Merge branch 'akpm' (patches from Andrew)
[cascardo/linux.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                 goto out;
341
342         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
343                                      value, len);
344         if (error == -ENODATA)
345                 goto out;
346
347         BUFFER_TRACE(is.iloc.bh, "get_write_access");
348         error = ext4_journal_get_write_access(handle, is.iloc.bh);
349         if (error)
350                 goto out;
351
352         /* Update the xttr entry. */
353         i.value = value;
354         i.value_len = len;
355
356         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
357         if (error)
358                 goto out;
359
360         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
361                                       (void *)ext4_raw_inode(&is.iloc));
362         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
363                                 le32_to_cpu(is.s.here->e_value_size);
364         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
365         get_bh(is.iloc.bh);
366         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
367
368 out:
369         kfree(value);
370         brelse(is.iloc.bh);
371         return error;
372 }
373
374 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
375                                     unsigned int len)
376 {
377         int ret, size;
378         struct ext4_inode_info *ei = EXT4_I(inode);
379
380         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
381                 return -ENOSPC;
382
383         size = ext4_get_max_inline_size(inode);
384         if (size < len)
385                 return -ENOSPC;
386
387         down_write(&EXT4_I(inode)->xattr_sem);
388
389         if (ei->i_inline_off)
390                 ret = ext4_update_inline_data(handle, inode, len);
391         else
392                 ret = ext4_create_inline_data(handle, inode, len);
393
394         up_write(&EXT4_I(inode)->xattr_sem);
395
396         return ret;
397 }
398
399 static int ext4_destroy_inline_data_nolock(handle_t *handle,
400                                            struct inode *inode)
401 {
402         struct ext4_inode_info *ei = EXT4_I(inode);
403         struct ext4_xattr_ibody_find is = {
404                 .s = { .not_found = 0, },
405         };
406         struct ext4_xattr_info i = {
407                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
408                 .name = EXT4_XATTR_SYSTEM_DATA,
409                 .value = NULL,
410                 .value_len = 0,
411         };
412         int error;
413
414         if (!ei->i_inline_off)
415                 return 0;
416
417         error = ext4_get_inode_loc(inode, &is.iloc);
418         if (error)
419                 return error;
420
421         error = ext4_xattr_ibody_find(inode, &i, &is);
422         if (error)
423                 goto out;
424
425         BUFFER_TRACE(is.iloc.bh, "get_write_access");
426         error = ext4_journal_get_write_access(handle, is.iloc.bh);
427         if (error)
428                 goto out;
429
430         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
431         if (error)
432                 goto out;
433
434         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
435                 0, EXT4_MIN_INLINE_DATA_SIZE);
436
437         if (ext4_has_feature_extents(inode->i_sb)) {
438                 if (S_ISDIR(inode->i_mode) ||
439                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
440                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
441                         ext4_ext_tree_init(handle, inode);
442                 }
443         }
444         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
445
446         get_bh(is.iloc.bh);
447         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
448
449         EXT4_I(inode)->i_inline_off = 0;
450         EXT4_I(inode)->i_inline_size = 0;
451         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
452 out:
453         brelse(is.iloc.bh);
454         if (error == -ENODATA)
455                 error = 0;
456         return error;
457 }
458
459 static int ext4_read_inline_page(struct inode *inode, struct page *page)
460 {
461         void *kaddr;
462         int ret = 0;
463         size_t len;
464         struct ext4_iloc iloc;
465
466         BUG_ON(!PageLocked(page));
467         BUG_ON(!ext4_has_inline_data(inode));
468         BUG_ON(page->index);
469
470         if (!EXT4_I(inode)->i_inline_off) {
471                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
472                              inode->i_ino);
473                 goto out;
474         }
475
476         ret = ext4_get_inode_loc(inode, &iloc);
477         if (ret)
478                 goto out;
479
480         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
481         kaddr = kmap_atomic(page);
482         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
483         flush_dcache_page(page);
484         kunmap_atomic(kaddr);
485         zero_user_segment(page, len, PAGE_SIZE);
486         SetPageUptodate(page);
487         brelse(iloc.bh);
488
489 out:
490         return ret;
491 }
492
493 int ext4_readpage_inline(struct inode *inode, struct page *page)
494 {
495         int ret = 0;
496
497         down_read(&EXT4_I(inode)->xattr_sem);
498         if (!ext4_has_inline_data(inode)) {
499                 up_read(&EXT4_I(inode)->xattr_sem);
500                 return -EAGAIN;
501         }
502
503         /*
504          * Current inline data can only exist in the 1st page,
505          * So for all the other pages, just set them uptodate.
506          */
507         if (!page->index)
508                 ret = ext4_read_inline_page(inode, page);
509         else if (!PageUptodate(page)) {
510                 zero_user_segment(page, 0, PAGE_SIZE);
511                 SetPageUptodate(page);
512         }
513
514         up_read(&EXT4_I(inode)->xattr_sem);
515
516         unlock_page(page);
517         return ret >= 0 ? 0 : ret;
518 }
519
520 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
521                                               struct inode *inode,
522                                               unsigned flags)
523 {
524         int ret, needed_blocks;
525         handle_t *handle = NULL;
526         int retries = 0, sem_held = 0;
527         struct page *page = NULL;
528         unsigned from, to;
529         struct ext4_iloc iloc;
530
531         if (!ext4_has_inline_data(inode)) {
532                 /*
533                  * clear the flag so that no new write
534                  * will trap here again.
535                  */
536                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
537                 return 0;
538         }
539
540         needed_blocks = ext4_writepage_trans_blocks(inode);
541
542         ret = ext4_get_inode_loc(inode, &iloc);
543         if (ret)
544                 return ret;
545
546 retry:
547         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
548         if (IS_ERR(handle)) {
549                 ret = PTR_ERR(handle);
550                 handle = NULL;
551                 goto out;
552         }
553
554         /* We cannot recurse into the filesystem as the transaction is already
555          * started */
556         flags |= AOP_FLAG_NOFS;
557
558         page = grab_cache_page_write_begin(mapping, 0, flags);
559         if (!page) {
560                 ret = -ENOMEM;
561                 goto out;
562         }
563
564         down_write(&EXT4_I(inode)->xattr_sem);
565         sem_held = 1;
566         /* If some one has already done this for us, just exit. */
567         if (!ext4_has_inline_data(inode)) {
568                 ret = 0;
569                 goto out;
570         }
571
572         from = 0;
573         to = ext4_get_inline_size(inode);
574         if (!PageUptodate(page)) {
575                 ret = ext4_read_inline_page(inode, page);
576                 if (ret < 0)
577                         goto out;
578         }
579
580         ret = ext4_destroy_inline_data_nolock(handle, inode);
581         if (ret)
582                 goto out;
583
584         if (ext4_should_dioread_nolock(inode)) {
585                 ret = __block_write_begin(page, from, to,
586                                           ext4_get_block_unwritten);
587         } else
588                 ret = __block_write_begin(page, from, to, ext4_get_block);
589
590         if (!ret && ext4_should_journal_data(inode)) {
591                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
592                                              from, to, NULL,
593                                              do_journal_get_write_access);
594         }
595
596         if (ret) {
597                 unlock_page(page);
598                 put_page(page);
599                 page = NULL;
600                 ext4_orphan_add(handle, inode);
601                 up_write(&EXT4_I(inode)->xattr_sem);
602                 sem_held = 0;
603                 ext4_journal_stop(handle);
604                 handle = NULL;
605                 ext4_truncate_failed_write(inode);
606                 /*
607                  * If truncate failed early the inode might
608                  * still be on the orphan list; we need to
609                  * make sure the inode is removed from the
610                  * orphan list in that case.
611                  */
612                 if (inode->i_nlink)
613                         ext4_orphan_del(NULL, inode);
614         }
615
616         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617                 goto retry;
618
619         if (page)
620                 block_commit_write(page, from, to);
621 out:
622         if (page) {
623                 unlock_page(page);
624                 put_page(page);
625         }
626         if (sem_held)
627                 up_write(&EXT4_I(inode)->xattr_sem);
628         if (handle)
629                 ext4_journal_stop(handle);
630         brelse(iloc.bh);
631         return ret;
632 }
633
634 /*
635  * Try to write data in the inode.
636  * If the inode has inline data, check whether the new write can be
637  * in the inode also. If not, create the page the handle, move the data
638  * to the page make it update and let the later codes create extent for it.
639  */
640 int ext4_try_to_write_inline_data(struct address_space *mapping,
641                                   struct inode *inode,
642                                   loff_t pos, unsigned len,
643                                   unsigned flags,
644                                   struct page **pagep)
645 {
646         int ret;
647         handle_t *handle;
648         struct page *page;
649         struct ext4_iloc iloc;
650
651         if (pos + len > ext4_get_max_inline_size(inode))
652                 goto convert;
653
654         ret = ext4_get_inode_loc(inode, &iloc);
655         if (ret)
656                 return ret;
657
658         /*
659          * The possible write could happen in the inode,
660          * so try to reserve the space in inode first.
661          */
662         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
663         if (IS_ERR(handle)) {
664                 ret = PTR_ERR(handle);
665                 handle = NULL;
666                 goto out;
667         }
668
669         ret = ext4_prepare_inline_data(handle, inode, pos + len);
670         if (ret && ret != -ENOSPC)
671                 goto out;
672
673         /* We don't have space in inline inode, so convert it to extent. */
674         if (ret == -ENOSPC) {
675                 ext4_journal_stop(handle);
676                 brelse(iloc.bh);
677                 goto convert;
678         }
679
680         flags |= AOP_FLAG_NOFS;
681
682         page = grab_cache_page_write_begin(mapping, 0, flags);
683         if (!page) {
684                 ret = -ENOMEM;
685                 goto out;
686         }
687
688         *pagep = page;
689         down_read(&EXT4_I(inode)->xattr_sem);
690         if (!ext4_has_inline_data(inode)) {
691                 ret = 0;
692                 unlock_page(page);
693                 put_page(page);
694                 goto out_up_read;
695         }
696
697         if (!PageUptodate(page)) {
698                 ret = ext4_read_inline_page(inode, page);
699                 if (ret < 0)
700                         goto out_up_read;
701         }
702
703         ret = 1;
704         handle = NULL;
705 out_up_read:
706         up_read(&EXT4_I(inode)->xattr_sem);
707 out:
708         if (handle)
709                 ext4_journal_stop(handle);
710         brelse(iloc.bh);
711         return ret;
712 convert:
713         return ext4_convert_inline_data_to_extent(mapping,
714                                                   inode, flags);
715 }
716
717 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
718                                unsigned copied, struct page *page)
719 {
720         int ret;
721         void *kaddr;
722         struct ext4_iloc iloc;
723
724         if (unlikely(copied < len)) {
725                 if (!PageUptodate(page)) {
726                         copied = 0;
727                         goto out;
728                 }
729         }
730
731         ret = ext4_get_inode_loc(inode, &iloc);
732         if (ret) {
733                 ext4_std_error(inode->i_sb, ret);
734                 copied = 0;
735                 goto out;
736         }
737
738         down_write(&EXT4_I(inode)->xattr_sem);
739         BUG_ON(!ext4_has_inline_data(inode));
740
741         kaddr = kmap_atomic(page);
742         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
743         kunmap_atomic(kaddr);
744         SetPageUptodate(page);
745         /* clear page dirty so that writepages wouldn't work for us. */
746         ClearPageDirty(page);
747
748         up_write(&EXT4_I(inode)->xattr_sem);
749         brelse(iloc.bh);
750 out:
751         return copied;
752 }
753
754 struct buffer_head *
755 ext4_journalled_write_inline_data(struct inode *inode,
756                                   unsigned len,
757                                   struct page *page)
758 {
759         int ret;
760         void *kaddr;
761         struct ext4_iloc iloc;
762
763         ret = ext4_get_inode_loc(inode, &iloc);
764         if (ret) {
765                 ext4_std_error(inode->i_sb, ret);
766                 return NULL;
767         }
768
769         down_write(&EXT4_I(inode)->xattr_sem);
770         kaddr = kmap_atomic(page);
771         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
772         kunmap_atomic(kaddr);
773         up_write(&EXT4_I(inode)->xattr_sem);
774
775         return iloc.bh;
776 }
777
778 /*
779  * Try to make the page cache and handle ready for the inline data case.
780  * We can call this function in 2 cases:
781  * 1. The inode is created and the first write exceeds inline size. We can
782  *    clear the inode state safely.
783  * 2. The inode has inline data, then we need to read the data, make it
784  *    update and dirty so that ext4_da_writepages can handle it. We don't
785  *    need to start the journal since the file's metatdata isn't changed now.
786  */
787 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
788                                                  struct inode *inode,
789                                                  unsigned flags,
790                                                  void **fsdata)
791 {
792         int ret = 0, inline_size;
793         struct page *page;
794
795         page = grab_cache_page_write_begin(mapping, 0, flags);
796         if (!page)
797                 return -ENOMEM;
798
799         down_read(&EXT4_I(inode)->xattr_sem);
800         if (!ext4_has_inline_data(inode)) {
801                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
802                 goto out;
803         }
804
805         inline_size = ext4_get_inline_size(inode);
806
807         if (!PageUptodate(page)) {
808                 ret = ext4_read_inline_page(inode, page);
809                 if (ret < 0)
810                         goto out;
811         }
812
813         ret = __block_write_begin(page, 0, inline_size,
814                                   ext4_da_get_block_prep);
815         if (ret) {
816                 up_read(&EXT4_I(inode)->xattr_sem);
817                 unlock_page(page);
818                 put_page(page);
819                 ext4_truncate_failed_write(inode);
820                 return ret;
821         }
822
823         SetPageDirty(page);
824         SetPageUptodate(page);
825         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
826         *fsdata = (void *)CONVERT_INLINE_DATA;
827
828 out:
829         up_read(&EXT4_I(inode)->xattr_sem);
830         if (page) {
831                 unlock_page(page);
832                 put_page(page);
833         }
834         return ret;
835 }
836
837 /*
838  * Prepare the write for the inline data.
839  * If the the data can be written into the inode, we just read
840  * the page and make it uptodate, and start the journal.
841  * Otherwise read the page, makes it dirty so that it can be
842  * handle in writepages(the i_disksize update is left to the
843  * normal ext4_da_write_end).
844  */
845 int ext4_da_write_inline_data_begin(struct address_space *mapping,
846                                     struct inode *inode,
847                                     loff_t pos, unsigned len,
848                                     unsigned flags,
849                                     struct page **pagep,
850                                     void **fsdata)
851 {
852         int ret, inline_size;
853         handle_t *handle;
854         struct page *page;
855         struct ext4_iloc iloc;
856         int retries;
857
858         ret = ext4_get_inode_loc(inode, &iloc);
859         if (ret)
860                 return ret;
861
862 retry_journal:
863         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
864         if (IS_ERR(handle)) {
865                 ret = PTR_ERR(handle);
866                 goto out;
867         }
868
869         inline_size = ext4_get_max_inline_size(inode);
870
871         ret = -ENOSPC;
872         if (inline_size >= pos + len) {
873                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
874                 if (ret && ret != -ENOSPC)
875                         goto out_journal;
876         }
877
878         /*
879          * We cannot recurse into the filesystem as the transaction
880          * is already started.
881          */
882         flags |= AOP_FLAG_NOFS;
883
884         if (ret == -ENOSPC) {
885                 ret = ext4_da_convert_inline_data_to_extent(mapping,
886                                                             inode,
887                                                             flags,
888                                                             fsdata);
889                 ext4_journal_stop(handle);
890                 if (ret == -ENOSPC &&
891                     ext4_should_retry_alloc(inode->i_sb, &retries))
892                         goto retry_journal;
893                 goto out;
894         }
895
896
897         page = grab_cache_page_write_begin(mapping, 0, flags);
898         if (!page) {
899                 ret = -ENOMEM;
900                 goto out_journal;
901         }
902
903         down_read(&EXT4_I(inode)->xattr_sem);
904         if (!ext4_has_inline_data(inode)) {
905                 ret = 0;
906                 goto out_release_page;
907         }
908
909         if (!PageUptodate(page)) {
910                 ret = ext4_read_inline_page(inode, page);
911                 if (ret < 0)
912                         goto out_release_page;
913         }
914
915         up_read(&EXT4_I(inode)->xattr_sem);
916         *pagep = page;
917         brelse(iloc.bh);
918         return 1;
919 out_release_page:
920         up_read(&EXT4_I(inode)->xattr_sem);
921         unlock_page(page);
922         put_page(page);
923 out_journal:
924         ext4_journal_stop(handle);
925 out:
926         brelse(iloc.bh);
927         return ret;
928 }
929
930 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
931                                   unsigned len, unsigned copied,
932                                   struct page *page)
933 {
934         int i_size_changed = 0;
935
936         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
937
938         /*
939          * No need to use i_size_read() here, the i_size
940          * cannot change under us because we hold i_mutex.
941          *
942          * But it's important to update i_size while still holding page lock:
943          * page writeout could otherwise come in and zero beyond i_size.
944          */
945         if (pos+copied > inode->i_size) {
946                 i_size_write(inode, pos+copied);
947                 i_size_changed = 1;
948         }
949         unlock_page(page);
950         put_page(page);
951
952         /*
953          * Don't mark the inode dirty under page lock. First, it unnecessarily
954          * makes the holding time of page lock longer. Second, it forces lock
955          * ordering of page lock and transaction start for journaling
956          * filesystems.
957          */
958         if (i_size_changed)
959                 mark_inode_dirty(inode);
960
961         return copied;
962 }
963
964 #ifdef INLINE_DIR_DEBUG
965 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
966                           void *inline_start, int inline_size)
967 {
968         int offset;
969         unsigned short de_len;
970         struct ext4_dir_entry_2 *de = inline_start;
971         void *dlimit = inline_start + inline_size;
972
973         trace_printk("inode %lu\n", dir->i_ino);
974         offset = 0;
975         while ((void *)de < dlimit) {
976                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
977                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
978                              offset, de_len, de->name_len, de->name,
979                              de->name_len, le32_to_cpu(de->inode));
980                 if (ext4_check_dir_entry(dir, NULL, de, bh,
981                                          inline_start, inline_size, offset))
982                         BUG();
983
984                 offset += de_len;
985                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
986         }
987 }
988 #else
989 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
990 #endif
991
992 /*
993  * Add a new entry into a inline dir.
994  * It will return -ENOSPC if no space is available, and -EIO
995  * and -EEXIST if directory entry already exists.
996  */
997 static int ext4_add_dirent_to_inline(handle_t *handle,
998                                      struct ext4_filename *fname,
999                                      struct inode *dir,
1000                                      struct inode *inode,
1001                                      struct ext4_iloc *iloc,
1002                                      void *inline_start, int inline_size)
1003 {
1004         int             err;
1005         struct ext4_dir_entry_2 *de;
1006
1007         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1008                                 inline_size, fname, &de);
1009         if (err)
1010                 return err;
1011
1012         BUFFER_TRACE(iloc->bh, "get_write_access");
1013         err = ext4_journal_get_write_access(handle, iloc->bh);
1014         if (err)
1015                 return err;
1016         ext4_insert_dentry(dir, inode, de, inline_size, fname);
1017
1018         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1019
1020         /*
1021          * XXX shouldn't update any times until successful
1022          * completion of syscall, but too many callers depend
1023          * on this.
1024          *
1025          * XXX similarly, too many callers depend on
1026          * ext4_new_inode() setting the times, but error
1027          * recovery deletes the inode, so the worst that can
1028          * happen is that the times are slightly out of date
1029          * and/or different from the directory change time.
1030          */
1031         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1032         ext4_update_dx_flag(dir);
1033         dir->i_version++;
1034         ext4_mark_inode_dirty(handle, dir);
1035         return 1;
1036 }
1037
1038 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1039                                        struct ext4_iloc *iloc)
1040 {
1041         struct ext4_xattr_entry *entry;
1042         struct ext4_xattr_ibody_header *header;
1043
1044         BUG_ON(!EXT4_I(inode)->i_inline_off);
1045
1046         header = IHDR(inode, ext4_raw_inode(iloc));
1047         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1048                                             EXT4_I(inode)->i_inline_off);
1049
1050         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1051 }
1052
1053 /* Set the final de to cover the whole block. */
1054 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1055 {
1056         struct ext4_dir_entry_2 *de, *prev_de;
1057         void *limit;
1058         int de_len;
1059
1060         de = (struct ext4_dir_entry_2 *)de_buf;
1061         if (old_size) {
1062                 limit = de_buf + old_size;
1063                 do {
1064                         prev_de = de;
1065                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1066                         de_buf += de_len;
1067                         de = (struct ext4_dir_entry_2 *)de_buf;
1068                 } while (de_buf < limit);
1069
1070                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1071                                                         old_size, new_size);
1072         } else {
1073                 /* this is just created, so create an empty entry. */
1074                 de->inode = 0;
1075                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1076         }
1077 }
1078
1079 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1080                                   struct ext4_iloc *iloc)
1081 {
1082         int ret;
1083         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1084         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1085
1086         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1087                 return -ENOSPC;
1088
1089         ret = ext4_update_inline_data(handle, dir,
1090                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1091         if (ret)
1092                 return ret;
1093
1094         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1095                              EXT4_I(dir)->i_inline_size -
1096                                                 EXT4_MIN_INLINE_DATA_SIZE);
1097         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1098         return 0;
1099 }
1100
1101 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1102                                      struct ext4_iloc *iloc,
1103                                      void *buf, int inline_size)
1104 {
1105         ext4_create_inline_data(handle, inode, inline_size);
1106         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1107         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1108 }
1109
1110 static int ext4_finish_convert_inline_dir(handle_t *handle,
1111                                           struct inode *inode,
1112                                           struct buffer_head *dir_block,
1113                                           void *buf,
1114                                           int inline_size)
1115 {
1116         int err, csum_size = 0, header_size = 0;
1117         struct ext4_dir_entry_2 *de;
1118         struct ext4_dir_entry_tail *t;
1119         void *target = dir_block->b_data;
1120
1121         /*
1122          * First create "." and ".." and then copy the dir information
1123          * back to the block.
1124          */
1125         de = (struct ext4_dir_entry_2 *)target;
1126         de = ext4_init_dot_dotdot(inode, de,
1127                 inode->i_sb->s_blocksize, csum_size,
1128                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1129         header_size = (void *)de - target;
1130
1131         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1132                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1133
1134         if (ext4_has_metadata_csum(inode->i_sb))
1135                 csum_size = sizeof(struct ext4_dir_entry_tail);
1136
1137         inode->i_size = inode->i_sb->s_blocksize;
1138         i_size_write(inode, inode->i_sb->s_blocksize);
1139         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1140         ext4_update_final_de(dir_block->b_data,
1141                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1142                         inode->i_sb->s_blocksize - csum_size);
1143
1144         if (csum_size) {
1145                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1146                                      inode->i_sb->s_blocksize);
1147                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1148         }
1149         set_buffer_uptodate(dir_block);
1150         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1151         if (err)
1152                 goto out;
1153         set_buffer_verified(dir_block);
1154 out:
1155         return err;
1156 }
1157
1158 static int ext4_convert_inline_data_nolock(handle_t *handle,
1159                                            struct inode *inode,
1160                                            struct ext4_iloc *iloc)
1161 {
1162         int error;
1163         void *buf = NULL;
1164         struct buffer_head *data_bh = NULL;
1165         struct ext4_map_blocks map;
1166         int inline_size;
1167
1168         inline_size = ext4_get_inline_size(inode);
1169         buf = kmalloc(inline_size, GFP_NOFS);
1170         if (!buf) {
1171                 error = -ENOMEM;
1172                 goto out;
1173         }
1174
1175         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1176         if (error < 0)
1177                 goto out;
1178
1179         /*
1180          * Make sure the inline directory entries pass checks before we try to
1181          * convert them, so that we avoid touching stuff that needs fsck.
1182          */
1183         if (S_ISDIR(inode->i_mode)) {
1184                 error = ext4_check_all_de(inode, iloc->bh,
1185                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1186                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1187                 if (error)
1188                         goto out;
1189         }
1190
1191         error = ext4_destroy_inline_data_nolock(handle, inode);
1192         if (error)
1193                 goto out;
1194
1195         map.m_lblk = 0;
1196         map.m_len = 1;
1197         map.m_flags = 0;
1198         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1199         if (error < 0)
1200                 goto out_restore;
1201         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1202                 error = -EIO;
1203                 goto out_restore;
1204         }
1205
1206         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1207         if (!data_bh) {
1208                 error = -ENOMEM;
1209                 goto out_restore;
1210         }
1211
1212         lock_buffer(data_bh);
1213         error = ext4_journal_get_create_access(handle, data_bh);
1214         if (error) {
1215                 unlock_buffer(data_bh);
1216                 error = -EIO;
1217                 goto out_restore;
1218         }
1219         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1220
1221         if (!S_ISDIR(inode->i_mode)) {
1222                 memcpy(data_bh->b_data, buf, inline_size);
1223                 set_buffer_uptodate(data_bh);
1224                 error = ext4_handle_dirty_metadata(handle,
1225                                                    inode, data_bh);
1226         } else {
1227                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1228                                                        buf, inline_size);
1229         }
1230
1231         unlock_buffer(data_bh);
1232 out_restore:
1233         if (error)
1234                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1235
1236 out:
1237         brelse(data_bh);
1238         kfree(buf);
1239         return error;
1240 }
1241
1242 /*
1243  * Try to add the new entry to the inline data.
1244  * If succeeds, return 0. If not, extended the inline dir and copied data to
1245  * the new created block.
1246  */
1247 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1248                               struct inode *dir, struct inode *inode)
1249 {
1250         int ret, inline_size;
1251         void *inline_start;
1252         struct ext4_iloc iloc;
1253
1254         ret = ext4_get_inode_loc(dir, &iloc);
1255         if (ret)
1256                 return ret;
1257
1258         down_write(&EXT4_I(dir)->xattr_sem);
1259         if (!ext4_has_inline_data(dir))
1260                 goto out;
1261
1262         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1263                                                  EXT4_INLINE_DOTDOT_SIZE;
1264         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1265
1266         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1267                                         inline_start, inline_size);
1268         if (ret != -ENOSPC)
1269                 goto out;
1270
1271         /* check whether it can be inserted to inline xattr space. */
1272         inline_size = EXT4_I(dir)->i_inline_size -
1273                         EXT4_MIN_INLINE_DATA_SIZE;
1274         if (!inline_size) {
1275                 /* Try to use the xattr space.*/
1276                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1277                 if (ret && ret != -ENOSPC)
1278                         goto out;
1279
1280                 inline_size = EXT4_I(dir)->i_inline_size -
1281                                 EXT4_MIN_INLINE_DATA_SIZE;
1282         }
1283
1284         if (inline_size) {
1285                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1286
1287                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1288                                                 inode, &iloc, inline_start,
1289                                                 inline_size);
1290
1291                 if (ret != -ENOSPC)
1292                         goto out;
1293         }
1294
1295         /*
1296          * The inline space is filled up, so create a new block for it.
1297          * As the extent tree will be created, we have to save the inline
1298          * dir first.
1299          */
1300         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1301
1302 out:
1303         ext4_mark_inode_dirty(handle, dir);
1304         up_write(&EXT4_I(dir)->xattr_sem);
1305         brelse(iloc.bh);
1306         return ret;
1307 }
1308
1309 /*
1310  * This function fills a red-black tree with information from an
1311  * inlined dir.  It returns the number directory entries loaded
1312  * into the tree.  If there is an error it is returned in err.
1313  */
1314 int htree_inlinedir_to_tree(struct file *dir_file,
1315                             struct inode *dir, ext4_lblk_t block,
1316                             struct dx_hash_info *hinfo,
1317                             __u32 start_hash, __u32 start_minor_hash,
1318                             int *has_inline_data)
1319 {
1320         int err = 0, count = 0;
1321         unsigned int parent_ino;
1322         int pos;
1323         struct ext4_dir_entry_2 *de;
1324         struct inode *inode = file_inode(dir_file);
1325         int ret, inline_size = 0;
1326         struct ext4_iloc iloc;
1327         void *dir_buf = NULL;
1328         struct ext4_dir_entry_2 fake;
1329         struct fscrypt_str tmp_str;
1330
1331         ret = ext4_get_inode_loc(inode, &iloc);
1332         if (ret)
1333                 return ret;
1334
1335         down_read(&EXT4_I(inode)->xattr_sem);
1336         if (!ext4_has_inline_data(inode)) {
1337                 up_read(&EXT4_I(inode)->xattr_sem);
1338                 *has_inline_data = 0;
1339                 goto out;
1340         }
1341
1342         inline_size = ext4_get_inline_size(inode);
1343         dir_buf = kmalloc(inline_size, GFP_NOFS);
1344         if (!dir_buf) {
1345                 ret = -ENOMEM;
1346                 up_read(&EXT4_I(inode)->xattr_sem);
1347                 goto out;
1348         }
1349
1350         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1351         up_read(&EXT4_I(inode)->xattr_sem);
1352         if (ret < 0)
1353                 goto out;
1354
1355         pos = 0;
1356         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1357         while (pos < inline_size) {
1358                 /*
1359                  * As inlined dir doesn't store any information about '.' and
1360                  * only the inode number of '..' is stored, we have to handle
1361                  * them differently.
1362                  */
1363                 if (pos == 0) {
1364                         fake.inode = cpu_to_le32(inode->i_ino);
1365                         fake.name_len = 1;
1366                         strcpy(fake.name, ".");
1367                         fake.rec_len = ext4_rec_len_to_disk(
1368                                                 EXT4_DIR_REC_LEN(fake.name_len),
1369                                                 inline_size);
1370                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1371                         de = &fake;
1372                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1373                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1374                         fake.inode = cpu_to_le32(parent_ino);
1375                         fake.name_len = 2;
1376                         strcpy(fake.name, "..");
1377                         fake.rec_len = ext4_rec_len_to_disk(
1378                                                 EXT4_DIR_REC_LEN(fake.name_len),
1379                                                 inline_size);
1380                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1381                         de = &fake;
1382                         pos = EXT4_INLINE_DOTDOT_SIZE;
1383                 } else {
1384                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1385                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1386                         if (ext4_check_dir_entry(inode, dir_file, de,
1387                                          iloc.bh, dir_buf,
1388                                          inline_size, pos)) {
1389                                 ret = count;
1390                                 goto out;
1391                         }
1392                 }
1393
1394                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1395                 if ((hinfo->hash < start_hash) ||
1396                     ((hinfo->hash == start_hash) &&
1397                      (hinfo->minor_hash < start_minor_hash)))
1398                         continue;
1399                 if (de->inode == 0)
1400                         continue;
1401                 tmp_str.name = de->name;
1402                 tmp_str.len = de->name_len;
1403                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1404                                               hinfo->minor_hash, de, &tmp_str);
1405                 if (err) {
1406                         count = err;
1407                         goto out;
1408                 }
1409                 count++;
1410         }
1411         ret = count;
1412 out:
1413         kfree(dir_buf);
1414         brelse(iloc.bh);
1415         return ret;
1416 }
1417
1418 /*
1419  * So this function is called when the volume is mkfsed with
1420  * dir_index disabled. In order to keep f_pos persistent
1421  * after we convert from an inlined dir to a blocked based,
1422  * we just pretend that we are a normal dir and return the
1423  * offset as if '.' and '..' really take place.
1424  *
1425  */
1426 int ext4_read_inline_dir(struct file *file,
1427                          struct dir_context *ctx,
1428                          int *has_inline_data)
1429 {
1430         unsigned int offset, parent_ino;
1431         int i;
1432         struct ext4_dir_entry_2 *de;
1433         struct super_block *sb;
1434         struct inode *inode = file_inode(file);
1435         int ret, inline_size = 0;
1436         struct ext4_iloc iloc;
1437         void *dir_buf = NULL;
1438         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1439
1440         ret = ext4_get_inode_loc(inode, &iloc);
1441         if (ret)
1442                 return ret;
1443
1444         down_read(&EXT4_I(inode)->xattr_sem);
1445         if (!ext4_has_inline_data(inode)) {
1446                 up_read(&EXT4_I(inode)->xattr_sem);
1447                 *has_inline_data = 0;
1448                 goto out;
1449         }
1450
1451         inline_size = ext4_get_inline_size(inode);
1452         dir_buf = kmalloc(inline_size, GFP_NOFS);
1453         if (!dir_buf) {
1454                 ret = -ENOMEM;
1455                 up_read(&EXT4_I(inode)->xattr_sem);
1456                 goto out;
1457         }
1458
1459         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1460         up_read(&EXT4_I(inode)->xattr_sem);
1461         if (ret < 0)
1462                 goto out;
1463
1464         ret = 0;
1465         sb = inode->i_sb;
1466         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1467         offset = ctx->pos;
1468
1469         /*
1470          * dotdot_offset and dotdot_size is the real offset and
1471          * size for ".." and "." if the dir is block based while
1472          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1473          * So we will use extra_offset and extra_size to indicate them
1474          * during the inline dir iteration.
1475          */
1476         dotdot_offset = EXT4_DIR_REC_LEN(1);
1477         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1478         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1479         extra_size = extra_offset + inline_size;
1480
1481         /*
1482          * If the version has changed since the last call to
1483          * readdir(2), then we might be pointing to an invalid
1484          * dirent right now.  Scan from the start of the inline
1485          * dir to make sure.
1486          */
1487         if (file->f_version != inode->i_version) {
1488                 for (i = 0; i < extra_size && i < offset;) {
1489                         /*
1490                          * "." is with offset 0 and
1491                          * ".." is dotdot_offset.
1492                          */
1493                         if (!i) {
1494                                 i = dotdot_offset;
1495                                 continue;
1496                         } else if (i == dotdot_offset) {
1497                                 i = dotdot_size;
1498                                 continue;
1499                         }
1500                         /* for other entry, the real offset in
1501                          * the buf has to be tuned accordingly.
1502                          */
1503                         de = (struct ext4_dir_entry_2 *)
1504                                 (dir_buf + i - extra_offset);
1505                         /* It's too expensive to do a full
1506                          * dirent test each time round this
1507                          * loop, but we do have to test at
1508                          * least that it is non-zero.  A
1509                          * failure will be detected in the
1510                          * dirent test below. */
1511                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1512                                 < EXT4_DIR_REC_LEN(1))
1513                                 break;
1514                         i += ext4_rec_len_from_disk(de->rec_len,
1515                                                     extra_size);
1516                 }
1517                 offset = i;
1518                 ctx->pos = offset;
1519                 file->f_version = inode->i_version;
1520         }
1521
1522         while (ctx->pos < extra_size) {
1523                 if (ctx->pos == 0) {
1524                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1525                                 goto out;
1526                         ctx->pos = dotdot_offset;
1527                         continue;
1528                 }
1529
1530                 if (ctx->pos == dotdot_offset) {
1531                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1532                                 goto out;
1533                         ctx->pos = dotdot_size;
1534                         continue;
1535                 }
1536
1537                 de = (struct ext4_dir_entry_2 *)
1538                         (dir_buf + ctx->pos - extra_offset);
1539                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1540                                          extra_size, ctx->pos))
1541                         goto out;
1542                 if (le32_to_cpu(de->inode)) {
1543                         if (!dir_emit(ctx, de->name, de->name_len,
1544                                       le32_to_cpu(de->inode),
1545                                       get_dtype(sb, de->file_type)))
1546                                 goto out;
1547                 }
1548                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1549         }
1550 out:
1551         kfree(dir_buf);
1552         brelse(iloc.bh);
1553         return ret;
1554 }
1555
1556 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1557                                         struct ext4_dir_entry_2 **parent_de,
1558                                         int *retval)
1559 {
1560         struct ext4_iloc iloc;
1561
1562         *retval = ext4_get_inode_loc(inode, &iloc);
1563         if (*retval)
1564                 return NULL;
1565
1566         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1567
1568         return iloc.bh;
1569 }
1570
1571 /*
1572  * Try to create the inline data for the new dir.
1573  * If it succeeds, return 0, otherwise return the error.
1574  * In case of ENOSPC, the caller should create the normal disk layout dir.
1575  */
1576 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1577                                struct inode *inode)
1578 {
1579         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1580         struct ext4_iloc iloc;
1581         struct ext4_dir_entry_2 *de;
1582
1583         ret = ext4_get_inode_loc(inode, &iloc);
1584         if (ret)
1585                 return ret;
1586
1587         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1588         if (ret)
1589                 goto out;
1590
1591         /*
1592          * For inline dir, we only save the inode information for the ".."
1593          * and create a fake dentry to cover the left space.
1594          */
1595         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1596         de->inode = cpu_to_le32(parent->i_ino);
1597         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1598         de->inode = 0;
1599         de->rec_len = ext4_rec_len_to_disk(
1600                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1601                                 inline_size);
1602         set_nlink(inode, 2);
1603         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1604 out:
1605         brelse(iloc.bh);
1606         return ret;
1607 }
1608
1609 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1610                                         struct ext4_filename *fname,
1611                                         const struct qstr *d_name,
1612                                         struct ext4_dir_entry_2 **res_dir,
1613                                         int *has_inline_data)
1614 {
1615         int ret;
1616         struct ext4_iloc iloc;
1617         void *inline_start;
1618         int inline_size;
1619
1620         if (ext4_get_inode_loc(dir, &iloc))
1621                 return NULL;
1622
1623         down_read(&EXT4_I(dir)->xattr_sem);
1624         if (!ext4_has_inline_data(dir)) {
1625                 *has_inline_data = 0;
1626                 goto out;
1627         }
1628
1629         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1630                                                 EXT4_INLINE_DOTDOT_SIZE;
1631         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1632         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1633                               dir, fname, d_name, 0, res_dir);
1634         if (ret == 1)
1635                 goto out_find;
1636         if (ret < 0)
1637                 goto out;
1638
1639         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1640                 goto out;
1641
1642         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1643         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1644
1645         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1646                               dir, fname, d_name, 0, res_dir);
1647         if (ret == 1)
1648                 goto out_find;
1649
1650 out:
1651         brelse(iloc.bh);
1652         iloc.bh = NULL;
1653 out_find:
1654         up_read(&EXT4_I(dir)->xattr_sem);
1655         return iloc.bh;
1656 }
1657
1658 int ext4_delete_inline_entry(handle_t *handle,
1659                              struct inode *dir,
1660                              struct ext4_dir_entry_2 *de_del,
1661                              struct buffer_head *bh,
1662                              int *has_inline_data)
1663 {
1664         int err, inline_size;
1665         struct ext4_iloc iloc;
1666         void *inline_start;
1667
1668         err = ext4_get_inode_loc(dir, &iloc);
1669         if (err)
1670                 return err;
1671
1672         down_write(&EXT4_I(dir)->xattr_sem);
1673         if (!ext4_has_inline_data(dir)) {
1674                 *has_inline_data = 0;
1675                 goto out;
1676         }
1677
1678         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1679                 EXT4_MIN_INLINE_DATA_SIZE) {
1680                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1681                                         EXT4_INLINE_DOTDOT_SIZE;
1682                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1683                                 EXT4_INLINE_DOTDOT_SIZE;
1684         } else {
1685                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1686                 inline_size = ext4_get_inline_size(dir) -
1687                                 EXT4_MIN_INLINE_DATA_SIZE;
1688         }
1689
1690         BUFFER_TRACE(bh, "get_write_access");
1691         err = ext4_journal_get_write_access(handle, bh);
1692         if (err)
1693                 goto out;
1694
1695         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1696                                         inline_start, inline_size, 0);
1697         if (err)
1698                 goto out;
1699
1700         err = ext4_mark_inode_dirty(handle, dir);
1701         if (unlikely(err))
1702                 goto out;
1703
1704         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1705 out:
1706         up_write(&EXT4_I(dir)->xattr_sem);
1707         brelse(iloc.bh);
1708         if (err != -ENOENT)
1709                 ext4_std_error(dir->i_sb, err);
1710         return err;
1711 }
1712
1713 /*
1714  * Get the inline dentry at offset.
1715  */
1716 static inline struct ext4_dir_entry_2 *
1717 ext4_get_inline_entry(struct inode *inode,
1718                       struct ext4_iloc *iloc,
1719                       unsigned int offset,
1720                       void **inline_start,
1721                       int *inline_size)
1722 {
1723         void *inline_pos;
1724
1725         BUG_ON(offset > ext4_get_inline_size(inode));
1726
1727         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1728                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1729                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1730         } else {
1731                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1732                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1733                 *inline_size = ext4_get_inline_size(inode) -
1734                                 EXT4_MIN_INLINE_DATA_SIZE;
1735         }
1736
1737         if (inline_start)
1738                 *inline_start = inline_pos;
1739         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1740 }
1741
1742 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1743 {
1744         int err, inline_size;
1745         struct ext4_iloc iloc;
1746         void *inline_pos;
1747         unsigned int offset;
1748         struct ext4_dir_entry_2 *de;
1749         bool ret = true;
1750
1751         err = ext4_get_inode_loc(dir, &iloc);
1752         if (err) {
1753                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1754                                  err, dir->i_ino);
1755                 return true;
1756         }
1757
1758         down_read(&EXT4_I(dir)->xattr_sem);
1759         if (!ext4_has_inline_data(dir)) {
1760                 *has_inline_data = 0;
1761                 goto out;
1762         }
1763
1764         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1765         if (!le32_to_cpu(de->inode)) {
1766                 ext4_warning(dir->i_sb,
1767                              "bad inline directory (dir #%lu) - no `..'",
1768                              dir->i_ino);
1769                 ret = true;
1770                 goto out;
1771         }
1772
1773         offset = EXT4_INLINE_DOTDOT_SIZE;
1774         while (offset < dir->i_size) {
1775                 de = ext4_get_inline_entry(dir, &iloc, offset,
1776                                            &inline_pos, &inline_size);
1777                 if (ext4_check_dir_entry(dir, NULL, de,
1778                                          iloc.bh, inline_pos,
1779                                          inline_size, offset)) {
1780                         ext4_warning(dir->i_sb,
1781                                      "bad inline directory (dir #%lu) - "
1782                                      "inode %u, rec_len %u, name_len %d"
1783                                      "inline size %d",
1784                                      dir->i_ino, le32_to_cpu(de->inode),
1785                                      le16_to_cpu(de->rec_len), de->name_len,
1786                                      inline_size);
1787                         ret = true;
1788                         goto out;
1789                 }
1790                 if (le32_to_cpu(de->inode)) {
1791                         ret = false;
1792                         goto out;
1793                 }
1794                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1795         }
1796
1797 out:
1798         up_read(&EXT4_I(dir)->xattr_sem);
1799         brelse(iloc.bh);
1800         return ret;
1801 }
1802
1803 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1804 {
1805         int ret;
1806
1807         down_write(&EXT4_I(inode)->xattr_sem);
1808         ret = ext4_destroy_inline_data_nolock(handle, inode);
1809         up_write(&EXT4_I(inode)->xattr_sem);
1810
1811         return ret;
1812 }
1813
1814 int ext4_inline_data_fiemap(struct inode *inode,
1815                             struct fiemap_extent_info *fieinfo,
1816                             int *has_inline, __u64 start, __u64 len)
1817 {
1818         __u64 physical = 0;
1819         __u64 inline_len;
1820         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1821                 FIEMAP_EXTENT_LAST;
1822         int error = 0;
1823         struct ext4_iloc iloc;
1824
1825         down_read(&EXT4_I(inode)->xattr_sem);
1826         if (!ext4_has_inline_data(inode)) {
1827                 *has_inline = 0;
1828                 goto out;
1829         }
1830         inline_len = min_t(size_t, ext4_get_inline_size(inode),
1831                            i_size_read(inode));
1832         if (start >= inline_len)
1833                 goto out;
1834         if (start + len < inline_len)
1835                 inline_len = start + len;
1836         inline_len -= start;
1837
1838         error = ext4_get_inode_loc(inode, &iloc);
1839         if (error)
1840                 goto out;
1841
1842         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1843         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1844         physical += offsetof(struct ext4_inode, i_block);
1845
1846         if (physical)
1847                 error = fiemap_fill_next_extent(fieinfo, start, physical,
1848                                                 inline_len, flags);
1849         brelse(iloc.bh);
1850 out:
1851         up_read(&EXT4_I(inode)->xattr_sem);
1852         return (error < 0 ? error : 0);
1853 }
1854
1855 /*
1856  * Called during xattr set, and if we can sparse space 'needed',
1857  * just create the extent tree evict the data to the outer block.
1858  *
1859  * We use jbd2 instead of page cache to move data to the 1st block
1860  * so that the whole transaction can be committed as a whole and
1861  * the data isn't lost because of the delayed page cache write.
1862  */
1863 int ext4_try_to_evict_inline_data(handle_t *handle,
1864                                   struct inode *inode,
1865                                   int needed)
1866 {
1867         int error;
1868         struct ext4_xattr_entry *entry;
1869         struct ext4_inode *raw_inode;
1870         struct ext4_iloc iloc;
1871
1872         error = ext4_get_inode_loc(inode, &iloc);
1873         if (error)
1874                 return error;
1875
1876         raw_inode = ext4_raw_inode(&iloc);
1877         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1878                                             EXT4_I(inode)->i_inline_off);
1879         if (EXT4_XATTR_LEN(entry->e_name_len) +
1880             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1881                 error = -ENOSPC;
1882                 goto out;
1883         }
1884
1885         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1886 out:
1887         brelse(iloc.bh);
1888         return error;
1889 }
1890
1891 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1892 {
1893         handle_t *handle;
1894         int inline_size, value_len, needed_blocks;
1895         size_t i_size;
1896         void *value = NULL;
1897         struct ext4_xattr_ibody_find is = {
1898                 .s = { .not_found = -ENODATA, },
1899         };
1900         struct ext4_xattr_info i = {
1901                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1902                 .name = EXT4_XATTR_SYSTEM_DATA,
1903         };
1904
1905
1906         needed_blocks = ext4_writepage_trans_blocks(inode);
1907         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1908         if (IS_ERR(handle))
1909                 return;
1910
1911         down_write(&EXT4_I(inode)->xattr_sem);
1912         if (!ext4_has_inline_data(inode)) {
1913                 *has_inline = 0;
1914                 ext4_journal_stop(handle);
1915                 return;
1916         }
1917
1918         if (ext4_orphan_add(handle, inode))
1919                 goto out;
1920
1921         if (ext4_get_inode_loc(inode, &is.iloc))
1922                 goto out;
1923
1924         down_write(&EXT4_I(inode)->i_data_sem);
1925         i_size = inode->i_size;
1926         inline_size = ext4_get_inline_size(inode);
1927         EXT4_I(inode)->i_disksize = i_size;
1928
1929         if (i_size < inline_size) {
1930                 /* Clear the content in the xattr space. */
1931                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1932                         if (ext4_xattr_ibody_find(inode, &i, &is))
1933                                 goto out_error;
1934
1935                         BUG_ON(is.s.not_found);
1936
1937                         value_len = le32_to_cpu(is.s.here->e_value_size);
1938                         value = kmalloc(value_len, GFP_NOFS);
1939                         if (!value)
1940                                 goto out_error;
1941
1942                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1943                                                 value, value_len))
1944                                 goto out_error;
1945
1946                         i.value = value;
1947                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1948                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1949                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1950                                 goto out_error;
1951                 }
1952
1953                 /* Clear the content within i_blocks. */
1954                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1955                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1956                         memset(p + i_size, 0,
1957                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1958                 }
1959
1960                 EXT4_I(inode)->i_inline_size = i_size <
1961                                         EXT4_MIN_INLINE_DATA_SIZE ?
1962                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1963         }
1964
1965 out_error:
1966         up_write(&EXT4_I(inode)->i_data_sem);
1967 out:
1968         brelse(is.iloc.bh);
1969         up_write(&EXT4_I(inode)->xattr_sem);
1970         kfree(value);
1971         if (inode->i_nlink)
1972                 ext4_orphan_del(handle, inode);
1973
1974         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1975         ext4_mark_inode_dirty(handle, inode);
1976         if (IS_SYNC(inode))
1977                 ext4_handle_sync(handle);
1978
1979         ext4_journal_stop(handle);
1980         return;
1981 }
1982
1983 int ext4_convert_inline_data(struct inode *inode)
1984 {
1985         int error, needed_blocks;
1986         handle_t *handle;
1987         struct ext4_iloc iloc;
1988
1989         if (!ext4_has_inline_data(inode)) {
1990                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1991                 return 0;
1992         }
1993
1994         needed_blocks = ext4_writepage_trans_blocks(inode);
1995
1996         iloc.bh = NULL;
1997         error = ext4_get_inode_loc(inode, &iloc);
1998         if (error)
1999                 return error;
2000
2001         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2002         if (IS_ERR(handle)) {
2003                 error = PTR_ERR(handle);
2004                 goto out_free;
2005         }
2006
2007         down_write(&EXT4_I(inode)->xattr_sem);
2008         if (!ext4_has_inline_data(inode)) {
2009                 up_write(&EXT4_I(inode)->xattr_sem);
2010                 goto out;
2011         }
2012
2013         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2014         up_write(&EXT4_I(inode)->xattr_sem);
2015 out:
2016         ext4_journal_stop(handle);
2017 out_free:
2018         brelse(iloc.bh);
2019         return error;
2020 }