Merge tag 'libnvdimm-for-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdim...
[linux-2.6-microblaze.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         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
38                         EXT4_GOOD_OLD_INODE_SIZE -
39                         EXT4_I(inode)->i_extra_isize -
40                         sizeof(struct ext4_xattr_ibody_header);
41
42         /*
43          * We need to subtract another sizeof(__u32) since an in-inode xattr
44          * needs an empty 4 bytes to indicate the gap between the xattr entry
45          * and the name/value pair.
46          */
47         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
48                 return EXT4_XATTR_SIZE(min_offs -
49                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
50                         EXT4_XATTR_ROUND - sizeof(__u32));
51
52         raw_inode = ext4_raw_inode(iloc);
53         header = IHDR(inode, raw_inode);
54         entry = IFIRST(header);
55
56         /* Compute min_offs. */
57         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
58                 if (!entry->e_value_inum && entry->e_value_size) {
59                         size_t offs = le16_to_cpu(entry->e_value_offs);
60                         if (offs < min_offs)
61                                 min_offs = offs;
62                 }
63         }
64         free = min_offs -
65                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
66
67         if (EXT4_I(inode)->i_inline_off) {
68                 entry = (struct ext4_xattr_entry *)
69                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
70
71                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
72                 goto out;
73         }
74
75         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
76
77         if (free > EXT4_XATTR_ROUND)
78                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
79         else
80                 free = 0;
81
82 out:
83         return free;
84 }
85
86 /*
87  * Get the maximum size we now can store in an inode.
88  * If we can't find the space for a xattr entry, don't use the space
89  * of the extents since we have no space to indicate the inline data.
90  */
91 int ext4_get_max_inline_size(struct inode *inode)
92 {
93         int error, max_inline_size;
94         struct ext4_iloc iloc;
95
96         if (EXT4_I(inode)->i_extra_isize == 0)
97                 return 0;
98
99         error = ext4_get_inode_loc(inode, &iloc);
100         if (error) {
101                 ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
102                                      "can't get inode location %lu",
103                                      inode->i_ino);
104                 return 0;
105         }
106
107         down_read(&EXT4_I(inode)->xattr_sem);
108         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
109         up_read(&EXT4_I(inode)->xattr_sem);
110
111         brelse(iloc.bh);
112
113         if (!max_inline_size)
114                 return 0;
115
116         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
117 }
118
119 /*
120  * this function does not take xattr_sem, which is OK because it is
121  * currently only used in a code path coming form ext4_iget, before
122  * the new inode has been unlocked
123  */
124 int ext4_find_inline_data_nolock(struct inode *inode)
125 {
126         struct ext4_xattr_ibody_find is = {
127                 .s = { .not_found = -ENODATA, },
128         };
129         struct ext4_xattr_info i = {
130                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
131                 .name = EXT4_XATTR_SYSTEM_DATA,
132         };
133         int error;
134
135         if (EXT4_I(inode)->i_extra_isize == 0)
136                 return 0;
137
138         error = ext4_get_inode_loc(inode, &is.iloc);
139         if (error)
140                 return error;
141
142         error = ext4_xattr_ibody_find(inode, &i, &is);
143         if (error)
144                 goto out;
145
146         if (!is.s.not_found) {
147                 if (is.s.here->e_value_inum) {
148                         EXT4_ERROR_INODE(inode, "inline data xattr refers "
149                                          "to an external xattr inode");
150                         error = -EFSCORRUPTED;
151                         goto out;
152                 }
153                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154                                         (void *)ext4_raw_inode(&is.iloc));
155                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156                                 le32_to_cpu(is.s.here->e_value_size);
157                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158         }
159 out:
160         brelse(is.iloc.bh);
161         return error;
162 }
163
164 static int ext4_read_inline_data(struct inode *inode, void *buffer,
165                                  unsigned int len,
166                                  struct ext4_iloc *iloc)
167 {
168         struct ext4_xattr_entry *entry;
169         struct ext4_xattr_ibody_header *header;
170         int cp_len = 0;
171         struct ext4_inode *raw_inode;
172
173         if (!len)
174                 return 0;
175
176         BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179                         len : EXT4_MIN_INLINE_DATA_SIZE;
180
181         raw_inode = ext4_raw_inode(iloc);
182         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184         len -= cp_len;
185         buffer += cp_len;
186
187         if (!len)
188                 goto out;
189
190         header = IHDR(inode, raw_inode);
191         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192                                             EXT4_I(inode)->i_inline_off);
193         len = min_t(unsigned int, len,
194                     (unsigned int)le32_to_cpu(entry->e_value_size));
195
196         memcpy(buffer,
197                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198         cp_len += len;
199
200 out:
201         return cp_len;
202 }
203
204 /*
205  * write the buffer to the inline inode.
206  * If 'create' is set, we don't need to do the extra copy in the xattr
207  * value since it is already handled by ext4_xattr_ibody_set.
208  * That saves us one memcpy.
209  */
210 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211                                    void *buffer, loff_t pos, unsigned int len)
212 {
213         struct ext4_xattr_entry *entry;
214         struct ext4_xattr_ibody_header *header;
215         struct ext4_inode *raw_inode;
216         int cp_len = 0;
217
218         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
219                 return;
220
221         BUG_ON(!EXT4_I(inode)->i_inline_off);
222         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
223
224         raw_inode = ext4_raw_inode(iloc);
225         buffer += pos;
226
227         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
228                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
229                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
230                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
231
232                 len -= cp_len;
233                 buffer += cp_len;
234                 pos += cp_len;
235         }
236
237         if (!len)
238                 return;
239
240         pos -= EXT4_MIN_INLINE_DATA_SIZE;
241         header = IHDR(inode, raw_inode);
242         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
243                                             EXT4_I(inode)->i_inline_off);
244
245         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
246                buffer, len);
247 }
248
249 static int ext4_create_inline_data(handle_t *handle,
250                                    struct inode *inode, unsigned len)
251 {
252         int error;
253         void *value = NULL;
254         struct ext4_xattr_ibody_find is = {
255                 .s = { .not_found = -ENODATA, },
256         };
257         struct ext4_xattr_info i = {
258                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
259                 .name = EXT4_XATTR_SYSTEM_DATA,
260         };
261
262         error = ext4_get_inode_loc(inode, &is.iloc);
263         if (error)
264                 return error;
265
266         BUFFER_TRACE(is.iloc.bh, "get_write_access");
267         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
268                                               EXT4_JTR_NONE);
269         if (error)
270                 goto out;
271
272         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
273                 value = EXT4_ZERO_XATTR_VALUE;
274                 len -= EXT4_MIN_INLINE_DATA_SIZE;
275         } else {
276                 value = "";
277                 len = 0;
278         }
279
280         /* Insert the xttr entry. */
281         i.value = value;
282         i.value_len = len;
283
284         error = ext4_xattr_ibody_find(inode, &i, &is);
285         if (error)
286                 goto out;
287
288         BUG_ON(!is.s.not_found);
289
290         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
291         if (error) {
292                 if (error == -ENOSPC)
293                         ext4_clear_inode_state(inode,
294                                                EXT4_STATE_MAY_INLINE_DATA);
295                 goto out;
296         }
297
298         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
299                 0, EXT4_MIN_INLINE_DATA_SIZE);
300
301         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
302                                       (void *)ext4_raw_inode(&is.iloc));
303         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
304         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
305         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
306         get_bh(is.iloc.bh);
307         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
308
309 out:
310         brelse(is.iloc.bh);
311         return error;
312 }
313
314 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
315                                    unsigned int len)
316 {
317         int error;
318         void *value = NULL;
319         struct ext4_xattr_ibody_find is = {
320                 .s = { .not_found = -ENODATA, },
321         };
322         struct ext4_xattr_info i = {
323                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
324                 .name = EXT4_XATTR_SYSTEM_DATA,
325         };
326
327         /* If the old space is ok, write the data directly. */
328         if (len <= EXT4_I(inode)->i_inline_size)
329                 return 0;
330
331         error = ext4_get_inode_loc(inode, &is.iloc);
332         if (error)
333                 return error;
334
335         error = ext4_xattr_ibody_find(inode, &i, &is);
336         if (error)
337                 goto out;
338
339         BUG_ON(is.s.not_found);
340
341         len -= EXT4_MIN_INLINE_DATA_SIZE;
342         value = kzalloc(len, GFP_NOFS);
343         if (!value) {
344                 error = -ENOMEM;
345                 goto out;
346         }
347
348         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
349                                      value, len);
350         if (error == -ENODATA)
351                 goto out;
352
353         BUFFER_TRACE(is.iloc.bh, "get_write_access");
354         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
355                                               EXT4_JTR_NONE);
356         if (error)
357                 goto out;
358
359         /* Update the xattr entry. */
360         i.value = value;
361         i.value_len = len;
362
363         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
364         if (error)
365                 goto out;
366
367         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
368                                       (void *)ext4_raw_inode(&is.iloc));
369         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
370                                 le32_to_cpu(is.s.here->e_value_size);
371         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
372         get_bh(is.iloc.bh);
373         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
374
375 out:
376         kfree(value);
377         brelse(is.iloc.bh);
378         return error;
379 }
380
381 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
382                                     unsigned int len)
383 {
384         int ret, size, no_expand;
385         struct ext4_inode_info *ei = EXT4_I(inode);
386
387         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
388                 return -ENOSPC;
389
390         size = ext4_get_max_inline_size(inode);
391         if (size < len)
392                 return -ENOSPC;
393
394         ext4_write_lock_xattr(inode, &no_expand);
395
396         if (ei->i_inline_off)
397                 ret = ext4_update_inline_data(handle, inode, len);
398         else
399                 ret = ext4_create_inline_data(handle, inode, len);
400
401         ext4_write_unlock_xattr(inode, &no_expand);
402         return ret;
403 }
404
405 static int ext4_destroy_inline_data_nolock(handle_t *handle,
406                                            struct inode *inode)
407 {
408         struct ext4_inode_info *ei = EXT4_I(inode);
409         struct ext4_xattr_ibody_find is = {
410                 .s = { .not_found = 0, },
411         };
412         struct ext4_xattr_info i = {
413                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
414                 .name = EXT4_XATTR_SYSTEM_DATA,
415                 .value = NULL,
416                 .value_len = 0,
417         };
418         int error;
419
420         if (!ei->i_inline_off)
421                 return 0;
422
423         error = ext4_get_inode_loc(inode, &is.iloc);
424         if (error)
425                 return error;
426
427         error = ext4_xattr_ibody_find(inode, &i, &is);
428         if (error)
429                 goto out;
430
431         BUFFER_TRACE(is.iloc.bh, "get_write_access");
432         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
433                                               EXT4_JTR_NONE);
434         if (error)
435                 goto out;
436
437         error = ext4_xattr_ibody_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, inode, 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, inode->i_sb, iloc.bh,
689                                             EXT4_JTR_NONE);
690         if (ret)
691                 goto out;
692
693         flags |= AOP_FLAG_NOFS;
694
695         page = grab_cache_page_write_begin(mapping, 0, flags);
696         if (!page) {
697                 ret = -ENOMEM;
698                 goto out;
699         }
700
701         *pagep = page;
702         down_read(&EXT4_I(inode)->xattr_sem);
703         if (!ext4_has_inline_data(inode)) {
704                 ret = 0;
705                 unlock_page(page);
706                 put_page(page);
707                 goto out_up_read;
708         }
709
710         if (!PageUptodate(page)) {
711                 ret = ext4_read_inline_page(inode, page);
712                 if (ret < 0) {
713                         unlock_page(page);
714                         put_page(page);
715                         goto out_up_read;
716                 }
717         }
718
719         ret = 1;
720         handle = NULL;
721 out_up_read:
722         up_read(&EXT4_I(inode)->xattr_sem);
723 out:
724         if (handle && (ret != 1))
725                 ext4_journal_stop(handle);
726         brelse(iloc.bh);
727         return ret;
728 convert:
729         return ext4_convert_inline_data_to_extent(mapping,
730                                                   inode, flags);
731 }
732
733 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
734                                unsigned copied, struct page *page)
735 {
736         int ret, no_expand;
737         void *kaddr;
738         struct ext4_iloc iloc;
739
740         if (unlikely(copied < len)) {
741                 if (!PageUptodate(page)) {
742                         copied = 0;
743                         goto out;
744                 }
745         }
746
747         ret = ext4_get_inode_loc(inode, &iloc);
748         if (ret) {
749                 ext4_std_error(inode->i_sb, ret);
750                 copied = 0;
751                 goto out;
752         }
753
754         ext4_write_lock_xattr(inode, &no_expand);
755         BUG_ON(!ext4_has_inline_data(inode));
756
757         /*
758          * ei->i_inline_off may have changed since ext4_write_begin()
759          * called ext4_try_to_write_inline_data()
760          */
761         (void) ext4_find_inline_data_nolock(inode);
762
763         kaddr = kmap_atomic(page);
764         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
765         kunmap_atomic(kaddr);
766         SetPageUptodate(page);
767         /* clear page dirty so that writepages wouldn't work for us. */
768         ClearPageDirty(page);
769
770         ext4_write_unlock_xattr(inode, &no_expand);
771         brelse(iloc.bh);
772         mark_inode_dirty(inode);
773 out:
774         return copied;
775 }
776
777 struct buffer_head *
778 ext4_journalled_write_inline_data(struct inode *inode,
779                                   unsigned len,
780                                   struct page *page)
781 {
782         int ret, no_expand;
783         void *kaddr;
784         struct ext4_iloc iloc;
785
786         ret = ext4_get_inode_loc(inode, &iloc);
787         if (ret) {
788                 ext4_std_error(inode->i_sb, ret);
789                 return NULL;
790         }
791
792         ext4_write_lock_xattr(inode, &no_expand);
793         kaddr = kmap_atomic(page);
794         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
795         kunmap_atomic(kaddr);
796         ext4_write_unlock_xattr(inode, &no_expand);
797
798         return iloc.bh;
799 }
800
801 /*
802  * Try to make the page cache and handle ready for the inline data case.
803  * We can call this function in 2 cases:
804  * 1. The inode is created and the first write exceeds inline size. We can
805  *    clear the inode state safely.
806  * 2. The inode has inline data, then we need to read the data, make it
807  *    update and dirty so that ext4_da_writepages can handle it. We don't
808  *    need to start the journal since the file's metadata isn't changed now.
809  */
810 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
811                                                  struct inode *inode,
812                                                  unsigned flags,
813                                                  void **fsdata)
814 {
815         int ret = 0, inline_size;
816         struct page *page;
817
818         page = grab_cache_page_write_begin(mapping, 0, flags);
819         if (!page)
820                 return -ENOMEM;
821
822         down_read(&EXT4_I(inode)->xattr_sem);
823         if (!ext4_has_inline_data(inode)) {
824                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
825                 goto out;
826         }
827
828         inline_size = ext4_get_inline_size(inode);
829
830         if (!PageUptodate(page)) {
831                 ret = ext4_read_inline_page(inode, page);
832                 if (ret < 0)
833                         goto out;
834         }
835
836         ret = __block_write_begin(page, 0, inline_size,
837                                   ext4_da_get_block_prep);
838         if (ret) {
839                 up_read(&EXT4_I(inode)->xattr_sem);
840                 unlock_page(page);
841                 put_page(page);
842                 ext4_truncate_failed_write(inode);
843                 return ret;
844         }
845
846         SetPageDirty(page);
847         SetPageUptodate(page);
848         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
849         *fsdata = (void *)CONVERT_INLINE_DATA;
850
851 out:
852         up_read(&EXT4_I(inode)->xattr_sem);
853         if (page) {
854                 unlock_page(page);
855                 put_page(page);
856         }
857         return ret;
858 }
859
860 /*
861  * Prepare the write for the inline data.
862  * If the data can be written into the inode, we just read
863  * the page and make it uptodate, and start the journal.
864  * Otherwise read the page, makes it dirty so that it can be
865  * handle in writepages(the i_disksize update is left to the
866  * normal ext4_da_write_end).
867  */
868 int ext4_da_write_inline_data_begin(struct address_space *mapping,
869                                     struct inode *inode,
870                                     loff_t pos, unsigned len,
871                                     unsigned flags,
872                                     struct page **pagep,
873                                     void **fsdata)
874 {
875         int ret, inline_size;
876         handle_t *handle;
877         struct page *page;
878         struct ext4_iloc iloc;
879         int retries = 0;
880
881         ret = ext4_get_inode_loc(inode, &iloc);
882         if (ret)
883                 return ret;
884
885 retry_journal:
886         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
887         if (IS_ERR(handle)) {
888                 ret = PTR_ERR(handle);
889                 goto out;
890         }
891
892         inline_size = ext4_get_max_inline_size(inode);
893
894         ret = -ENOSPC;
895         if (inline_size >= pos + len) {
896                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
897                 if (ret && ret != -ENOSPC)
898                         goto out_journal;
899         }
900
901         /*
902          * We cannot recurse into the filesystem as the transaction
903          * is already started.
904          */
905         flags |= AOP_FLAG_NOFS;
906
907         if (ret == -ENOSPC) {
908                 ext4_journal_stop(handle);
909                 ret = ext4_da_convert_inline_data_to_extent(mapping,
910                                                             inode,
911                                                             flags,
912                                                             fsdata);
913                 if (ret == -ENOSPC &&
914                     ext4_should_retry_alloc(inode->i_sb, &retries))
915                         goto retry_journal;
916                 goto out;
917         }
918
919         page = grab_cache_page_write_begin(mapping, 0, flags);
920         if (!page) {
921                 ret = -ENOMEM;
922                 goto out_journal;
923         }
924
925         down_read(&EXT4_I(inode)->xattr_sem);
926         if (!ext4_has_inline_data(inode)) {
927                 ret = 0;
928                 goto out_release_page;
929         }
930
931         if (!PageUptodate(page)) {
932                 ret = ext4_read_inline_page(inode, page);
933                 if (ret < 0)
934                         goto out_release_page;
935         }
936         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
937                                             EXT4_JTR_NONE);
938         if (ret)
939                 goto out_release_page;
940
941         up_read(&EXT4_I(inode)->xattr_sem);
942         *pagep = page;
943         brelse(iloc.bh);
944         return 1;
945 out_release_page:
946         up_read(&EXT4_I(inode)->xattr_sem);
947         unlock_page(page);
948         put_page(page);
949 out_journal:
950         ext4_journal_stop(handle);
951 out:
952         brelse(iloc.bh);
953         return ret;
954 }
955
956 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
957                                   unsigned len, unsigned copied,
958                                   struct page *page)
959 {
960         int ret;
961
962         ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
963         if (ret < 0) {
964                 unlock_page(page);
965                 put_page(page);
966                 return ret;
967         }
968         copied = ret;
969
970         /*
971          * No need to use i_size_read() here, the i_size
972          * cannot change under us because we hold i_mutex.
973          *
974          * But it's important to update i_size while still holding page lock:
975          * page writeout could otherwise come in and zero beyond i_size.
976          */
977         if (pos+copied > inode->i_size)
978                 i_size_write(inode, pos+copied);
979         unlock_page(page);
980         put_page(page);
981
982         /*
983          * Don't mark the inode dirty under page lock. First, it unnecessarily
984          * makes the holding time of page lock longer. Second, it forces lock
985          * ordering of page lock and transaction start for journaling
986          * filesystems.
987          */
988         mark_inode_dirty(inode);
989
990         return copied;
991 }
992
993 #ifdef INLINE_DIR_DEBUG
994 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
995                           void *inline_start, int inline_size)
996 {
997         int offset;
998         unsigned short de_len;
999         struct ext4_dir_entry_2 *de = inline_start;
1000         void *dlimit = inline_start + inline_size;
1001
1002         trace_printk("inode %lu\n", dir->i_ino);
1003         offset = 0;
1004         while ((void *)de < dlimit) {
1005                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1006                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1007                              offset, de_len, de->name_len, de->name,
1008                              de->name_len, le32_to_cpu(de->inode));
1009                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1010                                          inline_start, inline_size, offset))
1011                         BUG();
1012
1013                 offset += de_len;
1014                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1015         }
1016 }
1017 #else
1018 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1019 #endif
1020
1021 /*
1022  * Add a new entry into a inline dir.
1023  * It will return -ENOSPC if no space is available, and -EIO
1024  * and -EEXIST if directory entry already exists.
1025  */
1026 static int ext4_add_dirent_to_inline(handle_t *handle,
1027                                      struct ext4_filename *fname,
1028                                      struct inode *dir,
1029                                      struct inode *inode,
1030                                      struct ext4_iloc *iloc,
1031                                      void *inline_start, int inline_size)
1032 {
1033         int             err;
1034         struct ext4_dir_entry_2 *de;
1035
1036         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1037                                 inline_size, fname, &de);
1038         if (err)
1039                 return err;
1040
1041         BUFFER_TRACE(iloc->bh, "get_write_access");
1042         err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1043                                             EXT4_JTR_NONE);
1044         if (err)
1045                 return err;
1046         ext4_insert_dentry(dir, inode, de, inline_size, fname);
1047
1048         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1049
1050         /*
1051          * XXX shouldn't update any times until successful
1052          * completion of syscall, but too many callers depend
1053          * on this.
1054          *
1055          * XXX similarly, too many callers depend on
1056          * ext4_new_inode() setting the times, but error
1057          * recovery deletes the inode, so the worst that can
1058          * happen is that the times are slightly out of date
1059          * and/or different from the directory change time.
1060          */
1061         dir->i_mtime = dir->i_ctime = current_time(dir);
1062         ext4_update_dx_flag(dir);
1063         inode_inc_iversion(dir);
1064         return 1;
1065 }
1066
1067 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1068                                        struct ext4_iloc *iloc)
1069 {
1070         struct ext4_xattr_entry *entry;
1071         struct ext4_xattr_ibody_header *header;
1072
1073         BUG_ON(!EXT4_I(inode)->i_inline_off);
1074
1075         header = IHDR(inode, ext4_raw_inode(iloc));
1076         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1077                                             EXT4_I(inode)->i_inline_off);
1078
1079         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1080 }
1081
1082 /* Set the final de to cover the whole block. */
1083 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1084 {
1085         struct ext4_dir_entry_2 *de, *prev_de;
1086         void *limit;
1087         int de_len;
1088
1089         de = (struct ext4_dir_entry_2 *)de_buf;
1090         if (old_size) {
1091                 limit = de_buf + old_size;
1092                 do {
1093                         prev_de = de;
1094                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1095                         de_buf += de_len;
1096                         de = (struct ext4_dir_entry_2 *)de_buf;
1097                 } while (de_buf < limit);
1098
1099                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1100                                                         old_size, new_size);
1101         } else {
1102                 /* this is just created, so create an empty entry. */
1103                 de->inode = 0;
1104                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1105         }
1106 }
1107
1108 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1109                                   struct ext4_iloc *iloc)
1110 {
1111         int ret;
1112         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1113         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1114
1115         if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1116                 return -ENOSPC;
1117
1118         ret = ext4_update_inline_data(handle, dir,
1119                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1120         if (ret)
1121                 return ret;
1122
1123         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1124                              EXT4_I(dir)->i_inline_size -
1125                                                 EXT4_MIN_INLINE_DATA_SIZE);
1126         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1127         return 0;
1128 }
1129
1130 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1131                                      struct ext4_iloc *iloc,
1132                                      void *buf, int inline_size)
1133 {
1134         ext4_create_inline_data(handle, inode, inline_size);
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, inode->i_sb, data_bh,
1239                                                EXT4_JTR_NONE);
1240         if (error) {
1241                 unlock_buffer(data_bh);
1242                 error = -EIO;
1243                 goto out_restore;
1244         }
1245         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1246
1247         if (!S_ISDIR(inode->i_mode)) {
1248                 memcpy(data_bh->b_data, buf, inline_size);
1249                 set_buffer_uptodate(data_bh);
1250                 error = ext4_handle_dirty_metadata(handle,
1251                                                    inode, data_bh);
1252         } else {
1253                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1254                                                        buf, inline_size);
1255         }
1256
1257         unlock_buffer(data_bh);
1258 out_restore:
1259         if (error)
1260                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1261
1262 out:
1263         brelse(data_bh);
1264         kfree(buf);
1265         return error;
1266 }
1267
1268 /*
1269  * Try to add the new entry to the inline data.
1270  * If succeeds, return 0. If not, extended the inline dir and copied data to
1271  * the new created block.
1272  */
1273 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1274                               struct inode *dir, struct inode *inode)
1275 {
1276         int ret, ret2, inline_size, no_expand;
1277         void *inline_start;
1278         struct ext4_iloc iloc;
1279
1280         ret = ext4_get_inode_loc(dir, &iloc);
1281         if (ret)
1282                 return ret;
1283
1284         ext4_write_lock_xattr(dir, &no_expand);
1285         if (!ext4_has_inline_data(dir))
1286                 goto out;
1287
1288         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1289                                                  EXT4_INLINE_DOTDOT_SIZE;
1290         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1291
1292         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1293                                         inline_start, inline_size);
1294         if (ret != -ENOSPC)
1295                 goto out;
1296
1297         /* check whether it can be inserted to inline xattr space. */
1298         inline_size = EXT4_I(dir)->i_inline_size -
1299                         EXT4_MIN_INLINE_DATA_SIZE;
1300         if (!inline_size) {
1301                 /* Try to use the xattr space.*/
1302                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1303                 if (ret && ret != -ENOSPC)
1304                         goto out;
1305
1306                 inline_size = EXT4_I(dir)->i_inline_size -
1307                                 EXT4_MIN_INLINE_DATA_SIZE;
1308         }
1309
1310         if (inline_size) {
1311                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1312
1313                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1314                                                 inode, &iloc, inline_start,
1315                                                 inline_size);
1316
1317                 if (ret != -ENOSPC)
1318                         goto out;
1319         }
1320
1321         /*
1322          * The inline space is filled up, so create a new block for it.
1323          * As the extent tree will be created, we have to save the inline
1324          * dir first.
1325          */
1326         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1327
1328 out:
1329         ext4_write_unlock_xattr(dir, &no_expand);
1330         ret2 = ext4_mark_inode_dirty(handle, dir);
1331         if (unlikely(ret2 && !ret))
1332                 ret = ret2;
1333         brelse(iloc.bh);
1334         return ret;
1335 }
1336
1337 /*
1338  * This function fills a red-black tree with information from an
1339  * inlined dir.  It returns the number directory entries loaded
1340  * into the tree.  If there is an error it is returned in err.
1341  */
1342 int ext4_inlinedir_to_tree(struct file *dir_file,
1343                            struct inode *dir, ext4_lblk_t block,
1344                            struct dx_hash_info *hinfo,
1345                            __u32 start_hash, __u32 start_minor_hash,
1346                            int *has_inline_data)
1347 {
1348         int err = 0, count = 0;
1349         unsigned int parent_ino;
1350         int pos;
1351         struct ext4_dir_entry_2 *de;
1352         struct inode *inode = file_inode(dir_file);
1353         int ret, inline_size = 0;
1354         struct ext4_iloc iloc;
1355         void *dir_buf = NULL;
1356         struct ext4_dir_entry_2 fake;
1357         struct fscrypt_str tmp_str;
1358
1359         ret = ext4_get_inode_loc(inode, &iloc);
1360         if (ret)
1361                 return ret;
1362
1363         down_read(&EXT4_I(inode)->xattr_sem);
1364         if (!ext4_has_inline_data(inode)) {
1365                 up_read(&EXT4_I(inode)->xattr_sem);
1366                 *has_inline_data = 0;
1367                 goto out;
1368         }
1369
1370         inline_size = ext4_get_inline_size(inode);
1371         dir_buf = kmalloc(inline_size, GFP_NOFS);
1372         if (!dir_buf) {
1373                 ret = -ENOMEM;
1374                 up_read(&EXT4_I(inode)->xattr_sem);
1375                 goto out;
1376         }
1377
1378         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1379         up_read(&EXT4_I(inode)->xattr_sem);
1380         if (ret < 0)
1381                 goto out;
1382
1383         pos = 0;
1384         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1385         while (pos < inline_size) {
1386                 /*
1387                  * As inlined dir doesn't store any information about '.' and
1388                  * only the inode number of '..' is stored, we have to handle
1389                  * them differently.
1390                  */
1391                 if (pos == 0) {
1392                         fake.inode = cpu_to_le32(inode->i_ino);
1393                         fake.name_len = 1;
1394                         strcpy(fake.name, ".");
1395                         fake.rec_len = ext4_rec_len_to_disk(
1396                                           ext4_dir_rec_len(fake.name_len, NULL),
1397                                           inline_size);
1398                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1399                         de = &fake;
1400                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1401                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1402                         fake.inode = cpu_to_le32(parent_ino);
1403                         fake.name_len = 2;
1404                         strcpy(fake.name, "..");
1405                         fake.rec_len = ext4_rec_len_to_disk(
1406                                           ext4_dir_rec_len(fake.name_len, NULL),
1407                                           inline_size);
1408                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1409                         de = &fake;
1410                         pos = EXT4_INLINE_DOTDOT_SIZE;
1411                 } else {
1412                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1413                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1414                         if (ext4_check_dir_entry(inode, dir_file, de,
1415                                          iloc.bh, dir_buf,
1416                                          inline_size, pos)) {
1417                                 ret = count;
1418                                 goto out;
1419                         }
1420                 }
1421
1422                 if (ext4_hash_in_dirent(dir)) {
1423                         hinfo->hash = EXT4_DIRENT_HASH(de);
1424                         hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1425                 } else {
1426                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1427                 }
1428                 if ((hinfo->hash < start_hash) ||
1429                     ((hinfo->hash == start_hash) &&
1430                      (hinfo->minor_hash < start_minor_hash)))
1431                         continue;
1432                 if (de->inode == 0)
1433                         continue;
1434                 tmp_str.name = de->name;
1435                 tmp_str.len = de->name_len;
1436                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1437                                               hinfo->minor_hash, de, &tmp_str);
1438                 if (err) {
1439                         ret = err;
1440                         goto out;
1441                 }
1442                 count++;
1443         }
1444         ret = count;
1445 out:
1446         kfree(dir_buf);
1447         brelse(iloc.bh);
1448         return ret;
1449 }
1450
1451 /*
1452  * So this function is called when the volume is mkfsed with
1453  * dir_index disabled. In order to keep f_pos persistent
1454  * after we convert from an inlined dir to a blocked based,
1455  * we just pretend that we are a normal dir and return the
1456  * offset as if '.' and '..' really take place.
1457  *
1458  */
1459 int ext4_read_inline_dir(struct file *file,
1460                          struct dir_context *ctx,
1461                          int *has_inline_data)
1462 {
1463         unsigned int offset, parent_ino;
1464         int i;
1465         struct ext4_dir_entry_2 *de;
1466         struct super_block *sb;
1467         struct inode *inode = file_inode(file);
1468         int ret, inline_size = 0;
1469         struct ext4_iloc iloc;
1470         void *dir_buf = NULL;
1471         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1472
1473         ret = ext4_get_inode_loc(inode, &iloc);
1474         if (ret)
1475                 return ret;
1476
1477         down_read(&EXT4_I(inode)->xattr_sem);
1478         if (!ext4_has_inline_data(inode)) {
1479                 up_read(&EXT4_I(inode)->xattr_sem);
1480                 *has_inline_data = 0;
1481                 goto out;
1482         }
1483
1484         inline_size = ext4_get_inline_size(inode);
1485         dir_buf = kmalloc(inline_size, GFP_NOFS);
1486         if (!dir_buf) {
1487                 ret = -ENOMEM;
1488                 up_read(&EXT4_I(inode)->xattr_sem);
1489                 goto out;
1490         }
1491
1492         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1493         up_read(&EXT4_I(inode)->xattr_sem);
1494         if (ret < 0)
1495                 goto out;
1496
1497         ret = 0;
1498         sb = inode->i_sb;
1499         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1500         offset = ctx->pos;
1501
1502         /*
1503          * dotdot_offset and dotdot_size is the real offset and
1504          * size for ".." and "." if the dir is block based while
1505          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1506          * So we will use extra_offset and extra_size to indicate them
1507          * during the inline dir iteration.
1508          */
1509         dotdot_offset = ext4_dir_rec_len(1, NULL);
1510         dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1511         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1512         extra_size = extra_offset + inline_size;
1513
1514         /*
1515          * If the version has changed since the last call to
1516          * readdir(2), then we might be pointing to an invalid
1517          * dirent right now.  Scan from the start of the inline
1518          * dir to make sure.
1519          */
1520         if (!inode_eq_iversion(inode, file->f_version)) {
1521                 for (i = 0; i < extra_size && i < offset;) {
1522                         /*
1523                          * "." is with offset 0 and
1524                          * ".." is dotdot_offset.
1525                          */
1526                         if (!i) {
1527                                 i = dotdot_offset;
1528                                 continue;
1529                         } else if (i == dotdot_offset) {
1530                                 i = dotdot_size;
1531                                 continue;
1532                         }
1533                         /* for other entry, the real offset in
1534                          * the buf has to be tuned accordingly.
1535                          */
1536                         de = (struct ext4_dir_entry_2 *)
1537                                 (dir_buf + i - extra_offset);
1538                         /* It's too expensive to do a full
1539                          * dirent test each time round this
1540                          * loop, but we do have to test at
1541                          * least that it is non-zero.  A
1542                          * failure will be detected in the
1543                          * dirent test below. */
1544                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1545                                 < ext4_dir_rec_len(1, NULL))
1546                                 break;
1547                         i += ext4_rec_len_from_disk(de->rec_len,
1548                                                     extra_size);
1549                 }
1550                 offset = i;
1551                 ctx->pos = offset;
1552                 file->f_version = inode_query_iversion(inode);
1553         }
1554
1555         while (ctx->pos < extra_size) {
1556                 if (ctx->pos == 0) {
1557                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1558                                 goto out;
1559                         ctx->pos = dotdot_offset;
1560                         continue;
1561                 }
1562
1563                 if (ctx->pos == dotdot_offset) {
1564                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1565                                 goto out;
1566                         ctx->pos = dotdot_size;
1567                         continue;
1568                 }
1569
1570                 de = (struct ext4_dir_entry_2 *)
1571                         (dir_buf + ctx->pos - extra_offset);
1572                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1573                                          extra_size, ctx->pos))
1574                         goto out;
1575                 if (le32_to_cpu(de->inode)) {
1576                         if (!dir_emit(ctx, de->name, de->name_len,
1577                                       le32_to_cpu(de->inode),
1578                                       get_dtype(sb, de->file_type)))
1579                                 goto out;
1580                 }
1581                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1582         }
1583 out:
1584         kfree(dir_buf);
1585         brelse(iloc.bh);
1586         return ret;
1587 }
1588
1589 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1590                                         struct ext4_dir_entry_2 **parent_de,
1591                                         int *retval)
1592 {
1593         struct ext4_iloc iloc;
1594
1595         *retval = ext4_get_inode_loc(inode, &iloc);
1596         if (*retval)
1597                 return NULL;
1598
1599         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1600
1601         return iloc.bh;
1602 }
1603
1604 /*
1605  * Try to create the inline data for the new dir.
1606  * If it succeeds, return 0, otherwise return the error.
1607  * In case of ENOSPC, the caller should create the normal disk layout dir.
1608  */
1609 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1610                                struct inode *inode)
1611 {
1612         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1613         struct ext4_iloc iloc;
1614         struct ext4_dir_entry_2 *de;
1615
1616         ret = ext4_get_inode_loc(inode, &iloc);
1617         if (ret)
1618                 return ret;
1619
1620         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1621         if (ret)
1622                 goto out;
1623
1624         /*
1625          * For inline dir, we only save the inode information for the ".."
1626          * and create a fake dentry to cover the left space.
1627          */
1628         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1629         de->inode = cpu_to_le32(parent->i_ino);
1630         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1631         de->inode = 0;
1632         de->rec_len = ext4_rec_len_to_disk(
1633                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1634                                 inline_size);
1635         set_nlink(inode, 2);
1636         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1637 out:
1638         brelse(iloc.bh);
1639         return ret;
1640 }
1641
1642 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1643                                         struct ext4_filename *fname,
1644                                         struct ext4_dir_entry_2 **res_dir,
1645                                         int *has_inline_data)
1646 {
1647         int ret;
1648         struct ext4_iloc iloc;
1649         void *inline_start;
1650         int inline_size;
1651
1652         if (ext4_get_inode_loc(dir, &iloc))
1653                 return NULL;
1654
1655         down_read(&EXT4_I(dir)->xattr_sem);
1656         if (!ext4_has_inline_data(dir)) {
1657                 *has_inline_data = 0;
1658                 goto out;
1659         }
1660
1661         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1662                                                 EXT4_INLINE_DOTDOT_SIZE;
1663         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1664         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1665                               dir, fname, 0, res_dir);
1666         if (ret == 1)
1667                 goto out_find;
1668         if (ret < 0)
1669                 goto out;
1670
1671         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1672                 goto out;
1673
1674         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1675         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1676
1677         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1678                               dir, fname, 0, res_dir);
1679         if (ret == 1)
1680                 goto out_find;
1681
1682 out:
1683         brelse(iloc.bh);
1684         iloc.bh = NULL;
1685 out_find:
1686         up_read(&EXT4_I(dir)->xattr_sem);
1687         return iloc.bh;
1688 }
1689
1690 int ext4_delete_inline_entry(handle_t *handle,
1691                              struct inode *dir,
1692                              struct ext4_dir_entry_2 *de_del,
1693                              struct buffer_head *bh,
1694                              int *has_inline_data)
1695 {
1696         int err, inline_size, no_expand;
1697         struct ext4_iloc iloc;
1698         void *inline_start;
1699
1700         err = ext4_get_inode_loc(dir, &iloc);
1701         if (err)
1702                 return err;
1703
1704         ext4_write_lock_xattr(dir, &no_expand);
1705         if (!ext4_has_inline_data(dir)) {
1706                 *has_inline_data = 0;
1707                 goto out;
1708         }
1709
1710         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1711                 EXT4_MIN_INLINE_DATA_SIZE) {
1712                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1713                                         EXT4_INLINE_DOTDOT_SIZE;
1714                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1715                                 EXT4_INLINE_DOTDOT_SIZE;
1716         } else {
1717                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1718                 inline_size = ext4_get_inline_size(dir) -
1719                                 EXT4_MIN_INLINE_DATA_SIZE;
1720         }
1721
1722         BUFFER_TRACE(bh, "get_write_access");
1723         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1724                                             EXT4_JTR_NONE);
1725         if (err)
1726                 goto out;
1727
1728         err = ext4_generic_delete_entry(dir, de_del, bh,
1729                                         inline_start, inline_size, 0);
1730         if (err)
1731                 goto out;
1732
1733         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1734 out:
1735         ext4_write_unlock_xattr(dir, &no_expand);
1736         if (likely(err == 0))
1737                 err = ext4_mark_inode_dirty(handle, dir);
1738         brelse(iloc.bh);
1739         if (err != -ENOENT)
1740                 ext4_std_error(dir->i_sb, err);
1741         return err;
1742 }
1743
1744 /*
1745  * Get the inline dentry at offset.
1746  */
1747 static inline struct ext4_dir_entry_2 *
1748 ext4_get_inline_entry(struct inode *inode,
1749                       struct ext4_iloc *iloc,
1750                       unsigned int offset,
1751                       void **inline_start,
1752                       int *inline_size)
1753 {
1754         void *inline_pos;
1755
1756         BUG_ON(offset > ext4_get_inline_size(inode));
1757
1758         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1759                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1760                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1761         } else {
1762                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1763                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1764                 *inline_size = ext4_get_inline_size(inode) -
1765                                 EXT4_MIN_INLINE_DATA_SIZE;
1766         }
1767
1768         if (inline_start)
1769                 *inline_start = inline_pos;
1770         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1771 }
1772
1773 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1774 {
1775         int err, inline_size;
1776         struct ext4_iloc iloc;
1777         size_t inline_len;
1778         void *inline_pos;
1779         unsigned int offset;
1780         struct ext4_dir_entry_2 *de;
1781         bool ret = true;
1782
1783         err = ext4_get_inode_loc(dir, &iloc);
1784         if (err) {
1785                 EXT4_ERROR_INODE_ERR(dir, -err,
1786                                      "error %d getting inode %lu block",
1787                                      err, dir->i_ino);
1788                 return true;
1789         }
1790
1791         down_read(&EXT4_I(dir)->xattr_sem);
1792         if (!ext4_has_inline_data(dir)) {
1793                 *has_inline_data = 0;
1794                 goto out;
1795         }
1796
1797         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1798         if (!le32_to_cpu(de->inode)) {
1799                 ext4_warning(dir->i_sb,
1800                              "bad inline directory (dir #%lu) - no `..'",
1801                              dir->i_ino);
1802                 ret = true;
1803                 goto out;
1804         }
1805
1806         inline_len = ext4_get_inline_size(dir);
1807         offset = EXT4_INLINE_DOTDOT_SIZE;
1808         while (offset < inline_len) {
1809                 de = ext4_get_inline_entry(dir, &iloc, offset,
1810                                            &inline_pos, &inline_size);
1811                 if (ext4_check_dir_entry(dir, NULL, de,
1812                                          iloc.bh, inline_pos,
1813                                          inline_size, offset)) {
1814                         ext4_warning(dir->i_sb,
1815                                      "bad inline directory (dir #%lu) - "
1816                                      "inode %u, rec_len %u, name_len %d"
1817                                      "inline size %d",
1818                                      dir->i_ino, le32_to_cpu(de->inode),
1819                                      le16_to_cpu(de->rec_len), de->name_len,
1820                                      inline_size);
1821                         ret = true;
1822                         goto out;
1823                 }
1824                 if (le32_to_cpu(de->inode)) {
1825                         ret = false;
1826                         goto out;
1827                 }
1828                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1829         }
1830
1831 out:
1832         up_read(&EXT4_I(dir)->xattr_sem);
1833         brelse(iloc.bh);
1834         return ret;
1835 }
1836
1837 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1838 {
1839         int ret, no_expand;
1840
1841         ext4_write_lock_xattr(inode, &no_expand);
1842         ret = ext4_destroy_inline_data_nolock(handle, inode);
1843         ext4_write_unlock_xattr(inode, &no_expand);
1844
1845         return ret;
1846 }
1847
1848 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1849 {
1850         __u64 addr;
1851         int error = -EAGAIN;
1852         struct ext4_iloc iloc;
1853
1854         down_read(&EXT4_I(inode)->xattr_sem);
1855         if (!ext4_has_inline_data(inode))
1856                 goto out;
1857
1858         error = ext4_get_inode_loc(inode, &iloc);
1859         if (error)
1860                 goto out;
1861
1862         addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1863         addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1864         addr += offsetof(struct ext4_inode, i_block);
1865
1866         brelse(iloc.bh);
1867
1868         iomap->addr = addr;
1869         iomap->offset = 0;
1870         iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1871                               i_size_read(inode));
1872         iomap->type = IOMAP_INLINE;
1873         iomap->flags = 0;
1874
1875 out:
1876         up_read(&EXT4_I(inode)->xattr_sem);
1877         return error;
1878 }
1879
1880 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1881 {
1882         handle_t *handle;
1883         int inline_size, value_len, needed_blocks, no_expand, err = 0;
1884         size_t i_size;
1885         void *value = NULL;
1886         struct ext4_xattr_ibody_find is = {
1887                 .s = { .not_found = -ENODATA, },
1888         };
1889         struct ext4_xattr_info i = {
1890                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1891                 .name = EXT4_XATTR_SYSTEM_DATA,
1892         };
1893
1894
1895         needed_blocks = ext4_writepage_trans_blocks(inode);
1896         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1897         if (IS_ERR(handle))
1898                 return PTR_ERR(handle);
1899
1900         ext4_write_lock_xattr(inode, &no_expand);
1901         if (!ext4_has_inline_data(inode)) {
1902                 ext4_write_unlock_xattr(inode, &no_expand);
1903                 *has_inline = 0;
1904                 ext4_journal_stop(handle);
1905                 return 0;
1906         }
1907
1908         if ((err = ext4_orphan_add(handle, inode)) != 0)
1909                 goto out;
1910
1911         if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1912                 goto out;
1913
1914         down_write(&EXT4_I(inode)->i_data_sem);
1915         i_size = inode->i_size;
1916         inline_size = ext4_get_inline_size(inode);
1917         EXT4_I(inode)->i_disksize = i_size;
1918
1919         if (i_size < inline_size) {
1920                 /* Clear the content in the xattr space. */
1921                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1922                         if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1923                                 goto out_error;
1924
1925                         BUG_ON(is.s.not_found);
1926
1927                         value_len = le32_to_cpu(is.s.here->e_value_size);
1928                         value = kmalloc(value_len, GFP_NOFS);
1929                         if (!value) {
1930                                 err = -ENOMEM;
1931                                 goto out_error;
1932                         }
1933
1934                         err = ext4_xattr_ibody_get(inode, i.name_index,
1935                                                    i.name, value, value_len);
1936                         if (err <= 0)
1937                                 goto out_error;
1938
1939                         i.value = value;
1940                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1941                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1942                         err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1943                         if (err)
1944                                 goto out_error;
1945                 }
1946
1947                 /* Clear the content within i_blocks. */
1948                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1949                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1950                         memset(p + i_size, 0,
1951                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1952                 }
1953
1954                 EXT4_I(inode)->i_inline_size = i_size <
1955                                         EXT4_MIN_INLINE_DATA_SIZE ?
1956                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1957         }
1958
1959 out_error:
1960         up_write(&EXT4_I(inode)->i_data_sem);
1961 out:
1962         brelse(is.iloc.bh);
1963         ext4_write_unlock_xattr(inode, &no_expand);
1964         kfree(value);
1965         if (inode->i_nlink)
1966                 ext4_orphan_del(handle, inode);
1967
1968         if (err == 0) {
1969                 inode->i_mtime = inode->i_ctime = current_time(inode);
1970                 err = ext4_mark_inode_dirty(handle, inode);
1971                 if (IS_SYNC(inode))
1972                         ext4_handle_sync(handle);
1973         }
1974         ext4_journal_stop(handle);
1975         return err;
1976 }
1977
1978 int ext4_convert_inline_data(struct inode *inode)
1979 {
1980         int error, needed_blocks, no_expand;
1981         handle_t *handle;
1982         struct ext4_iloc iloc;
1983
1984         if (!ext4_has_inline_data(inode)) {
1985                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1986                 return 0;
1987         }
1988
1989         needed_blocks = ext4_writepage_trans_blocks(inode);
1990
1991         iloc.bh = NULL;
1992         error = ext4_get_inode_loc(inode, &iloc);
1993         if (error)
1994                 return error;
1995
1996         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1997         if (IS_ERR(handle)) {
1998                 error = PTR_ERR(handle);
1999                 goto out_free;
2000         }
2001
2002         ext4_write_lock_xattr(inode, &no_expand);
2003         if (ext4_has_inline_data(inode))
2004                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2005         ext4_write_unlock_xattr(inode, &no_expand);
2006         ext4_journal_stop(handle);
2007 out_free:
2008         brelse(iloc.bh);
2009         return error;
2010 }