]> Pileus Git - ~andy/linux/blob - fs/ext4/inline.c
ext4: retry allocation when inline->extent conversion failed
[~andy/linux] / fs / ext4 / inline.c
1 /*
2  * Copyright (c) 2012 Taobao.
3  * Written by Tao Ma <boyu.mt@taobao.com>
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of version 2.1 of the GNU Lesser General Public License
7  * as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  */
14 #include "ext4_jbd2.h"
15 #include "ext4.h"
16 #include "xattr.h"
17 #include "truncate.h"
18 #include <linux/fiemap.h>
19
20 #define EXT4_XATTR_SYSTEM_DATA  "data"
21 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
22 #define EXT4_INLINE_DOTDOT_OFFSET       2
23 #define EXT4_INLINE_DOTDOT_SIZE         4
24
25 int ext4_get_inline_size(struct inode *inode)
26 {
27         if (EXT4_I(inode)->i_inline_off)
28                 return EXT4_I(inode)->i_inline_size;
29
30         return 0;
31 }
32
33 static int get_max_inline_xattr_value_size(struct inode *inode,
34                                            struct ext4_iloc *iloc)
35 {
36         struct ext4_xattr_ibody_header *header;
37         struct ext4_xattr_entry *entry;
38         struct ext4_inode *raw_inode;
39         int free, min_offs;
40
41         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42                         EXT4_GOOD_OLD_INODE_SIZE -
43                         EXT4_I(inode)->i_extra_isize -
44                         sizeof(struct ext4_xattr_ibody_header);
45
46         /*
47          * We need to subtract another sizeof(__u32) since an in-inode xattr
48          * needs an empty 4 bytes to indicate the gap between the xattr entry
49          * and the name/value pair.
50          */
51         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52                 return EXT4_XATTR_SIZE(min_offs -
53                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54                         EXT4_XATTR_ROUND - sizeof(__u32));
55
56         raw_inode = ext4_raw_inode(iloc);
57         header = IHDR(inode, raw_inode);
58         entry = IFIRST(header);
59
60         /* Compute min_offs. */
61         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62                 if (!entry->e_value_block && entry->e_value_size) {
63                         size_t offs = le16_to_cpu(entry->e_value_offs);
64                         if (offs < min_offs)
65                                 min_offs = offs;
66                 }
67         }
68         free = min_offs -
69                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71         if (EXT4_I(inode)->i_inline_off) {
72                 entry = (struct ext4_xattr_entry *)
73                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76                 goto out;
77         }
78
79         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81         if (free > EXT4_XATTR_ROUND)
82                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83         else
84                 free = 0;
85
86 out:
87         return free;
88 }
89
90 /*
91  * Get the maximum size we now can store in an inode.
92  * If we can't find the space for a xattr entry, don't use the space
93  * of the extents since we have no space to indicate the inline data.
94  */
95 int ext4_get_max_inline_size(struct inode *inode)
96 {
97         int error, max_inline_size;
98         struct ext4_iloc iloc;
99
100         if (EXT4_I(inode)->i_extra_isize == 0)
101                 return 0;
102
103         error = ext4_get_inode_loc(inode, &iloc);
104         if (error) {
105                 ext4_error_inode(inode, __func__, __LINE__, 0,
106                                  "can't get inode location %lu",
107                                  inode->i_ino);
108                 return 0;
109         }
110
111         down_read(&EXT4_I(inode)->xattr_sem);
112         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113         up_read(&EXT4_I(inode)->xattr_sem);
114
115         brelse(iloc.bh);
116
117         if (!max_inline_size)
118                 return 0;
119
120         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121 }
122
123 int ext4_has_inline_data(struct inode *inode)
124 {
125         return ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) &&
126                EXT4_I(inode)->i_inline_off;
127 }
128
129 /*
130  * this function does not take xattr_sem, which is OK because it is
131  * currently only used in a code path coming form ext4_iget, before
132  * the new inode has been unlocked
133  */
134 int ext4_find_inline_data_nolock(struct inode *inode)
135 {
136         struct ext4_xattr_ibody_find is = {
137                 .s = { .not_found = -ENODATA, },
138         };
139         struct ext4_xattr_info i = {
140                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
141                 .name = EXT4_XATTR_SYSTEM_DATA,
142         };
143         int error;
144
145         if (EXT4_I(inode)->i_extra_isize == 0)
146                 return 0;
147
148         error = ext4_get_inode_loc(inode, &is.iloc);
149         if (error)
150                 return error;
151
152         error = ext4_xattr_ibody_find(inode, &i, &is);
153         if (error)
154                 goto out;
155
156         if (!is.s.not_found) {
157                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
158                                         (void *)ext4_raw_inode(&is.iloc));
159                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
160                                 le32_to_cpu(is.s.here->e_value_size);
161                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
162         }
163 out:
164         brelse(is.iloc.bh);
165         return error;
166 }
167
168 static int ext4_read_inline_data(struct inode *inode, void *buffer,
169                                  unsigned int len,
170                                  struct ext4_iloc *iloc)
171 {
172         struct ext4_xattr_entry *entry;
173         struct ext4_xattr_ibody_header *header;
174         int cp_len = 0;
175         struct ext4_inode *raw_inode;
176
177         if (!len)
178                 return 0;
179
180         BUG_ON(len > EXT4_I(inode)->i_inline_size);
181
182         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
183                         len : EXT4_MIN_INLINE_DATA_SIZE;
184
185         raw_inode = ext4_raw_inode(iloc);
186         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
187
188         len -= cp_len;
189         buffer += cp_len;
190
191         if (!len)
192                 goto out;
193
194         header = IHDR(inode, raw_inode);
195         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
196                                             EXT4_I(inode)->i_inline_off);
197         len = min_t(unsigned int, len,
198                     (unsigned int)le32_to_cpu(entry->e_value_size));
199
200         memcpy(buffer,
201                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
202         cp_len += len;
203
204 out:
205         return cp_len;
206 }
207
208 /*
209  * write the buffer to the inline inode.
210  * If 'create' is set, we don't need to do the extra copy in the xattr
211  * value since it is already handled by ext4_xattr_ibody_inline_set.
212  * That saves us one memcpy.
213  */
214 void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
215                             void *buffer, loff_t pos, unsigned int len)
216 {
217         struct ext4_xattr_entry *entry;
218         struct ext4_xattr_ibody_header *header;
219         struct ext4_inode *raw_inode;
220         int cp_len = 0;
221
222         BUG_ON(!EXT4_I(inode)->i_inline_off);
223         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
224
225         raw_inode = ext4_raw_inode(iloc);
226         buffer += pos;
227
228         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
229                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
230                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
231                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
232
233                 len -= cp_len;
234                 buffer += cp_len;
235                 pos += cp_len;
236         }
237
238         if (!len)
239                 return;
240
241         pos -= EXT4_MIN_INLINE_DATA_SIZE;
242         header = IHDR(inode, raw_inode);
243         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
244                                             EXT4_I(inode)->i_inline_off);
245
246         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
247                buffer, len);
248 }
249
250 static int ext4_create_inline_data(handle_t *handle,
251                                    struct inode *inode, unsigned len)
252 {
253         int error;
254         void *value = NULL;
255         struct ext4_xattr_ibody_find is = {
256                 .s = { .not_found = -ENODATA, },
257         };
258         struct ext4_xattr_info i = {
259                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
260                 .name = EXT4_XATTR_SYSTEM_DATA,
261         };
262
263         error = ext4_get_inode_loc(inode, &is.iloc);
264         if (error)
265                 return error;
266
267         error = ext4_journal_get_write_access(handle, is.iloc.bh);
268         if (error)
269                 goto out;
270
271         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272                 value = EXT4_ZERO_XATTR_VALUE;
273                 len -= EXT4_MIN_INLINE_DATA_SIZE;
274         } else {
275                 value = "";
276                 len = 0;
277         }
278
279         /* Insert the the xttr entry. */
280         i.value = value;
281         i.value_len = len;
282
283         error = ext4_xattr_ibody_find(inode, &i, &is);
284         if (error)
285                 goto out;
286
287         BUG_ON(!is.s.not_found);
288
289         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290         if (error) {
291                 if (error == -ENOSPC)
292                         ext4_clear_inode_state(inode,
293                                                EXT4_STATE_MAY_INLINE_DATA);
294                 goto out;
295         }
296
297         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298                 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301                                       (void *)ext4_raw_inode(&is.iloc));
302         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305         get_bh(is.iloc.bh);
306         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
307
308 out:
309         brelse(is.iloc.bh);
310         return error;
311 }
312
313 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
314                                    unsigned int len)
315 {
316         int error;
317         void *value = NULL;
318         struct ext4_xattr_ibody_find is = {
319                 .s = { .not_found = -ENODATA, },
320         };
321         struct ext4_xattr_info i = {
322                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
323                 .name = EXT4_XATTR_SYSTEM_DATA,
324         };
325
326         /* If the old space is ok, write the data directly. */
327         if (len <= EXT4_I(inode)->i_inline_size)
328                 return 0;
329
330         error = ext4_get_inode_loc(inode, &is.iloc);
331         if (error)
332                 return error;
333
334         error = ext4_xattr_ibody_find(inode, &i, &is);
335         if (error)
336                 goto out;
337
338         BUG_ON(is.s.not_found);
339
340         len -= EXT4_MIN_INLINE_DATA_SIZE;
341         value = kzalloc(len, GFP_NOFS);
342         if (!value)
343                 goto out;
344
345         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
346                                      value, len);
347         if (error == -ENODATA)
348                 goto out;
349
350         error = ext4_journal_get_write_access(handle, is.iloc.bh);
351         if (error)
352                 goto out;
353
354         /* Update the xttr entry. */
355         i.value = value;
356         i.value_len = len;
357
358         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
359         if (error)
360                 goto out;
361
362         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
363                                       (void *)ext4_raw_inode(&is.iloc));
364         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
365                                 le32_to_cpu(is.s.here->e_value_size);
366         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
367         get_bh(is.iloc.bh);
368         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
369
370 out:
371         kfree(value);
372         brelse(is.iloc.bh);
373         return error;
374 }
375
376 int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
377                              unsigned int len)
378 {
379         int ret, size;
380         struct ext4_inode_info *ei = EXT4_I(inode);
381
382         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
383                 return -ENOSPC;
384
385         size = ext4_get_max_inline_size(inode);
386         if (size < len)
387                 return -ENOSPC;
388
389         down_write(&EXT4_I(inode)->xattr_sem);
390
391         if (ei->i_inline_off)
392                 ret = ext4_update_inline_data(handle, inode, len);
393         else
394                 ret = ext4_create_inline_data(handle, inode, len);
395
396         up_write(&EXT4_I(inode)->xattr_sem);
397
398         return ret;
399 }
400
401 static int ext4_destroy_inline_data_nolock(handle_t *handle,
402                                            struct inode *inode)
403 {
404         struct ext4_inode_info *ei = EXT4_I(inode);
405         struct ext4_xattr_ibody_find is = {
406                 .s = { .not_found = 0, },
407         };
408         struct ext4_xattr_info i = {
409                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
410                 .name = EXT4_XATTR_SYSTEM_DATA,
411                 .value = NULL,
412                 .value_len = 0,
413         };
414         int error;
415
416         if (!ei->i_inline_off)
417                 return 0;
418
419         error = ext4_get_inode_loc(inode, &is.iloc);
420         if (error)
421                 return error;
422
423         error = ext4_xattr_ibody_find(inode, &i, &is);
424         if (error)
425                 goto out;
426
427         error = ext4_journal_get_write_access(handle, is.iloc.bh);
428         if (error)
429                 goto out;
430
431         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
432         if (error)
433                 goto out;
434
435         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
436                 0, EXT4_MIN_INLINE_DATA_SIZE);
437
438         if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
439                                       EXT4_FEATURE_INCOMPAT_EXTENTS)) {
440                 if (S_ISDIR(inode->i_mode) ||
441                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
442                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
443                         ext4_ext_tree_init(handle, inode);
444                 }
445         }
446         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
447
448         get_bh(is.iloc.bh);
449         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
450
451         EXT4_I(inode)->i_inline_off = 0;
452         EXT4_I(inode)->i_inline_size = 0;
453         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
454 out:
455         brelse(is.iloc.bh);
456         if (error == -ENODATA)
457                 error = 0;
458         return error;
459 }
460
461 static int ext4_read_inline_page(struct inode *inode, struct page *page)
462 {
463         void *kaddr;
464         int ret = 0;
465         size_t len;
466         struct ext4_iloc iloc;
467
468         BUG_ON(!PageLocked(page));
469         BUG_ON(!ext4_has_inline_data(inode));
470         BUG_ON(page->index);
471
472         if (!EXT4_I(inode)->i_inline_off) {
473                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
474                              inode->i_ino);
475                 goto out;
476         }
477
478         ret = ext4_get_inode_loc(inode, &iloc);
479         if (ret)
480                 goto out;
481
482         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
483         kaddr = kmap_atomic(page);
484         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
485         flush_dcache_page(page);
486         kunmap_atomic(kaddr);
487         zero_user_segment(page, len, PAGE_CACHE_SIZE);
488         SetPageUptodate(page);
489         brelse(iloc.bh);
490
491 out:
492         return ret;
493 }
494
495 int ext4_readpage_inline(struct inode *inode, struct page *page)
496 {
497         int ret = 0;
498
499         down_read(&EXT4_I(inode)->xattr_sem);
500         if (!ext4_has_inline_data(inode)) {
501                 up_read(&EXT4_I(inode)->xattr_sem);
502                 return -EAGAIN;
503         }
504
505         /*
506          * Current inline data can only exist in the 1st page,
507          * So for all the other pages, just set them uptodate.
508          */
509         if (!page->index)
510                 ret = ext4_read_inline_page(inode, page);
511         else if (!PageUptodate(page)) {
512                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
513                 SetPageUptodate(page);
514         }
515
516         up_read(&EXT4_I(inode)->xattr_sem);
517
518         unlock_page(page);
519         return ret >= 0 ? 0 : ret;
520 }
521
522 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
523                                               struct inode *inode,
524                                               unsigned flags)
525 {
526         int ret, needed_blocks;
527         handle_t *handle = NULL;
528         int retries = 0, sem_held = 0;
529         struct page *page = NULL;
530         unsigned from, to;
531         struct ext4_iloc iloc;
532
533         if (!ext4_has_inline_data(inode)) {
534                 /*
535                  * clear the flag so that no new write
536                  * will trap here again.
537                  */
538                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
539                 return 0;
540         }
541
542         needed_blocks = ext4_writepage_trans_blocks(inode);
543
544         ret = ext4_get_inode_loc(inode, &iloc);
545         if (ret)
546                 return ret;
547
548 retry:
549         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
550         if (IS_ERR(handle)) {
551                 ret = PTR_ERR(handle);
552                 handle = NULL;
553                 goto out;
554         }
555
556         /* We cannot recurse into the filesystem as the transaction is already
557          * started */
558         flags |= AOP_FLAG_NOFS;
559
560         page = grab_cache_page_write_begin(mapping, 0, flags);
561         if (!page) {
562                 ret = -ENOMEM;
563                 goto out;
564         }
565
566         down_write(&EXT4_I(inode)->xattr_sem);
567         sem_held = 1;
568         /* If some one has already done this for us, just exit. */
569         if (!ext4_has_inline_data(inode)) {
570                 ret = 0;
571                 goto out;
572         }
573
574         from = 0;
575         to = ext4_get_inline_size(inode);
576         if (!PageUptodate(page)) {
577                 ret = ext4_read_inline_page(inode, page);
578                 if (ret < 0)
579                         goto out;
580         }
581
582         ret = ext4_destroy_inline_data_nolock(handle, inode);
583         if (ret)
584                 goto out;
585
586         if (ext4_should_dioread_nolock(inode))
587                 ret = __block_write_begin(page, from, to, ext4_get_block_write);
588         else
589                 ret = __block_write_begin(page, from, to, ext4_get_block);
590
591         if (!ret && ext4_should_journal_data(inode)) {
592                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
593                                              from, to, NULL,
594                                              do_journal_get_write_access);
595         }
596
597         if (ret) {
598                 unlock_page(page);
599                 page_cache_release(page);
600                 ext4_orphan_add(handle, inode);
601                 up_write(&EXT4_I(inode)->xattr_sem);
602                 sem_held = 0;
603                 ext4_journal_stop(handle);
604                 handle = NULL;
605                 ext4_truncate_failed_write(inode);
606                 /*
607                  * If truncate failed early the inode might
608                  * still be on the orphan list; we need to
609                  * make sure the inode is removed from the
610                  * orphan list in that case.
611                  */
612                 if (inode->i_nlink)
613                         ext4_orphan_del(NULL, inode);
614         }
615
616         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617                 goto retry;
618
619         block_commit_write(page, from, to);
620 out:
621         if (page) {
622                 unlock_page(page);
623                 page_cache_release(page);
624         }
625         if (sem_held)
626                 up_write(&EXT4_I(inode)->xattr_sem);
627         if (handle)
628                 ext4_journal_stop(handle);
629         brelse(iloc.bh);
630         return ret;
631 }
632
633 /*
634  * Try to write data in the inode.
635  * If the inode has inline data, check whether the new write can be
636  * in the inode also. If not, create the page the handle, move the data
637  * to the page make it update and let the later codes create extent for it.
638  */
639 int ext4_try_to_write_inline_data(struct address_space *mapping,
640                                   struct inode *inode,
641                                   loff_t pos, unsigned len,
642                                   unsigned flags,
643                                   struct page **pagep)
644 {
645         int ret;
646         handle_t *handle;
647         struct page *page;
648         struct ext4_iloc iloc;
649
650         if (pos + len > ext4_get_max_inline_size(inode))
651                 goto convert;
652
653         ret = ext4_get_inode_loc(inode, &iloc);
654         if (ret)
655                 return ret;
656
657         /*
658          * The possible write could happen in the inode,
659          * so try to reserve the space in inode first.
660          */
661         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
662         if (IS_ERR(handle)) {
663                 ret = PTR_ERR(handle);
664                 handle = NULL;
665                 goto out;
666         }
667
668         ret = ext4_prepare_inline_data(handle, inode, pos + len);
669         if (ret && ret != -ENOSPC)
670                 goto out;
671
672         /* We don't have space in inline inode, so convert it to extent. */
673         if (ret == -ENOSPC) {
674                 ext4_journal_stop(handle);
675                 brelse(iloc.bh);
676                 goto convert;
677         }
678
679         flags |= AOP_FLAG_NOFS;
680
681         page = grab_cache_page_write_begin(mapping, 0, flags);
682         if (!page) {
683                 ret = -ENOMEM;
684                 goto out;
685         }
686
687         *pagep = page;
688         down_read(&EXT4_I(inode)->xattr_sem);
689         if (!ext4_has_inline_data(inode)) {
690                 ret = 0;
691                 unlock_page(page);
692                 page_cache_release(page);
693                 goto out_up_read;
694         }
695
696         if (!PageUptodate(page)) {
697                 ret = ext4_read_inline_page(inode, page);
698                 if (ret < 0)
699                         goto out_up_read;
700         }
701
702         ret = 1;
703         handle = NULL;
704 out_up_read:
705         up_read(&EXT4_I(inode)->xattr_sem);
706 out:
707         if (handle)
708                 ext4_journal_stop(handle);
709         brelse(iloc.bh);
710         return ret;
711 convert:
712         return ext4_convert_inline_data_to_extent(mapping,
713                                                   inode, flags);
714 }
715
716 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
717                                unsigned copied, struct page *page)
718 {
719         int ret;
720         void *kaddr;
721         struct ext4_iloc iloc;
722
723         if (unlikely(copied < len)) {
724                 if (!PageUptodate(page)) {
725                         copied = 0;
726                         goto out;
727                 }
728         }
729
730         ret = ext4_get_inode_loc(inode, &iloc);
731         if (ret) {
732                 ext4_std_error(inode->i_sb, ret);
733                 copied = 0;
734                 goto out;
735         }
736
737         down_write(&EXT4_I(inode)->xattr_sem);
738         BUG_ON(!ext4_has_inline_data(inode));
739
740         kaddr = kmap_atomic(page);
741         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
742         kunmap_atomic(kaddr);
743         SetPageUptodate(page);
744         /* clear page dirty so that writepages wouldn't work for us. */
745         ClearPageDirty(page);
746
747         up_write(&EXT4_I(inode)->xattr_sem);
748         brelse(iloc.bh);
749 out:
750         return copied;
751 }
752
753 struct buffer_head *
754 ext4_journalled_write_inline_data(struct inode *inode,
755                                   unsigned len,
756                                   struct page *page)
757 {
758         int ret;
759         void *kaddr;
760         struct ext4_iloc iloc;
761
762         ret = ext4_get_inode_loc(inode, &iloc);
763         if (ret) {
764                 ext4_std_error(inode->i_sb, ret);
765                 return NULL;
766         }
767
768         down_write(&EXT4_I(inode)->xattr_sem);
769         kaddr = kmap_atomic(page);
770         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
771         kunmap_atomic(kaddr);
772         up_write(&EXT4_I(inode)->xattr_sem);
773
774         return iloc.bh;
775 }
776
777 /*
778  * Try to make the page cache and handle ready for the inline data case.
779  * We can call this function in 2 cases:
780  * 1. The inode is created and the first write exceeds inline size. We can
781  *    clear the inode state safely.
782  * 2. The inode has inline data, then we need to read the data, make it
783  *    update and dirty so that ext4_da_writepages can handle it. We don't
784  *    need to start the journal since the file's metatdata isn't changed now.
785  */
786 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
787                                                  struct inode *inode,
788                                                  unsigned flags,
789                                                  void **fsdata)
790 {
791         int ret = 0, inline_size;
792         struct page *page;
793
794         page = grab_cache_page_write_begin(mapping, 0, flags);
795         if (!page)
796                 return -ENOMEM;
797
798         down_read(&EXT4_I(inode)->xattr_sem);
799         if (!ext4_has_inline_data(inode)) {
800                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
801                 goto out;
802         }
803
804         inline_size = ext4_get_inline_size(inode);
805
806         if (!PageUptodate(page)) {
807                 ret = ext4_read_inline_page(inode, page);
808                 if (ret < 0)
809                         goto out;
810         }
811
812         ret = __block_write_begin(page, 0, inline_size,
813                                   ext4_da_get_block_prep);
814         if (ret) {
815                 ext4_truncate_failed_write(inode);
816                 goto out;
817         }
818
819         SetPageDirty(page);
820         SetPageUptodate(page);
821         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
822         *fsdata = (void *)CONVERT_INLINE_DATA;
823
824 out:
825         up_read(&EXT4_I(inode)->xattr_sem);
826         if (page) {
827                 unlock_page(page);
828                 page_cache_release(page);
829         }
830         return ret;
831 }
832
833 /*
834  * Prepare the write for the inline data.
835  * If the the data can be written into the inode, we just read
836  * the page and make it uptodate, and start the journal.
837  * Otherwise read the page, makes it dirty so that it can be
838  * handle in writepages(the i_disksize update is left to the
839  * normal ext4_da_write_end).
840  */
841 int ext4_da_write_inline_data_begin(struct address_space *mapping,
842                                     struct inode *inode,
843                                     loff_t pos, unsigned len,
844                                     unsigned flags,
845                                     struct page **pagep,
846                                     void **fsdata)
847 {
848         int ret, inline_size;
849         handle_t *handle;
850         struct page *page;
851         struct ext4_iloc iloc;
852         int retries;
853
854         ret = ext4_get_inode_loc(inode, &iloc);
855         if (ret)
856                 return ret;
857
858 retry_journal:
859         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
860         if (IS_ERR(handle)) {
861                 ret = PTR_ERR(handle);
862                 handle = NULL;
863                 goto out;
864         }
865
866         inline_size = ext4_get_max_inline_size(inode);
867
868         ret = -ENOSPC;
869         if (inline_size >= pos + len) {
870                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
871                 if (ret && ret != -ENOSPC)
872                         goto out;
873         }
874
875         if (ret == -ENOSPC) {
876                 ret = ext4_da_convert_inline_data_to_extent(mapping,
877                                                             inode,
878                                                             flags,
879                                                             fsdata);
880                 ext4_journal_stop(handle);
881                 handle = NULL;
882                 if (ret == -ENOSPC &&
883                     ext4_should_retry_alloc(inode->i_sb, &retries))
884                         goto retry_journal;
885                 goto out;
886         }
887
888         /*
889          * We cannot recurse into the filesystem as the transaction
890          * is already started.
891          */
892         flags |= AOP_FLAG_NOFS;
893
894         page = grab_cache_page_write_begin(mapping, 0, flags);
895         if (!page) {
896                 ret = -ENOMEM;
897                 goto out;
898         }
899
900         down_read(&EXT4_I(inode)->xattr_sem);
901         if (!ext4_has_inline_data(inode)) {
902                 ret = 0;
903                 goto out_release_page;
904         }
905
906         if (!PageUptodate(page)) {
907                 ret = ext4_read_inline_page(inode, page);
908                 if (ret < 0)
909                         goto out_release_page;
910         }
911
912         up_read(&EXT4_I(inode)->xattr_sem);
913         *pagep = page;
914         handle = NULL;
915         brelse(iloc.bh);
916         return 1;
917 out_release_page:
918         up_read(&EXT4_I(inode)->xattr_sem);
919         unlock_page(page);
920         page_cache_release(page);
921 out:
922         if (handle)
923                 ext4_journal_stop(handle);
924         brelse(iloc.bh);
925         return ret;
926 }
927
928 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
929                                   unsigned len, unsigned copied,
930                                   struct page *page)
931 {
932         int i_size_changed = 0;
933
934         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
935
936         /*
937          * No need to use i_size_read() here, the i_size
938          * cannot change under us because we hold i_mutex.
939          *
940          * But it's important to update i_size while still holding page lock:
941          * page writeout could otherwise come in and zero beyond i_size.
942          */
943         if (pos+copied > inode->i_size) {
944                 i_size_write(inode, pos+copied);
945                 i_size_changed = 1;
946         }
947         unlock_page(page);
948         page_cache_release(page);
949
950         /*
951          * Don't mark the inode dirty under page lock. First, it unnecessarily
952          * makes the holding time of page lock longer. Second, it forces lock
953          * ordering of page lock and transaction start for journaling
954          * filesystems.
955          */
956         if (i_size_changed)
957                 mark_inode_dirty(inode);
958
959         return copied;
960 }
961
962 #ifdef INLINE_DIR_DEBUG
963 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
964                           void *inline_start, int inline_size)
965 {
966         int offset;
967         unsigned short de_len;
968         struct ext4_dir_entry_2 *de = inline_start;
969         void *dlimit = inline_start + inline_size;
970
971         trace_printk("inode %lu\n", dir->i_ino);
972         offset = 0;
973         while ((void *)de < dlimit) {
974                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
975                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
976                              offset, de_len, de->name_len, de->name,
977                              de->name_len, le32_to_cpu(de->inode));
978                 if (ext4_check_dir_entry(dir, NULL, de, bh,
979                                          inline_start, inline_size, offset))
980                         BUG();
981
982                 offset += de_len;
983                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
984         }
985 }
986 #else
987 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
988 #endif
989
990 /*
991  * Add a new entry into a inline dir.
992  * It will return -ENOSPC if no space is available, and -EIO
993  * and -EEXIST if directory entry already exists.
994  */
995 static int ext4_add_dirent_to_inline(handle_t *handle,
996                                      struct dentry *dentry,
997                                      struct inode *inode,
998                                      struct ext4_iloc *iloc,
999                                      void *inline_start, int inline_size)
1000 {
1001         struct inode    *dir = dentry->d_parent->d_inode;
1002         const char      *name = dentry->d_name.name;
1003         int             namelen = dentry->d_name.len;
1004         int             err;
1005         struct ext4_dir_entry_2 *de;
1006
1007         err = ext4_find_dest_de(dir, inode, iloc->bh,
1008                                 inline_start, inline_size,
1009                                 name, namelen, &de);
1010         if (err)
1011                 return err;
1012
1013         err = ext4_journal_get_write_access(handle, iloc->bh);
1014         if (err)
1015                 return err;
1016         ext4_insert_dentry(inode, de, inline_size, name, namelen);
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_RO_COMPAT_FEATURE(inode->i_sb,
1135                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1136                 csum_size = sizeof(struct ext4_dir_entry_tail);
1137
1138         inode->i_size = inode->i_sb->s_blocksize;
1139         i_size_write(inode, inode->i_sb->s_blocksize);
1140         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1141         ext4_update_final_de(dir_block->b_data,
1142                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1143                         inode->i_sb->s_blocksize - csum_size);
1144
1145         if (csum_size) {
1146                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1147                                      inode->i_sb->s_blocksize);
1148                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1149         }
1150         set_buffer_uptodate(dir_block);
1151         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1152         if (err)
1153                 goto out;
1154         set_buffer_verified(dir_block);
1155 out:
1156         return err;
1157 }
1158
1159 static int ext4_convert_inline_data_nolock(handle_t *handle,
1160                                            struct inode *inode,
1161                                            struct ext4_iloc *iloc)
1162 {
1163         int error;
1164         void *buf = NULL;
1165         struct buffer_head *data_bh = NULL;
1166         struct ext4_map_blocks map;
1167         int inline_size;
1168
1169         inline_size = ext4_get_inline_size(inode);
1170         buf = kmalloc(inline_size, GFP_NOFS);
1171         if (!buf) {
1172                 error = -ENOMEM;
1173                 goto out;
1174         }
1175
1176         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1177         if (error < 0)
1178                 goto out;
1179
1180         error = ext4_destroy_inline_data_nolock(handle, inode);
1181         if (error)
1182                 goto out;
1183
1184         map.m_lblk = 0;
1185         map.m_len = 1;
1186         map.m_flags = 0;
1187         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1188         if (error < 0)
1189                 goto out_restore;
1190         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1191                 error = -EIO;
1192                 goto out_restore;
1193         }
1194
1195         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1196         if (!data_bh) {
1197                 error = -ENOMEM;
1198                 goto out_restore;
1199         }
1200
1201         lock_buffer(data_bh);
1202         error = ext4_journal_get_create_access(handle, data_bh);
1203         if (error) {
1204                 unlock_buffer(data_bh);
1205                 error = -EIO;
1206                 goto out_restore;
1207         }
1208         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1209
1210         if (!S_ISDIR(inode->i_mode)) {
1211                 memcpy(data_bh->b_data, buf, inline_size);
1212                 set_buffer_uptodate(data_bh);
1213                 error = ext4_handle_dirty_metadata(handle,
1214                                                    inode, data_bh);
1215         } else {
1216                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1217                                                        buf, inline_size);
1218         }
1219
1220         unlock_buffer(data_bh);
1221 out_restore:
1222         if (error)
1223                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1224
1225 out:
1226         brelse(data_bh);
1227         kfree(buf);
1228         return error;
1229 }
1230
1231 /*
1232  * Try to add the new entry to the inline data.
1233  * If succeeds, return 0. If not, extended the inline dir and copied data to
1234  * the new created block.
1235  */
1236 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1237                               struct inode *inode)
1238 {
1239         int ret, inline_size;
1240         void *inline_start;
1241         struct ext4_iloc iloc;
1242         struct inode *dir = dentry->d_parent->d_inode;
1243
1244         ret = ext4_get_inode_loc(dir, &iloc);
1245         if (ret)
1246                 return ret;
1247
1248         down_write(&EXT4_I(dir)->xattr_sem);
1249         if (!ext4_has_inline_data(dir))
1250                 goto out;
1251
1252         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1253                                                  EXT4_INLINE_DOTDOT_SIZE;
1254         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1255
1256         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1257                                         inline_start, inline_size);
1258         if (ret != -ENOSPC)
1259                 goto out;
1260
1261         /* check whether it can be inserted to inline xattr space. */
1262         inline_size = EXT4_I(dir)->i_inline_size -
1263                         EXT4_MIN_INLINE_DATA_SIZE;
1264         if (!inline_size) {
1265                 /* Try to use the xattr space.*/
1266                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1267                 if (ret && ret != -ENOSPC)
1268                         goto out;
1269
1270                 inline_size = EXT4_I(dir)->i_inline_size -
1271                                 EXT4_MIN_INLINE_DATA_SIZE;
1272         }
1273
1274         if (inline_size) {
1275                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1276
1277                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1278                                                 inline_start, inline_size);
1279
1280                 if (ret != -ENOSPC)
1281                         goto out;
1282         }
1283
1284         /*
1285          * The inline space is filled up, so create a new block for it.
1286          * As the extent tree will be created, we have to save the inline
1287          * dir first.
1288          */
1289         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1290
1291 out:
1292         ext4_mark_inode_dirty(handle, dir);
1293         up_write(&EXT4_I(dir)->xattr_sem);
1294         brelse(iloc.bh);
1295         return ret;
1296 }
1297
1298 /*
1299  * This function fills a red-black tree with information from an
1300  * inlined dir.  It returns the number directory entries loaded
1301  * into the tree.  If there is an error it is returned in err.
1302  */
1303 int htree_inlinedir_to_tree(struct file *dir_file,
1304                             struct inode *dir, ext4_lblk_t block,
1305                             struct dx_hash_info *hinfo,
1306                             __u32 start_hash, __u32 start_minor_hash,
1307                             int *has_inline_data)
1308 {
1309         int err = 0, count = 0;
1310         unsigned int parent_ino;
1311         int pos;
1312         struct ext4_dir_entry_2 *de;
1313         struct inode *inode = file_inode(dir_file);
1314         int ret, inline_size = 0;
1315         struct ext4_iloc iloc;
1316         void *dir_buf = NULL;
1317         struct ext4_dir_entry_2 fake;
1318
1319         ret = ext4_get_inode_loc(inode, &iloc);
1320         if (ret)
1321                 return ret;
1322
1323         down_read(&EXT4_I(inode)->xattr_sem);
1324         if (!ext4_has_inline_data(inode)) {
1325                 up_read(&EXT4_I(inode)->xattr_sem);
1326                 *has_inline_data = 0;
1327                 goto out;
1328         }
1329
1330         inline_size = ext4_get_inline_size(inode);
1331         dir_buf = kmalloc(inline_size, GFP_NOFS);
1332         if (!dir_buf) {
1333                 ret = -ENOMEM;
1334                 up_read(&EXT4_I(inode)->xattr_sem);
1335                 goto out;
1336         }
1337
1338         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1339         up_read(&EXT4_I(inode)->xattr_sem);
1340         if (ret < 0)
1341                 goto out;
1342
1343         pos = 0;
1344         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1345         while (pos < inline_size) {
1346                 /*
1347                  * As inlined dir doesn't store any information about '.' and
1348                  * only the inode number of '..' is stored, we have to handle
1349                  * them differently.
1350                  */
1351                 if (pos == 0) {
1352                         fake.inode = cpu_to_le32(inode->i_ino);
1353                         fake.name_len = 1;
1354                         strcpy(fake.name, ".");
1355                         fake.rec_len = ext4_rec_len_to_disk(
1356                                                 EXT4_DIR_REC_LEN(fake.name_len),
1357                                                 inline_size);
1358                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1359                         de = &fake;
1360                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1361                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1362                         fake.inode = cpu_to_le32(parent_ino);
1363                         fake.name_len = 2;
1364                         strcpy(fake.name, "..");
1365                         fake.rec_len = ext4_rec_len_to_disk(
1366                                                 EXT4_DIR_REC_LEN(fake.name_len),
1367                                                 inline_size);
1368                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1369                         de = &fake;
1370                         pos = EXT4_INLINE_DOTDOT_SIZE;
1371                 } else {
1372                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1373                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1374                         if (ext4_check_dir_entry(inode, dir_file, de,
1375                                          iloc.bh, dir_buf,
1376                                          inline_size, pos)) {
1377                                 ret = count;
1378                                 goto out;
1379                         }
1380                 }
1381
1382                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1383                 if ((hinfo->hash < start_hash) ||
1384                     ((hinfo->hash == start_hash) &&
1385                      (hinfo->minor_hash < start_minor_hash)))
1386                         continue;
1387                 if (de->inode == 0)
1388                         continue;
1389                 err = ext4_htree_store_dirent(dir_file,
1390                                    hinfo->hash, hinfo->minor_hash, de);
1391                 if (err) {
1392                         count = err;
1393                         goto out;
1394                 }
1395                 count++;
1396         }
1397         ret = count;
1398 out:
1399         kfree(dir_buf);
1400         brelse(iloc.bh);
1401         return ret;
1402 }
1403
1404 /*
1405  * So this function is called when the volume is mkfsed with
1406  * dir_index disabled. In order to keep f_pos persistent
1407  * after we convert from an inlined dir to a blocked based,
1408  * we just pretend that we are a normal dir and return the
1409  * offset as if '.' and '..' really take place.
1410  *
1411  */
1412 int ext4_read_inline_dir(struct file *file,
1413                          struct dir_context *ctx,
1414                          int *has_inline_data)
1415 {
1416         unsigned int offset, parent_ino;
1417         int i;
1418         struct ext4_dir_entry_2 *de;
1419         struct super_block *sb;
1420         struct inode *inode = file_inode(file);
1421         int ret, inline_size = 0;
1422         struct ext4_iloc iloc;
1423         void *dir_buf = NULL;
1424         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1425
1426         ret = ext4_get_inode_loc(inode, &iloc);
1427         if (ret)
1428                 return ret;
1429
1430         down_read(&EXT4_I(inode)->xattr_sem);
1431         if (!ext4_has_inline_data(inode)) {
1432                 up_read(&EXT4_I(inode)->xattr_sem);
1433                 *has_inline_data = 0;
1434                 goto out;
1435         }
1436
1437         inline_size = ext4_get_inline_size(inode);
1438         dir_buf = kmalloc(inline_size, GFP_NOFS);
1439         if (!dir_buf) {
1440                 ret = -ENOMEM;
1441                 up_read(&EXT4_I(inode)->xattr_sem);
1442                 goto out;
1443         }
1444
1445         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1446         up_read(&EXT4_I(inode)->xattr_sem);
1447         if (ret < 0)
1448                 goto out;
1449
1450         ret = 0;
1451         sb = inode->i_sb;
1452         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1453         offset = ctx->pos;
1454
1455         /*
1456          * dotdot_offset and dotdot_size is the real offset and
1457          * size for ".." and "." if the dir is block based while
1458          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1459          * So we will use extra_offset and extra_size to indicate them
1460          * during the inline dir iteration.
1461          */
1462         dotdot_offset = EXT4_DIR_REC_LEN(1);
1463         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1464         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1465         extra_size = extra_offset + inline_size;
1466
1467         /*
1468          * If the version has changed since the last call to
1469          * readdir(2), then we might be pointing to an invalid
1470          * dirent right now.  Scan from the start of the inline
1471          * dir to make sure.
1472          */
1473         if (file->f_version != inode->i_version) {
1474                 for (i = 0; i < extra_size && i < offset;) {
1475                         /*
1476                          * "." is with offset 0 and
1477                          * ".." is dotdot_offset.
1478                          */
1479                         if (!i) {
1480                                 i = dotdot_offset;
1481                                 continue;
1482                         } else if (i == dotdot_offset) {
1483                                 i = dotdot_size;
1484                                 continue;
1485                         }
1486                         /* for other entry, the real offset in
1487                          * the buf has to be tuned accordingly.
1488                          */
1489                         de = (struct ext4_dir_entry_2 *)
1490                                 (dir_buf + i - extra_offset);
1491                         /* It's too expensive to do a full
1492                          * dirent test each time round this
1493                          * loop, but we do have to test at
1494                          * least that it is non-zero.  A
1495                          * failure will be detected in the
1496                          * dirent test below. */
1497                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1498                                 < EXT4_DIR_REC_LEN(1))
1499                                 break;
1500                         i += ext4_rec_len_from_disk(de->rec_len,
1501                                                     extra_size);
1502                 }
1503                 offset = i;
1504                 ctx->pos = offset;
1505                 file->f_version = inode->i_version;
1506         }
1507
1508         while (ctx->pos < extra_size) {
1509                 if (ctx->pos == 0) {
1510                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1511                                 goto out;
1512                         ctx->pos = dotdot_offset;
1513                         continue;
1514                 }
1515
1516                 if (ctx->pos == dotdot_offset) {
1517                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1518                                 goto out;
1519                         ctx->pos = dotdot_size;
1520                         continue;
1521                 }
1522
1523                 de = (struct ext4_dir_entry_2 *)
1524                         (dir_buf + ctx->pos - extra_offset);
1525                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1526                                          extra_size, ctx->pos))
1527                         goto out;
1528                 if (le32_to_cpu(de->inode)) {
1529                         if (!dir_emit(ctx, de->name, de->name_len,
1530                                       le32_to_cpu(de->inode),
1531                                       get_dtype(sb, de->file_type)))
1532                                 goto out;
1533                 }
1534                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1535         }
1536 out:
1537         kfree(dir_buf);
1538         brelse(iloc.bh);
1539         return ret;
1540 }
1541
1542 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1543                                         struct ext4_dir_entry_2 **parent_de,
1544                                         int *retval)
1545 {
1546         struct ext4_iloc iloc;
1547
1548         *retval = ext4_get_inode_loc(inode, &iloc);
1549         if (*retval)
1550                 return NULL;
1551
1552         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1553
1554         return iloc.bh;
1555 }
1556
1557 /*
1558  * Try to create the inline data for the new dir.
1559  * If it succeeds, return 0, otherwise return the error.
1560  * In case of ENOSPC, the caller should create the normal disk layout dir.
1561  */
1562 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1563                                struct inode *inode)
1564 {
1565         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1566         struct ext4_iloc iloc;
1567         struct ext4_dir_entry_2 *de;
1568
1569         ret = ext4_get_inode_loc(inode, &iloc);
1570         if (ret)
1571                 return ret;
1572
1573         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1574         if (ret)
1575                 goto out;
1576
1577         /*
1578          * For inline dir, we only save the inode information for the ".."
1579          * and create a fake dentry to cover the left space.
1580          */
1581         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1582         de->inode = cpu_to_le32(parent->i_ino);
1583         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1584         de->inode = 0;
1585         de->rec_len = ext4_rec_len_to_disk(
1586                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1587                                 inline_size);
1588         set_nlink(inode, 2);
1589         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1590 out:
1591         brelse(iloc.bh);
1592         return ret;
1593 }
1594
1595 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1596                                         const struct qstr *d_name,
1597                                         struct ext4_dir_entry_2 **res_dir,
1598                                         int *has_inline_data)
1599 {
1600         int ret;
1601         struct ext4_iloc iloc;
1602         void *inline_start;
1603         int inline_size;
1604
1605         if (ext4_get_inode_loc(dir, &iloc))
1606                 return NULL;
1607
1608         down_read(&EXT4_I(dir)->xattr_sem);
1609         if (!ext4_has_inline_data(dir)) {
1610                 *has_inline_data = 0;
1611                 goto out;
1612         }
1613
1614         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1615                                                 EXT4_INLINE_DOTDOT_SIZE;
1616         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1617         ret = search_dir(iloc.bh, inline_start, inline_size,
1618                          dir, d_name, 0, res_dir);
1619         if (ret == 1)
1620                 goto out_find;
1621         if (ret < 0)
1622                 goto out;
1623
1624         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1625                 goto out;
1626
1627         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1628         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1629
1630         ret = search_dir(iloc.bh, inline_start, inline_size,
1631                          dir, d_name, 0, res_dir);
1632         if (ret == 1)
1633                 goto out_find;
1634
1635 out:
1636         brelse(iloc.bh);
1637         iloc.bh = NULL;
1638 out_find:
1639         up_read(&EXT4_I(dir)->xattr_sem);
1640         return iloc.bh;
1641 }
1642
1643 int ext4_delete_inline_entry(handle_t *handle,
1644                              struct inode *dir,
1645                              struct ext4_dir_entry_2 *de_del,
1646                              struct buffer_head *bh,
1647                              int *has_inline_data)
1648 {
1649         int err, inline_size;
1650         struct ext4_iloc iloc;
1651         void *inline_start;
1652
1653         err = ext4_get_inode_loc(dir, &iloc);
1654         if (err)
1655                 return err;
1656
1657         down_write(&EXT4_I(dir)->xattr_sem);
1658         if (!ext4_has_inline_data(dir)) {
1659                 *has_inline_data = 0;
1660                 goto out;
1661         }
1662
1663         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1664                 EXT4_MIN_INLINE_DATA_SIZE) {
1665                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1666                                         EXT4_INLINE_DOTDOT_SIZE;
1667                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1668                                 EXT4_INLINE_DOTDOT_SIZE;
1669         } else {
1670                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1671                 inline_size = ext4_get_inline_size(dir) -
1672                                 EXT4_MIN_INLINE_DATA_SIZE;
1673         }
1674
1675         err = ext4_journal_get_write_access(handle, bh);
1676         if (err)
1677                 goto out;
1678
1679         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1680                                         inline_start, inline_size, 0);
1681         if (err)
1682                 goto out;
1683
1684         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1685         err = ext4_mark_inode_dirty(handle, dir);
1686         if (unlikely(err))
1687                 goto out;
1688
1689         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1690 out:
1691         up_write(&EXT4_I(dir)->xattr_sem);
1692         brelse(iloc.bh);
1693         if (err != -ENOENT)
1694                 ext4_std_error(dir->i_sb, err);
1695         return err;
1696 }
1697
1698 /*
1699  * Get the inline dentry at offset.
1700  */
1701 static inline struct ext4_dir_entry_2 *
1702 ext4_get_inline_entry(struct inode *inode,
1703                       struct ext4_iloc *iloc,
1704                       unsigned int offset,
1705                       void **inline_start,
1706                       int *inline_size)
1707 {
1708         void *inline_pos;
1709
1710         BUG_ON(offset > ext4_get_inline_size(inode));
1711
1712         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1713                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1714                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1715         } else {
1716                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1717                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1718                 *inline_size = ext4_get_inline_size(inode) -
1719                                 EXT4_MIN_INLINE_DATA_SIZE;
1720         }
1721
1722         if (inline_start)
1723                 *inline_start = inline_pos;
1724         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1725 }
1726
1727 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1728 {
1729         int err, inline_size;
1730         struct ext4_iloc iloc;
1731         void *inline_pos;
1732         unsigned int offset;
1733         struct ext4_dir_entry_2 *de;
1734         int ret = 1;
1735
1736         err = ext4_get_inode_loc(dir, &iloc);
1737         if (err) {
1738                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1739                                  err, dir->i_ino);
1740                 return 1;
1741         }
1742
1743         down_read(&EXT4_I(dir)->xattr_sem);
1744         if (!ext4_has_inline_data(dir)) {
1745                 *has_inline_data = 0;
1746                 goto out;
1747         }
1748
1749         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1750         if (!le32_to_cpu(de->inode)) {
1751                 ext4_warning(dir->i_sb,
1752                              "bad inline directory (dir #%lu) - no `..'",
1753                              dir->i_ino);
1754                 ret = 1;
1755                 goto out;
1756         }
1757
1758         offset = EXT4_INLINE_DOTDOT_SIZE;
1759         while (offset < dir->i_size) {
1760                 de = ext4_get_inline_entry(dir, &iloc, offset,
1761                                            &inline_pos, &inline_size);
1762                 if (ext4_check_dir_entry(dir, NULL, de,
1763                                          iloc.bh, inline_pos,
1764                                          inline_size, offset)) {
1765                         ext4_warning(dir->i_sb,
1766                                      "bad inline directory (dir #%lu) - "
1767                                      "inode %u, rec_len %u, name_len %d"
1768                                      "inline size %d\n",
1769                                      dir->i_ino, le32_to_cpu(de->inode),
1770                                      le16_to_cpu(de->rec_len), de->name_len,
1771                                      inline_size);
1772                         ret = 1;
1773                         goto out;
1774                 }
1775                 if (le32_to_cpu(de->inode)) {
1776                         ret = 0;
1777                         goto out;
1778                 }
1779                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1780         }
1781
1782 out:
1783         up_read(&EXT4_I(dir)->xattr_sem);
1784         brelse(iloc.bh);
1785         return ret;
1786 }
1787
1788 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1789 {
1790         int ret;
1791
1792         down_write(&EXT4_I(inode)->xattr_sem);
1793         ret = ext4_destroy_inline_data_nolock(handle, inode);
1794         up_write(&EXT4_I(inode)->xattr_sem);
1795
1796         return ret;
1797 }
1798
1799 int ext4_inline_data_fiemap(struct inode *inode,
1800                             struct fiemap_extent_info *fieinfo,
1801                             int *has_inline)
1802 {
1803         __u64 physical = 0;
1804         __u64 length;
1805         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1806         int error = 0;
1807         struct ext4_iloc iloc;
1808
1809         down_read(&EXT4_I(inode)->xattr_sem);
1810         if (!ext4_has_inline_data(inode)) {
1811                 *has_inline = 0;
1812                 goto out;
1813         }
1814
1815         error = ext4_get_inode_loc(inode, &iloc);
1816         if (error)
1817                 goto out;
1818
1819         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1820         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1821         physical += offsetof(struct ext4_inode, i_block);
1822         length = i_size_read(inode);
1823
1824         if (physical)
1825                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1826                                                 length, flags);
1827         brelse(iloc.bh);
1828 out:
1829         up_read(&EXT4_I(inode)->xattr_sem);
1830         return (error < 0 ? error : 0);
1831 }
1832
1833 /*
1834  * Called during xattr set, and if we can sparse space 'needed',
1835  * just create the extent tree evict the data to the outer block.
1836  *
1837  * We use jbd2 instead of page cache to move data to the 1st block
1838  * so that the whole transaction can be committed as a whole and
1839  * the data isn't lost because of the delayed page cache write.
1840  */
1841 int ext4_try_to_evict_inline_data(handle_t *handle,
1842                                   struct inode *inode,
1843                                   int needed)
1844 {
1845         int error;
1846         struct ext4_xattr_entry *entry;
1847         struct ext4_xattr_ibody_header *header;
1848         struct ext4_inode *raw_inode;
1849         struct ext4_iloc iloc;
1850
1851         error = ext4_get_inode_loc(inode, &iloc);
1852         if (error)
1853                 return error;
1854
1855         raw_inode = ext4_raw_inode(&iloc);
1856         header = IHDR(inode, raw_inode);
1857         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1858                                             EXT4_I(inode)->i_inline_off);
1859         if (EXT4_XATTR_LEN(entry->e_name_len) +
1860             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1861                 error = -ENOSPC;
1862                 goto out;
1863         }
1864
1865         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1866 out:
1867         brelse(iloc.bh);
1868         return error;
1869 }
1870
1871 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1872 {
1873         handle_t *handle;
1874         int inline_size, value_len, needed_blocks;
1875         size_t i_size;
1876         void *value = NULL;
1877         struct ext4_xattr_ibody_find is = {
1878                 .s = { .not_found = -ENODATA, },
1879         };
1880         struct ext4_xattr_info i = {
1881                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1882                 .name = EXT4_XATTR_SYSTEM_DATA,
1883         };
1884
1885
1886         needed_blocks = ext4_writepage_trans_blocks(inode);
1887         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1888         if (IS_ERR(handle))
1889                 return;
1890
1891         down_write(&EXT4_I(inode)->xattr_sem);
1892         if (!ext4_has_inline_data(inode)) {
1893                 *has_inline = 0;
1894                 ext4_journal_stop(handle);
1895                 return;
1896         }
1897
1898         if (ext4_orphan_add(handle, inode))
1899                 goto out;
1900
1901         if (ext4_get_inode_loc(inode, &is.iloc))
1902                 goto out;
1903
1904         down_write(&EXT4_I(inode)->i_data_sem);
1905         i_size = inode->i_size;
1906         inline_size = ext4_get_inline_size(inode);
1907         EXT4_I(inode)->i_disksize = i_size;
1908
1909         if (i_size < inline_size) {
1910                 /* Clear the content in the xattr space. */
1911                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1912                         if (ext4_xattr_ibody_find(inode, &i, &is))
1913                                 goto out_error;
1914
1915                         BUG_ON(is.s.not_found);
1916
1917                         value_len = le32_to_cpu(is.s.here->e_value_size);
1918                         value = kmalloc(value_len, GFP_NOFS);
1919                         if (!value)
1920                                 goto out_error;
1921
1922                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1923                                                 value, value_len))
1924                                 goto out_error;
1925
1926                         i.value = value;
1927                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1928                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1929                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1930                                 goto out_error;
1931                 }
1932
1933                 /* Clear the content within i_blocks. */
1934                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE)
1935                         memset(ext4_raw_inode(&is.iloc)->i_block + i_size, 0,
1936                                         EXT4_MIN_INLINE_DATA_SIZE - i_size);
1937
1938                 EXT4_I(inode)->i_inline_size = i_size <
1939                                         EXT4_MIN_INLINE_DATA_SIZE ?
1940                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1941         }
1942
1943 out_error:
1944         up_write(&EXT4_I(inode)->i_data_sem);
1945 out:
1946         brelse(is.iloc.bh);
1947         up_write(&EXT4_I(inode)->xattr_sem);
1948         kfree(value);
1949         if (inode->i_nlink)
1950                 ext4_orphan_del(handle, inode);
1951
1952         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1953         ext4_mark_inode_dirty(handle, inode);
1954         if (IS_SYNC(inode))
1955                 ext4_handle_sync(handle);
1956
1957         ext4_journal_stop(handle);
1958         return;
1959 }
1960
1961 int ext4_convert_inline_data(struct inode *inode)
1962 {
1963         int error, needed_blocks;
1964         handle_t *handle;
1965         struct ext4_iloc iloc;
1966
1967         if (!ext4_has_inline_data(inode)) {
1968                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1969                 return 0;
1970         }
1971
1972         needed_blocks = ext4_writepage_trans_blocks(inode);
1973
1974         iloc.bh = NULL;
1975         error = ext4_get_inode_loc(inode, &iloc);
1976         if (error)
1977                 return error;
1978
1979         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1980         if (IS_ERR(handle)) {
1981                 error = PTR_ERR(handle);
1982                 goto out_free;
1983         }
1984
1985         down_write(&EXT4_I(inode)->xattr_sem);
1986         if (!ext4_has_inline_data(inode)) {
1987                 up_write(&EXT4_I(inode)->xattr_sem);
1988                 goto out;
1989         }
1990
1991         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1992         up_write(&EXT4_I(inode)->xattr_sem);
1993 out:
1994         ext4_journal_stop(handle);
1995 out_free:
1996         brelse(iloc.bh);
1997         return error;
1998 }