Linux-libre 3.18.62-gnu
[librecmc/linux-libre.git] / fs / ext4 / inline.c
1 /*
2  * Copyright (c) 2012 Taobao.
3  * Written by Tao Ma <boyu.mt@taobao.com>
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of version 2.1 of the GNU Lesser General Public License
7  * as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  */
14 #include "ext4_jbd2.h"
15 #include "ext4.h"
16 #include "xattr.h"
17 #include "truncate.h"
18 #include <linux/fiemap.h>
19
20 #define EXT4_XATTR_SYSTEM_DATA  "data"
21 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
22 #define EXT4_INLINE_DOTDOT_OFFSET       2
23 #define EXT4_INLINE_DOTDOT_SIZE         4
24
25 static int ext4_get_inline_size(struct inode *inode)
26 {
27         if (EXT4_I(inode)->i_inline_off)
28                 return EXT4_I(inode)->i_inline_size;
29
30         return 0;
31 }
32
33 static int get_max_inline_xattr_value_size(struct inode *inode,
34                                            struct ext4_iloc *iloc)
35 {
36         struct ext4_xattr_ibody_header *header;
37         struct ext4_xattr_entry *entry;
38         struct ext4_inode *raw_inode;
39         int free, min_offs;
40
41         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42                         EXT4_GOOD_OLD_INODE_SIZE -
43                         EXT4_I(inode)->i_extra_isize -
44                         sizeof(struct ext4_xattr_ibody_header);
45
46         /*
47          * We need to subtract another sizeof(__u32) since an in-inode xattr
48          * needs an empty 4 bytes to indicate the gap between the xattr entry
49          * and the name/value pair.
50          */
51         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52                 return EXT4_XATTR_SIZE(min_offs -
53                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54                         EXT4_XATTR_ROUND - sizeof(__u32));
55
56         raw_inode = ext4_raw_inode(iloc);
57         header = IHDR(inode, raw_inode);
58         entry = IFIRST(header);
59
60         /* Compute min_offs. */
61         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62                 if (!entry->e_value_block && entry->e_value_size) {
63                         size_t offs = le16_to_cpu(entry->e_value_offs);
64                         if (offs < min_offs)
65                                 min_offs = offs;
66                 }
67         }
68         free = min_offs -
69                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71         if (EXT4_I(inode)->i_inline_off) {
72                 entry = (struct ext4_xattr_entry *)
73                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76                 goto out;
77         }
78
79         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81         if (free > EXT4_XATTR_ROUND)
82                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83         else
84                 free = 0;
85
86 out:
87         return free;
88 }
89
90 /*
91  * Get the maximum size we now can store in an inode.
92  * If we can't find the space for a xattr entry, don't use the space
93  * of the extents since we have no space to indicate the inline data.
94  */
95 int ext4_get_max_inline_size(struct inode *inode)
96 {
97         int error, max_inline_size;
98         struct ext4_iloc iloc;
99
100         if (EXT4_I(inode)->i_extra_isize == 0)
101                 return 0;
102
103         error = ext4_get_inode_loc(inode, &iloc);
104         if (error) {
105                 ext4_error_inode(inode, __func__, __LINE__, 0,
106                                  "can't get inode location %lu",
107                                  inode->i_ino);
108                 return 0;
109         }
110
111         down_read(&EXT4_I(inode)->xattr_sem);
112         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113         up_read(&EXT4_I(inode)->xattr_sem);
114
115         brelse(iloc.bh);
116
117         if (!max_inline_size)
118                 return 0;
119
120         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121 }
122
123 /*
124  * this function does not take xattr_sem, which is OK because it is
125  * currently only used in a code path coming form ext4_iget, before
126  * the new inode has been unlocked
127  */
128 int ext4_find_inline_data_nolock(struct inode *inode)
129 {
130         struct ext4_xattr_ibody_find is = {
131                 .s = { .not_found = -ENODATA, },
132         };
133         struct ext4_xattr_info i = {
134                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
135                 .name = EXT4_XATTR_SYSTEM_DATA,
136         };
137         int error;
138
139         if (EXT4_I(inode)->i_extra_isize == 0)
140                 return 0;
141
142         error = ext4_get_inode_loc(inode, &is.iloc);
143         if (error)
144                 return error;
145
146         error = ext4_xattr_ibody_find(inode, &i, &is);
147         if (error)
148                 goto out;
149
150         if (!is.s.not_found) {
151                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
152                                         (void *)ext4_raw_inode(&is.iloc));
153                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
154                                 le32_to_cpu(is.s.here->e_value_size);
155                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
156         }
157 out:
158         brelse(is.iloc.bh);
159         return error;
160 }
161
162 static int ext4_read_inline_data(struct inode *inode, void *buffer,
163                                  unsigned int len,
164                                  struct ext4_iloc *iloc)
165 {
166         struct ext4_xattr_entry *entry;
167         struct ext4_xattr_ibody_header *header;
168         int cp_len = 0;
169         struct ext4_inode *raw_inode;
170
171         if (!len)
172                 return 0;
173
174         BUG_ON(len > EXT4_I(inode)->i_inline_size);
175
176         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
177                         len : EXT4_MIN_INLINE_DATA_SIZE;
178
179         raw_inode = ext4_raw_inode(iloc);
180         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
181
182         len -= cp_len;
183         buffer += cp_len;
184
185         if (!len)
186                 goto out;
187
188         header = IHDR(inode, raw_inode);
189         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
190                                             EXT4_I(inode)->i_inline_off);
191         len = min_t(unsigned int, len,
192                     (unsigned int)le32_to_cpu(entry->e_value_size));
193
194         memcpy(buffer,
195                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
196         cp_len += len;
197
198 out:
199         return cp_len;
200 }
201
202 /*
203  * write the buffer to the inline inode.
204  * If 'create' is set, we don't need to do the extra copy in the xattr
205  * value since it is already handled by ext4_xattr_ibody_inline_set.
206  * That saves us one memcpy.
207  */
208 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
209                                    void *buffer, loff_t pos, unsigned int len)
210 {
211         struct ext4_xattr_entry *entry;
212         struct ext4_xattr_ibody_header *header;
213         struct ext4_inode *raw_inode;
214         int cp_len = 0;
215
216         BUG_ON(!EXT4_I(inode)->i_inline_off);
217         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
218
219         raw_inode = ext4_raw_inode(iloc);
220         buffer += pos;
221
222         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
223                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
224                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
225                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
226
227                 len -= cp_len;
228                 buffer += cp_len;
229                 pos += cp_len;
230         }
231
232         if (!len)
233                 return;
234
235         pos -= EXT4_MIN_INLINE_DATA_SIZE;
236         header = IHDR(inode, raw_inode);
237         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
238                                             EXT4_I(inode)->i_inline_off);
239
240         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
241                buffer, len);
242 }
243
244 static int ext4_create_inline_data(handle_t *handle,
245                                    struct inode *inode, unsigned len)
246 {
247         int error;
248         void *value = NULL;
249         struct ext4_xattr_ibody_find is = {
250                 .s = { .not_found = -ENODATA, },
251         };
252         struct ext4_xattr_info i = {
253                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
254                 .name = EXT4_XATTR_SYSTEM_DATA,
255         };
256
257         error = ext4_get_inode_loc(inode, &is.iloc);
258         if (error)
259                 return error;
260
261         BUFFER_TRACE(is.iloc.bh, "get_write_access");
262         error = ext4_journal_get_write_access(handle, is.iloc.bh);
263         if (error)
264                 goto out;
265
266         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
267                 value = EXT4_ZERO_XATTR_VALUE;
268                 len -= EXT4_MIN_INLINE_DATA_SIZE;
269         } else {
270                 value = "";
271                 len = 0;
272         }
273
274         /* Insert the the xttr entry. */
275         i.value = value;
276         i.value_len = len;
277
278         error = ext4_xattr_ibody_find(inode, &i, &is);
279         if (error)
280                 goto out;
281
282         BUG_ON(!is.s.not_found);
283
284         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
285         if (error) {
286                 if (error == -ENOSPC)
287                         ext4_clear_inode_state(inode,
288                                                EXT4_STATE_MAY_INLINE_DATA);
289                 goto out;
290         }
291
292         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
293                 0, EXT4_MIN_INLINE_DATA_SIZE);
294
295         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
296                                       (void *)ext4_raw_inode(&is.iloc));
297         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
298         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
299         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
300         get_bh(is.iloc.bh);
301         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
302
303 out:
304         brelse(is.iloc.bh);
305         return error;
306 }
307
308 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
309                                    unsigned int len)
310 {
311         int error;
312         void *value = NULL;
313         struct ext4_xattr_ibody_find is = {
314                 .s = { .not_found = -ENODATA, },
315         };
316         struct ext4_xattr_info i = {
317                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
318                 .name = EXT4_XATTR_SYSTEM_DATA,
319         };
320
321         /* If the old space is ok, write the data directly. */
322         if (len <= EXT4_I(inode)->i_inline_size)
323                 return 0;
324
325         error = ext4_get_inode_loc(inode, &is.iloc);
326         if (error)
327                 return error;
328
329         error = ext4_xattr_ibody_find(inode, &i, &is);
330         if (error)
331                 goto out;
332
333         BUG_ON(is.s.not_found);
334
335         len -= EXT4_MIN_INLINE_DATA_SIZE;
336         value = kzalloc(len, GFP_NOFS);
337         if (!value) {
338                 error = -ENOMEM;
339                 goto out;
340         }
341
342         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
343                                      value, len);
344         if (error == -ENODATA)
345                 goto out;
346
347         BUFFER_TRACE(is.iloc.bh, "get_write_access");
348         error = ext4_journal_get_write_access(handle, is.iloc.bh);
349         if (error)
350                 goto out;
351
352         /* Update the xttr entry. */
353         i.value = value;
354         i.value_len = len;
355
356         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
357         if (error)
358                 goto out;
359
360         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
361                                       (void *)ext4_raw_inode(&is.iloc));
362         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
363                                 le32_to_cpu(is.s.here->e_value_size);
364         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
365         get_bh(is.iloc.bh);
366         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
367
368 out:
369         kfree(value);
370         brelse(is.iloc.bh);
371         return error;
372 }
373
374 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
375                                     unsigned int len)
376 {
377         int ret, size;
378         struct ext4_inode_info *ei = EXT4_I(inode);
379
380         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
381                 return -ENOSPC;
382
383         size = ext4_get_max_inline_size(inode);
384         if (size < len)
385                 return -ENOSPC;
386
387         down_write(&EXT4_I(inode)->xattr_sem);
388
389         if (ei->i_inline_off)
390                 ret = ext4_update_inline_data(handle, inode, len);
391         else
392                 ret = ext4_create_inline_data(handle, inode, len);
393
394         up_write(&EXT4_I(inode)->xattr_sem);
395
396         return ret;
397 }
398
399 static int ext4_destroy_inline_data_nolock(handle_t *handle,
400                                            struct inode *inode)
401 {
402         struct ext4_inode_info *ei = EXT4_I(inode);
403         struct ext4_xattr_ibody_find is = {
404                 .s = { .not_found = 0, },
405         };
406         struct ext4_xattr_info i = {
407                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
408                 .name = EXT4_XATTR_SYSTEM_DATA,
409                 .value = NULL,
410                 .value_len = 0,
411         };
412         int error;
413
414         if (!ei->i_inline_off)
415                 return 0;
416
417         error = ext4_get_inode_loc(inode, &is.iloc);
418         if (error)
419                 return error;
420
421         error = ext4_xattr_ibody_find(inode, &i, &is);
422         if (error)
423                 goto out;
424
425         BUFFER_TRACE(is.iloc.bh, "get_write_access");
426         error = ext4_journal_get_write_access(handle, is.iloc.bh);
427         if (error)
428                 goto out;
429
430         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
431         if (error)
432                 goto out;
433
434         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
435                 0, EXT4_MIN_INLINE_DATA_SIZE);
436
437         if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
438                                       EXT4_FEATURE_INCOMPAT_EXTENTS)) {
439                 if (S_ISDIR(inode->i_mode) ||
440                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
441                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
442                         ext4_ext_tree_init(handle, inode);
443                 }
444         }
445         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
446
447         get_bh(is.iloc.bh);
448         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
449
450         EXT4_I(inode)->i_inline_off = 0;
451         EXT4_I(inode)->i_inline_size = 0;
452         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
453 out:
454         brelse(is.iloc.bh);
455         if (error == -ENODATA)
456                 error = 0;
457         return error;
458 }
459
460 static int ext4_read_inline_page(struct inode *inode, struct page *page)
461 {
462         void *kaddr;
463         int ret = 0;
464         size_t len;
465         struct ext4_iloc iloc;
466
467         BUG_ON(!PageLocked(page));
468         BUG_ON(!ext4_has_inline_data(inode));
469         BUG_ON(page->index);
470
471         if (!EXT4_I(inode)->i_inline_off) {
472                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
473                              inode->i_ino);
474                 goto out;
475         }
476
477         ret = ext4_get_inode_loc(inode, &iloc);
478         if (ret)
479                 goto out;
480
481         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
482         kaddr = kmap_atomic(page);
483         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
484         flush_dcache_page(page);
485         kunmap_atomic(kaddr);
486         zero_user_segment(page, len, PAGE_CACHE_SIZE);
487         SetPageUptodate(page);
488         brelse(iloc.bh);
489
490 out:
491         return ret;
492 }
493
494 int ext4_readpage_inline(struct inode *inode, struct page *page)
495 {
496         int ret = 0;
497
498         down_read(&EXT4_I(inode)->xattr_sem);
499         if (!ext4_has_inline_data(inode)) {
500                 up_read(&EXT4_I(inode)->xattr_sem);
501                 return -EAGAIN;
502         }
503
504         /*
505          * Current inline data can only exist in the 1st page,
506          * So for all the other pages, just set them uptodate.
507          */
508         if (!page->index)
509                 ret = ext4_read_inline_page(inode, page);
510         else if (!PageUptodate(page)) {
511                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
512                 SetPageUptodate(page);
513         }
514
515         up_read(&EXT4_I(inode)->xattr_sem);
516
517         unlock_page(page);
518         return ret >= 0 ? 0 : ret;
519 }
520
521 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
522                                               struct inode *inode,
523                                               unsigned flags)
524 {
525         int ret, needed_blocks;
526         handle_t *handle = NULL;
527         int retries = 0, sem_held = 0;
528         struct page *page = NULL;
529         unsigned from, to;
530         struct ext4_iloc iloc;
531
532         if (!ext4_has_inline_data(inode)) {
533                 /*
534                  * clear the flag so that no new write
535                  * will trap here again.
536                  */
537                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
538                 return 0;
539         }
540
541         needed_blocks = ext4_writepage_trans_blocks(inode);
542
543         ret = ext4_get_inode_loc(inode, &iloc);
544         if (ret)
545                 return ret;
546
547 retry:
548         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
549         if (IS_ERR(handle)) {
550                 ret = PTR_ERR(handle);
551                 handle = NULL;
552                 goto out;
553         }
554
555         /* We cannot recurse into the filesystem as the transaction is already
556          * started */
557         flags |= AOP_FLAG_NOFS;
558
559         page = grab_cache_page_write_begin(mapping, 0, flags);
560         if (!page) {
561                 ret = -ENOMEM;
562                 goto out;
563         }
564
565         down_write(&EXT4_I(inode)->xattr_sem);
566         sem_held = 1;
567         /* If some one has already done this for us, just exit. */
568         if (!ext4_has_inline_data(inode)) {
569                 ret = 0;
570                 goto out;
571         }
572
573         from = 0;
574         to = ext4_get_inline_size(inode);
575         if (!PageUptodate(page)) {
576                 ret = ext4_read_inline_page(inode, page);
577                 if (ret < 0)
578                         goto out;
579         }
580
581         ret = ext4_destroy_inline_data_nolock(handle, inode);
582         if (ret)
583                 goto out;
584
585         if (ext4_should_dioread_nolock(inode))
586                 ret = __block_write_begin(page, from, to, ext4_get_block_write);
587         else
588                 ret = __block_write_begin(page, from, to, ext4_get_block);
589
590         if (!ret && ext4_should_journal_data(inode)) {
591                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
592                                              from, to, NULL,
593                                              do_journal_get_write_access);
594         }
595
596         if (ret) {
597                 unlock_page(page);
598                 page_cache_release(page);
599                 page = NULL;
600                 ext4_orphan_add(handle, inode);
601                 up_write(&EXT4_I(inode)->xattr_sem);
602                 sem_held = 0;
603                 ext4_journal_stop(handle);
604                 handle = NULL;
605                 ext4_truncate_failed_write(inode);
606                 /*
607                  * If truncate failed early the inode might
608                  * still be on the orphan list; we need to
609                  * make sure the inode is removed from the
610                  * orphan list in that case.
611                  */
612                 if (inode->i_nlink)
613                         ext4_orphan_del(NULL, inode);
614         }
615
616         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617                 goto retry;
618
619         if (page)
620                 block_commit_write(page, from, to);
621 out:
622         if (page) {
623                 unlock_page(page);
624                 page_cache_release(page);
625         }
626         if (sem_held)
627                 up_write(&EXT4_I(inode)->xattr_sem);
628         if (handle)
629                 ext4_journal_stop(handle);
630         brelse(iloc.bh);
631         return ret;
632 }
633
634 /*
635  * Try to write data in the inode.
636  * If the inode has inline data, check whether the new write can be
637  * in the inode also. If not, create the page the handle, move the data
638  * to the page make it update and let the later codes create extent for it.
639  */
640 int ext4_try_to_write_inline_data(struct address_space *mapping,
641                                   struct inode *inode,
642                                   loff_t pos, unsigned len,
643                                   unsigned flags,
644                                   struct page **pagep)
645 {
646         int ret;
647         handle_t *handle;
648         struct page *page;
649         struct ext4_iloc iloc;
650
651         if (pos + len > ext4_get_max_inline_size(inode))
652                 goto convert;
653
654         ret = ext4_get_inode_loc(inode, &iloc);
655         if (ret)
656                 return ret;
657
658         /*
659          * The possible write could happen in the inode,
660          * so try to reserve the space in inode first.
661          */
662         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
663         if (IS_ERR(handle)) {
664                 ret = PTR_ERR(handle);
665                 handle = NULL;
666                 goto out;
667         }
668
669         ret = ext4_prepare_inline_data(handle, inode, pos + len);
670         if (ret && ret != -ENOSPC)
671                 goto out;
672
673         /* We don't have space in inline inode, so convert it to extent. */
674         if (ret == -ENOSPC) {
675                 ext4_journal_stop(handle);
676                 brelse(iloc.bh);
677                 goto convert;
678         }
679
680         flags |= AOP_FLAG_NOFS;
681
682         page = grab_cache_page_write_begin(mapping, 0, flags);
683         if (!page) {
684                 ret = -ENOMEM;
685                 goto out;
686         }
687
688         *pagep = page;
689         down_read(&EXT4_I(inode)->xattr_sem);
690         if (!ext4_has_inline_data(inode)) {
691                 ret = 0;
692                 unlock_page(page);
693                 page_cache_release(page);
694                 goto out_up_read;
695         }
696
697         if (!PageUptodate(page)) {
698                 ret = ext4_read_inline_page(inode, page);
699                 if (ret < 0)
700                         goto out_up_read;
701         }
702
703         ret = 1;
704         handle = NULL;
705 out_up_read:
706         up_read(&EXT4_I(inode)->xattr_sem);
707 out:
708         if (handle)
709                 ext4_journal_stop(handle);
710         brelse(iloc.bh);
711         return ret;
712 convert:
713         return ext4_convert_inline_data_to_extent(mapping,
714                                                   inode, flags);
715 }
716
717 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
718                                unsigned copied, struct page *page)
719 {
720         int ret;
721         void *kaddr;
722         struct ext4_iloc iloc;
723
724         if (unlikely(copied < len)) {
725                 if (!PageUptodate(page)) {
726                         copied = 0;
727                         goto out;
728                 }
729         }
730
731         ret = ext4_get_inode_loc(inode, &iloc);
732         if (ret) {
733                 ext4_std_error(inode->i_sb, ret);
734                 copied = 0;
735                 goto out;
736         }
737
738         down_write(&EXT4_I(inode)->xattr_sem);
739         BUG_ON(!ext4_has_inline_data(inode));
740
741         kaddr = kmap_atomic(page);
742         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
743         kunmap_atomic(kaddr);
744         SetPageUptodate(page);
745         /* clear page dirty so that writepages wouldn't work for us. */
746         ClearPageDirty(page);
747
748         up_write(&EXT4_I(inode)->xattr_sem);
749         brelse(iloc.bh);
750 out:
751         return copied;
752 }
753
754 struct buffer_head *
755 ext4_journalled_write_inline_data(struct inode *inode,
756                                   unsigned len,
757                                   struct page *page)
758 {
759         int ret;
760         void *kaddr;
761         struct ext4_iloc iloc;
762
763         ret = ext4_get_inode_loc(inode, &iloc);
764         if (ret) {
765                 ext4_std_error(inode->i_sb, ret);
766                 return NULL;
767         }
768
769         down_write(&EXT4_I(inode)->xattr_sem);
770         kaddr = kmap_atomic(page);
771         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
772         kunmap_atomic(kaddr);
773         up_write(&EXT4_I(inode)->xattr_sem);
774
775         return iloc.bh;
776 }
777
778 /*
779  * Try to make the page cache and handle ready for the inline data case.
780  * We can call this function in 2 cases:
781  * 1. The inode is created and the first write exceeds inline size. We can
782  *    clear the inode state safely.
783  * 2. The inode has inline data, then we need to read the data, make it
784  *    update and dirty so that ext4_da_writepages can handle it. We don't
785  *    need to start the journal since the file's metatdata isn't changed now.
786  */
787 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
788                                                  struct inode *inode,
789                                                  unsigned flags,
790                                                  void **fsdata)
791 {
792         int ret = 0, inline_size;
793         struct page *page;
794
795         page = grab_cache_page_write_begin(mapping, 0, flags);
796         if (!page)
797                 return -ENOMEM;
798
799         down_read(&EXT4_I(inode)->xattr_sem);
800         if (!ext4_has_inline_data(inode)) {
801                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
802                 goto out;
803         }
804
805         inline_size = ext4_get_inline_size(inode);
806
807         if (!PageUptodate(page)) {
808                 ret = ext4_read_inline_page(inode, page);
809                 if (ret < 0)
810                         goto out;
811         }
812
813         ret = __block_write_begin(page, 0, inline_size,
814                                   ext4_da_get_block_prep);
815         if (ret) {
816                 ext4_truncate_failed_write(inode);
817                 goto out;
818         }
819
820         SetPageDirty(page);
821         SetPageUptodate(page);
822         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
823         *fsdata = (void *)CONVERT_INLINE_DATA;
824
825 out:
826         up_read(&EXT4_I(inode)->xattr_sem);
827         if (page) {
828                 unlock_page(page);
829                 page_cache_release(page);
830         }
831         return ret;
832 }
833
834 /*
835  * Prepare the write for the inline data.
836  * If the the data can be written into the inode, we just read
837  * the page and make it uptodate, and start the journal.
838  * Otherwise read the page, makes it dirty so that it can be
839  * handle in writepages(the i_disksize update is left to the
840  * normal ext4_da_write_end).
841  */
842 int ext4_da_write_inline_data_begin(struct address_space *mapping,
843                                     struct inode *inode,
844                                     loff_t pos, unsigned len,
845                                     unsigned flags,
846                                     struct page **pagep,
847                                     void **fsdata)
848 {
849         int ret, inline_size;
850         handle_t *handle;
851         struct page *page;
852         struct ext4_iloc iloc;
853         int retries;
854
855         ret = ext4_get_inode_loc(inode, &iloc);
856         if (ret)
857                 return ret;
858
859 retry_journal:
860         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
861         if (IS_ERR(handle)) {
862                 ret = PTR_ERR(handle);
863                 goto out;
864         }
865
866         inline_size = ext4_get_max_inline_size(inode);
867
868         ret = -ENOSPC;
869         if (inline_size >= pos + len) {
870                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
871                 if (ret && ret != -ENOSPC)
872                         goto out_journal;
873         }
874
875         if (ret == -ENOSPC) {
876                 ret = ext4_da_convert_inline_data_to_extent(mapping,
877                                                             inode,
878                                                             flags,
879                                                             fsdata);
880                 ext4_journal_stop(handle);
881                 if (ret == -ENOSPC &&
882                     ext4_should_retry_alloc(inode->i_sb, &retries))
883                         goto retry_journal;
884                 goto out;
885         }
886
887         /*
888          * We cannot recurse into the filesystem as the transaction
889          * is already started.
890          */
891         flags |= AOP_FLAG_NOFS;
892
893         page = grab_cache_page_write_begin(mapping, 0, flags);
894         if (!page) {
895                 ret = -ENOMEM;
896                 goto out_journal;
897         }
898
899         down_read(&EXT4_I(inode)->xattr_sem);
900         if (!ext4_has_inline_data(inode)) {
901                 ret = 0;
902                 goto out_release_page;
903         }
904
905         if (!PageUptodate(page)) {
906                 ret = ext4_read_inline_page(inode, page);
907                 if (ret < 0)
908                         goto out_release_page;
909         }
910
911         up_read(&EXT4_I(inode)->xattr_sem);
912         *pagep = page;
913         brelse(iloc.bh);
914         return 1;
915 out_release_page:
916         up_read(&EXT4_I(inode)->xattr_sem);
917         unlock_page(page);
918         page_cache_release(page);
919 out_journal:
920         ext4_journal_stop(handle);
921 out:
922         brelse(iloc.bh);
923         return ret;
924 }
925
926 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
927                                   unsigned len, unsigned copied,
928                                   struct page *page)
929 {
930         int i_size_changed = 0;
931
932         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
933
934         /*
935          * No need to use i_size_read() here, the i_size
936          * cannot change under us because we hold i_mutex.
937          *
938          * But it's important to update i_size while still holding page lock:
939          * page writeout could otherwise come in and zero beyond i_size.
940          */
941         if (pos+copied > inode->i_size) {
942                 i_size_write(inode, pos+copied);
943                 i_size_changed = 1;
944         }
945         unlock_page(page);
946         page_cache_release(page);
947
948         /*
949          * Don't mark the inode dirty under page lock. First, it unnecessarily
950          * makes the holding time of page lock longer. Second, it forces lock
951          * ordering of page lock and transaction start for journaling
952          * filesystems.
953          */
954         if (i_size_changed)
955                 mark_inode_dirty(inode);
956
957         return copied;
958 }
959
960 #ifdef INLINE_DIR_DEBUG
961 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
962                           void *inline_start, int inline_size)
963 {
964         int offset;
965         unsigned short de_len;
966         struct ext4_dir_entry_2 *de = inline_start;
967         void *dlimit = inline_start + inline_size;
968
969         trace_printk("inode %lu\n", dir->i_ino);
970         offset = 0;
971         while ((void *)de < dlimit) {
972                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
973                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
974                              offset, de_len, de->name_len, de->name,
975                              de->name_len, le32_to_cpu(de->inode));
976                 if (ext4_check_dir_entry(dir, NULL, de, bh,
977                                          inline_start, inline_size, offset))
978                         BUG();
979
980                 offset += de_len;
981                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
982         }
983 }
984 #else
985 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
986 #endif
987
988 /*
989  * Add a new entry into a inline dir.
990  * It will return -ENOSPC if no space is available, and -EIO
991  * and -EEXIST if directory entry already exists.
992  */
993 static int ext4_add_dirent_to_inline(handle_t *handle,
994                                      struct dentry *dentry,
995                                      struct inode *inode,
996                                      struct ext4_iloc *iloc,
997                                      void *inline_start, int inline_size)
998 {
999         struct inode    *dir = dentry->d_parent->d_inode;
1000         const char      *name = dentry->d_name.name;
1001         int             namelen = dentry->d_name.len;
1002         int             err;
1003         struct ext4_dir_entry_2 *de;
1004
1005         err = ext4_find_dest_de(dir, inode, iloc->bh,
1006                                 inline_start, inline_size,
1007                                 name, namelen, &de);
1008         if (err)
1009                 return err;
1010
1011         BUFFER_TRACE(iloc->bh, "get_write_access");
1012         err = ext4_journal_get_write_access(handle, iloc->bh);
1013         if (err)
1014                 return err;
1015         ext4_insert_dentry(inode, de, inline_size, name, namelen);
1016
1017         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1018
1019         /*
1020          * XXX shouldn't update any times until successful
1021          * completion of syscall, but too many callers depend
1022          * on this.
1023          *
1024          * XXX similarly, too many callers depend on
1025          * ext4_new_inode() setting the times, but error
1026          * recovery deletes the inode, so the worst that can
1027          * happen is that the times are slightly out of date
1028          * and/or different from the directory change time.
1029          */
1030         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1031         ext4_update_dx_flag(dir);
1032         dir->i_version++;
1033         ext4_mark_inode_dirty(handle, dir);
1034         return 1;
1035 }
1036
1037 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1038                                        struct ext4_iloc *iloc)
1039 {
1040         struct ext4_xattr_entry *entry;
1041         struct ext4_xattr_ibody_header *header;
1042
1043         BUG_ON(!EXT4_I(inode)->i_inline_off);
1044
1045         header = IHDR(inode, ext4_raw_inode(iloc));
1046         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1047                                             EXT4_I(inode)->i_inline_off);
1048
1049         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1050 }
1051
1052 /* Set the final de to cover the whole block. */
1053 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1054 {
1055         struct ext4_dir_entry_2 *de, *prev_de;
1056         void *limit;
1057         int de_len;
1058
1059         de = (struct ext4_dir_entry_2 *)de_buf;
1060         if (old_size) {
1061                 limit = de_buf + old_size;
1062                 do {
1063                         prev_de = de;
1064                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1065                         de_buf += de_len;
1066                         de = (struct ext4_dir_entry_2 *)de_buf;
1067                 } while (de_buf < limit);
1068
1069                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1070                                                         old_size, new_size);
1071         } else {
1072                 /* this is just created, so create an empty entry. */
1073                 de->inode = 0;
1074                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1075         }
1076 }
1077
1078 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1079                                   struct ext4_iloc *iloc)
1080 {
1081         int ret;
1082         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1083         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1084
1085         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1086                 return -ENOSPC;
1087
1088         ret = ext4_update_inline_data(handle, dir,
1089                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1090         if (ret)
1091                 return ret;
1092
1093         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1094                              EXT4_I(dir)->i_inline_size -
1095                                                 EXT4_MIN_INLINE_DATA_SIZE);
1096         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1097         return 0;
1098 }
1099
1100 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1101                                      struct ext4_iloc *iloc,
1102                                      void *buf, int inline_size)
1103 {
1104         ext4_create_inline_data(handle, inode, inline_size);
1105         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1106         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1107 }
1108
1109 static int ext4_finish_convert_inline_dir(handle_t *handle,
1110                                           struct inode *inode,
1111                                           struct buffer_head *dir_block,
1112                                           void *buf,
1113                                           int inline_size)
1114 {
1115         int err, csum_size = 0, header_size = 0;
1116         struct ext4_dir_entry_2 *de;
1117         struct ext4_dir_entry_tail *t;
1118         void *target = dir_block->b_data;
1119
1120         /*
1121          * First create "." and ".." and then copy the dir information
1122          * back to the block.
1123          */
1124         de = (struct ext4_dir_entry_2 *)target;
1125         de = ext4_init_dot_dotdot(inode, de,
1126                 inode->i_sb->s_blocksize, csum_size,
1127                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1128         header_size = (void *)de - target;
1129
1130         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1131                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1132
1133         if (ext4_has_metadata_csum(inode->i_sb))
1134                 csum_size = sizeof(struct ext4_dir_entry_tail);
1135
1136         inode->i_size = inode->i_sb->s_blocksize;
1137         i_size_write(inode, inode->i_sb->s_blocksize);
1138         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1139         ext4_update_final_de(dir_block->b_data,
1140                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1141                         inode->i_sb->s_blocksize - csum_size);
1142
1143         if (csum_size) {
1144                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1145                                      inode->i_sb->s_blocksize);
1146                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1147         }
1148         set_buffer_uptodate(dir_block);
1149         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1150         if (err)
1151                 return err;
1152         set_buffer_verified(dir_block);
1153         return ext4_mark_inode_dirty(handle, inode);
1154 }
1155
1156 static int ext4_convert_inline_data_nolock(handle_t *handle,
1157                                            struct inode *inode,
1158                                            struct ext4_iloc *iloc)
1159 {
1160         int error;
1161         void *buf = NULL;
1162         struct buffer_head *data_bh = NULL;
1163         struct ext4_map_blocks map;
1164         int inline_size;
1165
1166         inline_size = ext4_get_inline_size(inode);
1167         buf = kmalloc(inline_size, GFP_NOFS);
1168         if (!buf) {
1169                 error = -ENOMEM;
1170                 goto out;
1171         }
1172
1173         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1174         if (error < 0)
1175                 goto out;
1176
1177         /*
1178          * Make sure the inline directory entries pass checks before we try to
1179          * convert them, so that we avoid touching stuff that needs fsck.
1180          */
1181         if (S_ISDIR(inode->i_mode)) {
1182                 error = ext4_check_all_de(inode, iloc->bh,
1183                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1184                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1185                 if (error)
1186                         goto out;
1187         }
1188
1189         error = ext4_destroy_inline_data_nolock(handle, inode);
1190         if (error)
1191                 goto out;
1192
1193         map.m_lblk = 0;
1194         map.m_len = 1;
1195         map.m_flags = 0;
1196         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1197         if (error < 0)
1198                 goto out_restore;
1199         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1200                 error = -EIO;
1201                 goto out_restore;
1202         }
1203
1204         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1205         if (!data_bh) {
1206                 error = -ENOMEM;
1207                 goto out_restore;
1208         }
1209
1210         lock_buffer(data_bh);
1211         error = ext4_journal_get_create_access(handle, data_bh);
1212         if (error) {
1213                 unlock_buffer(data_bh);
1214                 error = -EIO;
1215                 goto out_restore;
1216         }
1217         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1218
1219         if (!S_ISDIR(inode->i_mode)) {
1220                 memcpy(data_bh->b_data, buf, inline_size);
1221                 set_buffer_uptodate(data_bh);
1222                 error = ext4_handle_dirty_metadata(handle,
1223                                                    inode, data_bh);
1224         } else {
1225                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1226                                                        buf, inline_size);
1227         }
1228
1229         unlock_buffer(data_bh);
1230 out_restore:
1231         if (error)
1232                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1233
1234 out:
1235         brelse(data_bh);
1236         kfree(buf);
1237         return error;
1238 }
1239
1240 /*
1241  * Try to add the new entry to the inline data.
1242  * If succeeds, return 0. If not, extended the inline dir and copied data to
1243  * the new created block.
1244  */
1245 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1246                               struct inode *inode)
1247 {
1248         int ret, inline_size;
1249         void *inline_start;
1250         struct ext4_iloc iloc;
1251         struct inode *dir = dentry->d_parent->d_inode;
1252
1253         ret = ext4_get_inode_loc(dir, &iloc);
1254         if (ret)
1255                 return ret;
1256
1257         down_write(&EXT4_I(dir)->xattr_sem);
1258         if (!ext4_has_inline_data(dir))
1259                 goto out;
1260
1261         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1262                                                  EXT4_INLINE_DOTDOT_SIZE;
1263         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1264
1265         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1266                                         inline_start, inline_size);
1267         if (ret != -ENOSPC)
1268                 goto out;
1269
1270         /* check whether it can be inserted to inline xattr space. */
1271         inline_size = EXT4_I(dir)->i_inline_size -
1272                         EXT4_MIN_INLINE_DATA_SIZE;
1273         if (!inline_size) {
1274                 /* Try to use the xattr space.*/
1275                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1276                 if (ret && ret != -ENOSPC)
1277                         goto out;
1278
1279                 inline_size = EXT4_I(dir)->i_inline_size -
1280                                 EXT4_MIN_INLINE_DATA_SIZE;
1281         }
1282
1283         if (inline_size) {
1284                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1285
1286                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1287                                                 inline_start, inline_size);
1288
1289                 if (ret != -ENOSPC)
1290                         goto out;
1291         }
1292
1293         /*
1294          * The inline space is filled up, so create a new block for it.
1295          * As the extent tree will be created, we have to save the inline
1296          * dir first.
1297          */
1298         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1299
1300 out:
1301         ext4_mark_inode_dirty(handle, dir);
1302         up_write(&EXT4_I(dir)->xattr_sem);
1303         brelse(iloc.bh);
1304         return ret;
1305 }
1306
1307 /*
1308  * This function fills a red-black tree with information from an
1309  * inlined dir.  It returns the number directory entries loaded
1310  * into the tree.  If there is an error it is returned in err.
1311  */
1312 int htree_inlinedir_to_tree(struct file *dir_file,
1313                             struct inode *dir, ext4_lblk_t block,
1314                             struct dx_hash_info *hinfo,
1315                             __u32 start_hash, __u32 start_minor_hash,
1316                             int *has_inline_data)
1317 {
1318         int err = 0, count = 0;
1319         unsigned int parent_ino;
1320         int pos;
1321         struct ext4_dir_entry_2 *de;
1322         struct inode *inode = file_inode(dir_file);
1323         int ret, inline_size = 0;
1324         struct ext4_iloc iloc;
1325         void *dir_buf = NULL;
1326         struct ext4_dir_entry_2 fake;
1327
1328         ret = ext4_get_inode_loc(inode, &iloc);
1329         if (ret)
1330                 return ret;
1331
1332         down_read(&EXT4_I(inode)->xattr_sem);
1333         if (!ext4_has_inline_data(inode)) {
1334                 up_read(&EXT4_I(inode)->xattr_sem);
1335                 *has_inline_data = 0;
1336                 goto out;
1337         }
1338
1339         inline_size = ext4_get_inline_size(inode);
1340         dir_buf = kmalloc(inline_size, GFP_NOFS);
1341         if (!dir_buf) {
1342                 ret = -ENOMEM;
1343                 up_read(&EXT4_I(inode)->xattr_sem);
1344                 goto out;
1345         }
1346
1347         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1348         up_read(&EXT4_I(inode)->xattr_sem);
1349         if (ret < 0)
1350                 goto out;
1351
1352         pos = 0;
1353         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1354         while (pos < inline_size) {
1355                 /*
1356                  * As inlined dir doesn't store any information about '.' and
1357                  * only the inode number of '..' is stored, we have to handle
1358                  * them differently.
1359                  */
1360                 if (pos == 0) {
1361                         fake.inode = cpu_to_le32(inode->i_ino);
1362                         fake.name_len = 1;
1363                         strcpy(fake.name, ".");
1364                         fake.rec_len = ext4_rec_len_to_disk(
1365                                                 EXT4_DIR_REC_LEN(fake.name_len),
1366                                                 inline_size);
1367                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1368                         de = &fake;
1369                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1370                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1371                         fake.inode = cpu_to_le32(parent_ino);
1372                         fake.name_len = 2;
1373                         strcpy(fake.name, "..");
1374                         fake.rec_len = ext4_rec_len_to_disk(
1375                                                 EXT4_DIR_REC_LEN(fake.name_len),
1376                                                 inline_size);
1377                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1378                         de = &fake;
1379                         pos = EXT4_INLINE_DOTDOT_SIZE;
1380                 } else {
1381                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1382                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1383                         if (ext4_check_dir_entry(inode, dir_file, de,
1384                                          iloc.bh, dir_buf,
1385                                          inline_size, pos)) {
1386                                 ret = count;
1387                                 goto out;
1388                         }
1389                 }
1390
1391                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1392                 if ((hinfo->hash < start_hash) ||
1393                     ((hinfo->hash == start_hash) &&
1394                      (hinfo->minor_hash < start_minor_hash)))
1395                         continue;
1396                 if (de->inode == 0)
1397                         continue;
1398                 err = ext4_htree_store_dirent(dir_file,
1399                                    hinfo->hash, hinfo->minor_hash, de);
1400                 if (err) {
1401                         count = err;
1402                         goto out;
1403                 }
1404                 count++;
1405         }
1406         ret = count;
1407 out:
1408         kfree(dir_buf);
1409         brelse(iloc.bh);
1410         return ret;
1411 }
1412
1413 /*
1414  * So this function is called when the volume is mkfsed with
1415  * dir_index disabled. In order to keep f_pos persistent
1416  * after we convert from an inlined dir to a blocked based,
1417  * we just pretend that we are a normal dir and return the
1418  * offset as if '.' and '..' really take place.
1419  *
1420  */
1421 int ext4_read_inline_dir(struct file *file,
1422                          struct dir_context *ctx,
1423                          int *has_inline_data)
1424 {
1425         unsigned int offset, parent_ino;
1426         int i;
1427         struct ext4_dir_entry_2 *de;
1428         struct super_block *sb;
1429         struct inode *inode = file_inode(file);
1430         int ret, inline_size = 0;
1431         struct ext4_iloc iloc;
1432         void *dir_buf = NULL;
1433         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1434
1435         ret = ext4_get_inode_loc(inode, &iloc);
1436         if (ret)
1437                 return ret;
1438
1439         down_read(&EXT4_I(inode)->xattr_sem);
1440         if (!ext4_has_inline_data(inode)) {
1441                 up_read(&EXT4_I(inode)->xattr_sem);
1442                 *has_inline_data = 0;
1443                 goto out;
1444         }
1445
1446         inline_size = ext4_get_inline_size(inode);
1447         dir_buf = kmalloc(inline_size, GFP_NOFS);
1448         if (!dir_buf) {
1449                 ret = -ENOMEM;
1450                 up_read(&EXT4_I(inode)->xattr_sem);
1451                 goto out;
1452         }
1453
1454         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1455         up_read(&EXT4_I(inode)->xattr_sem);
1456         if (ret < 0)
1457                 goto out;
1458
1459         ret = 0;
1460         sb = inode->i_sb;
1461         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1462         offset = ctx->pos;
1463
1464         /*
1465          * dotdot_offset and dotdot_size is the real offset and
1466          * size for ".." and "." if the dir is block based while
1467          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1468          * So we will use extra_offset and extra_size to indicate them
1469          * during the inline dir iteration.
1470          */
1471         dotdot_offset = EXT4_DIR_REC_LEN(1);
1472         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1473         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1474         extra_size = extra_offset + inline_size;
1475
1476         /*
1477          * If the version has changed since the last call to
1478          * readdir(2), then we might be pointing to an invalid
1479          * dirent right now.  Scan from the start of the inline
1480          * dir to make sure.
1481          */
1482         if (file->f_version != inode->i_version) {
1483                 for (i = 0; i < extra_size && i < offset;) {
1484                         /*
1485                          * "." is with offset 0 and
1486                          * ".." is dotdot_offset.
1487                          */
1488                         if (!i) {
1489                                 i = dotdot_offset;
1490                                 continue;
1491                         } else if (i == dotdot_offset) {
1492                                 i = dotdot_size;
1493                                 continue;
1494                         }
1495                         /* for other entry, the real offset in
1496                          * the buf has to be tuned accordingly.
1497                          */
1498                         de = (struct ext4_dir_entry_2 *)
1499                                 (dir_buf + i - extra_offset);
1500                         /* It's too expensive to do a full
1501                          * dirent test each time round this
1502                          * loop, but we do have to test at
1503                          * least that it is non-zero.  A
1504                          * failure will be detected in the
1505                          * dirent test below. */
1506                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1507                                 < EXT4_DIR_REC_LEN(1))
1508                                 break;
1509                         i += ext4_rec_len_from_disk(de->rec_len,
1510                                                     extra_size);
1511                 }
1512                 offset = i;
1513                 ctx->pos = offset;
1514                 file->f_version = inode->i_version;
1515         }
1516
1517         while (ctx->pos < extra_size) {
1518                 if (ctx->pos == 0) {
1519                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1520                                 goto out;
1521                         ctx->pos = dotdot_offset;
1522                         continue;
1523                 }
1524
1525                 if (ctx->pos == dotdot_offset) {
1526                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1527                                 goto out;
1528                         ctx->pos = dotdot_size;
1529                         continue;
1530                 }
1531
1532                 de = (struct ext4_dir_entry_2 *)
1533                         (dir_buf + ctx->pos - extra_offset);
1534                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1535                                          extra_size, ctx->pos))
1536                         goto out;
1537                 if (le32_to_cpu(de->inode)) {
1538                         if (!dir_emit(ctx, de->name, de->name_len,
1539                                       le32_to_cpu(de->inode),
1540                                       get_dtype(sb, de->file_type)))
1541                                 goto out;
1542                 }
1543                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1544         }
1545 out:
1546         kfree(dir_buf);
1547         brelse(iloc.bh);
1548         return ret;
1549 }
1550
1551 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1552                                         struct ext4_dir_entry_2 **parent_de,
1553                                         int *retval)
1554 {
1555         struct ext4_iloc iloc;
1556
1557         *retval = ext4_get_inode_loc(inode, &iloc);
1558         if (*retval)
1559                 return NULL;
1560
1561         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1562
1563         return iloc.bh;
1564 }
1565
1566 /*
1567  * Try to create the inline data for the new dir.
1568  * If it succeeds, return 0, otherwise return the error.
1569  * In case of ENOSPC, the caller should create the normal disk layout dir.
1570  */
1571 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1572                                struct inode *inode)
1573 {
1574         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1575         struct ext4_iloc iloc;
1576         struct ext4_dir_entry_2 *de;
1577
1578         ret = ext4_get_inode_loc(inode, &iloc);
1579         if (ret)
1580                 return ret;
1581
1582         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1583         if (ret)
1584                 goto out;
1585
1586         /*
1587          * For inline dir, we only save the inode information for the ".."
1588          * and create a fake dentry to cover the left space.
1589          */
1590         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1591         de->inode = cpu_to_le32(parent->i_ino);
1592         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1593         de->inode = 0;
1594         de->rec_len = ext4_rec_len_to_disk(
1595                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1596                                 inline_size);
1597         set_nlink(inode, 2);
1598         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1599 out:
1600         brelse(iloc.bh);
1601         return ret;
1602 }
1603
1604 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1605                                         const struct qstr *d_name,
1606                                         struct ext4_dir_entry_2 **res_dir,
1607                                         int *has_inline_data)
1608 {
1609         int ret;
1610         struct ext4_iloc iloc;
1611         void *inline_start;
1612         int inline_size;
1613
1614         if (ext4_get_inode_loc(dir, &iloc))
1615                 return NULL;
1616
1617         down_read(&EXT4_I(dir)->xattr_sem);
1618         if (!ext4_has_inline_data(dir)) {
1619                 *has_inline_data = 0;
1620                 goto out;
1621         }
1622
1623         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1624                                                 EXT4_INLINE_DOTDOT_SIZE;
1625         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1626         ret = search_dir(iloc.bh, inline_start, inline_size,
1627                          dir, d_name, 0, res_dir);
1628         if (ret == 1)
1629                 goto out_find;
1630         if (ret < 0)
1631                 goto out;
1632
1633         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1634                 goto out;
1635
1636         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1637         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1638
1639         ret = search_dir(iloc.bh, inline_start, inline_size,
1640                          dir, d_name, 0, res_dir);
1641         if (ret == 1)
1642                 goto out_find;
1643
1644 out:
1645         brelse(iloc.bh);
1646         iloc.bh = NULL;
1647 out_find:
1648         up_read(&EXT4_I(dir)->xattr_sem);
1649         return iloc.bh;
1650 }
1651
1652 int ext4_delete_inline_entry(handle_t *handle,
1653                              struct inode *dir,
1654                              struct ext4_dir_entry_2 *de_del,
1655                              struct buffer_head *bh,
1656                              int *has_inline_data)
1657 {
1658         int err, inline_size;
1659         struct ext4_iloc iloc;
1660         void *inline_start;
1661
1662         err = ext4_get_inode_loc(dir, &iloc);
1663         if (err)
1664                 return err;
1665
1666         down_write(&EXT4_I(dir)->xattr_sem);
1667         if (!ext4_has_inline_data(dir)) {
1668                 *has_inline_data = 0;
1669                 goto out;
1670         }
1671
1672         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1673                 EXT4_MIN_INLINE_DATA_SIZE) {
1674                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1675                                         EXT4_INLINE_DOTDOT_SIZE;
1676                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1677                                 EXT4_INLINE_DOTDOT_SIZE;
1678         } else {
1679                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1680                 inline_size = ext4_get_inline_size(dir) -
1681                                 EXT4_MIN_INLINE_DATA_SIZE;
1682         }
1683
1684         BUFFER_TRACE(bh, "get_write_access");
1685         err = ext4_journal_get_write_access(handle, bh);
1686         if (err)
1687                 goto out;
1688
1689         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1690                                         inline_start, inline_size, 0);
1691         if (err)
1692                 goto out;
1693
1694         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1695         err = ext4_mark_inode_dirty(handle, dir);
1696         if (unlikely(err))
1697                 goto out;
1698
1699         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1700 out:
1701         up_write(&EXT4_I(dir)->xattr_sem);
1702         brelse(iloc.bh);
1703         if (err != -ENOENT)
1704                 ext4_std_error(dir->i_sb, err);
1705         return err;
1706 }
1707
1708 /*
1709  * Get the inline dentry at offset.
1710  */
1711 static inline struct ext4_dir_entry_2 *
1712 ext4_get_inline_entry(struct inode *inode,
1713                       struct ext4_iloc *iloc,
1714                       unsigned int offset,
1715                       void **inline_start,
1716                       int *inline_size)
1717 {
1718         void *inline_pos;
1719
1720         BUG_ON(offset > ext4_get_inline_size(inode));
1721
1722         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1723                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1724                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1725         } else {
1726                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1727                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1728                 *inline_size = ext4_get_inline_size(inode) -
1729                                 EXT4_MIN_INLINE_DATA_SIZE;
1730         }
1731
1732         if (inline_start)
1733                 *inline_start = inline_pos;
1734         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1735 }
1736
1737 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1738 {
1739         int err, inline_size;
1740         struct ext4_iloc iloc;
1741         void *inline_pos;
1742         unsigned int offset;
1743         struct ext4_dir_entry_2 *de;
1744         int ret = 1;
1745
1746         err = ext4_get_inode_loc(dir, &iloc);
1747         if (err) {
1748                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1749                                  err, dir->i_ino);
1750                 return 1;
1751         }
1752
1753         down_read(&EXT4_I(dir)->xattr_sem);
1754         if (!ext4_has_inline_data(dir)) {
1755                 *has_inline_data = 0;
1756                 goto out;
1757         }
1758
1759         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1760         if (!le32_to_cpu(de->inode)) {
1761                 ext4_warning(dir->i_sb,
1762                              "bad inline directory (dir #%lu) - no `..'",
1763                              dir->i_ino);
1764                 ret = 1;
1765                 goto out;
1766         }
1767
1768         offset = EXT4_INLINE_DOTDOT_SIZE;
1769         while (offset < dir->i_size) {
1770                 de = ext4_get_inline_entry(dir, &iloc, offset,
1771                                            &inline_pos, &inline_size);
1772                 if (ext4_check_dir_entry(dir, NULL, de,
1773                                          iloc.bh, inline_pos,
1774                                          inline_size, offset)) {
1775                         ext4_warning(dir->i_sb,
1776                                      "bad inline directory (dir #%lu) - "
1777                                      "inode %u, rec_len %u, name_len %d"
1778                                      "inline size %d\n",
1779                                      dir->i_ino, le32_to_cpu(de->inode),
1780                                      le16_to_cpu(de->rec_len), de->name_len,
1781                                      inline_size);
1782                         ret = 1;
1783                         goto out;
1784                 }
1785                 if (le32_to_cpu(de->inode)) {
1786                         ret = 0;
1787                         goto out;
1788                 }
1789                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1790         }
1791
1792 out:
1793         up_read(&EXT4_I(dir)->xattr_sem);
1794         brelse(iloc.bh);
1795         return ret;
1796 }
1797
1798 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1799 {
1800         int ret;
1801
1802         down_write(&EXT4_I(inode)->xattr_sem);
1803         ret = ext4_destroy_inline_data_nolock(handle, inode);
1804         up_write(&EXT4_I(inode)->xattr_sem);
1805
1806         return ret;
1807 }
1808
1809 int ext4_inline_data_fiemap(struct inode *inode,
1810                             struct fiemap_extent_info *fieinfo,
1811                             int *has_inline)
1812 {
1813         __u64 physical = 0;
1814         __u64 length;
1815         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1816         int error = 0;
1817         struct ext4_iloc iloc;
1818
1819         down_read(&EXT4_I(inode)->xattr_sem);
1820         if (!ext4_has_inline_data(inode)) {
1821                 *has_inline = 0;
1822                 goto out;
1823         }
1824
1825         error = ext4_get_inode_loc(inode, &iloc);
1826         if (error)
1827                 goto out;
1828
1829         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1830         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1831         physical += offsetof(struct ext4_inode, i_block);
1832         length = i_size_read(inode);
1833
1834         if (physical)
1835                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1836                                                 length, flags);
1837         brelse(iloc.bh);
1838 out:
1839         up_read(&EXT4_I(inode)->xattr_sem);
1840         return (error < 0 ? error : 0);
1841 }
1842
1843 /*
1844  * Called during xattr set, and if we can sparse space 'needed',
1845  * just create the extent tree evict the data to the outer block.
1846  *
1847  * We use jbd2 instead of page cache to move data to the 1st block
1848  * so that the whole transaction can be committed as a whole and
1849  * the data isn't lost because of the delayed page cache write.
1850  */
1851 int ext4_try_to_evict_inline_data(handle_t *handle,
1852                                   struct inode *inode,
1853                                   int needed)
1854 {
1855         int error;
1856         struct ext4_xattr_entry *entry;
1857         struct ext4_inode *raw_inode;
1858         struct ext4_iloc iloc;
1859
1860         error = ext4_get_inode_loc(inode, &iloc);
1861         if (error)
1862                 return error;
1863
1864         raw_inode = ext4_raw_inode(&iloc);
1865         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1866                                             EXT4_I(inode)->i_inline_off);
1867         if (EXT4_XATTR_LEN(entry->e_name_len) +
1868             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1869                 error = -ENOSPC;
1870                 goto out;
1871         }
1872
1873         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1874 out:
1875         brelse(iloc.bh);
1876         return error;
1877 }
1878
1879 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1880 {
1881         handle_t *handle;
1882         int inline_size, value_len, needed_blocks;
1883         size_t i_size;
1884         void *value = NULL;
1885         struct ext4_xattr_ibody_find is = {
1886                 .s = { .not_found = -ENODATA, },
1887         };
1888         struct ext4_xattr_info i = {
1889                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1890                 .name = EXT4_XATTR_SYSTEM_DATA,
1891         };
1892
1893
1894         needed_blocks = ext4_writepage_trans_blocks(inode);
1895         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1896         if (IS_ERR(handle))
1897                 return;
1898
1899         down_write(&EXT4_I(inode)->xattr_sem);
1900         if (!ext4_has_inline_data(inode)) {
1901                 *has_inline = 0;
1902                 ext4_journal_stop(handle);
1903                 return;
1904         }
1905
1906         if (ext4_orphan_add(handle, inode))
1907                 goto out;
1908
1909         if (ext4_get_inode_loc(inode, &is.iloc))
1910                 goto out;
1911
1912         down_write(&EXT4_I(inode)->i_data_sem);
1913         i_size = inode->i_size;
1914         inline_size = ext4_get_inline_size(inode);
1915         EXT4_I(inode)->i_disksize = i_size;
1916
1917         if (i_size < inline_size) {
1918                 /* Clear the content in the xattr space. */
1919                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1920                         if (ext4_xattr_ibody_find(inode, &i, &is))
1921                                 goto out_error;
1922
1923                         BUG_ON(is.s.not_found);
1924
1925                         value_len = le32_to_cpu(is.s.here->e_value_size);
1926                         value = kmalloc(value_len, GFP_NOFS);
1927                         if (!value)
1928                                 goto out_error;
1929
1930                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1931                                                 value, value_len))
1932                                 goto out_error;
1933
1934                         i.value = value;
1935                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1936                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1937                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1938                                 goto out_error;
1939                 }
1940
1941                 /* Clear the content within i_blocks. */
1942                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1943                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1944                         memset(p + i_size, 0,
1945                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1946                 }
1947
1948                 EXT4_I(inode)->i_inline_size = i_size <
1949                                         EXT4_MIN_INLINE_DATA_SIZE ?
1950                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1951         }
1952
1953 out_error:
1954         up_write(&EXT4_I(inode)->i_data_sem);
1955 out:
1956         brelse(is.iloc.bh);
1957         up_write(&EXT4_I(inode)->xattr_sem);
1958         kfree(value);
1959         if (inode->i_nlink)
1960                 ext4_orphan_del(handle, inode);
1961
1962         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1963         ext4_mark_inode_dirty(handle, inode);
1964         if (IS_SYNC(inode))
1965                 ext4_handle_sync(handle);
1966
1967         ext4_journal_stop(handle);
1968         return;
1969 }
1970
1971 int ext4_convert_inline_data(struct inode *inode)
1972 {
1973         int error, needed_blocks;
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         }
1981
1982         needed_blocks = ext4_writepage_trans_blocks(inode);
1983
1984         iloc.bh = NULL;
1985         error = ext4_get_inode_loc(inode, &iloc);
1986         if (error)
1987                 return error;
1988
1989         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1990         if (IS_ERR(handle)) {
1991                 error = PTR_ERR(handle);
1992                 goto out_free;
1993         }
1994
1995         down_write(&EXT4_I(inode)->xattr_sem);
1996         if (!ext4_has_inline_data(inode)) {
1997                 up_write(&EXT4_I(inode)->xattr_sem);
1998                 goto out;
1999         }
2000
2001         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2002         up_write(&EXT4_I(inode)->xattr_sem);
2003 out:
2004         ext4_journal_stop(handle);
2005 out_free:
2006         brelse(iloc.bh);
2007         return error;
2008 }