]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/ext4/inline.c
ext4: let add_dir_entry handle inline data properly
[karo-tx-linux.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
19 #define EXT4_XATTR_SYSTEM_DATA  "data"
20 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
21 #define EXT4_INLINE_DOTDOT_SIZE 4
22
23 int ext4_get_inline_size(struct inode *inode)
24 {
25         if (EXT4_I(inode)->i_inline_off)
26                 return EXT4_I(inode)->i_inline_size;
27
28         return 0;
29 }
30
31 static int get_max_inline_xattr_value_size(struct inode *inode,
32                                            struct ext4_iloc *iloc)
33 {
34         struct ext4_xattr_ibody_header *header;
35         struct ext4_xattr_entry *entry;
36         struct ext4_inode *raw_inode;
37         int free, min_offs;
38
39         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
40                         EXT4_GOOD_OLD_INODE_SIZE -
41                         EXT4_I(inode)->i_extra_isize -
42                         sizeof(struct ext4_xattr_ibody_header);
43
44         /*
45          * We need to subtract another sizeof(__u32) since an in-inode xattr
46          * needs an empty 4 bytes to indicate the gap between the xattr entry
47          * and the name/value pair.
48          */
49         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
50                 return EXT4_XATTR_SIZE(min_offs -
51                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
52                         EXT4_XATTR_ROUND - sizeof(__u32));
53
54         raw_inode = ext4_raw_inode(iloc);
55         header = IHDR(inode, raw_inode);
56         entry = IFIRST(header);
57
58         /* Compute min_offs. */
59         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
60                 if (!entry->e_value_block && entry->e_value_size) {
61                         size_t offs = le16_to_cpu(entry->e_value_offs);
62                         if (offs < min_offs)
63                                 min_offs = offs;
64                 }
65         }
66         free = min_offs -
67                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
68
69         if (EXT4_I(inode)->i_inline_off) {
70                 entry = (struct ext4_xattr_entry *)
71                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
72
73                 free += le32_to_cpu(entry->e_value_size);
74                 goto out;
75         }
76
77         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
78
79         if (free > EXT4_XATTR_ROUND)
80                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
81         else
82                 free = 0;
83
84 out:
85         return free;
86 }
87
88 /*
89  * Get the maximum size we now can store in an inode.
90  * If we can't find the space for a xattr entry, don't use the space
91  * of the extents since we have no space to indicate the inline data.
92  */
93 int ext4_get_max_inline_size(struct inode *inode)
94 {
95         int error, max_inline_size;
96         struct ext4_iloc iloc;
97
98         if (EXT4_I(inode)->i_extra_isize == 0)
99                 return 0;
100
101         error = ext4_get_inode_loc(inode, &iloc);
102         if (error) {
103                 ext4_error_inode(inode, __func__, __LINE__, 0,
104                                  "can't get inode location %lu",
105                                  inode->i_ino);
106                 return 0;
107         }
108
109         down_read(&EXT4_I(inode)->xattr_sem);
110         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
111         up_read(&EXT4_I(inode)->xattr_sem);
112
113         brelse(iloc.bh);
114
115         if (!max_inline_size)
116                 return 0;
117
118         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
119 }
120
121 int ext4_has_inline_data(struct inode *inode)
122 {
123         return ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) &&
124                EXT4_I(inode)->i_inline_off;
125 }
126
127 /*
128  * this function does not take xattr_sem, which is OK because it is
129  * currently only used in a code path coming form ext4_iget, before
130  * the new inode has been unlocked
131  */
132 int ext4_find_inline_data_nolock(struct inode *inode)
133 {
134         struct ext4_xattr_ibody_find is = {
135                 .s = { .not_found = -ENODATA, },
136         };
137         struct ext4_xattr_info i = {
138                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
139                 .name = EXT4_XATTR_SYSTEM_DATA,
140         };
141         int error;
142
143         if (EXT4_I(inode)->i_extra_isize == 0)
144                 return 0;
145
146         error = ext4_get_inode_loc(inode, &is.iloc);
147         if (error)
148                 return error;
149
150         error = ext4_xattr_ibody_find(inode, &i, &is);
151         if (error)
152                 goto out;
153
154         if (!is.s.not_found) {
155                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
156                                         (void *)ext4_raw_inode(&is.iloc));
157                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
158                                 le32_to_cpu(is.s.here->e_value_size);
159                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
160         }
161 out:
162         brelse(is.iloc.bh);
163         return error;
164 }
165
166 static int ext4_read_inline_data(struct inode *inode, void *buffer,
167                                  unsigned int len,
168                                  struct ext4_iloc *iloc)
169 {
170         struct ext4_xattr_entry *entry;
171         struct ext4_xattr_ibody_header *header;
172         int cp_len = 0;
173         struct ext4_inode *raw_inode;
174
175         if (!len)
176                 return 0;
177
178         BUG_ON(len > EXT4_I(inode)->i_inline_size);
179
180         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
181                         len : EXT4_MIN_INLINE_DATA_SIZE;
182
183         raw_inode = ext4_raw_inode(iloc);
184         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
185
186         len -= cp_len;
187         buffer += cp_len;
188
189         if (!len)
190                 goto out;
191
192         header = IHDR(inode, raw_inode);
193         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
194                                             EXT4_I(inode)->i_inline_off);
195         len = min_t(unsigned int, len,
196                     (unsigned int)le32_to_cpu(entry->e_value_size));
197
198         memcpy(buffer,
199                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
200         cp_len += len;
201
202 out:
203         return cp_len;
204 }
205
206 /*
207  * write the buffer to the inline inode.
208  * If 'create' is set, we don't need to do the extra copy in the xattr
209  * value since it is already handled by ext4_xattr_ibody_set. That saves
210  * us one memcpy.
211  */
212 void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
213                             void *buffer, loff_t pos, unsigned int len)
214 {
215         struct ext4_xattr_entry *entry;
216         struct ext4_xattr_ibody_header *header;
217         struct ext4_inode *raw_inode;
218         int cp_len = 0;
219
220         BUG_ON(!EXT4_I(inode)->i_inline_off);
221         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
222
223         raw_inode = ext4_raw_inode(iloc);
224         buffer += pos;
225
226         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
227                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
228                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
229                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
230
231                 len -= cp_len;
232                 buffer += cp_len;
233                 pos += cp_len;
234         }
235
236         if (!len)
237                 return;
238
239         pos -= EXT4_MIN_INLINE_DATA_SIZE;
240         header = IHDR(inode, raw_inode);
241         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
242                                             EXT4_I(inode)->i_inline_off);
243
244         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
245                buffer, len);
246 }
247
248 static int ext4_create_inline_data(handle_t *handle,
249                                    struct inode *inode, unsigned len)
250 {
251         int error;
252         void *value = NULL;
253         struct ext4_xattr_ibody_find is = {
254                 .s = { .not_found = -ENODATA, },
255         };
256         struct ext4_xattr_info i = {
257                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
258                 .name = EXT4_XATTR_SYSTEM_DATA,
259         };
260
261         error = ext4_get_inode_loc(inode, &is.iloc);
262         if (error)
263                 return error;
264
265         error = ext4_journal_get_write_access(handle, is.iloc.bh);
266         if (error)
267                 goto out;
268
269         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
270                 value = (void *)empty_zero_page;
271                 len -= EXT4_MIN_INLINE_DATA_SIZE;
272         } else {
273                 value = "";
274                 len = 0;
275         }
276
277         /* Insert the the xttr entry. */
278         i.value = value;
279         i.value_len = len;
280
281         error = ext4_xattr_ibody_find(inode, &i, &is);
282         if (error)
283                 goto out;
284
285         BUG_ON(!is.s.not_found);
286
287         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
288         if (error) {
289                 if (error == -ENOSPC)
290                         ext4_clear_inode_state(inode,
291                                                EXT4_STATE_MAY_INLINE_DATA);
292                 goto out;
293         }
294
295         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
296                 0, EXT4_MIN_INLINE_DATA_SIZE);
297
298         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
299                                       (void *)ext4_raw_inode(&is.iloc));
300         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
301         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
302         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
303         get_bh(is.iloc.bh);
304         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
305
306 out:
307         brelse(is.iloc.bh);
308         return error;
309 }
310
311 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
312                                    unsigned int len)
313 {
314         int error;
315         void *value = NULL;
316         struct ext4_xattr_ibody_find is = {
317                 .s = { .not_found = -ENODATA, },
318         };
319         struct ext4_xattr_info i = {
320                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
321                 .name = EXT4_XATTR_SYSTEM_DATA,
322         };
323
324         /* If the old space is ok, write the data directly. */
325         if (len <= EXT4_I(inode)->i_inline_size)
326                 return 0;
327
328         error = ext4_get_inode_loc(inode, &is.iloc);
329         if (error)
330                 return error;
331
332         error = ext4_xattr_ibody_find(inode, &i, &is);
333         if (error)
334                 goto out;
335
336         BUG_ON(is.s.not_found);
337
338         len -= EXT4_MIN_INLINE_DATA_SIZE;
339         value = kzalloc(len, GFP_NOFS);
340         if (!value)
341                 goto out;
342
343         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
344                                      value, len);
345         if (error == -ENODATA)
346                 goto out;
347
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_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 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         error = ext4_journal_get_write_access(handle, is.iloc.bh);
426         if (error)
427                 goto out;
428
429         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
430         if (error)
431                 goto out;
432
433         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
434                 0, EXT4_MIN_INLINE_DATA_SIZE);
435
436         if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
437                                       EXT4_FEATURE_INCOMPAT_EXTENTS)) {
438                 if (S_ISDIR(inode->i_mode) ||
439                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
440                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
441                         ext4_ext_tree_init(handle, inode);
442                 }
443         }
444         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
445
446         get_bh(is.iloc.bh);
447         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
448
449         EXT4_I(inode)->i_inline_off = 0;
450         EXT4_I(inode)->i_inline_size = 0;
451         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
452 out:
453         brelse(is.iloc.bh);
454         if (error == -ENODATA)
455                 error = 0;
456         return error;
457 }
458
459 static int ext4_read_inline_page(struct inode *inode, struct page *page)
460 {
461         void *kaddr;
462         int ret = 0;
463         size_t len;
464         struct ext4_iloc iloc;
465
466         BUG_ON(!PageLocked(page));
467         BUG_ON(!ext4_has_inline_data(inode));
468         BUG_ON(page->index);
469
470         if (!EXT4_I(inode)->i_inline_off) {
471                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
472                              inode->i_ino);
473                 goto out;
474         }
475
476         ret = ext4_get_inode_loc(inode, &iloc);
477         if (ret)
478                 goto out;
479
480         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
481         kaddr = kmap_atomic(page);
482         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
483         flush_dcache_page(page);
484         kunmap_atomic(kaddr);
485         zero_user_segment(page, len, PAGE_CACHE_SIZE);
486         SetPageUptodate(page);
487         brelse(iloc.bh);
488
489 out:
490         return ret;
491 }
492
493 int ext4_readpage_inline(struct inode *inode, struct page *page)
494 {
495         int ret = 0;
496
497         down_read(&EXT4_I(inode)->xattr_sem);
498         if (!ext4_has_inline_data(inode)) {
499                 up_read(&EXT4_I(inode)->xattr_sem);
500                 return -EAGAIN;
501         }
502
503         /*
504          * Current inline data can only exist in the 1st page,
505          * So for all the other pages, just set them uptodate.
506          */
507         if (!page->index)
508                 ret = ext4_read_inline_page(inode, page);
509         else if (!PageUptodate(page)) {
510                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
511                 SetPageUptodate(page);
512         }
513
514         up_read(&EXT4_I(inode)->xattr_sem);
515
516         unlock_page(page);
517         return ret >= 0 ? 0 : ret;
518 }
519
520 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
521                                               struct inode *inode,
522                                               unsigned flags)
523 {
524         int ret, needed_blocks;
525         handle_t *handle = NULL;
526         int retries = 0, sem_held = 0;
527         struct page *page = NULL;
528         unsigned from, to;
529         struct ext4_iloc iloc;
530
531         if (!ext4_has_inline_data(inode)) {
532                 /*
533                  * clear the flag so that no new write
534                  * will trap here again.
535                  */
536                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
537                 return 0;
538         }
539
540         needed_blocks = ext4_writepage_trans_blocks(inode);
541
542         ret = ext4_get_inode_loc(inode, &iloc);
543         if (ret)
544                 return ret;
545
546 retry:
547         handle = ext4_journal_start(inode, needed_blocks);
548         if (IS_ERR(handle)) {
549                 ret = PTR_ERR(handle);
550                 handle = NULL;
551                 goto out;
552         }
553
554         /* We cannot recurse into the filesystem as the transaction is already
555          * started */
556         flags |= AOP_FLAG_NOFS;
557
558         page = grab_cache_page_write_begin(mapping, 0, flags);
559         if (!page) {
560                 ret = -ENOMEM;
561                 goto out;
562         }
563
564         down_write(&EXT4_I(inode)->xattr_sem);
565         sem_held = 1;
566         /* If some one has already done this for us, just exit. */
567         if (!ext4_has_inline_data(inode)) {
568                 ret = 0;
569                 goto out;
570         }
571
572         from = 0;
573         to = ext4_get_inline_size(inode);
574         if (!PageUptodate(page)) {
575                 ret = ext4_read_inline_page(inode, page);
576                 if (ret < 0)
577                         goto out;
578         }
579
580         ret = ext4_destroy_inline_data_nolock(handle, inode);
581         if (ret)
582                 goto out;
583
584         if (ext4_should_dioread_nolock(inode))
585                 ret = __block_write_begin(page, from, to, ext4_get_block_write);
586         else
587                 ret = __block_write_begin(page, from, to, ext4_get_block);
588
589         if (!ret && ext4_should_journal_data(inode)) {
590                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
591                                              from, to, NULL,
592                                              do_journal_get_write_access);
593         }
594
595         if (ret) {
596                 unlock_page(page);
597                 page_cache_release(page);
598                 ext4_orphan_add(handle, inode);
599                 up_write(&EXT4_I(inode)->xattr_sem);
600                 sem_held = 0;
601                 ext4_journal_stop(handle);
602                 handle = NULL;
603                 ext4_truncate_failed_write(inode);
604                 /*
605                  * If truncate failed early the inode might
606                  * still be on the orphan list; we need to
607                  * make sure the inode is removed from the
608                  * orphan list in that case.
609                  */
610                 if (inode->i_nlink)
611                         ext4_orphan_del(NULL, inode);
612         }
613
614         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
615                 goto retry;
616
617         block_commit_write(page, from, to);
618 out:
619         if (page) {
620                 unlock_page(page);
621                 page_cache_release(page);
622         }
623         if (sem_held)
624                 up_write(&EXT4_I(inode)->xattr_sem);
625         if (handle)
626                 ext4_journal_stop(handle);
627         brelse(iloc.bh);
628         return ret;
629 }
630
631 /*
632  * Try to write data in the inode.
633  * If the inode has inline data, check whether the new write can be
634  * in the inode also. If not, create the page the handle, move the data
635  * to the page make it update and let the later codes create extent for it.
636  */
637 int ext4_try_to_write_inline_data(struct address_space *mapping,
638                                   struct inode *inode,
639                                   loff_t pos, unsigned len,
640                                   unsigned flags,
641                                   struct page **pagep)
642 {
643         int ret;
644         handle_t *handle;
645         struct page *page;
646         struct ext4_iloc iloc;
647
648         if (pos + len > ext4_get_max_inline_size(inode))
649                 goto convert;
650
651         ret = ext4_get_inode_loc(inode, &iloc);
652         if (ret)
653                 return ret;
654
655         /*
656          * The possible write could happen in the inode,
657          * so try to reserve the space in inode first.
658          */
659         handle = ext4_journal_start(inode, 1);
660         if (IS_ERR(handle)) {
661                 ret = PTR_ERR(handle);
662                 handle = NULL;
663                 goto out;
664         }
665
666         ret = ext4_prepare_inline_data(handle, inode, pos + len);
667         if (ret && ret != -ENOSPC)
668                 goto out;
669
670         /* We don't have space in inline inode, so convert it to extent. */
671         if (ret == -ENOSPC) {
672                 ext4_journal_stop(handle);
673                 brelse(iloc.bh);
674                 goto convert;
675         }
676
677         flags |= AOP_FLAG_NOFS;
678
679         page = grab_cache_page_write_begin(mapping, 0, flags);
680         if (!page) {
681                 ret = -ENOMEM;
682                 goto out;
683         }
684
685         *pagep = page;
686         down_read(&EXT4_I(inode)->xattr_sem);
687         if (!ext4_has_inline_data(inode)) {
688                 ret = 0;
689                 unlock_page(page);
690                 page_cache_release(page);
691                 goto out_up_read;
692         }
693
694         if (!PageUptodate(page)) {
695                 ret = ext4_read_inline_page(inode, page);
696                 if (ret < 0)
697                         goto out_up_read;
698         }
699
700         ret = 1;
701         handle = NULL;
702 out_up_read:
703         up_read(&EXT4_I(inode)->xattr_sem);
704 out:
705         if (handle)
706                 ext4_journal_stop(handle);
707         brelse(iloc.bh);
708         return ret;
709 convert:
710         return ext4_convert_inline_data_to_extent(mapping,
711                                                   inode, flags);
712 }
713
714 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
715                                unsigned copied, struct page *page)
716 {
717         int ret;
718         void *kaddr;
719         struct ext4_iloc iloc;
720
721         if (unlikely(copied < len)) {
722                 if (!PageUptodate(page)) {
723                         copied = 0;
724                         goto out;
725                 }
726         }
727
728         ret = ext4_get_inode_loc(inode, &iloc);
729         if (ret) {
730                 ext4_std_error(inode->i_sb, ret);
731                 copied = 0;
732                 goto out;
733         }
734
735         down_write(&EXT4_I(inode)->xattr_sem);
736         BUG_ON(!ext4_has_inline_data(inode));
737
738         kaddr = kmap_atomic(page);
739         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
740         kunmap_atomic(kaddr);
741         SetPageUptodate(page);
742         /* clear page dirty so that writepages wouldn't work for us. */
743         ClearPageDirty(page);
744
745         up_write(&EXT4_I(inode)->xattr_sem);
746         brelse(iloc.bh);
747 out:
748         return copied;
749 }
750
751 struct buffer_head *
752 ext4_journalled_write_inline_data(struct inode *inode,
753                                   unsigned len,
754                                   struct page *page)
755 {
756         int ret;
757         void *kaddr;
758         struct ext4_iloc iloc;
759
760         ret = ext4_get_inode_loc(inode, &iloc);
761         if (ret) {
762                 ext4_std_error(inode->i_sb, ret);
763                 return NULL;
764         }
765
766         down_write(&EXT4_I(inode)->xattr_sem);
767         kaddr = kmap_atomic(page);
768         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
769         kunmap_atomic(kaddr);
770         up_write(&EXT4_I(inode)->xattr_sem);
771
772         return iloc.bh;
773 }
774
775 /*
776  * Try to make the page cache and handle ready for the inline data case.
777  * We can call this function in 2 cases:
778  * 1. The inode is created and the first write exceeds inline size. We can
779  *    clear the inode state safely.
780  * 2. The inode has inline data, then we need to read the data, make it
781  *    update and dirty so that ext4_da_writepages can handle it. We don't
782  *    need to start the journal since the file's metatdata isn't changed now.
783  */
784 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
785                                                  struct inode *inode,
786                                                  unsigned flags,
787                                                  void **fsdata)
788 {
789         int ret = 0, inline_size;
790         struct page *page;
791
792         page = grab_cache_page_write_begin(mapping, 0, flags);
793         if (!page)
794                 return -ENOMEM;
795
796         down_read(&EXT4_I(inode)->xattr_sem);
797         if (!ext4_has_inline_data(inode)) {
798                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
799                 goto out;
800         }
801
802         inline_size = ext4_get_inline_size(inode);
803
804         if (!PageUptodate(page)) {
805                 ret = ext4_read_inline_page(inode, page);
806                 if (ret < 0)
807                         goto out;
808         }
809
810         ret = __block_write_begin(page, 0, inline_size,
811                                   ext4_da_get_block_prep);
812         if (ret) {
813                 ext4_truncate_failed_write(inode);
814                 goto out;
815         }
816
817         SetPageDirty(page);
818         SetPageUptodate(page);
819         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
820         *fsdata = (void *)CONVERT_INLINE_DATA;
821
822 out:
823         up_read(&EXT4_I(inode)->xattr_sem);
824         if (page) {
825                 unlock_page(page);
826                 page_cache_release(page);
827         }
828         return ret;
829 }
830
831 /*
832  * Prepare the write for the inline data.
833  * If the the data can be written into the inode, we just read
834  * the page and make it uptodate, and start the journal.
835  * Otherwise read the page, makes it dirty so that it can be
836  * handle in writepages(the i_disksize update is left to the
837  * normal ext4_da_write_end).
838  */
839 int ext4_da_write_inline_data_begin(struct address_space *mapping,
840                                     struct inode *inode,
841                                     loff_t pos, unsigned len,
842                                     unsigned flags,
843                                     struct page **pagep,
844                                     void **fsdata)
845 {
846         int ret, inline_size;
847         handle_t *handle;
848         struct page *page;
849         struct ext4_iloc iloc;
850
851         ret = ext4_get_inode_loc(inode, &iloc);
852         if (ret)
853                 return ret;
854
855         handle = ext4_journal_start(inode, 1);
856         if (IS_ERR(handle)) {
857                 ret = PTR_ERR(handle);
858                 handle = NULL;
859                 goto out;
860         }
861
862         inline_size = ext4_get_max_inline_size(inode);
863
864         ret = -ENOSPC;
865         if (inline_size >= pos + len) {
866                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
867                 if (ret && ret != -ENOSPC)
868                         goto out;
869         }
870
871         if (ret == -ENOSPC) {
872                 ret = ext4_da_convert_inline_data_to_extent(mapping,
873                                                             inode,
874                                                             flags,
875                                                             fsdata);
876                 goto out;
877         }
878
879         /*
880          * We cannot recurse into the filesystem as the transaction
881          * is already started.
882          */
883         flags |= AOP_FLAG_NOFS;
884
885         page = grab_cache_page_write_begin(mapping, 0, flags);
886         if (!page) {
887                 ret = -ENOMEM;
888                 goto out;
889         }
890
891         down_read(&EXT4_I(inode)->xattr_sem);
892         if (!ext4_has_inline_data(inode)) {
893                 ret = 0;
894                 goto out_release_page;
895         }
896
897         if (!PageUptodate(page)) {
898                 ret = ext4_read_inline_page(inode, page);
899                 if (ret < 0)
900                         goto out_release_page;
901         }
902
903         up_read(&EXT4_I(inode)->xattr_sem);
904         *pagep = page;
905         handle = NULL;
906         brelse(iloc.bh);
907         return 1;
908 out_release_page:
909         up_read(&EXT4_I(inode)->xattr_sem);
910         unlock_page(page);
911         page_cache_release(page);
912 out:
913         if (handle)
914                 ext4_journal_stop(handle);
915         brelse(iloc.bh);
916         return ret;
917 }
918
919 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
920                                   unsigned len, unsigned copied,
921                                   struct page *page)
922 {
923         int i_size_changed = 0;
924
925         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
926
927         /*
928          * No need to use i_size_read() here, the i_size
929          * cannot change under us because we hold i_mutex.
930          *
931          * But it's important to update i_size while still holding page lock:
932          * page writeout could otherwise come in and zero beyond i_size.
933          */
934         if (pos+copied > inode->i_size) {
935                 i_size_write(inode, pos+copied);
936                 i_size_changed = 1;
937         }
938         unlock_page(page);
939         page_cache_release(page);
940
941         /*
942          * Don't mark the inode dirty under page lock. First, it unnecessarily
943          * makes the holding time of page lock longer. Second, it forces lock
944          * ordering of page lock and transaction start for journaling
945          * filesystems.
946          */
947         if (i_size_changed)
948                 mark_inode_dirty(inode);
949
950         return copied;
951 }
952
953 #ifdef INLINE_DIR_DEBUG
954 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
955                           void *inline_start, int inline_size)
956 {
957         int offset;
958         unsigned short de_len;
959         struct ext4_dir_entry_2 *de = inline_start;
960         void *dlimit = inline_start + inline_size;
961
962         trace_printk("inode %lu\n", dir->i_ino);
963         offset = 0;
964         while ((void *)de < dlimit) {
965                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
966                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
967                              offset, de_len, de->name_len, de->name,
968                              de->name_len, le32_to_cpu(de->inode));
969                 if (ext4_check_dir_entry(dir, NULL, de, bh,
970                                          inline_start, inline_size, offset))
971                         BUG();
972
973                 offset += de_len;
974                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
975         }
976 }
977 #else
978 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
979 #endif
980
981 /*
982  * Add a new entry into a inline dir.
983  * It will return -ENOSPC if no space is available, and -EIO
984  * and -EEXIST if directory entry already exists.
985  */
986 static int ext4_add_dirent_to_inline(handle_t *handle,
987                                      struct dentry *dentry,
988                                      struct inode *inode,
989                                      struct ext4_iloc *iloc,
990                                      void *inline_start, int inline_size)
991 {
992         struct inode    *dir = dentry->d_parent->d_inode;
993         const char      *name = dentry->d_name.name;
994         int             namelen = dentry->d_name.len;
995         unsigned short  reclen;
996         int             err;
997         struct ext4_dir_entry_2 *de;
998
999         reclen = EXT4_DIR_REC_LEN(namelen);
1000         err = ext4_find_dest_de(dir, inode, iloc->bh,
1001                                 inline_start, inline_size,
1002                                 name, namelen, &de);
1003         if (err)
1004                 return err;
1005
1006         err = ext4_journal_get_write_access(handle, iloc->bh);
1007         if (err)
1008                 return err;
1009         ext4_insert_dentry(inode, de, inline_size, name, namelen);
1010
1011         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1012
1013         /*
1014          * XXX shouldn't update any times until successful
1015          * completion of syscall, but too many callers depend
1016          * on this.
1017          *
1018          * XXX similarly, too many callers depend on
1019          * ext4_new_inode() setting the times, but error
1020          * recovery deletes the inode, so the worst that can
1021          * happen is that the times are slightly out of date
1022          * and/or different from the directory change time.
1023          */
1024         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1025         ext4_update_dx_flag(dir);
1026         dir->i_version++;
1027         ext4_mark_inode_dirty(handle, dir);
1028         return 1;
1029 }
1030
1031 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1032                                        struct ext4_iloc *iloc)
1033 {
1034         struct ext4_xattr_entry *entry;
1035         struct ext4_xattr_ibody_header *header;
1036
1037         BUG_ON(!EXT4_I(inode)->i_inline_off);
1038
1039         header = IHDR(inode, ext4_raw_inode(iloc));
1040         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1041                                             EXT4_I(inode)->i_inline_off);
1042
1043         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1044 }
1045
1046 /* Set the final de to cover the whole block. */
1047 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1048 {
1049         struct ext4_dir_entry_2 *de, *prev_de;
1050         void *limit;
1051         int de_len;
1052
1053         de = (struct ext4_dir_entry_2 *)de_buf;
1054         if (old_size) {
1055                 limit = de_buf + old_size;
1056                 do {
1057                         prev_de = de;
1058                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1059                         de_buf += de_len;
1060                         de = (struct ext4_dir_entry_2 *)de_buf;
1061                 } while (de_buf < limit);
1062
1063                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1064                                                         old_size, new_size);
1065         } else {
1066                 /* this is just created, so create an empty entry. */
1067                 de->inode = 0;
1068                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1069         }
1070 }
1071
1072 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1073                                   struct ext4_iloc *iloc)
1074 {
1075         int ret;
1076         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1077         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1078
1079         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1080                 return -ENOSPC;
1081
1082         ret = ext4_update_inline_data(handle, dir,
1083                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1084         if (ret)
1085                 return ret;
1086
1087         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1088                              EXT4_I(dir)->i_inline_size -
1089                                                 EXT4_MIN_INLINE_DATA_SIZE);
1090         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1091         return 0;
1092 }
1093
1094 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1095                                      struct ext4_iloc *iloc,
1096                                      void *buf, int inline_size)
1097 {
1098         ext4_create_inline_data(handle, inode, inline_size);
1099         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1100         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1101 }
1102
1103 static int ext4_finish_convert_inline_dir(handle_t *handle,
1104                                           struct inode *inode,
1105                                           struct buffer_head *dir_block,
1106                                           void *buf,
1107                                           int inline_size)
1108 {
1109         int err, csum_size = 0, header_size = 0;
1110         struct ext4_dir_entry_2 *de;
1111         struct ext4_dir_entry_tail *t;
1112         void *target = dir_block->b_data;
1113
1114         /*
1115          * First create "." and ".." and then copy the dir information
1116          * back to the block.
1117          */
1118         de = (struct ext4_dir_entry_2 *)target;
1119         de = ext4_init_dot_dotdot(inode, de,
1120                 inode->i_sb->s_blocksize, csum_size,
1121                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1122         header_size = (void *)de - target;
1123
1124         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1125                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1126
1127         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1128                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1129                 csum_size = sizeof(struct ext4_dir_entry_tail);
1130
1131         inode->i_size = inode->i_sb->s_blocksize;
1132         i_size_write(inode, inode->i_sb->s_blocksize);
1133         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1134         ext4_update_final_de(dir_block->b_data,
1135                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1136                         inode->i_sb->s_blocksize - csum_size);
1137
1138         if (csum_size) {
1139                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1140                                      inode->i_sb->s_blocksize);
1141                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1142         }
1143         set_buffer_uptodate(dir_block);
1144         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1145         if (err)
1146                 goto out;
1147         set_buffer_verified(dir_block);
1148 out:
1149         return err;
1150 }
1151
1152 static int ext4_convert_inline_data_nolock(handle_t *handle,
1153                                            struct inode *inode,
1154                                            struct ext4_iloc *iloc)
1155 {
1156         int error;
1157         void *buf = NULL;
1158         struct buffer_head *data_bh = NULL;
1159         struct ext4_map_blocks map;
1160         int inline_size;
1161
1162         inline_size = ext4_get_inline_size(inode);
1163         buf = kmalloc(inline_size, GFP_NOFS);
1164         if (!buf) {
1165                 error = -ENOMEM;
1166                 goto out;
1167         }
1168
1169         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1170         if (error < 0)
1171                 goto out;
1172
1173         error = ext4_destroy_inline_data_nolock(handle, inode);
1174         if (error)
1175                 goto out;
1176
1177         map.m_lblk = 0;
1178         map.m_len = 1;
1179         map.m_flags = 0;
1180         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1181         if (error < 0)
1182                 goto out_restore;
1183         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1184                 error = -EIO;
1185                 goto out_restore;
1186         }
1187
1188         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1189         if (!data_bh) {
1190                 error = -EIO;
1191                 goto out_restore;
1192         }
1193
1194         lock_buffer(data_bh);
1195         error = ext4_journal_get_create_access(handle, data_bh);
1196         if (error) {
1197                 unlock_buffer(data_bh);
1198                 error = -EIO;
1199                 goto out_restore;
1200         }
1201         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1202
1203         if (!S_ISDIR(inode->i_mode)) {
1204                 memcpy(data_bh->b_data, buf, inline_size);
1205                 set_buffer_uptodate(data_bh);
1206                 error = ext4_handle_dirty_metadata(handle,
1207                                                    inode, data_bh);
1208         } else {
1209                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1210                                                        buf, inline_size);
1211         }
1212
1213         unlock_buffer(data_bh);
1214 out_restore:
1215         if (error)
1216                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1217
1218 out:
1219         brelse(data_bh);
1220         kfree(buf);
1221         return error;
1222 }
1223
1224 /*
1225  * Try to add the new entry to the inline data.
1226  * If succeeds, return 0. If not, extended the inline dir and copied data to
1227  * the new created block.
1228  */
1229 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1230                               struct inode *inode)
1231 {
1232         int ret, inline_size;
1233         void *inline_start;
1234         struct ext4_iloc iloc;
1235         struct inode *dir = dentry->d_parent->d_inode;
1236
1237         ret = ext4_get_inode_loc(dir, &iloc);
1238         if (ret)
1239                 return ret;
1240
1241         down_write(&EXT4_I(dir)->xattr_sem);
1242         if (!ext4_has_inline_data(dir))
1243                 goto out;
1244
1245         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1246                                                  EXT4_INLINE_DOTDOT_SIZE;
1247         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1248
1249         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1250                                         inline_start, inline_size);
1251         if (ret != -ENOSPC)
1252                 goto out;
1253
1254         /* check whether it can be inserted to inline xattr space. */
1255         inline_size = EXT4_I(dir)->i_inline_size -
1256                         EXT4_MIN_INLINE_DATA_SIZE;
1257         if (!inline_size) {
1258                 /* Try to use the xattr space.*/
1259                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1260                 if (ret && ret != -ENOSPC)
1261                         goto out;
1262
1263                 inline_size = EXT4_I(dir)->i_inline_size -
1264                                 EXT4_MIN_INLINE_DATA_SIZE;
1265         }
1266
1267         if (inline_size) {
1268                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1269
1270                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1271                                                 inline_start, inline_size);
1272
1273                 if (ret != -ENOSPC)
1274                         goto out;
1275         }
1276
1277         /*
1278          * The inline space is filled up, so create a new block for it.
1279          * As the extent tree will be created, we have to save the inline
1280          * dir first.
1281          */
1282         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1283
1284 out:
1285         ext4_mark_inode_dirty(handle, dir);
1286         up_write(&EXT4_I(dir)->xattr_sem);
1287         brelse(iloc.bh);
1288         return ret;
1289 }
1290
1291 /*
1292  * Try to create the inline data for the new dir.
1293  * If it succeeds, return 0, otherwise return the error.
1294  * In case of ENOSPC, the caller should create the normal disk layout dir.
1295  */
1296 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1297                                struct inode *inode)
1298 {
1299         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1300         struct ext4_iloc iloc;
1301         struct ext4_dir_entry_2 *de;
1302
1303         ret = ext4_get_inode_loc(inode, &iloc);
1304         if (ret)
1305                 return ret;
1306
1307         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1308         if (ret)
1309                 goto out;
1310
1311         /*
1312          * For inline dir, we only save the inode information for the ".."
1313          * and create a fake dentry to cover the left space.
1314          */
1315         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1316         de->inode = cpu_to_le32(parent->i_ino);
1317         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1318         de->inode = 0;
1319         de->rec_len = ext4_rec_len_to_disk(
1320                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1321                                 inline_size);
1322         set_nlink(inode, 2);
1323         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1324 out:
1325         brelse(iloc.bh);
1326         return ret;
1327 }
1328
1329 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1330 {
1331         int ret;
1332
1333         down_write(&EXT4_I(inode)->xattr_sem);
1334         ret = ext4_destroy_inline_data_nolock(handle, inode);
1335         up_write(&EXT4_I(inode)->xattr_sem);
1336
1337         return ret;
1338 }