]> Pileus Git - ~andy/linux/blob - fs/ext4/inline.c
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jesse/openvswitch
[~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                 goto out;
863         }
864
865         inline_size = ext4_get_max_inline_size(inode);
866
867         ret = -ENOSPC;
868         if (inline_size >= pos + len) {
869                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
870                 if (ret && ret != -ENOSPC)
871                         goto out_journal;
872         }
873
874         if (ret == -ENOSPC) {
875                 ret = ext4_da_convert_inline_data_to_extent(mapping,
876                                                             inode,
877                                                             flags,
878                                                             fsdata);
879                 ext4_journal_stop(handle);
880                 if (ret == -ENOSPC &&
881                     ext4_should_retry_alloc(inode->i_sb, &retries))
882                         goto retry_journal;
883                 goto out;
884         }
885
886         /*
887          * We cannot recurse into the filesystem as the transaction
888          * is already started.
889          */
890         flags |= AOP_FLAG_NOFS;
891
892         page = grab_cache_page_write_begin(mapping, 0, flags);
893         if (!page) {
894                 ret = -ENOMEM;
895                 goto out_journal;
896         }
897
898         down_read(&EXT4_I(inode)->xattr_sem);
899         if (!ext4_has_inline_data(inode)) {
900                 ret = 0;
901                 goto out_release_page;
902         }
903
904         if (!PageUptodate(page)) {
905                 ret = ext4_read_inline_page(inode, page);
906                 if (ret < 0)
907                         goto out_release_page;
908         }
909
910         up_read(&EXT4_I(inode)->xattr_sem);
911         *pagep = page;
912         brelse(iloc.bh);
913         return 1;
914 out_release_page:
915         up_read(&EXT4_I(inode)->xattr_sem);
916         unlock_page(page);
917         page_cache_release(page);
918 out_journal:
919         ext4_journal_stop(handle);
920 out:
921         brelse(iloc.bh);
922         return ret;
923 }
924
925 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
926                                   unsigned len, unsigned copied,
927                                   struct page *page)
928 {
929         int i_size_changed = 0;
930
931         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
932
933         /*
934          * No need to use i_size_read() here, the i_size
935          * cannot change under us because we hold i_mutex.
936          *
937          * But it's important to update i_size while still holding page lock:
938          * page writeout could otherwise come in and zero beyond i_size.
939          */
940         if (pos+copied > inode->i_size) {
941                 i_size_write(inode, pos+copied);
942                 i_size_changed = 1;
943         }
944         unlock_page(page);
945         page_cache_release(page);
946
947         /*
948          * Don't mark the inode dirty under page lock. First, it unnecessarily
949          * makes the holding time of page lock longer. Second, it forces lock
950          * ordering of page lock and transaction start for journaling
951          * filesystems.
952          */
953         if (i_size_changed)
954                 mark_inode_dirty(inode);
955
956         return copied;
957 }
958
959 #ifdef INLINE_DIR_DEBUG
960 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
961                           void *inline_start, int inline_size)
962 {
963         int offset;
964         unsigned short de_len;
965         struct ext4_dir_entry_2 *de = inline_start;
966         void *dlimit = inline_start + inline_size;
967
968         trace_printk("inode %lu\n", dir->i_ino);
969         offset = 0;
970         while ((void *)de < dlimit) {
971                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
972                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
973                              offset, de_len, de->name_len, de->name,
974                              de->name_len, le32_to_cpu(de->inode));
975                 if (ext4_check_dir_entry(dir, NULL, de, bh,
976                                          inline_start, inline_size, offset))
977                         BUG();
978
979                 offset += de_len;
980                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
981         }
982 }
983 #else
984 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
985 #endif
986
987 /*
988  * Add a new entry into a inline dir.
989  * It will return -ENOSPC if no space is available, and -EIO
990  * and -EEXIST if directory entry already exists.
991  */
992 static int ext4_add_dirent_to_inline(handle_t *handle,
993                                      struct dentry *dentry,
994                                      struct inode *inode,
995                                      struct ext4_iloc *iloc,
996                                      void *inline_start, int inline_size)
997 {
998         struct inode    *dir = dentry->d_parent->d_inode;
999         const char      *name = dentry->d_name.name;
1000         int             namelen = dentry->d_name.len;
1001         int             err;
1002         struct ext4_dir_entry_2 *de;
1003
1004         err = ext4_find_dest_de(dir, inode, iloc->bh,
1005                                 inline_start, inline_size,
1006                                 name, namelen, &de);
1007         if (err)
1008                 return err;
1009
1010         err = ext4_journal_get_write_access(handle, iloc->bh);
1011         if (err)
1012                 return err;
1013         ext4_insert_dentry(inode, de, inline_size, name, namelen);
1014
1015         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1016
1017         /*
1018          * XXX shouldn't update any times until successful
1019          * completion of syscall, but too many callers depend
1020          * on this.
1021          *
1022          * XXX similarly, too many callers depend on
1023          * ext4_new_inode() setting the times, but error
1024          * recovery deletes the inode, so the worst that can
1025          * happen is that the times are slightly out of date
1026          * and/or different from the directory change time.
1027          */
1028         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1029         ext4_update_dx_flag(dir);
1030         dir->i_version++;
1031         ext4_mark_inode_dirty(handle, dir);
1032         return 1;
1033 }
1034
1035 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1036                                        struct ext4_iloc *iloc)
1037 {
1038         struct ext4_xattr_entry *entry;
1039         struct ext4_xattr_ibody_header *header;
1040
1041         BUG_ON(!EXT4_I(inode)->i_inline_off);
1042
1043         header = IHDR(inode, ext4_raw_inode(iloc));
1044         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1045                                             EXT4_I(inode)->i_inline_off);
1046
1047         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1048 }
1049
1050 /* Set the final de to cover the whole block. */
1051 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1052 {
1053         struct ext4_dir_entry_2 *de, *prev_de;
1054         void *limit;
1055         int de_len;
1056
1057         de = (struct ext4_dir_entry_2 *)de_buf;
1058         if (old_size) {
1059                 limit = de_buf + old_size;
1060                 do {
1061                         prev_de = de;
1062                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1063                         de_buf += de_len;
1064                         de = (struct ext4_dir_entry_2 *)de_buf;
1065                 } while (de_buf < limit);
1066
1067                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1068                                                         old_size, new_size);
1069         } else {
1070                 /* this is just created, so create an empty entry. */
1071                 de->inode = 0;
1072                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1073         }
1074 }
1075
1076 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1077                                   struct ext4_iloc *iloc)
1078 {
1079         int ret;
1080         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1081         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1082
1083         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1084                 return -ENOSPC;
1085
1086         ret = ext4_update_inline_data(handle, dir,
1087                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1088         if (ret)
1089                 return ret;
1090
1091         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1092                              EXT4_I(dir)->i_inline_size -
1093                                                 EXT4_MIN_INLINE_DATA_SIZE);
1094         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1095         return 0;
1096 }
1097
1098 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1099                                      struct ext4_iloc *iloc,
1100                                      void *buf, int inline_size)
1101 {
1102         ext4_create_inline_data(handle, inode, inline_size);
1103         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1104         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1105 }
1106
1107 static int ext4_finish_convert_inline_dir(handle_t *handle,
1108                                           struct inode *inode,
1109                                           struct buffer_head *dir_block,
1110                                           void *buf,
1111                                           int inline_size)
1112 {
1113         int err, csum_size = 0, header_size = 0;
1114         struct ext4_dir_entry_2 *de;
1115         struct ext4_dir_entry_tail *t;
1116         void *target = dir_block->b_data;
1117
1118         /*
1119          * First create "." and ".." and then copy the dir information
1120          * back to the block.
1121          */
1122         de = (struct ext4_dir_entry_2 *)target;
1123         de = ext4_init_dot_dotdot(inode, de,
1124                 inode->i_sb->s_blocksize, csum_size,
1125                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1126         header_size = (void *)de - target;
1127
1128         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1129                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1130
1131         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1132                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1133                 csum_size = sizeof(struct ext4_dir_entry_tail);
1134
1135         inode->i_size = inode->i_sb->s_blocksize;
1136         i_size_write(inode, inode->i_sb->s_blocksize);
1137         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1138         ext4_update_final_de(dir_block->b_data,
1139                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1140                         inode->i_sb->s_blocksize - csum_size);
1141
1142         if (csum_size) {
1143                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1144                                      inode->i_sb->s_blocksize);
1145                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1146         }
1147         set_buffer_uptodate(dir_block);
1148         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1149         if (err)
1150                 goto out;
1151         set_buffer_verified(dir_block);
1152 out:
1153         return err;
1154 }
1155
1156 static int ext4_convert_inline_data_nolock(handle_t *handle,
1157                                            struct inode *inode,
1158                                            struct ext4_iloc *iloc)
1159 {
1160         int error;
1161         void *buf = NULL;
1162         struct buffer_head *data_bh = NULL;
1163         struct ext4_map_blocks map;
1164         int inline_size;
1165
1166         inline_size = ext4_get_inline_size(inode);
1167         buf = kmalloc(inline_size, GFP_NOFS);
1168         if (!buf) {
1169                 error = -ENOMEM;
1170                 goto out;
1171         }
1172
1173         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1174         if (error < 0)
1175                 goto out;
1176
1177         error = ext4_destroy_inline_data_nolock(handle, inode);
1178         if (error)
1179                 goto out;
1180
1181         map.m_lblk = 0;
1182         map.m_len = 1;
1183         map.m_flags = 0;
1184         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1185         if (error < 0)
1186                 goto out_restore;
1187         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1188                 error = -EIO;
1189                 goto out_restore;
1190         }
1191
1192         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1193         if (!data_bh) {
1194                 error = -ENOMEM;
1195                 goto out_restore;
1196         }
1197
1198         lock_buffer(data_bh);
1199         error = ext4_journal_get_create_access(handle, data_bh);
1200         if (error) {
1201                 unlock_buffer(data_bh);
1202                 error = -EIO;
1203                 goto out_restore;
1204         }
1205         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1206
1207         if (!S_ISDIR(inode->i_mode)) {
1208                 memcpy(data_bh->b_data, buf, inline_size);
1209                 set_buffer_uptodate(data_bh);
1210                 error = ext4_handle_dirty_metadata(handle,
1211                                                    inode, data_bh);
1212         } else {
1213                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1214                                                        buf, inline_size);
1215         }
1216
1217         unlock_buffer(data_bh);
1218 out_restore:
1219         if (error)
1220                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1221
1222 out:
1223         brelse(data_bh);
1224         kfree(buf);
1225         return error;
1226 }
1227
1228 /*
1229  * Try to add the new entry to the inline data.
1230  * If succeeds, return 0. If not, extended the inline dir and copied data to
1231  * the new created block.
1232  */
1233 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1234                               struct inode *inode)
1235 {
1236         int ret, inline_size;
1237         void *inline_start;
1238         struct ext4_iloc iloc;
1239         struct inode *dir = dentry->d_parent->d_inode;
1240
1241         ret = ext4_get_inode_loc(dir, &iloc);
1242         if (ret)
1243                 return ret;
1244
1245         down_write(&EXT4_I(dir)->xattr_sem);
1246         if (!ext4_has_inline_data(dir))
1247                 goto out;
1248
1249         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1250                                                  EXT4_INLINE_DOTDOT_SIZE;
1251         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1252
1253         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1254                                         inline_start, inline_size);
1255         if (ret != -ENOSPC)
1256                 goto out;
1257
1258         /* check whether it can be inserted to inline xattr space. */
1259         inline_size = EXT4_I(dir)->i_inline_size -
1260                         EXT4_MIN_INLINE_DATA_SIZE;
1261         if (!inline_size) {
1262                 /* Try to use the xattr space.*/
1263                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1264                 if (ret && ret != -ENOSPC)
1265                         goto out;
1266
1267                 inline_size = EXT4_I(dir)->i_inline_size -
1268                                 EXT4_MIN_INLINE_DATA_SIZE;
1269         }
1270
1271         if (inline_size) {
1272                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1273
1274                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1275                                                 inline_start, inline_size);
1276
1277                 if (ret != -ENOSPC)
1278                         goto out;
1279         }
1280
1281         /*
1282          * The inline space is filled up, so create a new block for it.
1283          * As the extent tree will be created, we have to save the inline
1284          * dir first.
1285          */
1286         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1287
1288 out:
1289         ext4_mark_inode_dirty(handle, dir);
1290         up_write(&EXT4_I(dir)->xattr_sem);
1291         brelse(iloc.bh);
1292         return ret;
1293 }
1294
1295 /*
1296  * This function fills a red-black tree with information from an
1297  * inlined dir.  It returns the number directory entries loaded
1298  * into the tree.  If there is an error it is returned in err.
1299  */
1300 int htree_inlinedir_to_tree(struct file *dir_file,
1301                             struct inode *dir, ext4_lblk_t block,
1302                             struct dx_hash_info *hinfo,
1303                             __u32 start_hash, __u32 start_minor_hash,
1304                             int *has_inline_data)
1305 {
1306         int err = 0, count = 0;
1307         unsigned int parent_ino;
1308         int pos;
1309         struct ext4_dir_entry_2 *de;
1310         struct inode *inode = file_inode(dir_file);
1311         int ret, inline_size = 0;
1312         struct ext4_iloc iloc;
1313         void *dir_buf = NULL;
1314         struct ext4_dir_entry_2 fake;
1315
1316         ret = ext4_get_inode_loc(inode, &iloc);
1317         if (ret)
1318                 return ret;
1319
1320         down_read(&EXT4_I(inode)->xattr_sem);
1321         if (!ext4_has_inline_data(inode)) {
1322                 up_read(&EXT4_I(inode)->xattr_sem);
1323                 *has_inline_data = 0;
1324                 goto out;
1325         }
1326
1327         inline_size = ext4_get_inline_size(inode);
1328         dir_buf = kmalloc(inline_size, GFP_NOFS);
1329         if (!dir_buf) {
1330                 ret = -ENOMEM;
1331                 up_read(&EXT4_I(inode)->xattr_sem);
1332                 goto out;
1333         }
1334
1335         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1336         up_read(&EXT4_I(inode)->xattr_sem);
1337         if (ret < 0)
1338                 goto out;
1339
1340         pos = 0;
1341         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1342         while (pos < inline_size) {
1343                 /*
1344                  * As inlined dir doesn't store any information about '.' and
1345                  * only the inode number of '..' is stored, we have to handle
1346                  * them differently.
1347                  */
1348                 if (pos == 0) {
1349                         fake.inode = cpu_to_le32(inode->i_ino);
1350                         fake.name_len = 1;
1351                         strcpy(fake.name, ".");
1352                         fake.rec_len = ext4_rec_len_to_disk(
1353                                                 EXT4_DIR_REC_LEN(fake.name_len),
1354                                                 inline_size);
1355                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1356                         de = &fake;
1357                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1358                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1359                         fake.inode = cpu_to_le32(parent_ino);
1360                         fake.name_len = 2;
1361                         strcpy(fake.name, "..");
1362                         fake.rec_len = ext4_rec_len_to_disk(
1363                                                 EXT4_DIR_REC_LEN(fake.name_len),
1364                                                 inline_size);
1365                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1366                         de = &fake;
1367                         pos = EXT4_INLINE_DOTDOT_SIZE;
1368                 } else {
1369                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1370                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1371                         if (ext4_check_dir_entry(inode, dir_file, de,
1372                                          iloc.bh, dir_buf,
1373                                          inline_size, pos)) {
1374                                 ret = count;
1375                                 goto out;
1376                         }
1377                 }
1378
1379                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1380                 if ((hinfo->hash < start_hash) ||
1381                     ((hinfo->hash == start_hash) &&
1382                      (hinfo->minor_hash < start_minor_hash)))
1383                         continue;
1384                 if (de->inode == 0)
1385                         continue;
1386                 err = ext4_htree_store_dirent(dir_file,
1387                                    hinfo->hash, hinfo->minor_hash, de);
1388                 if (err) {
1389                         count = err;
1390                         goto out;
1391                 }
1392                 count++;
1393         }
1394         ret = count;
1395 out:
1396         kfree(dir_buf);
1397         brelse(iloc.bh);
1398         return ret;
1399 }
1400
1401 /*
1402  * So this function is called when the volume is mkfsed with
1403  * dir_index disabled. In order to keep f_pos persistent
1404  * after we convert from an inlined dir to a blocked based,
1405  * we just pretend that we are a normal dir and return the
1406  * offset as if '.' and '..' really take place.
1407  *
1408  */
1409 int ext4_read_inline_dir(struct file *file,
1410                          struct dir_context *ctx,
1411                          int *has_inline_data)
1412 {
1413         unsigned int offset, parent_ino;
1414         int i;
1415         struct ext4_dir_entry_2 *de;
1416         struct super_block *sb;
1417         struct inode *inode = file_inode(file);
1418         int ret, inline_size = 0;
1419         struct ext4_iloc iloc;
1420         void *dir_buf = NULL;
1421         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1422
1423         ret = ext4_get_inode_loc(inode, &iloc);
1424         if (ret)
1425                 return ret;
1426
1427         down_read(&EXT4_I(inode)->xattr_sem);
1428         if (!ext4_has_inline_data(inode)) {
1429                 up_read(&EXT4_I(inode)->xattr_sem);
1430                 *has_inline_data = 0;
1431                 goto out;
1432         }
1433
1434         inline_size = ext4_get_inline_size(inode);
1435         dir_buf = kmalloc(inline_size, GFP_NOFS);
1436         if (!dir_buf) {
1437                 ret = -ENOMEM;
1438                 up_read(&EXT4_I(inode)->xattr_sem);
1439                 goto out;
1440         }
1441
1442         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1443         up_read(&EXT4_I(inode)->xattr_sem);
1444         if (ret < 0)
1445                 goto out;
1446
1447         ret = 0;
1448         sb = inode->i_sb;
1449         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1450         offset = ctx->pos;
1451
1452         /*
1453          * dotdot_offset and dotdot_size is the real offset and
1454          * size for ".." and "." if the dir is block based while
1455          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1456          * So we will use extra_offset and extra_size to indicate them
1457          * during the inline dir iteration.
1458          */
1459         dotdot_offset = EXT4_DIR_REC_LEN(1);
1460         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1461         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1462         extra_size = extra_offset + inline_size;
1463
1464         /*
1465          * If the version has changed since the last call to
1466          * readdir(2), then we might be pointing to an invalid
1467          * dirent right now.  Scan from the start of the inline
1468          * dir to make sure.
1469          */
1470         if (file->f_version != inode->i_version) {
1471                 for (i = 0; i < extra_size && i < offset;) {
1472                         /*
1473                          * "." is with offset 0 and
1474                          * ".." is dotdot_offset.
1475                          */
1476                         if (!i) {
1477                                 i = dotdot_offset;
1478                                 continue;
1479                         } else if (i == dotdot_offset) {
1480                                 i = dotdot_size;
1481                                 continue;
1482                         }
1483                         /* for other entry, the real offset in
1484                          * the buf has to be tuned accordingly.
1485                          */
1486                         de = (struct ext4_dir_entry_2 *)
1487                                 (dir_buf + i - extra_offset);
1488                         /* It's too expensive to do a full
1489                          * dirent test each time round this
1490                          * loop, but we do have to test at
1491                          * least that it is non-zero.  A
1492                          * failure will be detected in the
1493                          * dirent test below. */
1494                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1495                                 < EXT4_DIR_REC_LEN(1))
1496                                 break;
1497                         i += ext4_rec_len_from_disk(de->rec_len,
1498                                                     extra_size);
1499                 }
1500                 offset = i;
1501                 ctx->pos = offset;
1502                 file->f_version = inode->i_version;
1503         }
1504
1505         while (ctx->pos < extra_size) {
1506                 if (ctx->pos == 0) {
1507                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1508                                 goto out;
1509                         ctx->pos = dotdot_offset;
1510                         continue;
1511                 }
1512
1513                 if (ctx->pos == dotdot_offset) {
1514                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1515                                 goto out;
1516                         ctx->pos = dotdot_size;
1517                         continue;
1518                 }
1519
1520                 de = (struct ext4_dir_entry_2 *)
1521                         (dir_buf + ctx->pos - extra_offset);
1522                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1523                                          extra_size, ctx->pos))
1524                         goto out;
1525                 if (le32_to_cpu(de->inode)) {
1526                         if (!dir_emit(ctx, de->name, de->name_len,
1527                                       le32_to_cpu(de->inode),
1528                                       get_dtype(sb, de->file_type)))
1529                                 goto out;
1530                 }
1531                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1532         }
1533 out:
1534         kfree(dir_buf);
1535         brelse(iloc.bh);
1536         return ret;
1537 }
1538
1539 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1540                                         struct ext4_dir_entry_2 **parent_de,
1541                                         int *retval)
1542 {
1543         struct ext4_iloc iloc;
1544
1545         *retval = ext4_get_inode_loc(inode, &iloc);
1546         if (*retval)
1547                 return NULL;
1548
1549         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1550
1551         return iloc.bh;
1552 }
1553
1554 /*
1555  * Try to create the inline data for the new dir.
1556  * If it succeeds, return 0, otherwise return the error.
1557  * In case of ENOSPC, the caller should create the normal disk layout dir.
1558  */
1559 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1560                                struct inode *inode)
1561 {
1562         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1563         struct ext4_iloc iloc;
1564         struct ext4_dir_entry_2 *de;
1565
1566         ret = ext4_get_inode_loc(inode, &iloc);
1567         if (ret)
1568                 return ret;
1569
1570         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1571         if (ret)
1572                 goto out;
1573
1574         /*
1575          * For inline dir, we only save the inode information for the ".."
1576          * and create a fake dentry to cover the left space.
1577          */
1578         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1579         de->inode = cpu_to_le32(parent->i_ino);
1580         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1581         de->inode = 0;
1582         de->rec_len = ext4_rec_len_to_disk(
1583                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1584                                 inline_size);
1585         set_nlink(inode, 2);
1586         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1587 out:
1588         brelse(iloc.bh);
1589         return ret;
1590 }
1591
1592 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1593                                         const struct qstr *d_name,
1594                                         struct ext4_dir_entry_2 **res_dir,
1595                                         int *has_inline_data)
1596 {
1597         int ret;
1598         struct ext4_iloc iloc;
1599         void *inline_start;
1600         int inline_size;
1601
1602         if (ext4_get_inode_loc(dir, &iloc))
1603                 return NULL;
1604
1605         down_read(&EXT4_I(dir)->xattr_sem);
1606         if (!ext4_has_inline_data(dir)) {
1607                 *has_inline_data = 0;
1608                 goto out;
1609         }
1610
1611         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1612                                                 EXT4_INLINE_DOTDOT_SIZE;
1613         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1614         ret = search_dir(iloc.bh, inline_start, inline_size,
1615                          dir, d_name, 0, res_dir);
1616         if (ret == 1)
1617                 goto out_find;
1618         if (ret < 0)
1619                 goto out;
1620
1621         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1622                 goto out;
1623
1624         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1625         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1626
1627         ret = search_dir(iloc.bh, inline_start, inline_size,
1628                          dir, d_name, 0, res_dir);
1629         if (ret == 1)
1630                 goto out_find;
1631
1632 out:
1633         brelse(iloc.bh);
1634         iloc.bh = NULL;
1635 out_find:
1636         up_read(&EXT4_I(dir)->xattr_sem);
1637         return iloc.bh;
1638 }
1639
1640 int ext4_delete_inline_entry(handle_t *handle,
1641                              struct inode *dir,
1642                              struct ext4_dir_entry_2 *de_del,
1643                              struct buffer_head *bh,
1644                              int *has_inline_data)
1645 {
1646         int err, inline_size;
1647         struct ext4_iloc iloc;
1648         void *inline_start;
1649
1650         err = ext4_get_inode_loc(dir, &iloc);
1651         if (err)
1652                 return err;
1653
1654         down_write(&EXT4_I(dir)->xattr_sem);
1655         if (!ext4_has_inline_data(dir)) {
1656                 *has_inline_data = 0;
1657                 goto out;
1658         }
1659
1660         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1661                 EXT4_MIN_INLINE_DATA_SIZE) {
1662                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1663                                         EXT4_INLINE_DOTDOT_SIZE;
1664                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1665                                 EXT4_INLINE_DOTDOT_SIZE;
1666         } else {
1667                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1668                 inline_size = ext4_get_inline_size(dir) -
1669                                 EXT4_MIN_INLINE_DATA_SIZE;
1670         }
1671
1672         err = ext4_journal_get_write_access(handle, bh);
1673         if (err)
1674                 goto out;
1675
1676         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1677                                         inline_start, inline_size, 0);
1678         if (err)
1679                 goto out;
1680
1681         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1682         err = ext4_mark_inode_dirty(handle, dir);
1683         if (unlikely(err))
1684                 goto out;
1685
1686         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1687 out:
1688         up_write(&EXT4_I(dir)->xattr_sem);
1689         brelse(iloc.bh);
1690         if (err != -ENOENT)
1691                 ext4_std_error(dir->i_sb, err);
1692         return err;
1693 }
1694
1695 /*
1696  * Get the inline dentry at offset.
1697  */
1698 static inline struct ext4_dir_entry_2 *
1699 ext4_get_inline_entry(struct inode *inode,
1700                       struct ext4_iloc *iloc,
1701                       unsigned int offset,
1702                       void **inline_start,
1703                       int *inline_size)
1704 {
1705         void *inline_pos;
1706
1707         BUG_ON(offset > ext4_get_inline_size(inode));
1708
1709         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1710                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1711                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1712         } else {
1713                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1714                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1715                 *inline_size = ext4_get_inline_size(inode) -
1716                                 EXT4_MIN_INLINE_DATA_SIZE;
1717         }
1718
1719         if (inline_start)
1720                 *inline_start = inline_pos;
1721         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1722 }
1723
1724 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1725 {
1726         int err, inline_size;
1727         struct ext4_iloc iloc;
1728         void *inline_pos;
1729         unsigned int offset;
1730         struct ext4_dir_entry_2 *de;
1731         int ret = 1;
1732
1733         err = ext4_get_inode_loc(dir, &iloc);
1734         if (err) {
1735                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1736                                  err, dir->i_ino);
1737                 return 1;
1738         }
1739
1740         down_read(&EXT4_I(dir)->xattr_sem);
1741         if (!ext4_has_inline_data(dir)) {
1742                 *has_inline_data = 0;
1743                 goto out;
1744         }
1745
1746         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1747         if (!le32_to_cpu(de->inode)) {
1748                 ext4_warning(dir->i_sb,
1749                              "bad inline directory (dir #%lu) - no `..'",
1750                              dir->i_ino);
1751                 ret = 1;
1752                 goto out;
1753         }
1754
1755         offset = EXT4_INLINE_DOTDOT_SIZE;
1756         while (offset < dir->i_size) {
1757                 de = ext4_get_inline_entry(dir, &iloc, offset,
1758                                            &inline_pos, &inline_size);
1759                 if (ext4_check_dir_entry(dir, NULL, de,
1760                                          iloc.bh, inline_pos,
1761                                          inline_size, offset)) {
1762                         ext4_warning(dir->i_sb,
1763                                      "bad inline directory (dir #%lu) - "
1764                                      "inode %u, rec_len %u, name_len %d"
1765                                      "inline size %d\n",
1766                                      dir->i_ino, le32_to_cpu(de->inode),
1767                                      le16_to_cpu(de->rec_len), de->name_len,
1768                                      inline_size);
1769                         ret = 1;
1770                         goto out;
1771                 }
1772                 if (le32_to_cpu(de->inode)) {
1773                         ret = 0;
1774                         goto out;
1775                 }
1776                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1777         }
1778
1779 out:
1780         up_read(&EXT4_I(dir)->xattr_sem);
1781         brelse(iloc.bh);
1782         return ret;
1783 }
1784
1785 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1786 {
1787         int ret;
1788
1789         down_write(&EXT4_I(inode)->xattr_sem);
1790         ret = ext4_destroy_inline_data_nolock(handle, inode);
1791         up_write(&EXT4_I(inode)->xattr_sem);
1792
1793         return ret;
1794 }
1795
1796 int ext4_inline_data_fiemap(struct inode *inode,
1797                             struct fiemap_extent_info *fieinfo,
1798                             int *has_inline)
1799 {
1800         __u64 physical = 0;
1801         __u64 length;
1802         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1803         int error = 0;
1804         struct ext4_iloc iloc;
1805
1806         down_read(&EXT4_I(inode)->xattr_sem);
1807         if (!ext4_has_inline_data(inode)) {
1808                 *has_inline = 0;
1809                 goto out;
1810         }
1811
1812         error = ext4_get_inode_loc(inode, &iloc);
1813         if (error)
1814                 goto out;
1815
1816         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1817         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1818         physical += offsetof(struct ext4_inode, i_block);
1819         length = i_size_read(inode);
1820
1821         if (physical)
1822                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1823                                                 length, flags);
1824         brelse(iloc.bh);
1825 out:
1826         up_read(&EXT4_I(inode)->xattr_sem);
1827         return (error < 0 ? error : 0);
1828 }
1829
1830 /*
1831  * Called during xattr set, and if we can sparse space 'needed',
1832  * just create the extent tree evict the data to the outer block.
1833  *
1834  * We use jbd2 instead of page cache to move data to the 1st block
1835  * so that the whole transaction can be committed as a whole and
1836  * the data isn't lost because of the delayed page cache write.
1837  */
1838 int ext4_try_to_evict_inline_data(handle_t *handle,
1839                                   struct inode *inode,
1840                                   int needed)
1841 {
1842         int error;
1843         struct ext4_xattr_entry *entry;
1844         struct ext4_inode *raw_inode;
1845         struct ext4_iloc iloc;
1846
1847         error = ext4_get_inode_loc(inode, &iloc);
1848         if (error)
1849                 return error;
1850
1851         raw_inode = ext4_raw_inode(&iloc);
1852         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1853                                             EXT4_I(inode)->i_inline_off);
1854         if (EXT4_XATTR_LEN(entry->e_name_len) +
1855             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1856                 error = -ENOSPC;
1857                 goto out;
1858         }
1859
1860         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1861 out:
1862         brelse(iloc.bh);
1863         return error;
1864 }
1865
1866 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1867 {
1868         handle_t *handle;
1869         int inline_size, value_len, needed_blocks;
1870         size_t i_size;
1871         void *value = NULL;
1872         struct ext4_xattr_ibody_find is = {
1873                 .s = { .not_found = -ENODATA, },
1874         };
1875         struct ext4_xattr_info i = {
1876                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1877                 .name = EXT4_XATTR_SYSTEM_DATA,
1878         };
1879
1880
1881         needed_blocks = ext4_writepage_trans_blocks(inode);
1882         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1883         if (IS_ERR(handle))
1884                 return;
1885
1886         down_write(&EXT4_I(inode)->xattr_sem);
1887         if (!ext4_has_inline_data(inode)) {
1888                 *has_inline = 0;
1889                 ext4_journal_stop(handle);
1890                 return;
1891         }
1892
1893         if (ext4_orphan_add(handle, inode))
1894                 goto out;
1895
1896         if (ext4_get_inode_loc(inode, &is.iloc))
1897                 goto out;
1898
1899         down_write(&EXT4_I(inode)->i_data_sem);
1900         i_size = inode->i_size;
1901         inline_size = ext4_get_inline_size(inode);
1902         EXT4_I(inode)->i_disksize = i_size;
1903
1904         if (i_size < inline_size) {
1905                 /* Clear the content in the xattr space. */
1906                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1907                         if (ext4_xattr_ibody_find(inode, &i, &is))
1908                                 goto out_error;
1909
1910                         BUG_ON(is.s.not_found);
1911
1912                         value_len = le32_to_cpu(is.s.here->e_value_size);
1913                         value = kmalloc(value_len, GFP_NOFS);
1914                         if (!value)
1915                                 goto out_error;
1916
1917                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1918                                                 value, value_len))
1919                                 goto out_error;
1920
1921                         i.value = value;
1922                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1923                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1924                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1925                                 goto out_error;
1926                 }
1927
1928                 /* Clear the content within i_blocks. */
1929                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1930                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1931                         memset(p + i_size, 0,
1932                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1933                 }
1934
1935                 EXT4_I(inode)->i_inline_size = i_size <
1936                                         EXT4_MIN_INLINE_DATA_SIZE ?
1937                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1938         }
1939
1940 out_error:
1941         up_write(&EXT4_I(inode)->i_data_sem);
1942 out:
1943         brelse(is.iloc.bh);
1944         up_write(&EXT4_I(inode)->xattr_sem);
1945         kfree(value);
1946         if (inode->i_nlink)
1947                 ext4_orphan_del(handle, inode);
1948
1949         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1950         ext4_mark_inode_dirty(handle, inode);
1951         if (IS_SYNC(inode))
1952                 ext4_handle_sync(handle);
1953
1954         ext4_journal_stop(handle);
1955         return;
1956 }
1957
1958 int ext4_convert_inline_data(struct inode *inode)
1959 {
1960         int error, needed_blocks;
1961         handle_t *handle;
1962         struct ext4_iloc iloc;
1963
1964         if (!ext4_has_inline_data(inode)) {
1965                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1966                 return 0;
1967         }
1968
1969         needed_blocks = ext4_writepage_trans_blocks(inode);
1970
1971         iloc.bh = NULL;
1972         error = ext4_get_inode_loc(inode, &iloc);
1973         if (error)
1974                 return error;
1975
1976         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1977         if (IS_ERR(handle)) {
1978                 error = PTR_ERR(handle);
1979                 goto out_free;
1980         }
1981
1982         down_write(&EXT4_I(inode)->xattr_sem);
1983         if (!ext4_has_inline_data(inode)) {
1984                 up_write(&EXT4_I(inode)->xattr_sem);
1985                 goto out;
1986         }
1987
1988         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1989         up_write(&EXT4_I(inode)->xattr_sem);
1990 out:
1991         ext4_journal_stop(handle);
1992 out_free:
1993         brelse(iloc.bh);
1994         return error;
1995 }