6f5e292e86f725f1d2aadb184129f51b816b1920
[releases.git] / 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_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, 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, 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                 page = NULL;
601                 ext4_orphan_add(handle, inode);
602                 ext4_write_unlock_xattr(inode, &no_expand);
603                 sem_held = 0;
604                 ext4_journal_stop(handle);
605                 handle = NULL;
606                 ext4_truncate_failed_write(inode);
607                 /*
608                  * If truncate failed early the inode might
609                  * still be on the orphan list; we need to
610                  * make sure the inode is removed from the
611                  * orphan list in that case.
612                  */
613                 if (inode->i_nlink)
614                         ext4_orphan_del(NULL, inode);
615         }
616
617         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
618                 goto retry;
619
620         if (page)
621                 block_commit_write(page, from, to);
622 out:
623         if (page) {
624                 unlock_page(page);
625                 page_cache_release(page);
626         }
627         if (sem_held)
628                 ext4_write_unlock_xattr(inode, &no_expand);
629         if (handle)
630                 ext4_journal_stop(handle);
631         brelse(iloc.bh);
632         return ret;
633 }
634
635 /*
636  * Try to write data in the inode.
637  * If the inode has inline data, check whether the new write can be
638  * in the inode also. If not, create the page the handle, move the data
639  * to the page make it update and let the later codes create extent for it.
640  */
641 int ext4_try_to_write_inline_data(struct address_space *mapping,
642                                   struct inode *inode,
643                                   loff_t pos, unsigned len,
644                                   unsigned flags,
645                                   struct page **pagep)
646 {
647         int ret;
648         handle_t *handle;
649         struct page *page;
650         struct ext4_iloc iloc;
651
652         if (pos + len > ext4_get_max_inline_size(inode))
653                 goto convert;
654
655         ret = ext4_get_inode_loc(inode, &iloc);
656         if (ret)
657                 return ret;
658
659         /*
660          * The possible write could happen in the inode,
661          * so try to reserve the space in inode first.
662          */
663         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
664         if (IS_ERR(handle)) {
665                 ret = PTR_ERR(handle);
666                 handle = NULL;
667                 goto out;
668         }
669
670         ret = ext4_prepare_inline_data(handle, inode, pos + len);
671         if (ret && ret != -ENOSPC)
672                 goto out;
673
674         /* We don't have space in inline inode, so convert it to extent. */
675         if (ret == -ENOSPC) {
676                 ext4_journal_stop(handle);
677                 brelse(iloc.bh);
678                 goto convert;
679         }
680
681         ret = ext4_journal_get_write_access(handle, iloc.bh);
682         if (ret)
683                 goto out;
684
685         flags |= AOP_FLAG_NOFS;
686
687         page = grab_cache_page_write_begin(mapping, 0, flags);
688         if (!page) {
689                 ret = -ENOMEM;
690                 goto out;
691         }
692
693         *pagep = page;
694         down_read(&EXT4_I(inode)->xattr_sem);
695         if (!ext4_has_inline_data(inode)) {
696                 ret = 0;
697                 unlock_page(page);
698                 page_cache_release(page);
699                 goto out_up_read;
700         }
701
702         if (!PageUptodate(page)) {
703                 ret = ext4_read_inline_page(inode, page);
704                 if (ret < 0) {
705                         unlock_page(page);
706                         put_page(page);
707                         goto out_up_read;
708                 }
709         }
710
711         ret = 1;
712         handle = NULL;
713 out_up_read:
714         up_read(&EXT4_I(inode)->xattr_sem);
715 out:
716         if (handle && (ret != 1))
717                 ext4_journal_stop(handle);
718         brelse(iloc.bh);
719         return ret;
720 convert:
721         return ext4_convert_inline_data_to_extent(mapping,
722                                                   inode, flags);
723 }
724
725 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
726                                unsigned copied, struct page *page)
727 {
728         int ret, no_expand;
729         void *kaddr;
730         struct ext4_iloc iloc;
731
732         if (unlikely(copied < len)) {
733                 if (!PageUptodate(page)) {
734                         copied = 0;
735                         goto out;
736                 }
737         }
738
739         ret = ext4_get_inode_loc(inode, &iloc);
740         if (ret) {
741                 ext4_std_error(inode->i_sb, ret);
742                 copied = 0;
743                 goto out;
744         }
745
746         ext4_write_lock_xattr(inode, &no_expand);
747         BUG_ON(!ext4_has_inline_data(inode));
748
749         /*
750          * ei->i_inline_off may have changed since ext4_write_begin()
751          * called ext4_try_to_write_inline_data()
752          */
753         (void) ext4_find_inline_data_nolock(inode);
754
755         kaddr = kmap_atomic(page);
756         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
757         kunmap_atomic(kaddr);
758         SetPageUptodate(page);
759         /* clear page dirty so that writepages wouldn't work for us. */
760         ClearPageDirty(page);
761
762         ext4_write_unlock_xattr(inode, &no_expand);
763         brelse(iloc.bh);
764         mark_inode_dirty(inode);
765 out:
766         return copied;
767 }
768
769 struct buffer_head *
770 ext4_journalled_write_inline_data(struct inode *inode,
771                                   unsigned len,
772                                   struct page *page)
773 {
774         int ret, no_expand;
775         void *kaddr;
776         struct ext4_iloc iloc;
777
778         ret = ext4_get_inode_loc(inode, &iloc);
779         if (ret) {
780                 ext4_std_error(inode->i_sb, ret);
781                 return NULL;
782         }
783
784         ext4_write_lock_xattr(inode, &no_expand);
785         kaddr = kmap_atomic(page);
786         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
787         kunmap_atomic(kaddr);
788         ext4_write_unlock_xattr(inode, &no_expand);
789
790         return iloc.bh;
791 }
792
793 /*
794  * Try to make the page cache and handle ready for the inline data case.
795  * We can call this function in 2 cases:
796  * 1. The inode is created and the first write exceeds inline size. We can
797  *    clear the inode state safely.
798  * 2. The inode has inline data, then we need to read the data, make it
799  *    update and dirty so that ext4_da_writepages can handle it. We don't
800  *    need to start the journal since the file's metatdata isn't changed now.
801  */
802 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
803                                                  struct inode *inode,
804                                                  unsigned flags,
805                                                  void **fsdata)
806 {
807         int ret = 0, inline_size;
808         struct page *page;
809
810         page = grab_cache_page_write_begin(mapping, 0, flags);
811         if (!page)
812                 return -ENOMEM;
813
814         down_read(&EXT4_I(inode)->xattr_sem);
815         if (!ext4_has_inline_data(inode)) {
816                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
817                 goto out;
818         }
819
820         inline_size = ext4_get_inline_size(inode);
821
822         if (!PageUptodate(page)) {
823                 ret = ext4_read_inline_page(inode, page);
824                 if (ret < 0)
825                         goto out;
826         }
827
828         ret = __block_write_begin(page, 0, inline_size,
829                                   ext4_da_get_block_prep);
830         if (ret) {
831                 up_read(&EXT4_I(inode)->xattr_sem);
832                 unlock_page(page);
833                 page_cache_release(page);
834                 ext4_truncate_failed_write(inode);
835                 return ret;
836         }
837
838         SetPageDirty(page);
839         SetPageUptodate(page);
840         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
841         *fsdata = (void *)CONVERT_INLINE_DATA;
842
843 out:
844         up_read(&EXT4_I(inode)->xattr_sem);
845         if (page) {
846                 unlock_page(page);
847                 page_cache_release(page);
848         }
849         return ret;
850 }
851
852 /*
853  * Prepare the write for the inline data.
854  * If the the data can be written into the inode, we just read
855  * the page and make it uptodate, and start the journal.
856  * Otherwise read the page, makes it dirty so that it can be
857  * handle in writepages(the i_disksize update is left to the
858  * normal ext4_da_write_end).
859  */
860 int ext4_da_write_inline_data_begin(struct address_space *mapping,
861                                     struct inode *inode,
862                                     loff_t pos, unsigned len,
863                                     unsigned flags,
864                                     struct page **pagep,
865                                     void **fsdata)
866 {
867         int ret, inline_size;
868         handle_t *handle;
869         struct page *page;
870         struct ext4_iloc iloc;
871         int retries = 0;
872
873         ret = ext4_get_inode_loc(inode, &iloc);
874         if (ret)
875                 return ret;
876
877 retry_journal:
878         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
879         if (IS_ERR(handle)) {
880                 ret = PTR_ERR(handle);
881                 goto out;
882         }
883
884         inline_size = ext4_get_max_inline_size(inode);
885
886         ret = -ENOSPC;
887         if (inline_size >= pos + len) {
888                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
889                 if (ret && ret != -ENOSPC)
890                         goto out_journal;
891         }
892
893         /*
894          * We cannot recurse into the filesystem as the transaction
895          * is already started.
896          */
897         flags |= AOP_FLAG_NOFS;
898
899         if (ret == -ENOSPC) {
900                 ext4_journal_stop(handle);
901                 ret = ext4_da_convert_inline_data_to_extent(mapping,
902                                                             inode,
903                                                             flags,
904                                                             fsdata);
905                 if (ret == -ENOSPC &&
906                     ext4_should_retry_alloc(inode->i_sb, &retries))
907                         goto retry_journal;
908                 goto out;
909         }
910
911         page = grab_cache_page_write_begin(mapping, 0, flags);
912         if (!page) {
913                 ret = -ENOMEM;
914                 goto out_journal;
915         }
916
917         down_read(&EXT4_I(inode)->xattr_sem);
918         if (!ext4_has_inline_data(inode)) {
919                 ret = 0;
920                 goto out_release_page;
921         }
922
923         if (!PageUptodate(page)) {
924                 ret = ext4_read_inline_page(inode, page);
925                 if (ret < 0)
926                         goto out_release_page;
927         }
928         ret = ext4_journal_get_write_access(handle, iloc.bh);
929         if (ret)
930                 goto out_release_page;
931
932         up_read(&EXT4_I(inode)->xattr_sem);
933         *pagep = page;
934         brelse(iloc.bh);
935         return 1;
936 out_release_page:
937         up_read(&EXT4_I(inode)->xattr_sem);
938         unlock_page(page);
939         page_cache_release(page);
940 out_journal:
941         ext4_journal_stop(handle);
942 out:
943         brelse(iloc.bh);
944         return ret;
945 }
946
947 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
948                                   unsigned len, unsigned copied,
949                                   struct page *page)
950 {
951         int ret;
952
953         ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
954         if (ret < 0) {
955                 unlock_page(page);
956                 put_page(page);
957                 return ret;
958         }
959         copied = ret;
960
961         /*
962          * No need to use i_size_read() here, the i_size
963          * cannot change under us because we hold i_mutex.
964          *
965          * But it's important to update i_size while still holding page lock:
966          * page writeout could otherwise come in and zero beyond i_size.
967          */
968         if (pos+copied > inode->i_size)
969                 i_size_write(inode, pos+copied);
970         unlock_page(page);
971         page_cache_release(page);
972
973         /*
974          * Don't mark the inode dirty under page lock. First, it unnecessarily
975          * makes the holding time of page lock longer. Second, it forces lock
976          * ordering of page lock and transaction start for journaling
977          * filesystems.
978          */
979         mark_inode_dirty(inode);
980
981         return copied;
982 }
983
984 #ifdef INLINE_DIR_DEBUG
985 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
986                           void *inline_start, int inline_size)
987 {
988         int offset;
989         unsigned short de_len;
990         struct ext4_dir_entry_2 *de = inline_start;
991         void *dlimit = inline_start + inline_size;
992
993         trace_printk("inode %lu\n", dir->i_ino);
994         offset = 0;
995         while ((void *)de < dlimit) {
996                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
997                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
998                              offset, de_len, de->name_len, de->name,
999                              de->name_len, le32_to_cpu(de->inode));
1000                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1001                                          inline_start, inline_size, offset))
1002                         BUG();
1003
1004                 offset += de_len;
1005                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1006         }
1007 }
1008 #else
1009 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1010 #endif
1011
1012 /*
1013  * Add a new entry into a inline dir.
1014  * It will return -ENOSPC if no space is available, and -EIO
1015  * and -EEXIST if directory entry already exists.
1016  */
1017 static int ext4_add_dirent_to_inline(handle_t *handle,
1018                                      struct ext4_filename *fname,
1019                                      struct dentry *dentry,
1020                                      struct inode *inode,
1021                                      struct ext4_iloc *iloc,
1022                                      void *inline_start, int inline_size)
1023 {
1024         struct inode    *dir = d_inode(dentry->d_parent);
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 dentry *dentry, struct inode *inode)
1269 {
1270         int ret, inline_size, no_expand;
1271         void *inline_start;
1272         struct ext4_iloc iloc;
1273         struct inode *dir = d_inode(dentry->d_parent);
1274
1275         ret = ext4_get_inode_loc(dir, &iloc);
1276         if (ret)
1277                 return ret;
1278
1279         ext4_write_lock_xattr(dir, &no_expand);
1280         if (!ext4_has_inline_data(dir))
1281                 goto out;
1282
1283         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1284                                                  EXT4_INLINE_DOTDOT_SIZE;
1285         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1286
1287         ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1288                                         inline_start, inline_size);
1289         if (ret != -ENOSPC)
1290                 goto out;
1291
1292         /* check whether it can be inserted to inline xattr space. */
1293         inline_size = EXT4_I(dir)->i_inline_size -
1294                         EXT4_MIN_INLINE_DATA_SIZE;
1295         if (!inline_size) {
1296                 /* Try to use the xattr space.*/
1297                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1298                 if (ret && ret != -ENOSPC)
1299                         goto out;
1300
1301                 inline_size = EXT4_I(dir)->i_inline_size -
1302                                 EXT4_MIN_INLINE_DATA_SIZE;
1303         }
1304
1305         if (inline_size) {
1306                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1307
1308                 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1309                                                 inode, &iloc, inline_start,
1310                                                 inline_size);
1311
1312                 if (ret != -ENOSPC)
1313                         goto out;
1314         }
1315
1316         /*
1317          * The inline space is filled up, so create a new block for it.
1318          * As the extent tree will be created, we have to save the inline
1319          * dir first.
1320          */
1321         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1322
1323 out:
1324         ext4_mark_inode_dirty(handle, dir);
1325         ext4_write_unlock_xattr(dir, &no_expand);
1326         brelse(iloc.bh);
1327         return ret;
1328 }
1329
1330 /*
1331  * This function fills a red-black tree with information from an
1332  * inlined dir.  It returns the number directory entries loaded
1333  * into the tree.  If there is an error it is returned in err.
1334  */
1335 int htree_inlinedir_to_tree(struct file *dir_file,
1336                             struct inode *dir, ext4_lblk_t block,
1337                             struct dx_hash_info *hinfo,
1338                             __u32 start_hash, __u32 start_minor_hash,
1339                             int *has_inline_data)
1340 {
1341         int err = 0, count = 0;
1342         unsigned int parent_ino;
1343         int pos;
1344         struct ext4_dir_entry_2 *de;
1345         struct inode *inode = file_inode(dir_file);
1346         int ret, inline_size = 0;
1347         struct ext4_iloc iloc;
1348         void *dir_buf = NULL;
1349         struct ext4_dir_entry_2 fake;
1350         struct ext4_str tmp_str;
1351
1352         ret = ext4_get_inode_loc(inode, &iloc);
1353         if (ret)
1354                 return ret;
1355
1356         down_read(&EXT4_I(inode)->xattr_sem);
1357         if (!ext4_has_inline_data(inode)) {
1358                 up_read(&EXT4_I(inode)->xattr_sem);
1359                 *has_inline_data = 0;
1360                 goto out;
1361         }
1362
1363         inline_size = ext4_get_inline_size(inode);
1364         dir_buf = kmalloc(inline_size, GFP_NOFS);
1365         if (!dir_buf) {
1366                 ret = -ENOMEM;
1367                 up_read(&EXT4_I(inode)->xattr_sem);
1368                 goto out;
1369         }
1370
1371         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1372         up_read(&EXT4_I(inode)->xattr_sem);
1373         if (ret < 0)
1374                 goto out;
1375
1376         pos = 0;
1377         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1378         while (pos < inline_size) {
1379                 /*
1380                  * As inlined dir doesn't store any information about '.' and
1381                  * only the inode number of '..' is stored, we have to handle
1382                  * them differently.
1383                  */
1384                 if (pos == 0) {
1385                         fake.inode = cpu_to_le32(inode->i_ino);
1386                         fake.name_len = 1;
1387                         strcpy(fake.name, ".");
1388                         fake.rec_len = ext4_rec_len_to_disk(
1389                                                 EXT4_DIR_REC_LEN(fake.name_len),
1390                                                 inline_size);
1391                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1392                         de = &fake;
1393                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1394                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1395                         fake.inode = cpu_to_le32(parent_ino);
1396                         fake.name_len = 2;
1397                         strcpy(fake.name, "..");
1398                         fake.rec_len = ext4_rec_len_to_disk(
1399                                                 EXT4_DIR_REC_LEN(fake.name_len),
1400                                                 inline_size);
1401                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1402                         de = &fake;
1403                         pos = EXT4_INLINE_DOTDOT_SIZE;
1404                 } else {
1405                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1406                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1407                         if (ext4_check_dir_entry(inode, dir_file, de,
1408                                          iloc.bh, dir_buf,
1409                                          inline_size, pos)) {
1410                                 ret = count;
1411                                 goto out;
1412                         }
1413                 }
1414
1415                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1416                 if ((hinfo->hash < start_hash) ||
1417                     ((hinfo->hash == start_hash) &&
1418                      (hinfo->minor_hash < start_minor_hash)))
1419                         continue;
1420                 if (de->inode == 0)
1421                         continue;
1422                 tmp_str.name = de->name;
1423                 tmp_str.len = de->name_len;
1424                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1425                                               hinfo->minor_hash, de, &tmp_str);
1426                 if (err) {
1427                         ret = err;
1428                         goto out;
1429                 }
1430                 count++;
1431         }
1432         ret = count;
1433 out:
1434         kfree(dir_buf);
1435         brelse(iloc.bh);
1436         return ret;
1437 }
1438
1439 /*
1440  * So this function is called when the volume is mkfsed with
1441  * dir_index disabled. In order to keep f_pos persistent
1442  * after we convert from an inlined dir to a blocked based,
1443  * we just pretend that we are a normal dir and return the
1444  * offset as if '.' and '..' really take place.
1445  *
1446  */
1447 int ext4_read_inline_dir(struct file *file,
1448                          struct dir_context *ctx,
1449                          int *has_inline_data)
1450 {
1451         unsigned int offset, parent_ino;
1452         int i;
1453         struct ext4_dir_entry_2 *de;
1454         struct super_block *sb;
1455         struct inode *inode = file_inode(file);
1456         int ret, inline_size = 0;
1457         struct ext4_iloc iloc;
1458         void *dir_buf = NULL;
1459         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1460
1461         ret = ext4_get_inode_loc(inode, &iloc);
1462         if (ret)
1463                 return ret;
1464
1465         down_read(&EXT4_I(inode)->xattr_sem);
1466         if (!ext4_has_inline_data(inode)) {
1467                 up_read(&EXT4_I(inode)->xattr_sem);
1468                 *has_inline_data = 0;
1469                 goto out;
1470         }
1471
1472         inline_size = ext4_get_inline_size(inode);
1473         dir_buf = kmalloc(inline_size, GFP_NOFS);
1474         if (!dir_buf) {
1475                 ret = -ENOMEM;
1476                 up_read(&EXT4_I(inode)->xattr_sem);
1477                 goto out;
1478         }
1479
1480         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1481         up_read(&EXT4_I(inode)->xattr_sem);
1482         if (ret < 0)
1483                 goto out;
1484
1485         ret = 0;
1486         sb = inode->i_sb;
1487         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1488         offset = ctx->pos;
1489
1490         /*
1491          * dotdot_offset and dotdot_size is the real offset and
1492          * size for ".." and "." if the dir is block based while
1493          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1494          * So we will use extra_offset and extra_size to indicate them
1495          * during the inline dir iteration.
1496          */
1497         dotdot_offset = EXT4_DIR_REC_LEN(1);
1498         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1499         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1500         extra_size = extra_offset + inline_size;
1501
1502         /*
1503          * If the version has changed since the last call to
1504          * readdir(2), then we might be pointing to an invalid
1505          * dirent right now.  Scan from the start of the inline
1506          * dir to make sure.
1507          */
1508         if (file->f_version != inode->i_version) {
1509                 for (i = 0; i < extra_size && i < offset;) {
1510                         /*
1511                          * "." is with offset 0 and
1512                          * ".." is dotdot_offset.
1513                          */
1514                         if (!i) {
1515                                 i = dotdot_offset;
1516                                 continue;
1517                         } else if (i == dotdot_offset) {
1518                                 i = dotdot_size;
1519                                 continue;
1520                         }
1521                         /* for other entry, the real offset in
1522                          * the buf has to be tuned accordingly.
1523                          */
1524                         de = (struct ext4_dir_entry_2 *)
1525                                 (dir_buf + i - extra_offset);
1526                         /* It's too expensive to do a full
1527                          * dirent test each time round this
1528                          * loop, but we do have to test at
1529                          * least that it is non-zero.  A
1530                          * failure will be detected in the
1531                          * dirent test below. */
1532                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1533                                 < EXT4_DIR_REC_LEN(1))
1534                                 break;
1535                         i += ext4_rec_len_from_disk(de->rec_len,
1536                                                     extra_size);
1537                 }
1538                 offset = i;
1539                 ctx->pos = offset;
1540                 file->f_version = inode->i_version;
1541         }
1542
1543         while (ctx->pos < extra_size) {
1544                 if (ctx->pos == 0) {
1545                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1546                                 goto out;
1547                         ctx->pos = dotdot_offset;
1548                         continue;
1549                 }
1550
1551                 if (ctx->pos == dotdot_offset) {
1552                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1553                                 goto out;
1554                         ctx->pos = dotdot_size;
1555                         continue;
1556                 }
1557
1558                 de = (struct ext4_dir_entry_2 *)
1559                         (dir_buf + ctx->pos - extra_offset);
1560                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1561                                          extra_size, ctx->pos))
1562                         goto out;
1563                 if (le32_to_cpu(de->inode)) {
1564                         if (!dir_emit(ctx, de->name, de->name_len,
1565                                       le32_to_cpu(de->inode),
1566                                       get_dtype(sb, de->file_type)))
1567                                 goto out;
1568                 }
1569                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1570         }
1571 out:
1572         kfree(dir_buf);
1573         brelse(iloc.bh);
1574         return ret;
1575 }
1576
1577 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1578                                         struct ext4_dir_entry_2 **parent_de,
1579                                         int *retval)
1580 {
1581         struct ext4_iloc iloc;
1582
1583         *retval = ext4_get_inode_loc(inode, &iloc);
1584         if (*retval)
1585                 return NULL;
1586
1587         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1588
1589         return iloc.bh;
1590 }
1591
1592 /*
1593  * Try to create the inline data for the new dir.
1594  * If it succeeds, return 0, otherwise return the error.
1595  * In case of ENOSPC, the caller should create the normal disk layout dir.
1596  */
1597 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1598                                struct inode *inode)
1599 {
1600         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1601         struct ext4_iloc iloc;
1602         struct ext4_dir_entry_2 *de;
1603
1604         ret = ext4_get_inode_loc(inode, &iloc);
1605         if (ret)
1606                 return ret;
1607
1608         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1609         if (ret)
1610                 goto out;
1611
1612         /*
1613          * For inline dir, we only save the inode information for the ".."
1614          * and create a fake dentry to cover the left space.
1615          */
1616         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1617         de->inode = cpu_to_le32(parent->i_ino);
1618         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1619         de->inode = 0;
1620         de->rec_len = ext4_rec_len_to_disk(
1621                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1622                                 inline_size);
1623         set_nlink(inode, 2);
1624         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1625 out:
1626         brelse(iloc.bh);
1627         return ret;
1628 }
1629
1630 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1631                                         struct ext4_filename *fname,
1632                                         const struct qstr *d_name,
1633                                         struct ext4_dir_entry_2 **res_dir,
1634                                         int *has_inline_data)
1635 {
1636         int ret;
1637         struct ext4_iloc iloc;
1638         void *inline_start;
1639         int inline_size;
1640
1641         if (ext4_get_inode_loc(dir, &iloc))
1642                 return NULL;
1643
1644         down_read(&EXT4_I(dir)->xattr_sem);
1645         if (!ext4_has_inline_data(dir)) {
1646                 *has_inline_data = 0;
1647                 goto out;
1648         }
1649
1650         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1651                                                 EXT4_INLINE_DOTDOT_SIZE;
1652         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1653         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1654                               dir, fname, d_name, 0, res_dir);
1655         if (ret == 1)
1656                 goto out_find;
1657         if (ret < 0)
1658                 goto out;
1659
1660         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1661                 goto out;
1662
1663         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1664         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1665
1666         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1667                               dir, fname, d_name, 0, res_dir);
1668         if (ret == 1)
1669                 goto out_find;
1670
1671 out:
1672         brelse(iloc.bh);
1673         iloc.bh = NULL;
1674 out_find:
1675         up_read(&EXT4_I(dir)->xattr_sem);
1676         return iloc.bh;
1677 }
1678
1679 int ext4_delete_inline_entry(handle_t *handle,
1680                              struct inode *dir,
1681                              struct ext4_dir_entry_2 *de_del,
1682                              struct buffer_head *bh,
1683                              int *has_inline_data)
1684 {
1685         int err, inline_size, no_expand;
1686         struct ext4_iloc iloc;
1687         void *inline_start;
1688
1689         err = ext4_get_inode_loc(dir, &iloc);
1690         if (err)
1691                 return err;
1692
1693         ext4_write_lock_xattr(dir, &no_expand);
1694         if (!ext4_has_inline_data(dir)) {
1695                 *has_inline_data = 0;
1696                 goto out;
1697         }
1698
1699         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1700                 EXT4_MIN_INLINE_DATA_SIZE) {
1701                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1702                                         EXT4_INLINE_DOTDOT_SIZE;
1703                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1704                                 EXT4_INLINE_DOTDOT_SIZE;
1705         } else {
1706                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1707                 inline_size = ext4_get_inline_size(dir) -
1708                                 EXT4_MIN_INLINE_DATA_SIZE;
1709         }
1710
1711         BUFFER_TRACE(bh, "get_write_access");
1712         err = ext4_journal_get_write_access(handle, bh);
1713         if (err)
1714                 goto out;
1715
1716         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1717                                         inline_start, inline_size, 0);
1718         if (err)
1719                 goto out;
1720
1721         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1722         err = ext4_mark_inode_dirty(handle, dir);
1723         if (unlikely(err))
1724                 goto out;
1725
1726         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1727 out:
1728         ext4_write_unlock_xattr(dir, &no_expand);
1729         brelse(iloc.bh);
1730         if (err != -ENOENT)
1731                 ext4_std_error(dir->i_sb, err);
1732         return err;
1733 }
1734
1735 /*
1736  * Get the inline dentry at offset.
1737  */
1738 static inline struct ext4_dir_entry_2 *
1739 ext4_get_inline_entry(struct inode *inode,
1740                       struct ext4_iloc *iloc,
1741                       unsigned int offset,
1742                       void **inline_start,
1743                       int *inline_size)
1744 {
1745         void *inline_pos;
1746
1747         BUG_ON(offset > ext4_get_inline_size(inode));
1748
1749         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1750                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1751                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1752         } else {
1753                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1754                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1755                 *inline_size = ext4_get_inline_size(inode) -
1756                                 EXT4_MIN_INLINE_DATA_SIZE;
1757         }
1758
1759         if (inline_start)
1760                 *inline_start = inline_pos;
1761         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1762 }
1763
1764 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1765 {
1766         int err, inline_size;
1767         struct ext4_iloc iloc;
1768         size_t inline_len;
1769         void *inline_pos;
1770         unsigned int offset;
1771         struct ext4_dir_entry_2 *de;
1772         int ret = 1;
1773
1774         err = ext4_get_inode_loc(dir, &iloc);
1775         if (err) {
1776                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1777                                  err, dir->i_ino);
1778                 return 1;
1779         }
1780
1781         down_read(&EXT4_I(dir)->xattr_sem);
1782         if (!ext4_has_inline_data(dir)) {
1783                 *has_inline_data = 0;
1784                 goto out;
1785         }
1786
1787         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1788         if (!le32_to_cpu(de->inode)) {
1789                 ext4_warning(dir->i_sb,
1790                              "bad inline directory (dir #%lu) - no `..'",
1791                              dir->i_ino);
1792                 ret = 1;
1793                 goto out;
1794         }
1795
1796         inline_len = ext4_get_inline_size(dir);
1797         offset = EXT4_INLINE_DOTDOT_SIZE;
1798         while (offset < inline_len) {
1799                 de = ext4_get_inline_entry(dir, &iloc, offset,
1800                                            &inline_pos, &inline_size);
1801                 if (ext4_check_dir_entry(dir, NULL, de,
1802                                          iloc.bh, inline_pos,
1803                                          inline_size, offset)) {
1804                         ext4_warning(dir->i_sb,
1805                                      "bad inline directory (dir #%lu) - "
1806                                      "inode %u, rec_len %u, name_len %d"
1807                                      "inline size %d\n",
1808                                      dir->i_ino, le32_to_cpu(de->inode),
1809                                      le16_to_cpu(de->rec_len), de->name_len,
1810                                      inline_size);
1811                         ret = 1;
1812                         goto out;
1813                 }
1814                 if (le32_to_cpu(de->inode)) {
1815                         ret = 0;
1816                         goto out;
1817                 }
1818                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1819         }
1820
1821 out:
1822         up_read(&EXT4_I(dir)->xattr_sem);
1823         brelse(iloc.bh);
1824         return ret;
1825 }
1826
1827 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1828 {
1829         int ret, no_expand;
1830
1831         ext4_write_lock_xattr(inode, &no_expand);
1832         ret = ext4_destroy_inline_data_nolock(handle, inode);
1833         ext4_write_unlock_xattr(inode, &no_expand);
1834
1835         return ret;
1836 }
1837
1838 int ext4_inline_data_fiemap(struct inode *inode,
1839                             struct fiemap_extent_info *fieinfo,
1840                             int *has_inline, __u64 start, __u64 len)
1841 {
1842         __u64 physical = 0;
1843         __u64 inline_len;
1844         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1845                 FIEMAP_EXTENT_LAST;
1846         int error = 0;
1847         struct ext4_iloc iloc;
1848
1849         down_read(&EXT4_I(inode)->xattr_sem);
1850         if (!ext4_has_inline_data(inode)) {
1851                 *has_inline = 0;
1852                 goto out;
1853         }
1854         inline_len = min_t(size_t, ext4_get_inline_size(inode),
1855                            i_size_read(inode));
1856         if (start >= inline_len)
1857                 goto out;
1858         if (start + len < inline_len)
1859                 inline_len = start + len;
1860         inline_len -= start;
1861
1862         error = ext4_get_inode_loc(inode, &iloc);
1863         if (error)
1864                 goto out;
1865
1866         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1867         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1868         physical += offsetof(struct ext4_inode, i_block);
1869
1870         brelse(iloc.bh);
1871 out:
1872         up_read(&EXT4_I(inode)->xattr_sem);
1873         if (physical)
1874                 error = fiemap_fill_next_extent(fieinfo, start, physical,
1875                                                 inline_len, flags);
1876         return (error < 0 ? error : 0);
1877 }
1878
1879 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1880 {
1881         handle_t *handle;
1882         int inline_size, value_len, needed_blocks, no_expand;
1883         size_t i_size;
1884         void *value = NULL;
1885         struct ext4_xattr_ibody_find is = {
1886                 .s = { .not_found = -ENODATA, },
1887         };
1888         struct ext4_xattr_info i = {
1889                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1890                 .name = EXT4_XATTR_SYSTEM_DATA,
1891         };
1892
1893
1894         needed_blocks = ext4_writepage_trans_blocks(inode);
1895         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1896         if (IS_ERR(handle))
1897                 return;
1898
1899         ext4_write_lock_xattr(inode, &no_expand);
1900         if (!ext4_has_inline_data(inode)) {
1901                 ext4_write_unlock_xattr(inode, &no_expand);
1902                 *has_inline = 0;
1903                 ext4_journal_stop(handle);
1904                 return;
1905         }
1906
1907         if (ext4_orphan_add(handle, inode))
1908                 goto out;
1909
1910         if (ext4_get_inode_loc(inode, &is.iloc))
1911                 goto out;
1912
1913         down_write(&EXT4_I(inode)->i_data_sem);
1914         i_size = inode->i_size;
1915         inline_size = ext4_get_inline_size(inode);
1916         EXT4_I(inode)->i_disksize = i_size;
1917
1918         if (i_size < inline_size) {
1919                 /* Clear the content in the xattr space. */
1920                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1921                         if (ext4_xattr_ibody_find(inode, &i, &is))
1922                                 goto out_error;
1923
1924                         BUG_ON(is.s.not_found);
1925
1926                         value_len = le32_to_cpu(is.s.here->e_value_size);
1927                         value = kmalloc(value_len, GFP_NOFS);
1928                         if (!value)
1929                                 goto out_error;
1930
1931                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1932                                                 value, value_len))
1933                                 goto out_error;
1934
1935                         i.value = value;
1936                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1937                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1938                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1939                                 goto out_error;
1940                 }
1941
1942                 /* Clear the content within i_blocks. */
1943                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1944                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1945                         memset(p + i_size, 0,
1946                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1947                 }
1948
1949                 EXT4_I(inode)->i_inline_size = i_size <
1950                                         EXT4_MIN_INLINE_DATA_SIZE ?
1951                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1952         }
1953
1954 out_error:
1955         up_write(&EXT4_I(inode)->i_data_sem);
1956 out:
1957         brelse(is.iloc.bh);
1958         ext4_write_unlock_xattr(inode, &no_expand);
1959         kfree(value);
1960         if (inode->i_nlink)
1961                 ext4_orphan_del(handle, inode);
1962
1963         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1964         ext4_mark_inode_dirty(handle, inode);
1965         if (IS_SYNC(inode))
1966                 ext4_handle_sync(handle);
1967
1968         ext4_journal_stop(handle);
1969         return;
1970 }
1971
1972 int ext4_convert_inline_data(struct inode *inode)
1973 {
1974         int error, needed_blocks, no_expand;
1975         handle_t *handle;
1976         struct ext4_iloc iloc;
1977
1978         if (!ext4_has_inline_data(inode)) {
1979                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1980                 return 0;
1981         }
1982
1983         needed_blocks = ext4_writepage_trans_blocks(inode);
1984
1985         iloc.bh = NULL;
1986         error = ext4_get_inode_loc(inode, &iloc);
1987         if (error)
1988                 return error;
1989
1990         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1991         if (IS_ERR(handle)) {
1992                 error = PTR_ERR(handle);
1993                 goto out_free;
1994         }
1995
1996         ext4_write_lock_xattr(inode, &no_expand);
1997         if (ext4_has_inline_data(inode))
1998                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1999         ext4_write_unlock_xattr(inode, &no_expand);
2000         ext4_journal_stop(handle);
2001 out_free:
2002         brelse(iloc.bh);
2003         return error;
2004 }