]> git.samba.org - sfrench/cifs-2.6.git/blob - fs/ext4/inline.c
Merge tag 'renesas-fixes2-for-v3.16' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.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 int ext4_has_inline_data(struct inode *inode)
124 {
125         return ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) &&
126                EXT4_I(inode)->i_inline_off;
127 }
128
129 /*
130  * this function does not take xattr_sem, which is OK because it is
131  * currently only used in a code path coming form ext4_iget, before
132  * the new inode has been unlocked
133  */
134 int ext4_find_inline_data_nolock(struct inode *inode)
135 {
136         struct ext4_xattr_ibody_find is = {
137                 .s = { .not_found = -ENODATA, },
138         };
139         struct ext4_xattr_info i = {
140                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
141                 .name = EXT4_XATTR_SYSTEM_DATA,
142         };
143         int error;
144
145         if (EXT4_I(inode)->i_extra_isize == 0)
146                 return 0;
147
148         error = ext4_get_inode_loc(inode, &is.iloc);
149         if (error)
150                 return error;
151
152         error = ext4_xattr_ibody_find(inode, &i, &is);
153         if (error)
154                 goto out;
155
156         if (!is.s.not_found) {
157                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
158                                         (void *)ext4_raw_inode(&is.iloc));
159                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
160                                 le32_to_cpu(is.s.here->e_value_size);
161                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
162         }
163 out:
164         brelse(is.iloc.bh);
165         return error;
166 }
167
168 static int ext4_read_inline_data(struct inode *inode, void *buffer,
169                                  unsigned int len,
170                                  struct ext4_iloc *iloc)
171 {
172         struct ext4_xattr_entry *entry;
173         struct ext4_xattr_ibody_header *header;
174         int cp_len = 0;
175         struct ext4_inode *raw_inode;
176
177         if (!len)
178                 return 0;
179
180         BUG_ON(len > EXT4_I(inode)->i_inline_size);
181
182         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
183                         len : EXT4_MIN_INLINE_DATA_SIZE;
184
185         raw_inode = ext4_raw_inode(iloc);
186         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
187
188         len -= cp_len;
189         buffer += cp_len;
190
191         if (!len)
192                 goto out;
193
194         header = IHDR(inode, raw_inode);
195         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
196                                             EXT4_I(inode)->i_inline_off);
197         len = min_t(unsigned int, len,
198                     (unsigned int)le32_to_cpu(entry->e_value_size));
199
200         memcpy(buffer,
201                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
202         cp_len += len;
203
204 out:
205         return cp_len;
206 }
207
208 /*
209  * write the buffer to the inline inode.
210  * If 'create' is set, we don't need to do the extra copy in the xattr
211  * value since it is already handled by ext4_xattr_ibody_inline_set.
212  * That saves us one memcpy.
213  */
214 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
215                                    void *buffer, loff_t pos, unsigned int len)
216 {
217         struct ext4_xattr_entry *entry;
218         struct ext4_xattr_ibody_header *header;
219         struct ext4_inode *raw_inode;
220         int cp_len = 0;
221
222         BUG_ON(!EXT4_I(inode)->i_inline_off);
223         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
224
225         raw_inode = ext4_raw_inode(iloc);
226         buffer += pos;
227
228         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
229                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
230                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
231                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
232
233                 len -= cp_len;
234                 buffer += cp_len;
235                 pos += cp_len;
236         }
237
238         if (!len)
239                 return;
240
241         pos -= EXT4_MIN_INLINE_DATA_SIZE;
242         header = IHDR(inode, raw_inode);
243         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
244                                             EXT4_I(inode)->i_inline_off);
245
246         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
247                buffer, len);
248 }
249
250 static int ext4_create_inline_data(handle_t *handle,
251                                    struct inode *inode, unsigned len)
252 {
253         int error;
254         void *value = NULL;
255         struct ext4_xattr_ibody_find is = {
256                 .s = { .not_found = -ENODATA, },
257         };
258         struct ext4_xattr_info i = {
259                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
260                 .name = EXT4_XATTR_SYSTEM_DATA,
261         };
262
263         error = ext4_get_inode_loc(inode, &is.iloc);
264         if (error)
265                 return error;
266
267         BUFFER_TRACE(is.iloc.bh, "get_write_access");
268         error = ext4_journal_get_write_access(handle, is.iloc.bh);
269         if (error)
270                 goto out;
271
272         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
273                 value = EXT4_ZERO_XATTR_VALUE;
274                 len -= EXT4_MIN_INLINE_DATA_SIZE;
275         } else {
276                 value = "";
277                 len = 0;
278         }
279
280         /* Insert the the xttr entry. */
281         i.value = value;
282         i.value_len = len;
283
284         error = ext4_xattr_ibody_find(inode, &i, &is);
285         if (error)
286                 goto out;
287
288         BUG_ON(!is.s.not_found);
289
290         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
291         if (error) {
292                 if (error == -ENOSPC)
293                         ext4_clear_inode_state(inode,
294                                                EXT4_STATE_MAY_INLINE_DATA);
295                 goto out;
296         }
297
298         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
299                 0, EXT4_MIN_INLINE_DATA_SIZE);
300
301         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
302                                       (void *)ext4_raw_inode(&is.iloc));
303         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
304         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
305         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
306         get_bh(is.iloc.bh);
307         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
308
309 out:
310         brelse(is.iloc.bh);
311         return error;
312 }
313
314 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
315                                    unsigned int len)
316 {
317         int error;
318         void *value = NULL;
319         struct ext4_xattr_ibody_find is = {
320                 .s = { .not_found = -ENODATA, },
321         };
322         struct ext4_xattr_info i = {
323                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
324                 .name = EXT4_XATTR_SYSTEM_DATA,
325         };
326
327         /* If the old space is ok, write the data directly. */
328         if (len <= EXT4_I(inode)->i_inline_size)
329                 return 0;
330
331         error = ext4_get_inode_loc(inode, &is.iloc);
332         if (error)
333                 return error;
334
335         error = ext4_xattr_ibody_find(inode, &i, &is);
336         if (error)
337                 goto out;
338
339         BUG_ON(is.s.not_found);
340
341         len -= EXT4_MIN_INLINE_DATA_SIZE;
342         value = kzalloc(len, GFP_NOFS);
343         if (!value)
344                 goto out;
345
346         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
347                                      value, len);
348         if (error == -ENODATA)
349                 goto out;
350
351         BUFFER_TRACE(is.iloc.bh, "get_write_access");
352         error = ext4_journal_get_write_access(handle, is.iloc.bh);
353         if (error)
354                 goto out;
355
356         /* Update the xttr entry. */
357         i.value = value;
358         i.value_len = len;
359
360         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
361         if (error)
362                 goto out;
363
364         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
365                                       (void *)ext4_raw_inode(&is.iloc));
366         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
367                                 le32_to_cpu(is.s.here->e_value_size);
368         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
369         get_bh(is.iloc.bh);
370         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
371
372 out:
373         kfree(value);
374         brelse(is.iloc.bh);
375         return error;
376 }
377
378 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
379                                     unsigned int len)
380 {
381         int ret, size;
382         struct ext4_inode_info *ei = EXT4_I(inode);
383
384         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
385                 return -ENOSPC;
386
387         size = ext4_get_max_inline_size(inode);
388         if (size < len)
389                 return -ENOSPC;
390
391         down_write(&EXT4_I(inode)->xattr_sem);
392
393         if (ei->i_inline_off)
394                 ret = ext4_update_inline_data(handle, inode, len);
395         else
396                 ret = ext4_create_inline_data(handle, inode, len);
397
398         up_write(&EXT4_I(inode)->xattr_sem);
399
400         return ret;
401 }
402
403 static int ext4_destroy_inline_data_nolock(handle_t *handle,
404                                            struct inode *inode)
405 {
406         struct ext4_inode_info *ei = EXT4_I(inode);
407         struct ext4_xattr_ibody_find is = {
408                 .s = { .not_found = 0, },
409         };
410         struct ext4_xattr_info i = {
411                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
412                 .name = EXT4_XATTR_SYSTEM_DATA,
413                 .value = NULL,
414                 .value_len = 0,
415         };
416         int error;
417
418         if (!ei->i_inline_off)
419                 return 0;
420
421         error = ext4_get_inode_loc(inode, &is.iloc);
422         if (error)
423                 return error;
424
425         error = ext4_xattr_ibody_find(inode, &i, &is);
426         if (error)
427                 goto out;
428
429         BUFFER_TRACE(is.iloc.bh, "get_write_access");
430         error = ext4_journal_get_write_access(handle, is.iloc.bh);
431         if (error)
432                 goto out;
433
434         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
435         if (error)
436                 goto out;
437
438         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
439                 0, EXT4_MIN_INLINE_DATA_SIZE);
440
441         if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
442                                       EXT4_FEATURE_INCOMPAT_EXTENTS)) {
443                 if (S_ISDIR(inode->i_mode) ||
444                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
445                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
446                         ext4_ext_tree_init(handle, inode);
447                 }
448         }
449         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
450
451         get_bh(is.iloc.bh);
452         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
453
454         EXT4_I(inode)->i_inline_off = 0;
455         EXT4_I(inode)->i_inline_size = 0;
456         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
457 out:
458         brelse(is.iloc.bh);
459         if (error == -ENODATA)
460                 error = 0;
461         return error;
462 }
463
464 static int ext4_read_inline_page(struct inode *inode, struct page *page)
465 {
466         void *kaddr;
467         int ret = 0;
468         size_t len;
469         struct ext4_iloc iloc;
470
471         BUG_ON(!PageLocked(page));
472         BUG_ON(!ext4_has_inline_data(inode));
473         BUG_ON(page->index);
474
475         if (!EXT4_I(inode)->i_inline_off) {
476                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
477                              inode->i_ino);
478                 goto out;
479         }
480
481         ret = ext4_get_inode_loc(inode, &iloc);
482         if (ret)
483                 goto out;
484
485         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
486         kaddr = kmap_atomic(page);
487         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
488         flush_dcache_page(page);
489         kunmap_atomic(kaddr);
490         zero_user_segment(page, len, PAGE_CACHE_SIZE);
491         SetPageUptodate(page);
492         brelse(iloc.bh);
493
494 out:
495         return ret;
496 }
497
498 int ext4_readpage_inline(struct inode *inode, struct page *page)
499 {
500         int ret = 0;
501
502         down_read(&EXT4_I(inode)->xattr_sem);
503         if (!ext4_has_inline_data(inode)) {
504                 up_read(&EXT4_I(inode)->xattr_sem);
505                 return -EAGAIN;
506         }
507
508         /*
509          * Current inline data can only exist in the 1st page,
510          * So for all the other pages, just set them uptodate.
511          */
512         if (!page->index)
513                 ret = ext4_read_inline_page(inode, page);
514         else if (!PageUptodate(page)) {
515                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
516                 SetPageUptodate(page);
517         }
518
519         up_read(&EXT4_I(inode)->xattr_sem);
520
521         unlock_page(page);
522         return ret >= 0 ? 0 : ret;
523 }
524
525 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
526                                               struct inode *inode,
527                                               unsigned flags)
528 {
529         int ret, needed_blocks;
530         handle_t *handle = NULL;
531         int retries = 0, sem_held = 0;
532         struct page *page = NULL;
533         unsigned from, to;
534         struct ext4_iloc iloc;
535
536         if (!ext4_has_inline_data(inode)) {
537                 /*
538                  * clear the flag so that no new write
539                  * will trap here again.
540                  */
541                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
542                 return 0;
543         }
544
545         needed_blocks = ext4_writepage_trans_blocks(inode);
546
547         ret = ext4_get_inode_loc(inode, &iloc);
548         if (ret)
549                 return ret;
550
551 retry:
552         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
553         if (IS_ERR(handle)) {
554                 ret = PTR_ERR(handle);
555                 handle = NULL;
556                 goto out;
557         }
558
559         /* We cannot recurse into the filesystem as the transaction is already
560          * started */
561         flags |= AOP_FLAG_NOFS;
562
563         page = grab_cache_page_write_begin(mapping, 0, flags);
564         if (!page) {
565                 ret = -ENOMEM;
566                 goto out;
567         }
568
569         down_write(&EXT4_I(inode)->xattr_sem);
570         sem_held = 1;
571         /* If some one has already done this for us, just exit. */
572         if (!ext4_has_inline_data(inode)) {
573                 ret = 0;
574                 goto out;
575         }
576
577         from = 0;
578         to = ext4_get_inline_size(inode);
579         if (!PageUptodate(page)) {
580                 ret = ext4_read_inline_page(inode, page);
581                 if (ret < 0)
582                         goto out;
583         }
584
585         ret = ext4_destroy_inline_data_nolock(handle, inode);
586         if (ret)
587                 goto out;
588
589         if (ext4_should_dioread_nolock(inode))
590                 ret = __block_write_begin(page, from, to, ext4_get_block_write);
591         else
592                 ret = __block_write_begin(page, from, to, ext4_get_block);
593
594         if (!ret && ext4_should_journal_data(inode)) {
595                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
596                                              from, to, NULL,
597                                              do_journal_get_write_access);
598         }
599
600         if (ret) {
601                 unlock_page(page);
602                 page_cache_release(page);
603                 ext4_orphan_add(handle, inode);
604                 up_write(&EXT4_I(inode)->xattr_sem);
605                 sem_held = 0;
606                 ext4_journal_stop(handle);
607                 handle = NULL;
608                 ext4_truncate_failed_write(inode);
609                 /*
610                  * If truncate failed early the inode might
611                  * still be on the orphan list; we need to
612                  * make sure the inode is removed from the
613                  * orphan list in that case.
614                  */
615                 if (inode->i_nlink)
616                         ext4_orphan_del(NULL, inode);
617         }
618
619         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
620                 goto retry;
621
622         block_commit_write(page, from, to);
623 out:
624         if (page) {
625                 unlock_page(page);
626                 page_cache_release(page);
627         }
628         if (sem_held)
629                 up_write(&EXT4_I(inode)->xattr_sem);
630         if (handle)
631                 ext4_journal_stop(handle);
632         brelse(iloc.bh);
633         return ret;
634 }
635
636 /*
637  * Try to write data in the inode.
638  * If the inode has inline data, check whether the new write can be
639  * in the inode also. If not, create the page the handle, move the data
640  * to the page make it update and let the later codes create extent for it.
641  */
642 int ext4_try_to_write_inline_data(struct address_space *mapping,
643                                   struct inode *inode,
644                                   loff_t pos, unsigned len,
645                                   unsigned flags,
646                                   struct page **pagep)
647 {
648         int ret;
649         handle_t *handle;
650         struct page *page;
651         struct ext4_iloc iloc;
652
653         if (pos + len > ext4_get_max_inline_size(inode))
654                 goto convert;
655
656         ret = ext4_get_inode_loc(inode, &iloc);
657         if (ret)
658                 return ret;
659
660         /*
661          * The possible write could happen in the inode,
662          * so try to reserve the space in inode first.
663          */
664         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
665         if (IS_ERR(handle)) {
666                 ret = PTR_ERR(handle);
667                 handle = NULL;
668                 goto out;
669         }
670
671         ret = ext4_prepare_inline_data(handle, inode, pos + len);
672         if (ret && ret != -ENOSPC)
673                 goto out;
674
675         /* We don't have space in inline inode, so convert it to extent. */
676         if (ret == -ENOSPC) {
677                 ext4_journal_stop(handle);
678                 brelse(iloc.bh);
679                 goto convert;
680         }
681
682         flags |= AOP_FLAG_NOFS;
683
684         page = grab_cache_page_write_begin(mapping, 0, flags);
685         if (!page) {
686                 ret = -ENOMEM;
687                 goto out;
688         }
689
690         *pagep = page;
691         down_read(&EXT4_I(inode)->xattr_sem);
692         if (!ext4_has_inline_data(inode)) {
693                 ret = 0;
694                 unlock_page(page);
695                 page_cache_release(page);
696                 goto out_up_read;
697         }
698
699         if (!PageUptodate(page)) {
700                 ret = ext4_read_inline_page(inode, page);
701                 if (ret < 0)
702                         goto out_up_read;
703         }
704
705         ret = 1;
706         handle = NULL;
707 out_up_read:
708         up_read(&EXT4_I(inode)->xattr_sem);
709 out:
710         if (handle)
711                 ext4_journal_stop(handle);
712         brelse(iloc.bh);
713         return ret;
714 convert:
715         return ext4_convert_inline_data_to_extent(mapping,
716                                                   inode, flags);
717 }
718
719 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
720                                unsigned copied, struct page *page)
721 {
722         int ret;
723         void *kaddr;
724         struct ext4_iloc iloc;
725
726         if (unlikely(copied < len)) {
727                 if (!PageUptodate(page)) {
728                         copied = 0;
729                         goto out;
730                 }
731         }
732
733         ret = ext4_get_inode_loc(inode, &iloc);
734         if (ret) {
735                 ext4_std_error(inode->i_sb, ret);
736                 copied = 0;
737                 goto out;
738         }
739
740         down_write(&EXT4_I(inode)->xattr_sem);
741         BUG_ON(!ext4_has_inline_data(inode));
742
743         kaddr = kmap_atomic(page);
744         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
745         kunmap_atomic(kaddr);
746         SetPageUptodate(page);
747         /* clear page dirty so that writepages wouldn't work for us. */
748         ClearPageDirty(page);
749
750         up_write(&EXT4_I(inode)->xattr_sem);
751         brelse(iloc.bh);
752 out:
753         return copied;
754 }
755
756 struct buffer_head *
757 ext4_journalled_write_inline_data(struct inode *inode,
758                                   unsigned len,
759                                   struct page *page)
760 {
761         int ret;
762         void *kaddr;
763         struct ext4_iloc iloc;
764
765         ret = ext4_get_inode_loc(inode, &iloc);
766         if (ret) {
767                 ext4_std_error(inode->i_sb, ret);
768                 return NULL;
769         }
770
771         down_write(&EXT4_I(inode)->xattr_sem);
772         kaddr = kmap_atomic(page);
773         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
774         kunmap_atomic(kaddr);
775         up_write(&EXT4_I(inode)->xattr_sem);
776
777         return iloc.bh;
778 }
779
780 /*
781  * Try to make the page cache and handle ready for the inline data case.
782  * We can call this function in 2 cases:
783  * 1. The inode is created and the first write exceeds inline size. We can
784  *    clear the inode state safely.
785  * 2. The inode has inline data, then we need to read the data, make it
786  *    update and dirty so that ext4_da_writepages can handle it. We don't
787  *    need to start the journal since the file's metatdata isn't changed now.
788  */
789 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
790                                                  struct inode *inode,
791                                                  unsigned flags,
792                                                  void **fsdata)
793 {
794         int ret = 0, inline_size;
795         struct page *page;
796
797         page = grab_cache_page_write_begin(mapping, 0, flags);
798         if (!page)
799                 return -ENOMEM;
800
801         down_read(&EXT4_I(inode)->xattr_sem);
802         if (!ext4_has_inline_data(inode)) {
803                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
804                 goto out;
805         }
806
807         inline_size = ext4_get_inline_size(inode);
808
809         if (!PageUptodate(page)) {
810                 ret = ext4_read_inline_page(inode, page);
811                 if (ret < 0)
812                         goto out;
813         }
814
815         ret = __block_write_begin(page, 0, inline_size,
816                                   ext4_da_get_block_prep);
817         if (ret) {
818                 ext4_truncate_failed_write(inode);
819                 goto out;
820         }
821
822         SetPageDirty(page);
823         SetPageUptodate(page);
824         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
825         *fsdata = (void *)CONVERT_INLINE_DATA;
826
827 out:
828         up_read(&EXT4_I(inode)->xattr_sem);
829         if (page) {
830                 unlock_page(page);
831                 page_cache_release(page);
832         }
833         return ret;
834 }
835
836 /*
837  * Prepare the write for the inline data.
838  * If the the data can be written into the inode, we just read
839  * the page and make it uptodate, and start the journal.
840  * Otherwise read the page, makes it dirty so that it can be
841  * handle in writepages(the i_disksize update is left to the
842  * normal ext4_da_write_end).
843  */
844 int ext4_da_write_inline_data_begin(struct address_space *mapping,
845                                     struct inode *inode,
846                                     loff_t pos, unsigned len,
847                                     unsigned flags,
848                                     struct page **pagep,
849                                     void **fsdata)
850 {
851         int ret, inline_size;
852         handle_t *handle;
853         struct page *page;
854         struct ext4_iloc iloc;
855         int retries;
856
857         ret = ext4_get_inode_loc(inode, &iloc);
858         if (ret)
859                 return ret;
860
861 retry_journal:
862         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
863         if (IS_ERR(handle)) {
864                 ret = PTR_ERR(handle);
865                 goto out;
866         }
867
868         inline_size = ext4_get_max_inline_size(inode);
869
870         ret = -ENOSPC;
871         if (inline_size >= pos + len) {
872                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
873                 if (ret && ret != -ENOSPC)
874                         goto out_journal;
875         }
876
877         if (ret == -ENOSPC) {
878                 ret = ext4_da_convert_inline_data_to_extent(mapping,
879                                                             inode,
880                                                             flags,
881                                                             fsdata);
882                 ext4_journal_stop(handle);
883                 if (ret == -ENOSPC &&
884                     ext4_should_retry_alloc(inode->i_sb, &retries))
885                         goto retry_journal;
886                 goto out;
887         }
888
889         /*
890          * We cannot recurse into the filesystem as the transaction
891          * is already started.
892          */
893         flags |= AOP_FLAG_NOFS;
894
895         page = grab_cache_page_write_begin(mapping, 0, flags);
896         if (!page) {
897                 ret = -ENOMEM;
898                 goto out_journal;
899         }
900
901         down_read(&EXT4_I(inode)->xattr_sem);
902         if (!ext4_has_inline_data(inode)) {
903                 ret = 0;
904                 goto out_release_page;
905         }
906
907         if (!PageUptodate(page)) {
908                 ret = ext4_read_inline_page(inode, page);
909                 if (ret < 0)
910                         goto out_release_page;
911         }
912
913         up_read(&EXT4_I(inode)->xattr_sem);
914         *pagep = page;
915         brelse(iloc.bh);
916         return 1;
917 out_release_page:
918         up_read(&EXT4_I(inode)->xattr_sem);
919         unlock_page(page);
920         page_cache_release(page);
921 out_journal:
922         ext4_journal_stop(handle);
923 out:
924         brelse(iloc.bh);
925         return ret;
926 }
927
928 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
929                                   unsigned len, unsigned copied,
930                                   struct page *page)
931 {
932         int i_size_changed = 0;
933
934         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
935
936         /*
937          * No need to use i_size_read() here, the i_size
938          * cannot change under us because we hold i_mutex.
939          *
940          * But it's important to update i_size while still holding page lock:
941          * page writeout could otherwise come in and zero beyond i_size.
942          */
943         if (pos+copied > inode->i_size) {
944                 i_size_write(inode, pos+copied);
945                 i_size_changed = 1;
946         }
947         unlock_page(page);
948         page_cache_release(page);
949
950         /*
951          * Don't mark the inode dirty under page lock. First, it unnecessarily
952          * makes the holding time of page lock longer. Second, it forces lock
953          * ordering of page lock and transaction start for journaling
954          * filesystems.
955          */
956         if (i_size_changed)
957                 mark_inode_dirty(inode);
958
959         return copied;
960 }
961
962 #ifdef INLINE_DIR_DEBUG
963 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
964                           void *inline_start, int inline_size)
965 {
966         int offset;
967         unsigned short de_len;
968         struct ext4_dir_entry_2 *de = inline_start;
969         void *dlimit = inline_start + inline_size;
970
971         trace_printk("inode %lu\n", dir->i_ino);
972         offset = 0;
973         while ((void *)de < dlimit) {
974                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
975                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
976                              offset, de_len, de->name_len, de->name,
977                              de->name_len, le32_to_cpu(de->inode));
978                 if (ext4_check_dir_entry(dir, NULL, de, bh,
979                                          inline_start, inline_size, offset))
980                         BUG();
981
982                 offset += de_len;
983                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
984         }
985 }
986 #else
987 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
988 #endif
989
990 /*
991  * Add a new entry into a inline dir.
992  * It will return -ENOSPC if no space is available, and -EIO
993  * and -EEXIST if directory entry already exists.
994  */
995 static int ext4_add_dirent_to_inline(handle_t *handle,
996                                      struct dentry *dentry,
997                                      struct inode *inode,
998                                      struct ext4_iloc *iloc,
999                                      void *inline_start, int inline_size)
1000 {
1001         struct inode    *dir = dentry->d_parent->d_inode;
1002         const char      *name = dentry->d_name.name;
1003         int             namelen = dentry->d_name.len;
1004         int             err;
1005         struct ext4_dir_entry_2 *de;
1006
1007         err = ext4_find_dest_de(dir, inode, iloc->bh,
1008                                 inline_start, inline_size,
1009                                 name, namelen, &de);
1010         if (err)
1011                 return err;
1012
1013         BUFFER_TRACE(iloc->bh, "get_write_access");
1014         err = ext4_journal_get_write_access(handle, iloc->bh);
1015         if (err)
1016                 return err;
1017         ext4_insert_dentry(inode, de, inline_size, name, namelen);
1018
1019         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1020
1021         /*
1022          * XXX shouldn't update any times until successful
1023          * completion of syscall, but too many callers depend
1024          * on this.
1025          *
1026          * XXX similarly, too many callers depend on
1027          * ext4_new_inode() setting the times, but error
1028          * recovery deletes the inode, so the worst that can
1029          * happen is that the times are slightly out of date
1030          * and/or different from the directory change time.
1031          */
1032         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1033         ext4_update_dx_flag(dir);
1034         dir->i_version++;
1035         ext4_mark_inode_dirty(handle, dir);
1036         return 1;
1037 }
1038
1039 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1040                                        struct ext4_iloc *iloc)
1041 {
1042         struct ext4_xattr_entry *entry;
1043         struct ext4_xattr_ibody_header *header;
1044
1045         BUG_ON(!EXT4_I(inode)->i_inline_off);
1046
1047         header = IHDR(inode, ext4_raw_inode(iloc));
1048         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1049                                             EXT4_I(inode)->i_inline_off);
1050
1051         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1052 }
1053
1054 /* Set the final de to cover the whole block. */
1055 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1056 {
1057         struct ext4_dir_entry_2 *de, *prev_de;
1058         void *limit;
1059         int de_len;
1060
1061         de = (struct ext4_dir_entry_2 *)de_buf;
1062         if (old_size) {
1063                 limit = de_buf + old_size;
1064                 do {
1065                         prev_de = de;
1066                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1067                         de_buf += de_len;
1068                         de = (struct ext4_dir_entry_2 *)de_buf;
1069                 } while (de_buf < limit);
1070
1071                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1072                                                         old_size, new_size);
1073         } else {
1074                 /* this is just created, so create an empty entry. */
1075                 de->inode = 0;
1076                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1077         }
1078 }
1079
1080 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1081                                   struct ext4_iloc *iloc)
1082 {
1083         int ret;
1084         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1085         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1086
1087         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1088                 return -ENOSPC;
1089
1090         ret = ext4_update_inline_data(handle, dir,
1091                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1092         if (ret)
1093                 return ret;
1094
1095         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1096                              EXT4_I(dir)->i_inline_size -
1097                                                 EXT4_MIN_INLINE_DATA_SIZE);
1098         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1099         return 0;
1100 }
1101
1102 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1103                                      struct ext4_iloc *iloc,
1104                                      void *buf, int inline_size)
1105 {
1106         ext4_create_inline_data(handle, inode, inline_size);
1107         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1108         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1109 }
1110
1111 static int ext4_finish_convert_inline_dir(handle_t *handle,
1112                                           struct inode *inode,
1113                                           struct buffer_head *dir_block,
1114                                           void *buf,
1115                                           int inline_size)
1116 {
1117         int err, csum_size = 0, header_size = 0;
1118         struct ext4_dir_entry_2 *de;
1119         struct ext4_dir_entry_tail *t;
1120         void *target = dir_block->b_data;
1121
1122         /*
1123          * First create "." and ".." and then copy the dir information
1124          * back to the block.
1125          */
1126         de = (struct ext4_dir_entry_2 *)target;
1127         de = ext4_init_dot_dotdot(inode, de,
1128                 inode->i_sb->s_blocksize, csum_size,
1129                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1130         header_size = (void *)de - target;
1131
1132         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1133                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1134
1135         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1136                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1137                 csum_size = sizeof(struct ext4_dir_entry_tail);
1138
1139         inode->i_size = inode->i_sb->s_blocksize;
1140         i_size_write(inode, inode->i_sb->s_blocksize);
1141         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1142         ext4_update_final_de(dir_block->b_data,
1143                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1144                         inode->i_sb->s_blocksize - csum_size);
1145
1146         if (csum_size) {
1147                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1148                                      inode->i_sb->s_blocksize);
1149                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1150         }
1151         set_buffer_uptodate(dir_block);
1152         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1153         if (err)
1154                 goto out;
1155         set_buffer_verified(dir_block);
1156 out:
1157         return err;
1158 }
1159
1160 static int ext4_convert_inline_data_nolock(handle_t *handle,
1161                                            struct inode *inode,
1162                                            struct ext4_iloc *iloc)
1163 {
1164         int error;
1165         void *buf = NULL;
1166         struct buffer_head *data_bh = NULL;
1167         struct ext4_map_blocks map;
1168         int inline_size;
1169
1170         inline_size = ext4_get_inline_size(inode);
1171         buf = kmalloc(inline_size, GFP_NOFS);
1172         if (!buf) {
1173                 error = -ENOMEM;
1174                 goto out;
1175         }
1176
1177         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1178         if (error < 0)
1179                 goto out;
1180
1181         error = ext4_destroy_inline_data_nolock(handle, inode);
1182         if (error)
1183                 goto out;
1184
1185         map.m_lblk = 0;
1186         map.m_len = 1;
1187         map.m_flags = 0;
1188         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1189         if (error < 0)
1190                 goto out_restore;
1191         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1192                 error = -EIO;
1193                 goto out_restore;
1194         }
1195
1196         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1197         if (!data_bh) {
1198                 error = -ENOMEM;
1199                 goto out_restore;
1200         }
1201
1202         lock_buffer(data_bh);
1203         error = ext4_journal_get_create_access(handle, data_bh);
1204         if (error) {
1205                 unlock_buffer(data_bh);
1206                 error = -EIO;
1207                 goto out_restore;
1208         }
1209         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1210
1211         if (!S_ISDIR(inode->i_mode)) {
1212                 memcpy(data_bh->b_data, buf, inline_size);
1213                 set_buffer_uptodate(data_bh);
1214                 error = ext4_handle_dirty_metadata(handle,
1215                                                    inode, data_bh);
1216         } else {
1217                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1218                                                        buf, inline_size);
1219         }
1220
1221         unlock_buffer(data_bh);
1222 out_restore:
1223         if (error)
1224                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1225
1226 out:
1227         brelse(data_bh);
1228         kfree(buf);
1229         return error;
1230 }
1231
1232 /*
1233  * Try to add the new entry to the inline data.
1234  * If succeeds, return 0. If not, extended the inline dir and copied data to
1235  * the new created block.
1236  */
1237 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1238                               struct inode *inode)
1239 {
1240         int ret, inline_size;
1241         void *inline_start;
1242         struct ext4_iloc iloc;
1243         struct inode *dir = dentry->d_parent->d_inode;
1244
1245         ret = ext4_get_inode_loc(dir, &iloc);
1246         if (ret)
1247                 return ret;
1248
1249         down_write(&EXT4_I(dir)->xattr_sem);
1250         if (!ext4_has_inline_data(dir))
1251                 goto out;
1252
1253         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1254                                                  EXT4_INLINE_DOTDOT_SIZE;
1255         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1256
1257         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1258                                         inline_start, inline_size);
1259         if (ret != -ENOSPC)
1260                 goto out;
1261
1262         /* check whether it can be inserted to inline xattr space. */
1263         inline_size = EXT4_I(dir)->i_inline_size -
1264                         EXT4_MIN_INLINE_DATA_SIZE;
1265         if (!inline_size) {
1266                 /* Try to use the xattr space.*/
1267                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1268                 if (ret && ret != -ENOSPC)
1269                         goto out;
1270
1271                 inline_size = EXT4_I(dir)->i_inline_size -
1272                                 EXT4_MIN_INLINE_DATA_SIZE;
1273         }
1274
1275         if (inline_size) {
1276                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1277
1278                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1279                                                 inline_start, inline_size);
1280
1281                 if (ret != -ENOSPC)
1282                         goto out;
1283         }
1284
1285         /*
1286          * The inline space is filled up, so create a new block for it.
1287          * As the extent tree will be created, we have to save the inline
1288          * dir first.
1289          */
1290         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1291
1292 out:
1293         ext4_mark_inode_dirty(handle, dir);
1294         up_write(&EXT4_I(dir)->xattr_sem);
1295         brelse(iloc.bh);
1296         return ret;
1297 }
1298
1299 /*
1300  * This function fills a red-black tree with information from an
1301  * inlined dir.  It returns the number directory entries loaded
1302  * into the tree.  If there is an error it is returned in err.
1303  */
1304 int htree_inlinedir_to_tree(struct file *dir_file,
1305                             struct inode *dir, ext4_lblk_t block,
1306                             struct dx_hash_info *hinfo,
1307                             __u32 start_hash, __u32 start_minor_hash,
1308                             int *has_inline_data)
1309 {
1310         int err = 0, count = 0;
1311         unsigned int parent_ino;
1312         int pos;
1313         struct ext4_dir_entry_2 *de;
1314         struct inode *inode = file_inode(dir_file);
1315         int ret, inline_size = 0;
1316         struct ext4_iloc iloc;
1317         void *dir_buf = NULL;
1318         struct ext4_dir_entry_2 fake;
1319
1320         ret = ext4_get_inode_loc(inode, &iloc);
1321         if (ret)
1322                 return ret;
1323
1324         down_read(&EXT4_I(inode)->xattr_sem);
1325         if (!ext4_has_inline_data(inode)) {
1326                 up_read(&EXT4_I(inode)->xattr_sem);
1327                 *has_inline_data = 0;
1328                 goto out;
1329         }
1330
1331         inline_size = ext4_get_inline_size(inode);
1332         dir_buf = kmalloc(inline_size, GFP_NOFS);
1333         if (!dir_buf) {
1334                 ret = -ENOMEM;
1335                 up_read(&EXT4_I(inode)->xattr_sem);
1336                 goto out;
1337         }
1338
1339         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1340         up_read(&EXT4_I(inode)->xattr_sem);
1341         if (ret < 0)
1342                 goto out;
1343
1344         pos = 0;
1345         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1346         while (pos < inline_size) {
1347                 /*
1348                  * As inlined dir doesn't store any information about '.' and
1349                  * only the inode number of '..' is stored, we have to handle
1350                  * them differently.
1351                  */
1352                 if (pos == 0) {
1353                         fake.inode = cpu_to_le32(inode->i_ino);
1354                         fake.name_len = 1;
1355                         strcpy(fake.name, ".");
1356                         fake.rec_len = ext4_rec_len_to_disk(
1357                                                 EXT4_DIR_REC_LEN(fake.name_len),
1358                                                 inline_size);
1359                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1360                         de = &fake;
1361                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1362                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1363                         fake.inode = cpu_to_le32(parent_ino);
1364                         fake.name_len = 2;
1365                         strcpy(fake.name, "..");
1366                         fake.rec_len = ext4_rec_len_to_disk(
1367                                                 EXT4_DIR_REC_LEN(fake.name_len),
1368                                                 inline_size);
1369                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1370                         de = &fake;
1371                         pos = EXT4_INLINE_DOTDOT_SIZE;
1372                 } else {
1373                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1374                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1375                         if (ext4_check_dir_entry(inode, dir_file, de,
1376                                          iloc.bh, dir_buf,
1377                                          inline_size, pos)) {
1378                                 ret = count;
1379                                 goto out;
1380                         }
1381                 }
1382
1383                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1384                 if ((hinfo->hash < start_hash) ||
1385                     ((hinfo->hash == start_hash) &&
1386                      (hinfo->minor_hash < start_minor_hash)))
1387                         continue;
1388                 if (de->inode == 0)
1389                         continue;
1390                 err = ext4_htree_store_dirent(dir_file,
1391                                    hinfo->hash, hinfo->minor_hash, de);
1392                 if (err) {
1393                         count = err;
1394                         goto out;
1395                 }
1396                 count++;
1397         }
1398         ret = count;
1399 out:
1400         kfree(dir_buf);
1401         brelse(iloc.bh);
1402         return ret;
1403 }
1404
1405 /*
1406  * So this function is called when the volume is mkfsed with
1407  * dir_index disabled. In order to keep f_pos persistent
1408  * after we convert from an inlined dir to a blocked based,
1409  * we just pretend that we are a normal dir and return the
1410  * offset as if '.' and '..' really take place.
1411  *
1412  */
1413 int ext4_read_inline_dir(struct file *file,
1414                          struct dir_context *ctx,
1415                          int *has_inline_data)
1416 {
1417         unsigned int offset, parent_ino;
1418         int i;
1419         struct ext4_dir_entry_2 *de;
1420         struct super_block *sb;
1421         struct inode *inode = file_inode(file);
1422         int ret, inline_size = 0;
1423         struct ext4_iloc iloc;
1424         void *dir_buf = NULL;
1425         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1426
1427         ret = ext4_get_inode_loc(inode, &iloc);
1428         if (ret)
1429                 return ret;
1430
1431         down_read(&EXT4_I(inode)->xattr_sem);
1432         if (!ext4_has_inline_data(inode)) {
1433                 up_read(&EXT4_I(inode)->xattr_sem);
1434                 *has_inline_data = 0;
1435                 goto out;
1436         }
1437
1438         inline_size = ext4_get_inline_size(inode);
1439         dir_buf = kmalloc(inline_size, GFP_NOFS);
1440         if (!dir_buf) {
1441                 ret = -ENOMEM;
1442                 up_read(&EXT4_I(inode)->xattr_sem);
1443                 goto out;
1444         }
1445
1446         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1447         up_read(&EXT4_I(inode)->xattr_sem);
1448         if (ret < 0)
1449                 goto out;
1450
1451         ret = 0;
1452         sb = inode->i_sb;
1453         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1454         offset = ctx->pos;
1455
1456         /*
1457          * dotdot_offset and dotdot_size is the real offset and
1458          * size for ".." and "." if the dir is block based while
1459          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1460          * So we will use extra_offset and extra_size to indicate them
1461          * during the inline dir iteration.
1462          */
1463         dotdot_offset = EXT4_DIR_REC_LEN(1);
1464         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1465         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1466         extra_size = extra_offset + inline_size;
1467
1468         /*
1469          * If the version has changed since the last call to
1470          * readdir(2), then we might be pointing to an invalid
1471          * dirent right now.  Scan from the start of the inline
1472          * dir to make sure.
1473          */
1474         if (file->f_version != inode->i_version) {
1475                 for (i = 0; i < extra_size && i < offset;) {
1476                         /*
1477                          * "." is with offset 0 and
1478                          * ".." is dotdot_offset.
1479                          */
1480                         if (!i) {
1481                                 i = dotdot_offset;
1482                                 continue;
1483                         } else if (i == dotdot_offset) {
1484                                 i = dotdot_size;
1485                                 continue;
1486                         }
1487                         /* for other entry, the real offset in
1488                          * the buf has to be tuned accordingly.
1489                          */
1490                         de = (struct ext4_dir_entry_2 *)
1491                                 (dir_buf + i - extra_offset);
1492                         /* It's too expensive to do a full
1493                          * dirent test each time round this
1494                          * loop, but we do have to test at
1495                          * least that it is non-zero.  A
1496                          * failure will be detected in the
1497                          * dirent test below. */
1498                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1499                                 < EXT4_DIR_REC_LEN(1))
1500                                 break;
1501                         i += ext4_rec_len_from_disk(de->rec_len,
1502                                                     extra_size);
1503                 }
1504                 offset = i;
1505                 ctx->pos = offset;
1506                 file->f_version = inode->i_version;
1507         }
1508
1509         while (ctx->pos < extra_size) {
1510                 if (ctx->pos == 0) {
1511                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1512                                 goto out;
1513                         ctx->pos = dotdot_offset;
1514                         continue;
1515                 }
1516
1517                 if (ctx->pos == dotdot_offset) {
1518                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1519                                 goto out;
1520                         ctx->pos = dotdot_size;
1521                         continue;
1522                 }
1523
1524                 de = (struct ext4_dir_entry_2 *)
1525                         (dir_buf + ctx->pos - extra_offset);
1526                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1527                                          extra_size, ctx->pos))
1528                         goto out;
1529                 if (le32_to_cpu(de->inode)) {
1530                         if (!dir_emit(ctx, de->name, de->name_len,
1531                                       le32_to_cpu(de->inode),
1532                                       get_dtype(sb, de->file_type)))
1533                                 goto out;
1534                 }
1535                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1536         }
1537 out:
1538         kfree(dir_buf);
1539         brelse(iloc.bh);
1540         return ret;
1541 }
1542
1543 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1544                                         struct ext4_dir_entry_2 **parent_de,
1545                                         int *retval)
1546 {
1547         struct ext4_iloc iloc;
1548
1549         *retval = ext4_get_inode_loc(inode, &iloc);
1550         if (*retval)
1551                 return NULL;
1552
1553         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1554
1555         return iloc.bh;
1556 }
1557
1558 /*
1559  * Try to create the inline data for the new dir.
1560  * If it succeeds, return 0, otherwise return the error.
1561  * In case of ENOSPC, the caller should create the normal disk layout dir.
1562  */
1563 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1564                                struct inode *inode)
1565 {
1566         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1567         struct ext4_iloc iloc;
1568         struct ext4_dir_entry_2 *de;
1569
1570         ret = ext4_get_inode_loc(inode, &iloc);
1571         if (ret)
1572                 return ret;
1573
1574         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1575         if (ret)
1576                 goto out;
1577
1578         /*
1579          * For inline dir, we only save the inode information for the ".."
1580          * and create a fake dentry to cover the left space.
1581          */
1582         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1583         de->inode = cpu_to_le32(parent->i_ino);
1584         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1585         de->inode = 0;
1586         de->rec_len = ext4_rec_len_to_disk(
1587                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1588                                 inline_size);
1589         set_nlink(inode, 2);
1590         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1591 out:
1592         brelse(iloc.bh);
1593         return ret;
1594 }
1595
1596 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1597                                         const struct qstr *d_name,
1598                                         struct ext4_dir_entry_2 **res_dir,
1599                                         int *has_inline_data)
1600 {
1601         int ret;
1602         struct ext4_iloc iloc;
1603         void *inline_start;
1604         int inline_size;
1605
1606         if (ext4_get_inode_loc(dir, &iloc))
1607                 return NULL;
1608
1609         down_read(&EXT4_I(dir)->xattr_sem);
1610         if (!ext4_has_inline_data(dir)) {
1611                 *has_inline_data = 0;
1612                 goto out;
1613         }
1614
1615         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1616                                                 EXT4_INLINE_DOTDOT_SIZE;
1617         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1618         ret = search_dir(iloc.bh, inline_start, inline_size,
1619                          dir, d_name, 0, res_dir);
1620         if (ret == 1)
1621                 goto out_find;
1622         if (ret < 0)
1623                 goto out;
1624
1625         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1626                 goto out;
1627
1628         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1629         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1630
1631         ret = search_dir(iloc.bh, inline_start, inline_size,
1632                          dir, d_name, 0, res_dir);
1633         if (ret == 1)
1634                 goto out_find;
1635
1636 out:
1637         brelse(iloc.bh);
1638         iloc.bh = NULL;
1639 out_find:
1640         up_read(&EXT4_I(dir)->xattr_sem);
1641         return iloc.bh;
1642 }
1643
1644 int ext4_delete_inline_entry(handle_t *handle,
1645                              struct inode *dir,
1646                              struct ext4_dir_entry_2 *de_del,
1647                              struct buffer_head *bh,
1648                              int *has_inline_data)
1649 {
1650         int err, inline_size;
1651         struct ext4_iloc iloc;
1652         void *inline_start;
1653
1654         err = ext4_get_inode_loc(dir, &iloc);
1655         if (err)
1656                 return err;
1657
1658         down_write(&EXT4_I(dir)->xattr_sem);
1659         if (!ext4_has_inline_data(dir)) {
1660                 *has_inline_data = 0;
1661                 goto out;
1662         }
1663
1664         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1665                 EXT4_MIN_INLINE_DATA_SIZE) {
1666                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1667                                         EXT4_INLINE_DOTDOT_SIZE;
1668                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1669                                 EXT4_INLINE_DOTDOT_SIZE;
1670         } else {
1671                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1672                 inline_size = ext4_get_inline_size(dir) -
1673                                 EXT4_MIN_INLINE_DATA_SIZE;
1674         }
1675
1676         BUFFER_TRACE(bh, "get_write_access");
1677         err = ext4_journal_get_write_access(handle, bh);
1678         if (err)
1679                 goto out;
1680
1681         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1682                                         inline_start, inline_size, 0);
1683         if (err)
1684                 goto out;
1685
1686         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1687         err = ext4_mark_inode_dirty(handle, dir);
1688         if (unlikely(err))
1689                 goto out;
1690
1691         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1692 out:
1693         up_write(&EXT4_I(dir)->xattr_sem);
1694         brelse(iloc.bh);
1695         if (err != -ENOENT)
1696                 ext4_std_error(dir->i_sb, err);
1697         return err;
1698 }
1699
1700 /*
1701  * Get the inline dentry at offset.
1702  */
1703 static inline struct ext4_dir_entry_2 *
1704 ext4_get_inline_entry(struct inode *inode,
1705                       struct ext4_iloc *iloc,
1706                       unsigned int offset,
1707                       void **inline_start,
1708                       int *inline_size)
1709 {
1710         void *inline_pos;
1711
1712         BUG_ON(offset > ext4_get_inline_size(inode));
1713
1714         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1715                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1716                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1717         } else {
1718                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1719                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1720                 *inline_size = ext4_get_inline_size(inode) -
1721                                 EXT4_MIN_INLINE_DATA_SIZE;
1722         }
1723
1724         if (inline_start)
1725                 *inline_start = inline_pos;
1726         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1727 }
1728
1729 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1730 {
1731         int err, inline_size;
1732         struct ext4_iloc iloc;
1733         void *inline_pos;
1734         unsigned int offset;
1735         struct ext4_dir_entry_2 *de;
1736         int ret = 1;
1737
1738         err = ext4_get_inode_loc(dir, &iloc);
1739         if (err) {
1740                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1741                                  err, dir->i_ino);
1742                 return 1;
1743         }
1744
1745         down_read(&EXT4_I(dir)->xattr_sem);
1746         if (!ext4_has_inline_data(dir)) {
1747                 *has_inline_data = 0;
1748                 goto out;
1749         }
1750
1751         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1752         if (!le32_to_cpu(de->inode)) {
1753                 ext4_warning(dir->i_sb,
1754                              "bad inline directory (dir #%lu) - no `..'",
1755                              dir->i_ino);
1756                 ret = 1;
1757                 goto out;
1758         }
1759
1760         offset = EXT4_INLINE_DOTDOT_SIZE;
1761         while (offset < dir->i_size) {
1762                 de = ext4_get_inline_entry(dir, &iloc, offset,
1763                                            &inline_pos, &inline_size);
1764                 if (ext4_check_dir_entry(dir, NULL, de,
1765                                          iloc.bh, inline_pos,
1766                                          inline_size, offset)) {
1767                         ext4_warning(dir->i_sb,
1768                                      "bad inline directory (dir #%lu) - "
1769                                      "inode %u, rec_len %u, name_len %d"
1770                                      "inline size %d\n",
1771                                      dir->i_ino, le32_to_cpu(de->inode),
1772                                      le16_to_cpu(de->rec_len), de->name_len,
1773                                      inline_size);
1774                         ret = 1;
1775                         goto out;
1776                 }
1777                 if (le32_to_cpu(de->inode)) {
1778                         ret = 0;
1779                         goto out;
1780                 }
1781                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1782         }
1783
1784 out:
1785         up_read(&EXT4_I(dir)->xattr_sem);
1786         brelse(iloc.bh);
1787         return ret;
1788 }
1789
1790 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1791 {
1792         int ret;
1793
1794         down_write(&EXT4_I(inode)->xattr_sem);
1795         ret = ext4_destroy_inline_data_nolock(handle, inode);
1796         up_write(&EXT4_I(inode)->xattr_sem);
1797
1798         return ret;
1799 }
1800
1801 int ext4_inline_data_fiemap(struct inode *inode,
1802                             struct fiemap_extent_info *fieinfo,
1803                             int *has_inline)
1804 {
1805         __u64 physical = 0;
1806         __u64 length;
1807         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1808         int error = 0;
1809         struct ext4_iloc iloc;
1810
1811         down_read(&EXT4_I(inode)->xattr_sem);
1812         if (!ext4_has_inline_data(inode)) {
1813                 *has_inline = 0;
1814                 goto out;
1815         }
1816
1817         error = ext4_get_inode_loc(inode, &iloc);
1818         if (error)
1819                 goto out;
1820
1821         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1822         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1823         physical += offsetof(struct ext4_inode, i_block);
1824         length = i_size_read(inode);
1825
1826         if (physical)
1827                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1828                                                 length, flags);
1829         brelse(iloc.bh);
1830 out:
1831         up_read(&EXT4_I(inode)->xattr_sem);
1832         return (error < 0 ? error : 0);
1833 }
1834
1835 /*
1836  * Called during xattr set, and if we can sparse space 'needed',
1837  * just create the extent tree evict the data to the outer block.
1838  *
1839  * We use jbd2 instead of page cache to move data to the 1st block
1840  * so that the whole transaction can be committed as a whole and
1841  * the data isn't lost because of the delayed page cache write.
1842  */
1843 int ext4_try_to_evict_inline_data(handle_t *handle,
1844                                   struct inode *inode,
1845                                   int needed)
1846 {
1847         int error;
1848         struct ext4_xattr_entry *entry;
1849         struct ext4_inode *raw_inode;
1850         struct ext4_iloc iloc;
1851
1852         error = ext4_get_inode_loc(inode, &iloc);
1853         if (error)
1854                 return error;
1855
1856         raw_inode = ext4_raw_inode(&iloc);
1857         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1858                                             EXT4_I(inode)->i_inline_off);
1859         if (EXT4_XATTR_LEN(entry->e_name_len) +
1860             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1861                 error = -ENOSPC;
1862                 goto out;
1863         }
1864
1865         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1866 out:
1867         brelse(iloc.bh);
1868         return error;
1869 }
1870
1871 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1872 {
1873         handle_t *handle;
1874         int inline_size, value_len, needed_blocks;
1875         size_t i_size;
1876         void *value = NULL;
1877         struct ext4_xattr_ibody_find is = {
1878                 .s = { .not_found = -ENODATA, },
1879         };
1880         struct ext4_xattr_info i = {
1881                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1882                 .name = EXT4_XATTR_SYSTEM_DATA,
1883         };
1884
1885
1886         needed_blocks = ext4_writepage_trans_blocks(inode);
1887         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1888         if (IS_ERR(handle))
1889                 return;
1890
1891         down_write(&EXT4_I(inode)->xattr_sem);
1892         if (!ext4_has_inline_data(inode)) {
1893                 *has_inline = 0;
1894                 ext4_journal_stop(handle);
1895                 return;
1896         }
1897
1898         if (ext4_orphan_add(handle, inode))
1899                 goto out;
1900
1901         if (ext4_get_inode_loc(inode, &is.iloc))
1902                 goto out;
1903
1904         down_write(&EXT4_I(inode)->i_data_sem);
1905         i_size = inode->i_size;
1906         inline_size = ext4_get_inline_size(inode);
1907         EXT4_I(inode)->i_disksize = i_size;
1908
1909         if (i_size < inline_size) {
1910                 /* Clear the content in the xattr space. */
1911                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1912                         if (ext4_xattr_ibody_find(inode, &i, &is))
1913                                 goto out_error;
1914
1915                         BUG_ON(is.s.not_found);
1916
1917                         value_len = le32_to_cpu(is.s.here->e_value_size);
1918                         value = kmalloc(value_len, GFP_NOFS);
1919                         if (!value)
1920                                 goto out_error;
1921
1922                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1923                                                 value, value_len))
1924                                 goto out_error;
1925
1926                         i.value = value;
1927                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1928                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1929                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1930                                 goto out_error;
1931                 }
1932
1933                 /* Clear the content within i_blocks. */
1934                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1935                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1936                         memset(p + i_size, 0,
1937                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1938                 }
1939
1940                 EXT4_I(inode)->i_inline_size = i_size <
1941                                         EXT4_MIN_INLINE_DATA_SIZE ?
1942                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1943         }
1944
1945 out_error:
1946         up_write(&EXT4_I(inode)->i_data_sem);
1947 out:
1948         brelse(is.iloc.bh);
1949         up_write(&EXT4_I(inode)->xattr_sem);
1950         kfree(value);
1951         if (inode->i_nlink)
1952                 ext4_orphan_del(handle, inode);
1953
1954         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1955         ext4_mark_inode_dirty(handle, inode);
1956         if (IS_SYNC(inode))
1957                 ext4_handle_sync(handle);
1958
1959         ext4_journal_stop(handle);
1960         return;
1961 }
1962
1963 int ext4_convert_inline_data(struct inode *inode)
1964 {
1965         int error, needed_blocks;
1966         handle_t *handle;
1967         struct ext4_iloc iloc;
1968
1969         if (!ext4_has_inline_data(inode)) {
1970                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1971                 return 0;
1972         }
1973
1974         needed_blocks = ext4_writepage_trans_blocks(inode);
1975
1976         iloc.bh = NULL;
1977         error = ext4_get_inode_loc(inode, &iloc);
1978         if (error)
1979                 return error;
1980
1981         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1982         if (IS_ERR(handle)) {
1983                 error = PTR_ERR(handle);
1984                 goto out_free;
1985         }
1986
1987         down_write(&EXT4_I(inode)->xattr_sem);
1988         if (!ext4_has_inline_data(inode)) {
1989                 up_write(&EXT4_I(inode)->xattr_sem);
1990                 goto out;
1991         }
1992
1993         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1994         up_write(&EXT4_I(inode)->xattr_sem);
1995 out:
1996         ext4_journal_stop(handle);
1997 out_free:
1998         brelse(iloc.bh);
1999         return error;
2000 }