
1259         return error;
1260 }
1261
1262 /*
1263  * Try to add the new entry to the inline data.
1264  * If succeeds, return 0. If not, extended the inline dir and copied data to
1265  * the new created block.
1266  */
1267 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1268                               struct inode *dir, struct inode *inode)
1269 {
1270         int ret, inline_size, no_expand;
1271         void *inline_start;
1272         struct ext4_iloc iloc;
1273
1274         ret = ext4_get_inode_loc(dir, &iloc);
1275         if (ret)
1276                 return ret;
1277
1278         ext4_write_lock_xattr(dir, &no_expand);
1279         if (!ext4_has_inline_data(dir))
1280                 goto out;
1281
1282         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1283                                                  EXT4_INLINE_DOTDOT_SIZE;
1284         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1285
1286         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1287                                         inline_start, inline_size);
1288         if (ret != -ENOSPC)
1289                 goto out;
1290
1291         /* check whether it can be inserted to inline xattr space. */
1292         inline_size = EXT4_I(dir)->i_inline_size -
1293                         EXT4_MIN_INLINE_DATA_SIZE;
1294         if (!inline_size) {
1295                 /* Try to use the xattr space.*/
1296                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1297                 if (ret && ret != -ENOSPC)
1298                         goto out;
1299
1300                 inline_size = EXT4_I(dir)->i_inline_size -
1301                                 EXT4_MIN_INLINE_DATA_SIZE;
1302         }
1303
1304         if (inline_size) {
1305                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1306
1307                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1308                                                 inode, &iloc, inline_start,
1309                                                 inline_size);
1310
1311                 if (ret != -ENOSPC)
1312                         goto out;
1313         }
1314
1315         /*
1316          * The inline space is filled up, so create a new block for it.
1317          * As the extent tree will be created, we have to save the inline
1318          * dir first.
1319          */
1320         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1321
1322 out:
1323         ext4_mark_inode_dirty(handle, dir);
1324         ext4_write_unlock_xattr(dir, &no_expand);
1325         brelse(iloc.bh);
1326         return ret;
1327 }
1328
1329 /*
1330  * This function fills a red-black tree with information from an
1331  * inlined dir.  It returns the number directory entries loaded
1332  * into the tree.  If there is an error it is returned in err.
1333  */
1334 int htree_inlinedir_to_tree(struct file *dir_file,
1335                             struct inode *dir, ext4_lblk_t block,
1336                             struct dx_hash_info *hinfo,
1337                             __u32 start_hash, __u32 start_minor_hash,
1338                             int *has_inline_data)
1339 {
1340         int err = 0, count = 0;
1341         unsigned int parent_ino;
1342         int pos;
1343         struct ext4_dir_entry_2 *de;
1344         struct inode *inode = file_inode(dir_file);
1345         int ret, inline_size = 0;
1346         struct ext4_iloc iloc;
1347         void *dir_buf = NULL;
1348         struct ext4_dir_entry_2 fake;
1349         struct fscrypt_str tmp_str;
1350
1351         ret = ext4_get_inode_loc(inode, &iloc);
1352         if (ret)
1353                 return ret;
1354
1355         down_read(&EXT4_I(inode)->xattr_sem);
1356         if (!ext4_has_inline_data(inode)) {
1357                 up_read(&EXT4_I(inode)->xattr_sem);
1358                 *has_inline_data = 0;
1359                 goto out;
1360         }
1361
1362         inline_size = ext4_get_inline_size(inode);
1363         dir_buf = kmalloc(inline_size, GFP_NOFS);
1364         if (!dir_buf) {
1365                 ret = -ENOMEM;
1366                 up_read(&EXT4_I(inode)->xattr_sem);
1367                 goto out;
1368         }
1369
1370         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1371         up_read(&EXT4_I(inode)->xattr_sem);
1372         if (ret < 0)
1373                 goto out;
1374
1375         pos = 0;
1376         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1377         while (pos < inline_size) {
1378                 /*
1379                  * As inlined dir doesn't store any information about '.' and
1380                  * only the inode number of '..' is stored, we have to handle
1381                  * them differently.
1382                  */
1383                 if (pos == 0) {
1384                         fake.inode = cpu_to_le32(inode->i_ino);
1385                         fake.name_len = 1;
1386                         strcpy(fake.name, ".");
1387                         fake.rec_len = ext4_rec_len_to_disk(
1388                                                 EXT4_DIR_REC_LEN(fake.name_len),
1389                                                 inline_size);
1390                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1391                         de = &fake;
1392                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1393                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1394                         fake.inode = cpu_to_le32(parent_ino);
1395                         fake.name_len = 2;
1396                         strcpy(fake.name, "..");
1397                         fake.rec_len = ext4_rec_len_to_disk(
1398                                                 EXT4_DIR_REC_LEN(fake.name_len),
1399                                                 inline_size);
1400                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1401                         de = &fake;
1402                         pos = EXT4_INLINE_DOTDOT_SIZE;
1403                 } else {
1404                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1405                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1406                         if (ext4_check_dir_entry(inode, dir_file, de,
1407                                          iloc.bh, dir_buf,
1408                                          inline_size, pos)) {
1409                                 ret = count;
1410                                 goto out;
1411                         }
1412                 }
1413
1414                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1415                 if ((hinfo->hash < start_hash) ||
1416                     ((hinfo->hash == start_hash) &&
1417                      (hinfo->minor_hash < start_minor_hash)))
1418                         continue;
1419                 if (de->inode == 0)
1420                         continue;
1421                 tmp_str.name = de->name;
1422                 tmp_str.len = de->name_len;
1423                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1424                                               hinfo->minor_hash, de, &tmp_str);
1425                 if (err) {
1426                         ret = err;
1427                         goto out;
1428                 }
1429                 count++;
1430         }
1431         ret = count;
1432 out:
1433         kfree(dir_buf);
1434         brelse(iloc.bh);
1435         return ret;
1436 }
1437
1438 /*
1439  * So this function is called when the volume is mkfsed with
1440  * dir_index disabled. In order to keep f_pos persistent
1441  * after we convert from an inlined dir to a blocked based,
1442  * we just pretend that we are a normal dir and return the
1443  * offset as if '.' and '..' really take place.
1444  *
1445  */
1446 int ext4_read_inline_dir(struct file *file,
1447                          struct dir_context *ctx,
1448                          int *has_inline_data)
1449 {
1450         unsigned int offset, parent_ino;
1451         int i;
1452         struct ext4_dir_entry_2 *de;
1453         struct super_block *sb;
1454         struct inode *inode = file_inode(file);
1455         int ret, inline_size = 0;
1456         struct ext4_iloc iloc;
1457         void *dir_buf = NULL;
1458         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1459
1460         ret = ext4_get_inode_loc(inode, &iloc);
1461         if (ret)
1462                 return ret;
1463
1464         down_read(&EXT4_I(inode)->xattr_sem);
1465         if (!ext4_has_inline_data(inode)) {
1466                 up_read(&EXT4_I(inode)->xattr_sem);
1467                 *has_inline_data = 0;
1468                 goto out;
1469         }
1470
1471         inline_size = ext4_get_inline_size(inode);
1472         dir_buf = kmalloc(inline_size, GFP_NOFS);
1473         if (!dir_buf) {
1474                 ret = -ENOMEM;
1475                 up_read(&EXT4_I(inode)->xattr_sem);
1476                 goto out;
1477         }
1478
1479         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1480         up_read(&EXT4_I(inode)->xattr_sem);
1481         if (ret < 0)
1482                 goto out;
1483
1484         ret = 0;
1485         sb = inode->i_sb;
1486         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1487         offset = ctx->pos;
1488
1489         /*
1490          * dotdot_offset and dotdot_size is the real offset and
1491          * size for ".." and "." if the dir is block based while
1492          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1493          * So we will use extra_offset and extra_size to indicate them
1494          * during the inline dir iteration.
1495          */
1496         dotdot_offset = EXT4_DIR_REC_LEN(1);
1497         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1498         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1499         extra_size = extra_offset + inline_size;
1500
1501         /*
1502          * If the version has changed since the last call to
1503          * readdir(2), then we might be pointing to an invalid
1504          * dirent right now.  Scan from the start of the inline
1505          * dir to make sure.
1506          */
1507         if (file->f_version != inode->i_version) {
1508                 for (i = 0; i < extra_size && i < offset;) {
1509                         /*
1510                          * "." is with offset 0 and
1511                          * ".." is dotdot_offset.
1512                          */
1513                         if (!i) {
1514                                 i = dotdot_offset;
1515                                 continue;
1516                         } else if (i == dotdot_offset) {
1517                                 i = dotdot_size;
1518                                 continue;
1519                         }
1520                         /* for other entry, the real offset in
1521                          * the buf has to be tuned accordingly.
1522                          */
1523                         de = (struct ext4_dir_entry_2 *)
1524                                 (dir_buf + i - extra_offset);
1525                         /* It's too expensive to do a full
1526                          * dirent test each time round this
1527                          * loop, but we do have to test at
1528                          * least that it is non-zero.  A
1529                          * failure will be detected in the
1530                          * dirent test below. */
1531                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1532                                 < EXT4_DIR_REC_LEN(1))
1533                                 break;
1534                         i += ext4_rec_len_from_disk(de->rec_len,
1535                                                     extra_size);
1536                 }
1537                 offset = i;
1538                 ctx->pos = offset;
1539                 file->f_version = inode->i_version;
1540         }
1541
1542         while (ctx->pos < extra_size) {
1543                 if (ctx->pos == 0) {
1544                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1545                                 goto out;
1546                         ctx->pos = dotdot_offset;
1547                         continue;
1548                 }
1549
1550                 if (ctx->pos == dotdot_offset) {
1551                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1552                                 goto out;
1553                         ctx->pos = dotdot_size;
1554                         continue;
1555                 }
1556
1557                 de = (struct ext4_dir_entry_2 *)
1558                         (dir_buf + ctx->pos - extra_offset);
1559                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1560                                          extra_size, ctx->pos))
1561                         goto out;
1562                 if (le32_to_cpu(de->inode)) {
1563                         if (!dir_emit(ctx, de->name, de->name_len,
1564                                       le32_to_cpu(de->inode),
1565                                       get_dtype(sb, de->file_type)))
1566                                 goto out;
1567                 }
1568                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1569         }
1570 out:
1571         kfree(dir_buf);
1572         brelse(iloc.bh);
1573         return ret;
1574 }
1575
1576 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1577                                         struct ext4_dir_entry_2 **parent_de,
1578                                         int *retval)
1579 {
1580         struct ext4_iloc iloc;
1581
1582         *retval = ext4_get_inode_loc(inode, &iloc);
1583         if (*retval)
1584                 return NULL;
1585
1586         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1587
1588         return iloc.bh;
1589 }
1590
1591 /*
1592  * Try to create the inline data for the new dir.
1593  * If it succeeds, return 0, otherwise return the error.
1594  * In case of ENOSPC, the caller should create the normal disk layout dir.
1595  */
1596 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1597                                struct inode *inode)
1598 {
1599         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1600         struct ext4_iloc iloc;
1601         struct ext4_dir_entry_2 *de;
1602
1603         ret = ext4_get_inode_loc(inode, &iloc);
1604         if (ret)
1605                 return ret;
1606
1607         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1608         if (ret)
1609                 goto out;
1610
1611         /*
1612          * For inline dir, we only save the inode information for the ".."
1613          * and create a fake dentry to cover the left space.
1614          */
1615         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1616         de->inode = cpu_to_le32(parent->i_ino);
1617         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1618         de->inode = 0;
1619         de->rec_len = ext4_rec_len_to_disk(
1620                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1621                                 inline_size);
1622         set_nlink(inode, 2);
1623         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1624 out:
1625         brelse(iloc.bh);
1626         return ret;
1627 }
1628
1629 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1630                                         struct ext4_filename *fname,
1631                                         const struct qstr *d_name,
1632                                         struct ext4_dir_entry_2 **res_dir,
1633                                         int *has_inline_data)
1634 {
1635         int ret;
1636         struct ext4_iloc iloc;
1637         void *inline_start;
1638         int inline_size;
1639
1640         if (ext4_get_inode_loc(dir, &iloc))
1641                 return NULL;
1642
1643         down_read(&EXT4_I(dir)->xattr_sem);
1644         if (!ext4_has_inline_data(dir)) {
1645                 *has_inline_data = 0;
1646                 goto out;
1647         }
1648
1649         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1650                                                 EXT4_INLINE_DOTDOT_SIZE;
1651         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1652         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1653                               dir, fname, d_name, 0, res_dir);
1654         if (ret == 1)
1655                 goto out_find;
1656         if (ret < 0)
1657                 goto out;
1658
1659         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1660                 goto out;
1661
1662         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1663         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1664
1665         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1666                               dir, fname, d_name, 0, res_dir);
1667         if (ret == 1)
1668                 goto out_find;
1669
1670 out:
1671         brelse(iloc.bh);
1672         iloc.bh = NULL;
1673 out_find:
1674         up_read(&EXT4_I(dir)->xattr_sem);
1675         return iloc.bh;
1676 }
1677
1678 int ext4_delete_inline_entry(handle_t *handle,
1679                              struct inode *dir,
1680                              struct ext4_dir_entry_2 *de_del,
1681                              struct buffer_head *bh,
1682                              int *has_inline_data)
1683 {
1684         int err, inline_size, no_expand;
1685         struct ext4_iloc iloc;
1686         void *inline_start;
1687
1688         err = ext4_get_inode_loc(dir, &iloc);
1689         if (err)
1690                 return err;
1691
1692         ext4_write_lock_xattr(dir, &no_expand);
1693         if (!ext4_has_inline_data(dir)) {
1694                 *has_inline_data = 0;
1695                 goto out;
1696         }
1697
1698         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1699                 EXT4_MIN_INLINE_DATA_SIZE) {
1700                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1701                                         EXT4_INLINE_DOTDOT_SIZE;
1702                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1703                                 EXT4_INLINE_DOTDOT_SIZE;
1704         } else {
1705                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1706                 inline_size = ext4_get_inline_size(dir) -
1707                                 EXT4_MIN_INLINE_DATA_SIZE;
1708         }
1709
1710         BUFFER_TRACE(bh, "get_write_access");
1711         err = ext4_journal_get_write_access(handle, bh);
1712         if (err)
1713                 goto out;
1714
1715         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1716                                         inline_start, inline_size, 0);
1717         if (err)
1718                 goto out;
1719
1720         err = ext4_mark_inode_dirty(handle, dir);
1721         if (unlikely(err))
1722                 goto out;
1723
1724         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1725 out:
1726         ext4_write_unlock_xattr(dir, &no_expand);
1727         brelse(iloc.bh);
1728         if (err != -ENOENT)
1729                 ext4_std_error(dir->i_sb, err);
1730         return err;
1731 }
1732
1733 /*
1734  * Get the inline dentry at offset.
1735  */
1736 static inline struct ext4_dir_entry_2 *
1737 ext4_get_inline_entry(struct inode *inode,
1738                       struct ext4_iloc *iloc,
1739                       unsigned int offset,
1740                       void **inline_start,
1741                       int *inline_size)
1742 {
1743         void *inline_pos;
1744
1745         BUG_ON(offset > ext4_get_inline_size(inode));
1746
1747         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1748                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1749                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1750         } else {
1751                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1752                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1753                 *inline_size = ext4_get_inline_size(inode) -
1754                                 EXT4_MIN_INLINE_DATA_SIZE;
1755         }
1756
1757         if (inline_start)
1758                 *inline_start = inline_pos;
1759         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1760 }
1761
1762 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1763 {
1764         int err, inline_size;
1765         struct ext4_iloc iloc;
1766         size_t inline_len;
1767         void *inline_pos;
1768         unsigned int offset;
1769         struct ext4_dir_entry_2 *de;
1770         bool ret = true;
1771
1772         err = ext4_get_inode_loc(dir, &iloc);
1773         if (err) {
1774                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1775                                  err, dir->i_ino);
1776                 return true;
1777         }
1778
1779         down_read(&EXT4_I(dir)->xattr_sem);
1780         if (!ext4_has_inline_data(dir)) {
1781                 *has_inline_data = 0;
1782                 goto out;
1783         }
1784
1785         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1786         if (!le32_to_cpu(de->inode)) {
1787                 ext4_warning(dir->i_sb,
1788                              "bad inline directory (dir #%lu) - no `..'",
1789                              dir->i_ino);
1790                 ret = true;
1791                 goto out;
1792         }
1793
1794         inline_len = ext4_get_inline_size(dir);
1795         offset = EXT4_INLINE_DOTDOT_SIZE;
1796         while (offset < inline_len) {
1797                 de = ext4_get_inline_entry(dir, &iloc, offset,
1798                                            &inline_pos, &inline_size);
1799                 if (ext4_check_dir_entry(dir, NULL, de,
1800                                          iloc.bh, inline_pos,
1801                                          inline_size, offset)) {
1802                         ext4_warning(dir->i_sb,
1803                                      "bad inline directory (dir #%lu) - "
1804                                      "inode %u, rec_len %u, name_len %d"
1805                                      "inline size %d",
1806                                      dir->i_ino, le32_to_cpu(de->inode),
1807                                      le16_to_cpu(de->rec_len), de->name_len,
1808                                      inline_size);
1809                         ret = true;
1810                         goto out;
1811                 }
1812                 if (le32_to_cpu(de->inode)) {
1813                         ret = false;
1814                         goto out;
1815                 }
1816                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1817         }
1818
1819 out:
1820         up_read(&EXT4_I(dir)->xattr_sem);
1821         brelse(iloc.bh);
1822         return ret;
1823 }
1824
1825 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1826 {
1827         int ret, no_expand;
1828
1829         ext4_write_lock_xattr(inode, &no_expand);
1830         ret = ext4_destroy_inline_data_nolock(handle, inode);
1831         ext4_write_unlock_xattr(inode, &no_expand);
1832
1833         return ret;
1834 }
1835
1836 int ext4_inline_data_fiemap(struct inode *inode,
1837                             struct fiemap_extent_info *fieinfo,
1838                             int *has_inline, __u64 start, __u64 len)
1839 {
1840         __u64 physical = 0;
1841         __u64 inline_len;
1842         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1843                 FIEMAP_EXTENT_LAST;
1844         int error = 0;
1845         struct ext4_iloc iloc;
1846
1847         down_read(&EXT4_I(inode)->xattr_sem);
1848         if (!ext4_has_inline_data(inode)) {
1849                 *has_inline = 0;
1850                 goto out;
1851         }
1852         inline_len = min_t(size_t, ext4_get_inline_size(inode),
1853                            i_size_read(inode));
1854         if (start >= inline_len)
1855                 goto out;
1856         if (start + len < inline_len)
1857                 inline_len = start + len;
1858         inline_len -= start;
1859
1860         error = ext4_get_inode_loc(inode, &iloc);
1861         if (error)
1862                 goto out;
1863
1864         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1865         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1866         physical += offsetof(struct ext4_inode, i_block);
1867
1868         brelse(iloc.bh);
1869 out:
1870         up_read(&EXT4_I(inode)->xattr_sem);
1871         if (physical)
1872                 error = fiemap_fill_next_extent(fieinfo, start, physical,
1873                                                 inline_len, flags);
1874         return (error < 0 ? error : 0);
1875 }
1876
1877 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1878 {
1879         handle_t *handle;
1880         int inline_size, value_len, needed_blocks, no_expand;
1881         size_t i_size;
1882         void *value = NULL;
1883         struct ext4_xattr_ibody_find is = {
1884                 .s = { .not_found = -ENODATA, },
1885         };
1886         struct ext4_xattr_info i = {
1887                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1888                 .name = EXT4_XATTR_SYSTEM_DATA,
1889         };
1890
1891
1892         needed_blocks = ext4_writepage_trans_blocks(inode);
1893         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1894         if (IS_ERR(handle))
1895                 return;
1896
1897         ext4_write_lock_xattr(inode, &no_expand);
1898         if (!ext4_has_inline_data(inode)) {
1899                 ext4_write_unlock_xattr(inode, &no_expand);
1900                 *has_inline = 0;
1901                 ext4_journal_stop(handle);
1902                 return;
1903         }
1904
1905         if (ext4_orphan_add(handle, inode))
1906                 goto out;
1907
1908         if (ext4_get_inode_loc(inode, &is.iloc))
1909                 goto out;
1910
1911         down_write(&EXT4_I(inode)->i_data_sem);
1912         i_size = inode->i_size;
1913         inline_size = ext4_get_inline_size(inode);
1914         EXT4_I(inode)->i_disksize = i_size;
1915
1916         if (i_size < inline_size) {
1917                 /* Clear the content in the xattr space. */
1918                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1919                         if (ext4_xattr_ibody_find(inode, &i, &is))
1920                                 goto out_error;
1921
1922                         BUG_ON(is.s.not_found);
1923
1924                         value_len = le32_to_cpu(is.s.here->e_value_size);
1925                         value = kmalloc(value_len, GFP_NOFS);
1926                         if (!value)
1927                                 goto out_error;
1928
1929                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1930                                                 value, value_len))
1931                                 goto out_error;
1932
1933                         i.value = value;
1934                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1935                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1936                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1937                                 goto out_error;
1938                 }
1939
1940                 /* Clear the content within i_blocks. */
1941                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1942                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1943                         memset(p + i_size, 0,
1944                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1945                 }
1946
1947                 EXT4_I(inode)->i_inline_size = i_size <
1948                                         EXT4_MIN_INLINE_DATA_SIZE ?
1949                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1950         }
1951
1952 out_error:
1953         up_write(&EXT4_I(inode)->i_data_sem);
1954 out:
1955         brelse(is.iloc.bh);
1956         ext4_write_unlock_xattr(inode, &no_expand);
1957         kfree(value);
1958         if (inode->i_nlink)
1959                 ext4_orphan_del(handle, inode);
1960
1961         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1962         ext4_mark_inode_dirty(handle, inode);
1963         if (IS_SYNC(inode))
1964                 ext4_handle_sync(handle);
1965
1966         ext4_journal_stop(handle);
1967         return;
1968 }
1969
1970 int ext4_convert_inline_data(struct inode *inode)
1971 {
1972         int error, needed_blocks, no_expand;
1973         handle_t *handle;
1974         struct ext4_iloc iloc;
1975
1976         if (!ext4_has_inline_data(inode)) {
1977                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1978                 return 0;
1979         }
1980
1981         needed_blocks = ext4_writepage_trans_blocks(inode);
1982
1983         iloc.bh = NULL;
1984         error = ext4_get_inode_loc(inode, &iloc);
1985         if (error)
1986                 return error;
1987
1988         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1989         if (IS_ERR(handle)) {
1990                 error = PTR_ERR(handle);
1991                 goto out_free;
1992         }
1993
1994         ext4_write_lock_xattr(inode, &no_expand);
1995         if (ext4_has_inline_data(inode))
1996                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1997         ext4_write_unlock_xattr(inode, &no_expand);
1998         ext4_journal_stop(handle);
1999 out_free:
2000         brelse(iloc.bh);
2001         return error;
2002 }