GNU Linux-libre 5.10.153-gnu1
[releases.git] / fs / ext4 / inline.c
1 // SPDX-License-Identifier: LGPL-2.1
2 /*
3  * Copyright (c) 2012 Taobao.
4  * Written by Tao Ma <boyu.mt@taobao.com>
5  */
6
7 #include <linux/iomap.h>
8 #include <linux/fiemap.h>
9 #include <linux/iversion.h>
10
11 #include "ext4_jbd2.h"
12 #include "ext4.h"
13 #include "xattr.h"
14 #include "truncate.h"
15
16 #define EXT4_XATTR_SYSTEM_DATA  "data"
17 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
18 #define EXT4_INLINE_DOTDOT_OFFSET       2
19 #define EXT4_INLINE_DOTDOT_SIZE         4
20
21 static int ext4_get_inline_size(struct inode *inode)
22 {
23         if (EXT4_I(inode)->i_inline_off)
24                 return EXT4_I(inode)->i_inline_size;
25
26         return 0;
27 }
28
29 static int get_max_inline_xattr_value_size(struct inode *inode,
30                                            struct ext4_iloc *iloc)
31 {
32         struct ext4_xattr_ibody_header *header;
33         struct ext4_xattr_entry *entry;
34         struct ext4_inode *raw_inode;
35         int free, min_offs;
36
37         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
38                 return 0;
39
40         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
41                         EXT4_GOOD_OLD_INODE_SIZE -
42                         EXT4_I(inode)->i_extra_isize -
43                         sizeof(struct ext4_xattr_ibody_header);
44
45         /*
46          * We need to subtract another sizeof(__u32) since an in-inode xattr
47          * needs an empty 4 bytes to indicate the gap between the xattr entry
48          * and the name/value pair.
49          */
50         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
51                 return EXT4_XATTR_SIZE(min_offs -
52                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
53                         EXT4_XATTR_ROUND - sizeof(__u32));
54
55         raw_inode = ext4_raw_inode(iloc);
56         header = IHDR(inode, raw_inode);
57         entry = IFIRST(header);
58
59         /* Compute min_offs. */
60         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
61                 if (!entry->e_value_inum && entry->e_value_size) {
62                         size_t offs = le16_to_cpu(entry->e_value_offs);
63                         if (offs < min_offs)
64                                 min_offs = offs;
65                 }
66         }
67         free = min_offs -
68                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
69
70         if (EXT4_I(inode)->i_inline_off) {
71                 entry = (struct ext4_xattr_entry *)
72                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
73
74                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
75                 goto out;
76         }
77
78         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
79
80         if (free > EXT4_XATTR_ROUND)
81                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
82         else
83                 free = 0;
84
85 out:
86         return free;
87 }
88
89 /*
90  * Get the maximum size we now can store in an inode.
91  * If we can't find the space for a xattr entry, don't use the space
92  * of the extents since we have no space to indicate the inline data.
93  */
94 int ext4_get_max_inline_size(struct inode *inode)
95 {
96         int error, max_inline_size;
97         struct ext4_iloc iloc;
98
99         if (EXT4_I(inode)->i_extra_isize == 0)
100                 return 0;
101
102         error = ext4_get_inode_loc(inode, &iloc);
103         if (error) {
104                 ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
105                                      "can't get inode location %lu",
106                                      inode->i_ino);
107                 return 0;
108         }
109
110         down_read(&EXT4_I(inode)->xattr_sem);
111         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
112         up_read(&EXT4_I(inode)->xattr_sem);
113
114         brelse(iloc.bh);
115
116         if (!max_inline_size)
117                 return 0;
118
119         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
120 }
121
122 /*
123  * this function does not take xattr_sem, which is OK because it is
124  * currently only used in a code path coming form ext4_iget, before
125  * the new inode has been unlocked
126  */
127 int ext4_find_inline_data_nolock(struct inode *inode)
128 {
129         struct ext4_xattr_ibody_find is = {
130                 .s = { .not_found = -ENODATA, },
131         };
132         struct ext4_xattr_info i = {
133                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
134                 .name = EXT4_XATTR_SYSTEM_DATA,
135         };
136         int error;
137
138         if (EXT4_I(inode)->i_extra_isize == 0)
139                 return 0;
140
141         error = ext4_get_inode_loc(inode, &is.iloc);
142         if (error)
143                 return error;
144
145         error = ext4_xattr_ibody_find(inode, &i, &is);
146         if (error)
147                 goto out;
148
149         if (!is.s.not_found) {
150                 if (is.s.here->e_value_inum) {
151                         EXT4_ERROR_INODE(inode, "inline data xattr refers "
152                                          "to an external xattr inode");
153                         error = -EFSCORRUPTED;
154                         goto out;
155                 }
156                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
157                                         (void *)ext4_raw_inode(&is.iloc));
158                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
159                                 le32_to_cpu(is.s.here->e_value_size);
160                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
161         }
162 out:
163         brelse(is.iloc.bh);
164         return error;
165 }
166
167 static int ext4_read_inline_data(struct inode *inode, void *buffer,
168                                  unsigned int len,
169                                  struct ext4_iloc *iloc)
170 {
171         struct ext4_xattr_entry *entry;
172         struct ext4_xattr_ibody_header *header;
173         int cp_len = 0;
174         struct ext4_inode *raw_inode;
175
176         if (!len)
177                 return 0;
178
179         BUG_ON(len > EXT4_I(inode)->i_inline_size);
180
181         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
182                         len : EXT4_MIN_INLINE_DATA_SIZE;
183
184         raw_inode = ext4_raw_inode(iloc);
185         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
186
187         len -= cp_len;
188         buffer += cp_len;
189
190         if (!len)
191                 goto out;
192
193         header = IHDR(inode, raw_inode);
194         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
195                                             EXT4_I(inode)->i_inline_off);
196         len = min_t(unsigned int, len,
197                     (unsigned int)le32_to_cpu(entry->e_value_size));
198
199         memcpy(buffer,
200                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
201         cp_len += len;
202
203 out:
204         return cp_len;
205 }
206
207 /*
208  * write the buffer to the inline inode.
209  * If 'create' is set, we don't need to do the extra copy in the xattr
210  * value since it is already handled by ext4_xattr_ibody_inline_set.
211  * That saves us one memcpy.
212  */
213 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
214                                    void *buffer, loff_t pos, unsigned int len)
215 {
216         struct ext4_xattr_entry *entry;
217         struct ext4_xattr_ibody_header *header;
218         struct ext4_inode *raw_inode;
219         int cp_len = 0;
220
221         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
222                 return;
223
224         BUG_ON(!EXT4_I(inode)->i_inline_off);
225         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
226
227         raw_inode = ext4_raw_inode(iloc);
228         buffer += pos;
229
230         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
231                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
232                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
233                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
234
235                 len -= cp_len;
236                 buffer += cp_len;
237                 pos += cp_len;
238         }
239
240         if (!len)
241                 return;
242
243         pos -= EXT4_MIN_INLINE_DATA_SIZE;
244         header = IHDR(inode, raw_inode);
245         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
246                                             EXT4_I(inode)->i_inline_off);
247
248         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
249                buffer, len);
250 }
251
252 static int ext4_create_inline_data(handle_t *handle,
253                                    struct inode *inode, unsigned len)
254 {
255         int error;
256         void *value = NULL;
257         struct ext4_xattr_ibody_find is = {
258                 .s = { .not_found = -ENODATA, },
259         };
260         struct ext4_xattr_info i = {
261                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
262                 .name = EXT4_XATTR_SYSTEM_DATA,
263         };
264
265         error = ext4_get_inode_loc(inode, &is.iloc);
266         if (error)
267                 return error;
268
269         BUFFER_TRACE(is.iloc.bh, "get_write_access");
270         error = ext4_journal_get_write_access(handle, is.iloc.bh);
271         if (error)
272                 goto out;
273
274         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
275                 value = EXT4_ZERO_XATTR_VALUE;
276                 len -= EXT4_MIN_INLINE_DATA_SIZE;
277         } else {
278                 value = "";
279                 len = 0;
280         }
281
282         /* Insert the xttr entry. */
283         i.value = value;
284         i.value_len = len;
285
286         error = ext4_xattr_ibody_find(inode, &i, &is);
287         if (error)
288                 goto out;
289
290         BUG_ON(!is.s.not_found);
291
292         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
293         if (error) {
294                 if (error == -ENOSPC)
295                         ext4_clear_inode_state(inode,
296                                                EXT4_STATE_MAY_INLINE_DATA);
297                 goto out;
298         }
299
300         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
301                 0, EXT4_MIN_INLINE_DATA_SIZE);
302
303         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
304                                       (void *)ext4_raw_inode(&is.iloc));
305         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
306         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
307         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
308         get_bh(is.iloc.bh);
309         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
310
311 out:
312         brelse(is.iloc.bh);
313         return error;
314 }
315
316 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
317                                    unsigned int len)
318 {
319         int error;
320         void *value = NULL;
321         struct ext4_xattr_ibody_find is = {
322                 .s = { .not_found = -ENODATA, },
323         };
324         struct ext4_xattr_info i = {
325                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
326                 .name = EXT4_XATTR_SYSTEM_DATA,
327         };
328
329         /* If the old space is ok, write the data directly. */
330         if (len <= EXT4_I(inode)->i_inline_size)
331                 return 0;
332
333         error = ext4_get_inode_loc(inode, &is.iloc);
334         if (error)
335                 return error;
336
337         error = ext4_xattr_ibody_find(inode, &i, &is);
338         if (error)
339                 goto out;
340
341         BUG_ON(is.s.not_found);
342
343         len -= EXT4_MIN_INLINE_DATA_SIZE;
344         value = kzalloc(len, GFP_NOFS);
345         if (!value) {
346                 error = -ENOMEM;
347                 goto out;
348         }
349
350         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
351                                      value, len);
352         if (error == -ENODATA)
353                 goto out;
354
355         BUFFER_TRACE(is.iloc.bh, "get_write_access");
356         error = ext4_journal_get_write_access(handle, is.iloc.bh);
357         if (error)
358                 goto out;
359
360         /* Update the xattr entry. */
361         i.value = value;
362         i.value_len = len;
363
364         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
365         if (error)
366                 goto out;
367
368         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
369                                       (void *)ext4_raw_inode(&is.iloc));
370         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
371                                 le32_to_cpu(is.s.here->e_value_size);
372         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
373         get_bh(is.iloc.bh);
374         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
375
376 out:
377         kfree(value);
378         brelse(is.iloc.bh);
379         return error;
380 }
381
382 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
383                                     unsigned int len)
384 {
385         int ret, size, no_expand;
386         struct ext4_inode_info *ei = EXT4_I(inode);
387
388         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
389                 return -ENOSPC;
390
391         size = ext4_get_max_inline_size(inode);
392         if (size < len)
393                 return -ENOSPC;
394
395         ext4_write_lock_xattr(inode, &no_expand);
396
397         if (ei->i_inline_off)
398                 ret = ext4_update_inline_data(handle, inode, len);
399         else
400                 ret = ext4_create_inline_data(handle, inode, len);
401
402         ext4_write_unlock_xattr(inode, &no_expand);
403         return ret;
404 }
405
406 static int ext4_destroy_inline_data_nolock(handle_t *handle,
407                                            struct inode *inode)
408 {
409         struct ext4_inode_info *ei = EXT4_I(inode);
410         struct ext4_xattr_ibody_find is = {
411                 .s = { .not_found = 0, },
412         };
413         struct ext4_xattr_info i = {
414                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
415                 .name = EXT4_XATTR_SYSTEM_DATA,
416                 .value = NULL,
417                 .value_len = 0,
418         };
419         int error;
420
421         if (!ei->i_inline_off)
422                 return 0;
423
424         error = ext4_get_inode_loc(inode, &is.iloc);
425         if (error)
426                 return error;
427
428         error = ext4_xattr_ibody_find(inode, &i, &is);
429         if (error)
430                 goto out;
431
432         BUFFER_TRACE(is.iloc.bh, "get_write_access");
433         error = ext4_journal_get_write_access(handle, is.iloc.bh);
434         if (error)
435                 goto out;
436
437         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
438         if (error)
439                 goto out;
440
441         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
442                 0, EXT4_MIN_INLINE_DATA_SIZE);
443         memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
444
445         if (ext4_has_feature_extents(inode->i_sb)) {
446                 if (S_ISDIR(inode->i_mode) ||
447                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
448                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
449                         ext4_ext_tree_init(handle, inode);
450                 }
451         }
452         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
453
454         get_bh(is.iloc.bh);
455         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
456
457         EXT4_I(inode)->i_inline_off = 0;
458         EXT4_I(inode)->i_inline_size = 0;
459         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
460 out:
461         brelse(is.iloc.bh);
462         if (error == -ENODATA)
463                 error = 0;
464         return error;
465 }
466
467 static int ext4_read_inline_page(struct inode *inode, struct page *page)
468 {
469         void *kaddr;
470         int ret = 0;
471         size_t len;
472         struct ext4_iloc iloc;
473
474         BUG_ON(!PageLocked(page));
475         BUG_ON(!ext4_has_inline_data(inode));
476         BUG_ON(page->index);
477
478         if (!EXT4_I(inode)->i_inline_off) {
479                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
480                              inode->i_ino);
481                 goto out;
482         }
483
484         ret = ext4_get_inode_loc(inode, &iloc);
485         if (ret)
486                 goto out;
487
488         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
489         kaddr = kmap_atomic(page);
490         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
491         flush_dcache_page(page);
492         kunmap_atomic(kaddr);
493         zero_user_segment(page, len, PAGE_SIZE);
494         SetPageUptodate(page);
495         brelse(iloc.bh);
496
497 out:
498         return ret;
499 }
500
501 int ext4_readpage_inline(struct inode *inode, struct page *page)
502 {
503         int ret = 0;
504
505         down_read(&EXT4_I(inode)->xattr_sem);
506         if (!ext4_has_inline_data(inode)) {
507                 up_read(&EXT4_I(inode)->xattr_sem);
508                 return -EAGAIN;
509         }
510
511         /*
512          * Current inline data can only exist in the 1st page,
513          * So for all the other pages, just set them uptodate.
514          */
515         if (!page->index)
516                 ret = ext4_read_inline_page(inode, page);
517         else if (!PageUptodate(page)) {
518                 zero_user_segment(page, 0, PAGE_SIZE);
519                 SetPageUptodate(page);
520         }
521
522         up_read(&EXT4_I(inode)->xattr_sem);
523
524         unlock_page(page);
525         return ret >= 0 ? 0 : ret;
526 }
527
528 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
529                                               struct inode *inode,
530                                               unsigned flags)
531 {
532         int ret, needed_blocks, no_expand;
533         handle_t *handle = NULL;
534         int retries = 0, sem_held = 0;
535         struct page *page = NULL;
536         unsigned from, to;
537         struct ext4_iloc iloc;
538
539         if (!ext4_has_inline_data(inode)) {
540                 /*
541                  * clear the flag so that no new write
542                  * will trap here again.
543                  */
544                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
545                 return 0;
546         }
547
548         needed_blocks = ext4_writepage_trans_blocks(inode);
549
550         ret = ext4_get_inode_loc(inode, &iloc);
551         if (ret)
552                 return ret;
553
554 retry:
555         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
556         if (IS_ERR(handle)) {
557                 ret = PTR_ERR(handle);
558                 handle = NULL;
559                 goto out;
560         }
561
562         /* We cannot recurse into the filesystem as the transaction is already
563          * started */
564         flags |= AOP_FLAG_NOFS;
565
566         page = grab_cache_page_write_begin(mapping, 0, flags);
567         if (!page) {
568                 ret = -ENOMEM;
569                 goto out;
570         }
571
572         ext4_write_lock_xattr(inode, &no_expand);
573         sem_held = 1;
574         /* If some one has already done this for us, just exit. */
575         if (!ext4_has_inline_data(inode)) {
576                 ret = 0;
577                 goto out;
578         }
579
580         from = 0;
581         to = ext4_get_inline_size(inode);
582         if (!PageUptodate(page)) {
583                 ret = ext4_read_inline_page(inode, page);
584                 if (ret < 0)
585                         goto out;
586         }
587
588         ret = ext4_destroy_inline_data_nolock(handle, inode);
589         if (ret)
590                 goto out;
591
592         if (ext4_should_dioread_nolock(inode)) {
593                 ret = __block_write_begin(page, from, to,
594                                           ext4_get_block_unwritten);
595         } else
596                 ret = __block_write_begin(page, from, to, ext4_get_block);
597
598         if (!ret && ext4_should_journal_data(inode)) {
599                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
600                                              from, to, NULL,
601                                              do_journal_get_write_access);
602         }
603
604         if (ret) {
605                 unlock_page(page);
606                 put_page(page);
607                 page = NULL;
608                 ext4_orphan_add(handle, inode);
609                 ext4_write_unlock_xattr(inode, &no_expand);
610                 sem_held = 0;
611                 ext4_journal_stop(handle);
612                 handle = NULL;
613                 ext4_truncate_failed_write(inode);
614                 /*
615                  * If truncate failed early the inode might
616                  * still be on the orphan list; we need to
617                  * make sure the inode is removed from the
618                  * orphan list in that case.
619                  */
620                 if (inode->i_nlink)
621                         ext4_orphan_del(NULL, inode);
622         }
623
624         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
625                 goto retry;
626
627         if (page)
628                 block_commit_write(page, from, to);
629 out:
630         if (page) {
631                 unlock_page(page);
632                 put_page(page);
633         }
634         if (sem_held)
635                 ext4_write_unlock_xattr(inode, &no_expand);
636         if (handle)
637                 ext4_journal_stop(handle);
638         brelse(iloc.bh);
639         return ret;
640 }
641
642 /*
643  * Try to write data in the inode.
644  * If the inode has inline data, check whether the new write can be
645  * in the inode also. If not, create the page the handle, move the data
646  * to the page make it update and let the later codes create extent for it.
647  */
648 int ext4_try_to_write_inline_data(struct address_space *mapping,
649                                   struct inode *inode,
650                                   loff_t pos, unsigned len,
651                                   unsigned flags,
652                                   struct page **pagep)
653 {
654         int ret;
655         handle_t *handle;
656         struct page *page;
657         struct ext4_iloc iloc;
658
659         if (pos + len > ext4_get_max_inline_size(inode))
660                 goto convert;
661
662         ret = ext4_get_inode_loc(inode, &iloc);
663         if (ret)
664                 return ret;
665
666         /*
667          * The possible write could happen in the inode,
668          * so try to reserve the space in inode first.
669          */
670         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
671         if (IS_ERR(handle)) {
672                 ret = PTR_ERR(handle);
673                 handle = NULL;
674                 goto out;
675         }
676
677         ret = ext4_prepare_inline_data(handle, inode, pos + len);
678         if (ret && ret != -ENOSPC)
679                 goto out;
680
681         /* We don't have space in inline inode, so convert it to extent. */
682         if (ret == -ENOSPC) {
683                 ext4_journal_stop(handle);
684                 brelse(iloc.bh);
685                 goto convert;
686         }
687
688         ret = ext4_journal_get_write_access(handle, iloc.bh);
689         if (ret)
690                 goto out;
691
692         flags |= AOP_FLAG_NOFS;
693
694         page = grab_cache_page_write_begin(mapping, 0, flags);
695         if (!page) {
696                 ret = -ENOMEM;
697                 goto out;
698         }
699
700         *pagep = page;
701         down_read(&EXT4_I(inode)->xattr_sem);
702         if (!ext4_has_inline_data(inode)) {
703                 ret = 0;
704                 unlock_page(page);
705                 put_page(page);
706                 goto out_up_read;
707         }
708
709         if (!PageUptodate(page)) {
710                 ret = ext4_read_inline_page(inode, page);
711                 if (ret < 0) {
712                         unlock_page(page);
713                         put_page(page);
714                         goto out_up_read;
715                 }
716         }
717
718         ret = 1;
719         handle = NULL;
720 out_up_read:
721         up_read(&EXT4_I(inode)->xattr_sem);
722 out:
723         if (handle && (ret != 1))
724                 ext4_journal_stop(handle);
725         brelse(iloc.bh);
726         return ret;
727 convert:
728         return ext4_convert_inline_data_to_extent(mapping,
729                                                   inode, flags);
730 }
731
732 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
733                                unsigned copied, struct page *page)
734 {
735         int ret, no_expand;
736         void *kaddr;
737         struct ext4_iloc iloc;
738
739         if (unlikely(copied < len) && !PageUptodate(page))
740                 return 0;
741
742         ret = ext4_get_inode_loc(inode, &iloc);
743         if (ret) {
744                 ext4_std_error(inode->i_sb, ret);
745                 return ret;
746         }
747
748         ext4_write_lock_xattr(inode, &no_expand);
749         BUG_ON(!ext4_has_inline_data(inode));
750
751         /*
752          * ei->i_inline_off may have changed since ext4_write_begin()
753          * called ext4_try_to_write_inline_data()
754          */
755         (void) ext4_find_inline_data_nolock(inode);
756
757         kaddr = kmap_atomic(page);
758         ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
759         kunmap_atomic(kaddr);
760         SetPageUptodate(page);
761         /* clear page dirty so that writepages wouldn't work for us. */
762         ClearPageDirty(page);
763
764         ext4_write_unlock_xattr(inode, &no_expand);
765         brelse(iloc.bh);
766         mark_inode_dirty(inode);
767
768         return copied;
769 }
770
771 struct buffer_head *
772 ext4_journalled_write_inline_data(struct inode *inode,
773                                   unsigned len,
774                                   struct page *page)
775 {
776         int ret, no_expand;
777         void *kaddr;
778         struct ext4_iloc iloc;
779
780         ret = ext4_get_inode_loc(inode, &iloc);
781         if (ret) {
782                 ext4_std_error(inode->i_sb, ret);
783                 return NULL;
784         }
785
786         ext4_write_lock_xattr(inode, &no_expand);
787         kaddr = kmap_atomic(page);
788         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
789         kunmap_atomic(kaddr);
790         ext4_write_unlock_xattr(inode, &no_expand);
791
792         return iloc.bh;
793 }
794
795 /*
796  * Try to make the page cache and handle ready for the inline data case.
797  * We can call this function in 2 cases:
798  * 1. The inode is created and the first write exceeds inline size. We can
799  *    clear the inode state safely.
800  * 2. The inode has inline data, then we need to read the data, make it
801  *    update and dirty so that ext4_da_writepages can handle it. We don't
802  *    need to start the journal since the file's metatdata isn't changed now.
803  */
804 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
805                                                  struct inode *inode,
806                                                  unsigned flags,
807                                                  void **fsdata)
808 {
809         int ret = 0, inline_size;
810         struct page *page;
811
812         page = grab_cache_page_write_begin(mapping, 0, flags);
813         if (!page)
814                 return -ENOMEM;
815
816         down_read(&EXT4_I(inode)->xattr_sem);
817         if (!ext4_has_inline_data(inode)) {
818                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
819                 goto out;
820         }
821
822         inline_size = ext4_get_inline_size(inode);
823
824         if (!PageUptodate(page)) {
825                 ret = ext4_read_inline_page(inode, page);
826                 if (ret < 0)
827                         goto out;
828         }
829
830         ret = __block_write_begin(page, 0, inline_size,
831                                   ext4_da_get_block_prep);
832         if (ret) {
833                 up_read(&EXT4_I(inode)->xattr_sem);
834                 unlock_page(page);
835                 put_page(page);
836                 ext4_truncate_failed_write(inode);
837                 return ret;
838         }
839
840         SetPageDirty(page);
841         SetPageUptodate(page);
842         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
843         *fsdata = (void *)CONVERT_INLINE_DATA;
844
845 out:
846         up_read(&EXT4_I(inode)->xattr_sem);
847         if (page) {
848                 unlock_page(page);
849                 put_page(page);
850         }
851         return ret;
852 }
853
854 /*
855  * Prepare the write for the inline data.
856  * If the data can be written into the inode, we just read
857  * the page and make it uptodate, and start the journal.
858  * Otherwise read the page, makes it dirty so that it can be
859  * handle in writepages(the i_disksize update is left to the
860  * normal ext4_da_write_end).
861  */
862 int ext4_da_write_inline_data_begin(struct address_space *mapping,
863                                     struct inode *inode,
864                                     loff_t pos, unsigned len,
865                                     unsigned flags,
866                                     struct page **pagep,
867                                     void **fsdata)
868 {
869         int ret, inline_size;
870         handle_t *handle;
871         struct page *page;
872         struct ext4_iloc iloc;
873         int retries = 0;
874
875         ret = ext4_get_inode_loc(inode, &iloc);
876         if (ret)
877                 return ret;
878
879 retry_journal:
880         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
881         if (IS_ERR(handle)) {
882                 ret = PTR_ERR(handle);
883                 goto out;
884         }
885
886         inline_size = ext4_get_max_inline_size(inode);
887
888         ret = -ENOSPC;
889         if (inline_size >= pos + len) {
890                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
891                 if (ret && ret != -ENOSPC)
892                         goto out_journal;
893         }
894
895         /*
896          * We cannot recurse into the filesystem as the transaction
897          * is already started.
898          */
899         flags |= AOP_FLAG_NOFS;
900
901         if (ret == -ENOSPC) {
902                 ext4_journal_stop(handle);
903                 ret = ext4_da_convert_inline_data_to_extent(mapping,
904                                                             inode,
905                                                             flags,
906                                                             fsdata);
907                 if (ret == -ENOSPC &&
908                     ext4_should_retry_alloc(inode->i_sb, &retries))
909                         goto retry_journal;
910                 goto out;
911         }
912
913         page = grab_cache_page_write_begin(mapping, 0, flags);
914         if (!page) {
915                 ret = -ENOMEM;
916                 goto out_journal;
917         }
918
919         down_read(&EXT4_I(inode)->xattr_sem);
920         if (!ext4_has_inline_data(inode)) {
921                 ret = 0;
922                 goto out_release_page;
923         }
924
925         if (!PageUptodate(page)) {
926                 ret = ext4_read_inline_page(inode, page);
927                 if (ret < 0)
928                         goto out_release_page;
929         }
930         ret = ext4_journal_get_write_access(handle, iloc.bh);
931         if (ret)
932                 goto out_release_page;
933
934         up_read(&EXT4_I(inode)->xattr_sem);
935         *pagep = page;
936         brelse(iloc.bh);
937         return 1;
938 out_release_page:
939         up_read(&EXT4_I(inode)->xattr_sem);
940         unlock_page(page);
941         put_page(page);
942 out_journal:
943         ext4_journal_stop(handle);
944 out:
945         brelse(iloc.bh);
946         return ret;
947 }
948
949 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
950                                   unsigned len, unsigned copied,
951                                   struct page *page)
952 {
953         int ret;
954
955         ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
956         if (ret < 0) {
957                 unlock_page(page);
958                 put_page(page);
959                 return ret;
960         }
961         copied = ret;
962
963         /*
964          * No need to use i_size_read() here, the i_size
965          * cannot change under us because we hold i_mutex.
966          *
967          * But it's important to update i_size while still holding page lock:
968          * page writeout could otherwise come in and zero beyond i_size.
969          */
970         if (pos+copied > inode->i_size)
971                 i_size_write(inode, pos+copied);
972         unlock_page(page);
973         put_page(page);
974
975         /*
976          * Don't mark the inode dirty under page lock. First, it unnecessarily
977          * makes the holding time of page lock longer. Second, it forces lock
978          * ordering of page lock and transaction start for journaling
979          * filesystems.
980          */
981         mark_inode_dirty(inode);
982
983         return copied;
984 }
985
986 #ifdef INLINE_DIR_DEBUG
987 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
988                           void *inline_start, int inline_size)
989 {
990         int offset;
991         unsigned short de_len;
992         struct ext4_dir_entry_2 *de = inline_start;
993         void *dlimit = inline_start + inline_size;
994
995         trace_printk("inode %lu\n", dir->i_ino);
996         offset = 0;
997         while ((void *)de < dlimit) {
998                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
999                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1000                              offset, de_len, de->name_len, de->name,
1001                              de->name_len, le32_to_cpu(de->inode));
1002                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1003                                          inline_start, inline_size, offset))
1004                         BUG();
1005
1006                 offset += de_len;
1007                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1008         }
1009 }
1010 #else
1011 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1012 #endif
1013
1014 /*
1015  * Add a new entry into a inline dir.
1016  * It will return -ENOSPC if no space is available, and -EIO
1017  * and -EEXIST if directory entry already exists.
1018  */
1019 static int ext4_add_dirent_to_inline(handle_t *handle,
1020                                      struct ext4_filename *fname,
1021                                      struct inode *dir,
1022                                      struct inode *inode,
1023                                      struct ext4_iloc *iloc,
1024                                      void *inline_start, int inline_size)
1025 {
1026         int             err;
1027         struct ext4_dir_entry_2 *de;
1028
1029         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1030                                 inline_size, fname, &de);
1031         if (err)
1032                 return err;
1033
1034         BUFFER_TRACE(iloc->bh, "get_write_access");
1035         err = ext4_journal_get_write_access(handle, iloc->bh);
1036         if (err)
1037                 return err;
1038         ext4_insert_dentry(inode, de, inline_size, fname);
1039
1040         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1041
1042         /*
1043          * XXX shouldn't update any times until successful
1044          * completion of syscall, but too many callers depend
1045          * on this.
1046          *
1047          * XXX similarly, too many callers depend on
1048          * ext4_new_inode() setting the times, but error
1049          * recovery deletes the inode, so the worst that can
1050          * happen is that the times are slightly out of date
1051          * and/or different from the directory change time.
1052          */
1053         dir->i_mtime = dir->i_ctime = current_time(dir);
1054         ext4_update_dx_flag(dir);
1055         inode_inc_iversion(dir);
1056         return 1;
1057 }
1058
1059 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1060                                        struct ext4_iloc *iloc)
1061 {
1062         struct ext4_xattr_entry *entry;
1063         struct ext4_xattr_ibody_header *header;
1064
1065         BUG_ON(!EXT4_I(inode)->i_inline_off);
1066
1067         header = IHDR(inode, ext4_raw_inode(iloc));
1068         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1069                                             EXT4_I(inode)->i_inline_off);
1070
1071         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1072 }
1073
1074 /* Set the final de to cover the whole block. */
1075 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1076 {
1077         struct ext4_dir_entry_2 *de, *prev_de;
1078         void *limit;
1079         int de_len;
1080
1081         de = (struct ext4_dir_entry_2 *)de_buf;
1082         if (old_size) {
1083                 limit = de_buf + old_size;
1084                 do {
1085                         prev_de = de;
1086                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1087                         de_buf += de_len;
1088                         de = (struct ext4_dir_entry_2 *)de_buf;
1089                 } while (de_buf < limit);
1090
1091                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1092                                                         old_size, new_size);
1093         } else {
1094                 /* this is just created, so create an empty entry. */
1095                 de->inode = 0;
1096                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1097         }
1098 }
1099
1100 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1101                                   struct ext4_iloc *iloc)
1102 {
1103         int ret;
1104         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1105         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1106
1107         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1108                 return -ENOSPC;
1109
1110         ret = ext4_update_inline_data(handle, dir,
1111                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1112         if (ret)
1113                 return ret;
1114
1115         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1116                              EXT4_I(dir)->i_inline_size -
1117                                                 EXT4_MIN_INLINE_DATA_SIZE);
1118         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1119         return 0;
1120 }
1121
1122 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1123                                      struct ext4_iloc *iloc,
1124                                      void *buf, int inline_size)
1125 {
1126         int ret;
1127
1128         ret = ext4_create_inline_data(handle, inode, inline_size);
1129         if (ret) {
1130                 ext4_msg(inode->i_sb, KERN_EMERG,
1131                         "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1132                         inode->i_ino, ret);
1133                 return;
1134         }
1135         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1136         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1137 }
1138
1139 static int ext4_finish_convert_inline_dir(handle_t *handle,
1140                                           struct inode *inode,
1141                                           struct buffer_head *dir_block,
1142                                           void *buf,
1143                                           int inline_size)
1144 {
1145         int err, csum_size = 0, header_size = 0;
1146         struct ext4_dir_entry_2 *de;
1147         void *target = dir_block->b_data;
1148
1149         /*
1150          * First create "." and ".." and then copy the dir information
1151          * back to the block.
1152          */
1153         de = (struct ext4_dir_entry_2 *)target;
1154         de = ext4_init_dot_dotdot(inode, de,
1155                 inode->i_sb->s_blocksize, csum_size,
1156                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1157         header_size = (void *)de - target;
1158
1159         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1160                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1161
1162         if (ext4_has_metadata_csum(inode->i_sb))
1163                 csum_size = sizeof(struct ext4_dir_entry_tail);
1164
1165         inode->i_size = inode->i_sb->s_blocksize;
1166         i_size_write(inode, inode->i_sb->s_blocksize);
1167         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1168         ext4_update_final_de(dir_block->b_data,
1169                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1170                         inode->i_sb->s_blocksize - csum_size);
1171
1172         if (csum_size)
1173                 ext4_initialize_dirent_tail(dir_block,
1174                                             inode->i_sb->s_blocksize);
1175         set_buffer_uptodate(dir_block);
1176         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1177         if (err)
1178                 return err;
1179         set_buffer_verified(dir_block);
1180         return ext4_mark_inode_dirty(handle, inode);
1181 }
1182
1183 static int ext4_convert_inline_data_nolock(handle_t *handle,
1184                                            struct inode *inode,
1185                                            struct ext4_iloc *iloc)
1186 {
1187         int error;
1188         void *buf = NULL;
1189         struct buffer_head *data_bh = NULL;
1190         struct ext4_map_blocks map;
1191         int inline_size;
1192
1193         inline_size = ext4_get_inline_size(inode);
1194         buf = kmalloc(inline_size, GFP_NOFS);
1195         if (!buf) {
1196                 error = -ENOMEM;
1197                 goto out;
1198         }
1199
1200         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1201         if (error < 0)
1202                 goto out;
1203
1204         /*
1205          * Make sure the inline directory entries pass checks before we try to
1206          * convert them, so that we avoid touching stuff that needs fsck.
1207          */
1208         if (S_ISDIR(inode->i_mode)) {
1209                 error = ext4_check_all_de(inode, iloc->bh,
1210                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1211                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1212                 if (error)
1213                         goto out;
1214         }
1215
1216         error = ext4_destroy_inline_data_nolock(handle, inode);
1217         if (error)
1218                 goto out;
1219
1220         map.m_lblk = 0;
1221         map.m_len = 1;
1222         map.m_flags = 0;
1223         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1224         if (error < 0)
1225                 goto out_restore;
1226         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1227                 error = -EIO;
1228                 goto out_restore;
1229         }
1230
1231         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1232         if (!data_bh) {
1233                 error = -ENOMEM;
1234                 goto out_restore;
1235         }
1236
1237         lock_buffer(data_bh);
1238         error = ext4_journal_get_create_access(handle, data_bh);
1239         if (error) {
1240                 unlock_buffer(data_bh);
1241                 error = -EIO;
1242                 goto out_restore;
1243         }
1244         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1245
1246         if (!S_ISDIR(inode->i_mode)) {
1247                 memcpy(data_bh->b_data, buf, inline_size);
1248                 set_buffer_uptodate(data_bh);
1249                 error = ext4_handle_dirty_metadata(handle,
1250                                                    inode, data_bh);
1251         } else {
1252                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1253                                                        buf, inline_size);
1254         }
1255
1256         unlock_buffer(data_bh);
1257 out_restore:
1258         if (error)
1259                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1260
1261 out:
1262         brelse(data_bh);
1263         kfree(buf);
1264         return error;
1265 }
1266
1267 /*
1268  * Try to add the new entry to the inline data.
1269  * If succeeds, return 0. If not, extended the inline dir and copied data to
1270  * the new created block.
1271  */
1272 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1273                               struct inode *dir, struct inode *inode)
1274 {
1275         int ret, ret2, inline_size, no_expand;
1276         void *inline_start;
1277         struct ext4_iloc iloc;
1278
1279         ret = ext4_get_inode_loc(dir, &iloc);
1280         if (ret)
1281                 return ret;
1282
1283         ext4_write_lock_xattr(dir, &no_expand);
1284         if (!ext4_has_inline_data(dir))
1285                 goto out;
1286
1287         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1288                                                  EXT4_INLINE_DOTDOT_SIZE;
1289         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1290
1291         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1292                                         inline_start, inline_size);
1293         if (ret != -ENOSPC)
1294                 goto out;
1295
1296         /* check whether it can be inserted to inline xattr space. */
1297         inline_size = EXT4_I(dir)->i_inline_size -
1298                         EXT4_MIN_INLINE_DATA_SIZE;
1299         if (!inline_size) {
1300                 /* Try to use the xattr space.*/
1301                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1302                 if (ret && ret != -ENOSPC)
1303                         goto out;
1304
1305                 inline_size = EXT4_I(dir)->i_inline_size -
1306                                 EXT4_MIN_INLINE_DATA_SIZE;
1307         }
1308
1309         if (inline_size) {
1310                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1311
1312                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1313                                                 inode, &iloc, inline_start,
1314                                                 inline_size);
1315
1316                 if (ret != -ENOSPC)
1317                         goto out;
1318         }
1319
1320         /*
1321          * The inline space is filled up, so create a new block for it.
1322          * As the extent tree will be created, we have to save the inline
1323          * dir first.
1324          */
1325         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1326
1327 out:
1328         ext4_write_unlock_xattr(dir, &no_expand);
1329         ret2 = ext4_mark_inode_dirty(handle, dir);
1330         if (unlikely(ret2 && !ret))
1331                 ret = ret2;
1332         brelse(iloc.bh);
1333         return ret;
1334 }
1335
1336 /*
1337  * This function fills a red-black tree with information from an
1338  * inlined dir.  It returns the number directory entries loaded
1339  * into the tree.  If there is an error it is returned in err.
1340  */
1341 int ext4_inlinedir_to_tree(struct file *dir_file,
1342                            struct inode *dir, ext4_lblk_t block,
1343                            struct dx_hash_info *hinfo,
1344                            __u32 start_hash, __u32 start_minor_hash,
1345                            int *has_inline_data)
1346 {
1347         int err = 0, count = 0;
1348         unsigned int parent_ino;
1349         int pos;
1350         struct ext4_dir_entry_2 *de;
1351         struct inode *inode = file_inode(dir_file);
1352         int ret, inline_size = 0;
1353         struct ext4_iloc iloc;
1354         void *dir_buf = NULL;
1355         struct ext4_dir_entry_2 fake;
1356         struct fscrypt_str tmp_str;
1357
1358         ret = ext4_get_inode_loc(inode, &iloc);
1359         if (ret)
1360                 return ret;
1361
1362         down_read(&EXT4_I(inode)->xattr_sem);
1363         if (!ext4_has_inline_data(inode)) {
1364                 up_read(&EXT4_I(inode)->xattr_sem);
1365                 *has_inline_data = 0;
1366                 goto out;
1367         }
1368
1369         inline_size = ext4_get_inline_size(inode);
1370         dir_buf = kmalloc(inline_size, GFP_NOFS);
1371         if (!dir_buf) {
1372                 ret = -ENOMEM;
1373                 up_read(&EXT4_I(inode)->xattr_sem);
1374                 goto out;
1375         }
1376
1377         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1378         up_read(&EXT4_I(inode)->xattr_sem);
1379         if (ret < 0)
1380                 goto out;
1381
1382         pos = 0;
1383         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1384         while (pos < inline_size) {
1385                 /*
1386                  * As inlined dir doesn't store any information about '.' and
1387                  * only the inode number of '..' is stored, we have to handle
1388                  * them differently.
1389                  */
1390                 if (pos == 0) {
1391                         fake.inode = cpu_to_le32(inode->i_ino);
1392                         fake.name_len = 1;
1393                         strcpy(fake.name, ".");
1394                         fake.rec_len = ext4_rec_len_to_disk(
1395                                                 EXT4_DIR_REC_LEN(fake.name_len),
1396                                                 inline_size);
1397                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1398                         de = &fake;
1399                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1400                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1401                         fake.inode = cpu_to_le32(parent_ino);
1402                         fake.name_len = 2;
1403                         strcpy(fake.name, "..");
1404                         fake.rec_len = ext4_rec_len_to_disk(
1405                                                 EXT4_DIR_REC_LEN(fake.name_len),
1406                                                 inline_size);
1407                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1408                         de = &fake;
1409                         pos = EXT4_INLINE_DOTDOT_SIZE;
1410                 } else {
1411                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1412                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1413                         if (ext4_check_dir_entry(inode, dir_file, de,
1414                                          iloc.bh, dir_buf,
1415                                          inline_size, pos)) {
1416                                 ret = count;
1417                                 goto out;
1418                         }
1419                 }
1420
1421                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1422                 if ((hinfo->hash < start_hash) ||
1423                     ((hinfo->hash == start_hash) &&
1424                      (hinfo->minor_hash < start_minor_hash)))
1425                         continue;
1426                 if (de->inode == 0)
1427                         continue;
1428                 tmp_str.name = de->name;
1429                 tmp_str.len = de->name_len;
1430                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1431                                               hinfo->minor_hash, de, &tmp_str);
1432                 if (err) {
1433                         ret = err;
1434                         goto out;
1435                 }
1436                 count++;
1437         }
1438         ret = count;
1439 out:
1440         kfree(dir_buf);
1441         brelse(iloc.bh);
1442         return ret;
1443 }
1444
1445 /*
1446  * So this function is called when the volume is mkfsed with
1447  * dir_index disabled. In order to keep f_pos persistent
1448  * after we convert from an inlined dir to a blocked based,
1449  * we just pretend that we are a normal dir and return the
1450  * offset as if '.' and '..' really take place.
1451  *
1452  */
1453 int ext4_read_inline_dir(struct file *file,
1454                          struct dir_context *ctx,
1455                          int *has_inline_data)
1456 {
1457         unsigned int offset, parent_ino;
1458         int i;
1459         struct ext4_dir_entry_2 *de;
1460         struct super_block *sb;
1461         struct inode *inode = file_inode(file);
1462         int ret, inline_size = 0;
1463         struct ext4_iloc iloc;
1464         void *dir_buf = NULL;
1465         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1466
1467         ret = ext4_get_inode_loc(inode, &iloc);
1468         if (ret)
1469                 return ret;
1470
1471         down_read(&EXT4_I(inode)->xattr_sem);
1472         if (!ext4_has_inline_data(inode)) {
1473                 up_read(&EXT4_I(inode)->xattr_sem);
1474                 *has_inline_data = 0;
1475                 goto out;
1476         }
1477
1478         inline_size = ext4_get_inline_size(inode);
1479         dir_buf = kmalloc(inline_size, GFP_NOFS);
1480         if (!dir_buf) {
1481                 ret = -ENOMEM;
1482                 up_read(&EXT4_I(inode)->xattr_sem);
1483                 goto out;
1484         }
1485
1486         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1487         up_read(&EXT4_I(inode)->xattr_sem);
1488         if (ret < 0)
1489                 goto out;
1490
1491         ret = 0;
1492         sb = inode->i_sb;
1493         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1494         offset = ctx->pos;
1495
1496         /*
1497          * dotdot_offset and dotdot_size is the real offset and
1498          * size for ".." and "." if the dir is block based while
1499          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1500          * So we will use extra_offset and extra_size to indicate them
1501          * during the inline dir iteration.
1502          */
1503         dotdot_offset = EXT4_DIR_REC_LEN(1);
1504         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1505         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1506         extra_size = extra_offset + inline_size;
1507
1508         /*
1509          * If the version has changed since the last call to
1510          * readdir(2), then we might be pointing to an invalid
1511          * dirent right now.  Scan from the start of the inline
1512          * dir to make sure.
1513          */
1514         if (!inode_eq_iversion(inode, file->f_version)) {
1515                 for (i = 0; i < extra_size && i < offset;) {
1516                         /*
1517                          * "." is with offset 0 and
1518                          * ".." is dotdot_offset.
1519                          */
1520                         if (!i) {
1521                                 i = dotdot_offset;
1522                                 continue;
1523                         } else if (i == dotdot_offset) {
1524                                 i = dotdot_size;
1525                                 continue;
1526                         }
1527                         /* for other entry, the real offset in
1528                          * the buf has to be tuned accordingly.
1529                          */
1530                         de = (struct ext4_dir_entry_2 *)
1531                                 (dir_buf + i - extra_offset);
1532                         /* It's too expensive to do a full
1533                          * dirent test each time round this
1534                          * loop, but we do have to test at
1535                          * least that it is non-zero.  A
1536                          * failure will be detected in the
1537                          * dirent test below. */
1538                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1539                                 < EXT4_DIR_REC_LEN(1))
1540                                 break;
1541                         i += ext4_rec_len_from_disk(de->rec_len,
1542                                                     extra_size);
1543                 }
1544                 offset = i;
1545                 ctx->pos = offset;
1546                 file->f_version = inode_query_iversion(inode);
1547         }
1548
1549         while (ctx->pos < extra_size) {
1550                 if (ctx->pos == 0) {
1551                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1552                                 goto out;
1553                         ctx->pos = dotdot_offset;
1554                         continue;
1555                 }
1556
1557                 if (ctx->pos == dotdot_offset) {
1558                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1559                                 goto out;
1560                         ctx->pos = dotdot_size;
1561                         continue;
1562                 }
1563
1564                 de = (struct ext4_dir_entry_2 *)
1565                         (dir_buf + ctx->pos - extra_offset);
1566                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1567                                          extra_size, ctx->pos))
1568                         goto out;
1569                 if (le32_to_cpu(de->inode)) {
1570                         if (!dir_emit(ctx, de->name, de->name_len,
1571                                       le32_to_cpu(de->inode),
1572                                       get_dtype(sb, de->file_type)))
1573                                 goto out;
1574                 }
1575                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1576         }
1577 out:
1578         kfree(dir_buf);
1579         brelse(iloc.bh);
1580         return ret;
1581 }
1582
1583 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1584                                         struct ext4_dir_entry_2 **parent_de,
1585                                         int *retval)
1586 {
1587         struct ext4_iloc iloc;
1588
1589         *retval = ext4_get_inode_loc(inode, &iloc);
1590         if (*retval)
1591                 return NULL;
1592
1593         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1594
1595         return iloc.bh;
1596 }
1597
1598 /*
1599  * Try to create the inline data for the new dir.
1600  * If it succeeds, return 0, otherwise return the error.
1601  * In case of ENOSPC, the caller should create the normal disk layout dir.
1602  */
1603 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1604                                struct inode *inode)
1605 {
1606         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1607         struct ext4_iloc iloc;
1608         struct ext4_dir_entry_2 *de;
1609
1610         ret = ext4_get_inode_loc(inode, &iloc);
1611         if (ret)
1612                 return ret;
1613
1614         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1615         if (ret)
1616                 goto out;
1617
1618         /*
1619          * For inline dir, we only save the inode information for the ".."
1620          * and create a fake dentry to cover the left space.
1621          */
1622         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1623         de->inode = cpu_to_le32(parent->i_ino);
1624         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1625         de->inode = 0;
1626         de->rec_len = ext4_rec_len_to_disk(
1627                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1628                                 inline_size);
1629         set_nlink(inode, 2);
1630         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1631 out:
1632         brelse(iloc.bh);
1633         return ret;
1634 }
1635
1636 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1637                                         struct ext4_filename *fname,
1638                                         struct ext4_dir_entry_2 **res_dir,
1639                                         int *has_inline_data)
1640 {
1641         int ret;
1642         struct ext4_iloc iloc;
1643         void *inline_start;
1644         int inline_size;
1645
1646         if (ext4_get_inode_loc(dir, &iloc))
1647                 return NULL;
1648
1649         down_read(&EXT4_I(dir)->xattr_sem);
1650         if (!ext4_has_inline_data(dir)) {
1651                 *has_inline_data = 0;
1652                 goto out;
1653         }
1654
1655         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1656                                                 EXT4_INLINE_DOTDOT_SIZE;
1657         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1658         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1659                               dir, fname, 0, res_dir);
1660         if (ret == 1)
1661                 goto out_find;
1662         if (ret < 0)
1663                 goto out;
1664
1665         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1666                 goto out;
1667
1668         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1669         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1670
1671         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1672                               dir, fname, 0, res_dir);
1673         if (ret == 1)
1674                 goto out_find;
1675
1676 out:
1677         brelse(iloc.bh);
1678         iloc.bh = NULL;
1679 out_find:
1680         up_read(&EXT4_I(dir)->xattr_sem);
1681         return iloc.bh;
1682 }
1683
1684 int ext4_delete_inline_entry(handle_t *handle,
1685                              struct inode *dir,
1686                              struct ext4_dir_entry_2 *de_del,
1687                              struct buffer_head *bh,
1688                              int *has_inline_data)
1689 {
1690         int err, inline_size, no_expand;
1691         struct ext4_iloc iloc;
1692         void *inline_start;
1693
1694         err = ext4_get_inode_loc(dir, &iloc);
1695         if (err)
1696                 return err;
1697
1698         ext4_write_lock_xattr(dir, &no_expand);
1699         if (!ext4_has_inline_data(dir)) {
1700                 *has_inline_data = 0;
1701                 goto out;
1702         }
1703
1704         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1705                 EXT4_MIN_INLINE_DATA_SIZE) {
1706                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1707                                         EXT4_INLINE_DOTDOT_SIZE;
1708                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1709                                 EXT4_INLINE_DOTDOT_SIZE;
1710         } else {
1711                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1712                 inline_size = ext4_get_inline_size(dir) -
1713                                 EXT4_MIN_INLINE_DATA_SIZE;
1714         }
1715
1716         BUFFER_TRACE(bh, "get_write_access");
1717         err = ext4_journal_get_write_access(handle, bh);
1718         if (err)
1719                 goto out;
1720
1721         err = ext4_generic_delete_entry(dir, de_del, bh,
1722                                         inline_start, inline_size, 0);
1723         if (err)
1724                 goto out;
1725
1726         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1727 out:
1728         ext4_write_unlock_xattr(dir, &no_expand);
1729         if (likely(err == 0))
1730                 err = ext4_mark_inode_dirty(handle, dir);
1731         brelse(iloc.bh);
1732         if (err != -ENOENT)
1733                 ext4_std_error(dir->i_sb, err);
1734         return err;
1735 }
1736
1737 /*
1738  * Get the inline dentry at offset.
1739  */
1740 static inline struct ext4_dir_entry_2 *
1741 ext4_get_inline_entry(struct inode *inode,
1742                       struct ext4_iloc *iloc,
1743                       unsigned int offset,
1744                       void **inline_start,
1745                       int *inline_size)
1746 {
1747         void *inline_pos;
1748
1749         BUG_ON(offset > ext4_get_inline_size(inode));
1750
1751         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1752                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1753                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1754         } else {
1755                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1756                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1757                 *inline_size = ext4_get_inline_size(inode) -
1758                                 EXT4_MIN_INLINE_DATA_SIZE;
1759         }
1760
1761         if (inline_start)
1762                 *inline_start = inline_pos;
1763         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1764 }
1765
1766 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1767 {
1768         int err, inline_size;
1769         struct ext4_iloc iloc;
1770         size_t inline_len;
1771         void *inline_pos;
1772         unsigned int offset;
1773         struct ext4_dir_entry_2 *de;
1774         bool ret = false;
1775
1776         err = ext4_get_inode_loc(dir, &iloc);
1777         if (err) {
1778                 EXT4_ERROR_INODE_ERR(dir, -err,
1779                                      "error %d getting inode %lu block",
1780                                      err, dir->i_ino);
1781                 return false;
1782         }
1783
1784         down_read(&EXT4_I(dir)->xattr_sem);
1785         if (!ext4_has_inline_data(dir)) {
1786                 *has_inline_data = 0;
1787                 ret = true;
1788                 goto out;
1789         }
1790
1791         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1792         if (!le32_to_cpu(de->inode)) {
1793                 ext4_warning(dir->i_sb,
1794                              "bad inline directory (dir #%lu) - no `..'",
1795                              dir->i_ino);
1796                 goto out;
1797         }
1798
1799         inline_len = ext4_get_inline_size(dir);
1800         offset = EXT4_INLINE_DOTDOT_SIZE;
1801         while (offset < inline_len) {
1802                 de = ext4_get_inline_entry(dir, &iloc, offset,
1803                                            &inline_pos, &inline_size);
1804                 if (ext4_check_dir_entry(dir, NULL, de,
1805                                          iloc.bh, inline_pos,
1806                                          inline_size, offset)) {
1807                         ext4_warning(dir->i_sb,
1808                                      "bad inline directory (dir #%lu) - "
1809                                      "inode %u, rec_len %u, name_len %d"
1810                                      "inline size %d",
1811                                      dir->i_ino, le32_to_cpu(de->inode),
1812                                      le16_to_cpu(de->rec_len), de->name_len,
1813                                      inline_size);
1814                         goto out;
1815                 }
1816                 if (le32_to_cpu(de->inode)) {
1817                         goto out;
1818                 }
1819                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1820         }
1821
1822         ret = true;
1823 out:
1824         up_read(&EXT4_I(dir)->xattr_sem);
1825         brelse(iloc.bh);
1826         return ret;
1827 }
1828
1829 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1830 {
1831         int ret, no_expand;
1832
1833         ext4_write_lock_xattr(inode, &no_expand);
1834         ret = ext4_destroy_inline_data_nolock(handle, inode);
1835         ext4_write_unlock_xattr(inode, &no_expand);
1836
1837         return ret;
1838 }
1839
1840 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1841 {
1842         __u64 addr;
1843         int error = -EAGAIN;
1844         struct ext4_iloc iloc;
1845
1846         down_read(&EXT4_I(inode)->xattr_sem);
1847         if (!ext4_has_inline_data(inode))
1848                 goto out;
1849
1850         error = ext4_get_inode_loc(inode, &iloc);
1851         if (error)
1852                 goto out;
1853
1854         addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1855         addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1856         addr += offsetof(struct ext4_inode, i_block);
1857
1858         brelse(iloc.bh);
1859
1860         iomap->addr = addr;
1861         iomap->offset = 0;
1862         iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1863                               i_size_read(inode));
1864         iomap->type = IOMAP_INLINE;
1865         iomap->flags = 0;
1866
1867 out:
1868         up_read(&EXT4_I(inode)->xattr_sem);
1869         return error;
1870 }
1871
1872 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1873 {
1874         handle_t *handle;
1875         int inline_size, value_len, needed_blocks, no_expand, err = 0;
1876         size_t i_size;
1877         void *value = NULL;
1878         struct ext4_xattr_ibody_find is = {
1879                 .s = { .not_found = -ENODATA, },
1880         };
1881         struct ext4_xattr_info i = {
1882                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1883                 .name = EXT4_XATTR_SYSTEM_DATA,
1884         };
1885
1886
1887         needed_blocks = ext4_writepage_trans_blocks(inode);
1888         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1889         if (IS_ERR(handle))
1890                 return PTR_ERR(handle);
1891
1892         ext4_write_lock_xattr(inode, &no_expand);
1893         if (!ext4_has_inline_data(inode)) {
1894                 ext4_write_unlock_xattr(inode, &no_expand);
1895                 *has_inline = 0;
1896                 ext4_journal_stop(handle);
1897                 return 0;
1898         }
1899
1900         if ((err = ext4_orphan_add(handle, inode)) != 0)
1901                 goto out;
1902
1903         if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1904                 goto out;
1905
1906         down_write(&EXT4_I(inode)->i_data_sem);
1907         i_size = inode->i_size;
1908         inline_size = ext4_get_inline_size(inode);
1909         EXT4_I(inode)->i_disksize = i_size;
1910
1911         if (i_size < inline_size) {
1912                 /* Clear the content in the xattr space. */
1913                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1914                         if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1915                                 goto out_error;
1916
1917                         BUG_ON(is.s.not_found);
1918
1919                         value_len = le32_to_cpu(is.s.here->e_value_size);
1920                         value = kmalloc(value_len, GFP_NOFS);
1921                         if (!value) {
1922                                 err = -ENOMEM;
1923                                 goto out_error;
1924                         }
1925
1926                         err = ext4_xattr_ibody_get(inode, i.name_index,
1927                                                    i.name, value, value_len);
1928                         if (err <= 0)
1929                                 goto out_error;
1930
1931                         i.value = value;
1932                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1933                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1934                         err = ext4_xattr_ibody_inline_set(handle, inode,
1935                                                           &i, &is);
1936                         if (err)
1937                                 goto out_error;
1938                 }
1939
1940                 /* Clear the content within i_blocks. */
1941                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1942                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1943                         memset(p + i_size, 0,
1944                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1945                 }
1946
1947                 EXT4_I(inode)->i_inline_size = i_size <
1948                                         EXT4_MIN_INLINE_DATA_SIZE ?
1949                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1950         }
1951
1952 out_error:
1953         up_write(&EXT4_I(inode)->i_data_sem);
1954 out:
1955         brelse(is.iloc.bh);
1956         ext4_write_unlock_xattr(inode, &no_expand);
1957         kfree(value);
1958         if (inode->i_nlink)
1959                 ext4_orphan_del(handle, inode);
1960
1961         if (err == 0) {
1962                 inode->i_mtime = inode->i_ctime = current_time(inode);
1963                 err = ext4_mark_inode_dirty(handle, inode);
1964                 if (IS_SYNC(inode))
1965                         ext4_handle_sync(handle);
1966         }
1967         ext4_journal_stop(handle);
1968         return err;
1969 }
1970
1971 int ext4_convert_inline_data(struct inode *inode)
1972 {
1973         int error, needed_blocks, no_expand;
1974         handle_t *handle;
1975         struct ext4_iloc iloc;
1976
1977         if (!ext4_has_inline_data(inode)) {
1978                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1979                 return 0;
1980         } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1981                 /*
1982                  * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
1983                  * cleared. This means we are in the middle of moving of
1984                  * inline data to delay allocated block. Just force writeout
1985                  * here to finish conversion.
1986                  */
1987                 error = filemap_flush(inode->i_mapping);
1988                 if (error)
1989                         return error;
1990                 if (!ext4_has_inline_data(inode))
1991                         return 0;
1992         }
1993
1994         needed_blocks = ext4_writepage_trans_blocks(inode);
1995
1996         iloc.bh = NULL;
1997         error = ext4_get_inode_loc(inode, &iloc);
1998         if (error)
1999                 return error;
2000
2001         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2002         if (IS_ERR(handle)) {
2003                 error = PTR_ERR(handle);
2004                 goto out_free;
2005         }
2006
2007         ext4_write_lock_xattr(inode, &no_expand);
2008         if (ext4_has_inline_data(inode))
2009                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2010         ext4_write_unlock_xattr(inode, &no_expand);
2011         ext4_journal_stop(handle);
2012 out_free:
2013         brelse(iloc.bh);
2014         return error;
2015 }