Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[cascardo/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 #include <linux/fiemap.h>
19
20 #define EXT4_XATTR_SYSTEM_DATA  "data"
21 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
22 #define EXT4_INLINE_DOTDOT_OFFSET       2
23 #define EXT4_INLINE_DOTDOT_SIZE         4
24
25 static int ext4_get_inline_size(struct inode *inode)
26 {
27         if (EXT4_I(inode)->i_inline_off)
28                 return EXT4_I(inode)->i_inline_size;
29
30         return 0;
31 }
32
33 static int get_max_inline_xattr_value_size(struct inode *inode,
34                                            struct ext4_iloc *iloc)
35 {
36         struct ext4_xattr_ibody_header *header;
37         struct ext4_xattr_entry *entry;
38         struct ext4_inode *raw_inode;
39         int free, min_offs;
40
41         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42                         EXT4_GOOD_OLD_INODE_SIZE -
43                         EXT4_I(inode)->i_extra_isize -
44                         sizeof(struct ext4_xattr_ibody_header);
45
46         /*
47          * We need to subtract another sizeof(__u32) since an in-inode xattr
48          * needs an empty 4 bytes to indicate the gap between the xattr entry
49          * and the name/value pair.
50          */
51         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52                 return EXT4_XATTR_SIZE(min_offs -
53                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54                         EXT4_XATTR_ROUND - sizeof(__u32));
55
56         raw_inode = ext4_raw_inode(iloc);
57         header = IHDR(inode, raw_inode);
58         entry = IFIRST(header);
59
60         /* Compute min_offs. */
61         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62                 if (!entry->e_value_block && entry->e_value_size) {
63                         size_t offs = le16_to_cpu(entry->e_value_offs);
64                         if (offs < min_offs)
65                                 min_offs = offs;
66                 }
67         }
68         free = min_offs -
69                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71         if (EXT4_I(inode)->i_inline_off) {
72                 entry = (struct ext4_xattr_entry *)
73                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76                 goto out;
77         }
78
79         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81         if (free > EXT4_XATTR_ROUND)
82                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83         else
84                 free = 0;
85
86 out:
87         return free;
88 }
89
90 /*
91  * Get the maximum size we now can store in an inode.
92  * If we can't find the space for a xattr entry, don't use the space
93  * of the extents since we have no space to indicate the inline data.
94  */
95 int ext4_get_max_inline_size(struct inode *inode)
96 {
97         int error, max_inline_size;
98         struct ext4_iloc iloc;
99
100         if (EXT4_I(inode)->i_extra_isize == 0)
101                 return 0;
102
103         error = ext4_get_inode_loc(inode, &iloc);
104         if (error) {
105                 ext4_error_inode(inode, __func__, __LINE__, 0,
106                                  "can't get inode location %lu",
107                                  inode->i_ino);
108                 return 0;
109         }
110
111         down_read(&EXT4_I(inode)->xattr_sem);
112         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113         up_read(&EXT4_I(inode)->xattr_sem);
114
115         brelse(iloc.bh);
116
117         if (!max_inline_size)
118                 return 0;
119
120         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121 }
122
123 /*
124  * this function does not take xattr_sem, which is OK because it is
125  * currently only used in a code path coming form ext4_iget, before
126  * the new inode has been unlocked
127  */
128 int ext4_find_inline_data_nolock(struct inode *inode)
129 {
130         struct ext4_xattr_ibody_find is = {
131                 .s = { .not_found = -ENODATA, },
132         };
133         struct ext4_xattr_info i = {
134                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
135                 .name = EXT4_XATTR_SYSTEM_DATA,
136         };
137         int error;
138
139         if (EXT4_I(inode)->i_extra_isize == 0)
140                 return 0;
141
142         error = ext4_get_inode_loc(inode, &is.iloc);
143         if (error)
144                 return error;
145
146         error = ext4_xattr_ibody_find(inode, &i, &is);
147         if (error)
148                 goto out;
149
150         if (!is.s.not_found) {
151                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
152                                         (void *)ext4_raw_inode(&is.iloc));
153                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
154                                 le32_to_cpu(is.s.here->e_value_size);
155                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
156         }
157 out:
158         brelse(is.iloc.bh);
159         return error;
160 }
161
162 static int ext4_read_inline_data(struct inode *inode, void *buffer,
163                                  unsigned int len,
164                                  struct ext4_iloc *iloc)
165 {
166         struct ext4_xattr_entry *entry;
167         struct ext4_xattr_ibody_header *header;
168         int cp_len = 0;
169         struct ext4_inode *raw_inode;
170
171         if (!len)
172                 return 0;
173
174         BUG_ON(len > EXT4_I(inode)->i_inline_size);
175
176         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
177                         len : EXT4_MIN_INLINE_DATA_SIZE;
178
179         raw_inode = ext4_raw_inode(iloc);
180         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
181
182         len -= cp_len;
183         buffer += cp_len;
184
185         if (!len)
186                 goto out;
187
188         header = IHDR(inode, raw_inode);
189         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
190                                             EXT4_I(inode)->i_inline_off);
191         len = min_t(unsigned int, len,
192                     (unsigned int)le32_to_cpu(entry->e_value_size));
193
194         memcpy(buffer,
195                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
196         cp_len += len;
197
198 out:
199         return cp_len;
200 }
201
202 /*
203  * write the buffer to the inline inode.
204  * If 'create' is set, we don't need to do the extra copy in the xattr
205  * value since it is already handled by ext4_xattr_ibody_inline_set.
206  * That saves us one memcpy.
207  */
208 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
209                                    void *buffer, loff_t pos, unsigned int len)
210 {
211         struct ext4_xattr_entry *entry;
212         struct ext4_xattr_ibody_header *header;
213         struct ext4_inode *raw_inode;
214         int cp_len = 0;
215
216         BUG_ON(!EXT4_I(inode)->i_inline_off);
217         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
218
219         raw_inode = ext4_raw_inode(iloc);
220         buffer += pos;
221
222         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
223                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
224                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
225                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
226
227                 len -= cp_len;
228                 buffer += cp_len;
229                 pos += cp_len;
230         }
231
232         if (!len)
233                 return;
234
235         pos -= EXT4_MIN_INLINE_DATA_SIZE;
236         header = IHDR(inode, raw_inode);
237         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
238                                             EXT4_I(inode)->i_inline_off);
239
240         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
241                buffer, len);
242 }
243
244 static int ext4_create_inline_data(handle_t *handle,
245                                    struct inode *inode, unsigned len)
246 {
247         int error;
248         void *value = NULL;
249         struct ext4_xattr_ibody_find is = {
250                 .s = { .not_found = -ENODATA, },
251         };
252         struct ext4_xattr_info i = {
253                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
254                 .name = EXT4_XATTR_SYSTEM_DATA,
255         };
256
257         error = ext4_get_inode_loc(inode, &is.iloc);
258         if (error)
259                 return error;
260
261         BUFFER_TRACE(is.iloc.bh, "get_write_access");
262         error = ext4_journal_get_write_access(handle, is.iloc.bh);
263         if (error)
264                 goto out;
265
266         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
267                 value = EXT4_ZERO_XATTR_VALUE;
268                 len -= EXT4_MIN_INLINE_DATA_SIZE;
269         } else {
270                 value = "";
271                 len = 0;
272         }
273
274         /* Insert the the xttr entry. */
275         i.value = value;
276         i.value_len = len;
277
278         error = ext4_xattr_ibody_find(inode, &i, &is);
279         if (error)
280                 goto out;
281
282         BUG_ON(!is.s.not_found);
283
284         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
285         if (error) {
286                 if (error == -ENOSPC)
287                         ext4_clear_inode_state(inode,
288                                                EXT4_STATE_MAY_INLINE_DATA);
289                 goto out;
290         }
291
292         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
293                 0, EXT4_MIN_INLINE_DATA_SIZE);
294
295         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
296                                       (void *)ext4_raw_inode(&is.iloc));
297         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
298         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
299         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
300         get_bh(is.iloc.bh);
301         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
302
303 out:
304         brelse(is.iloc.bh);
305         return error;
306 }
307
308 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
309                                    unsigned int len)
310 {
311         int error;
312         void *value = NULL;
313         struct ext4_xattr_ibody_find is = {
314                 .s = { .not_found = -ENODATA, },
315         };
316         struct ext4_xattr_info i = {
317                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
318                 .name = EXT4_XATTR_SYSTEM_DATA,
319         };
320
321         /* If the old space is ok, write the data directly. */
322         if (len <= EXT4_I(inode)->i_inline_size)
323                 return 0;
324
325         error = ext4_get_inode_loc(inode, &is.iloc);
326         if (error)
327                 return error;
328
329         error = ext4_xattr_ibody_find(inode, &i, &is);
330         if (error)
331                 goto out;
332
333         BUG_ON(is.s.not_found);
334
335         len -= EXT4_MIN_INLINE_DATA_SIZE;
336         value = kzalloc(len, GFP_NOFS);
337         if (!value)
338                 goto out;
339
340         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
341                                      value, len);
342         if (error == -ENODATA)
343                 goto out;
344
345         BUFFER_TRACE(is.iloc.bh, "get_write_access");
346         error = ext4_journal_get_write_access(handle, is.iloc.bh);
347         if (error)
348                 goto out;
349
350         /* Update the xttr entry. */
351         i.value = value;
352         i.value_len = len;
353
354         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
355         if (error)
356                 goto out;
357
358         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
359                                       (void *)ext4_raw_inode(&is.iloc));
360         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
361                                 le32_to_cpu(is.s.here->e_value_size);
362         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
363         get_bh(is.iloc.bh);
364         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
365
366 out:
367         kfree(value);
368         brelse(is.iloc.bh);
369         return error;
370 }
371
372 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
373                                     unsigned int len)
374 {
375         int ret, size;
376         struct ext4_inode_info *ei = EXT4_I(inode);
377
378         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
379                 return -ENOSPC;
380
381         size = ext4_get_max_inline_size(inode);
382         if (size < len)
383                 return -ENOSPC;
384
385         down_write(&EXT4_I(inode)->xattr_sem);
386
387         if (ei->i_inline_off)
388                 ret = ext4_update_inline_data(handle, inode, len);
389         else
390                 ret = ext4_create_inline_data(handle, inode, len);
391
392         up_write(&EXT4_I(inode)->xattr_sem);
393
394         return ret;
395 }
396
397 static int ext4_destroy_inline_data_nolock(handle_t *handle,
398                                            struct inode *inode)
399 {
400         struct ext4_inode_info *ei = EXT4_I(inode);
401         struct ext4_xattr_ibody_find is = {
402                 .s = { .not_found = 0, },
403         };
404         struct ext4_xattr_info i = {
405                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
406                 .name = EXT4_XATTR_SYSTEM_DATA,
407                 .value = NULL,
408                 .value_len = 0,
409         };
410         int error;
411
412         if (!ei->i_inline_off)
413                 return 0;
414
415         error = ext4_get_inode_loc(inode, &is.iloc);
416         if (error)
417                 return error;
418
419         error = ext4_xattr_ibody_find(inode, &i, &is);
420         if (error)
421                 goto out;
422
423         BUFFER_TRACE(is.iloc.bh, "get_write_access");
424         error = ext4_journal_get_write_access(handle, is.iloc.bh);
425         if (error)
426                 goto out;
427
428         error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
429         if (error)
430                 goto out;
431
432         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
433                 0, EXT4_MIN_INLINE_DATA_SIZE);
434
435         if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
436                                       EXT4_FEATURE_INCOMPAT_EXTENTS)) {
437                 if (S_ISDIR(inode->i_mode) ||
438                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
439                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
440                         ext4_ext_tree_init(handle, inode);
441                 }
442         }
443         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
444
445         get_bh(is.iloc.bh);
446         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
447
448         EXT4_I(inode)->i_inline_off = 0;
449         EXT4_I(inode)->i_inline_size = 0;
450         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
451 out:
452         brelse(is.iloc.bh);
453         if (error == -ENODATA)
454                 error = 0;
455         return error;
456 }
457
458 static int ext4_read_inline_page(struct inode *inode, struct page *page)
459 {
460         void *kaddr;
461         int ret = 0;
462         size_t len;
463         struct ext4_iloc iloc;
464
465         BUG_ON(!PageLocked(page));
466         BUG_ON(!ext4_has_inline_data(inode));
467         BUG_ON(page->index);
468
469         if (!EXT4_I(inode)->i_inline_off) {
470                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
471                              inode->i_ino);
472                 goto out;
473         }
474
475         ret = ext4_get_inode_loc(inode, &iloc);
476         if (ret)
477                 goto out;
478
479         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
480         kaddr = kmap_atomic(page);
481         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
482         flush_dcache_page(page);
483         kunmap_atomic(kaddr);
484         zero_user_segment(page, len, PAGE_CACHE_SIZE);
485         SetPageUptodate(page);
486         brelse(iloc.bh);
487
488 out:
489         return ret;
490 }
491
492 int ext4_readpage_inline(struct inode *inode, struct page *page)
493 {
494         int ret = 0;
495
496         down_read(&EXT4_I(inode)->xattr_sem);
497         if (!ext4_has_inline_data(inode)) {
498                 up_read(&EXT4_I(inode)->xattr_sem);
499                 return -EAGAIN;
500         }
501
502         /*
503          * Current inline data can only exist in the 1st page,
504          * So for all the other pages, just set them uptodate.
505          */
506         if (!page->index)
507                 ret = ext4_read_inline_page(inode, page);
508         else if (!PageUptodate(page)) {
509                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
510                 SetPageUptodate(page);
511         }
512
513         up_read(&EXT4_I(inode)->xattr_sem);
514
515         unlock_page(page);
516         return ret >= 0 ? 0 : ret;
517 }
518
519 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
520                                               struct inode *inode,
521                                               unsigned flags)
522 {
523         int ret, needed_blocks;
524         handle_t *handle = NULL;
525         int retries = 0, sem_held = 0;
526         struct page *page = NULL;
527         unsigned from, to;
528         struct ext4_iloc iloc;
529
530         if (!ext4_has_inline_data(inode)) {
531                 /*
532                  * clear the flag so that no new write
533                  * will trap here again.
534                  */
535                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
536                 return 0;
537         }
538
539         needed_blocks = ext4_writepage_trans_blocks(inode);
540
541         ret = ext4_get_inode_loc(inode, &iloc);
542         if (ret)
543                 return ret;
544
545 retry:
546         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
547         if (IS_ERR(handle)) {
548                 ret = PTR_ERR(handle);
549                 handle = NULL;
550                 goto out;
551         }
552
553         /* We cannot recurse into the filesystem as the transaction is already
554          * started */
555         flags |= AOP_FLAG_NOFS;
556
557         page = grab_cache_page_write_begin(mapping, 0, flags);
558         if (!page) {
559                 ret = -ENOMEM;
560                 goto out;
561         }
562
563         down_write(&EXT4_I(inode)->xattr_sem);
564         sem_held = 1;
565         /* If some one has already done this for us, just exit. */
566         if (!ext4_has_inline_data(inode)) {
567                 ret = 0;
568                 goto out;
569         }
570
571         from = 0;
572         to = ext4_get_inline_size(inode);
573         if (!PageUptodate(page)) {
574                 ret = ext4_read_inline_page(inode, page);
575                 if (ret < 0)
576                         goto out;
577         }
578
579         ret = ext4_destroy_inline_data_nolock(handle, inode);
580         if (ret)
581                 goto out;
582
583         if (ext4_should_dioread_nolock(inode))
584                 ret = __block_write_begin(page, from, to, ext4_get_block_write);
585         else
586                 ret = __block_write_begin(page, from, to, ext4_get_block);
587
588         if (!ret && ext4_should_journal_data(inode)) {
589                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
590                                              from, to, NULL,
591                                              do_journal_get_write_access);
592         }
593
594         if (ret) {
595                 unlock_page(page);
596                 page_cache_release(page);
597                 page = NULL;
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         if (page)
618                 block_commit_write(page, from, to);
619 out:
620         if (page) {
621                 unlock_page(page);
622                 page_cache_release(page);
623         }
624         if (sem_held)
625                 up_write(&EXT4_I(inode)->xattr_sem);
626         if (handle)
627                 ext4_journal_stop(handle);
628         brelse(iloc.bh);
629         return ret;
630 }
631
632 /*
633  * Try to write data in the inode.
634  * If the inode has inline data, check whether the new write can be
635  * in the inode also. If not, create the page the handle, move the data
636  * to the page make it update and let the later codes create extent for it.
637  */
638 int ext4_try_to_write_inline_data(struct address_space *mapping,
639                                   struct inode *inode,
640                                   loff_t pos, unsigned len,
641                                   unsigned flags,
642                                   struct page **pagep)
643 {
644         int ret;
645         handle_t *handle;
646         struct page *page;
647         struct ext4_iloc iloc;
648
649         if (pos + len > ext4_get_max_inline_size(inode))
650                 goto convert;
651
652         ret = ext4_get_inode_loc(inode, &iloc);
653         if (ret)
654                 return ret;
655
656         /*
657          * The possible write could happen in the inode,
658          * so try to reserve the space in inode first.
659          */
660         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
661         if (IS_ERR(handle)) {
662                 ret = PTR_ERR(handle);
663                 handle = NULL;
664                 goto out;
665         }
666
667         ret = ext4_prepare_inline_data(handle, inode, pos + len);
668         if (ret && ret != -ENOSPC)
669                 goto out;
670
671         /* We don't have space in inline inode, so convert it to extent. */
672         if (ret == -ENOSPC) {
673                 ext4_journal_stop(handle);
674                 brelse(iloc.bh);
675                 goto convert;
676         }
677
678         flags |= AOP_FLAG_NOFS;
679
680         page = grab_cache_page_write_begin(mapping, 0, flags);
681         if (!page) {
682                 ret = -ENOMEM;
683                 goto out;
684         }
685
686         *pagep = page;
687         down_read(&EXT4_I(inode)->xattr_sem);
688         if (!ext4_has_inline_data(inode)) {
689                 ret = 0;
690                 unlock_page(page);
691                 page_cache_release(page);
692                 goto out_up_read;
693         }
694
695         if (!PageUptodate(page)) {
696                 ret = ext4_read_inline_page(inode, page);
697                 if (ret < 0)
698                         goto out_up_read;
699         }
700
701         ret = 1;
702         handle = NULL;
703 out_up_read:
704         up_read(&EXT4_I(inode)->xattr_sem);
705 out:
706         if (handle)
707                 ext4_journal_stop(handle);
708         brelse(iloc.bh);
709         return ret;
710 convert:
711         return ext4_convert_inline_data_to_extent(mapping,
712                                                   inode, flags);
713 }
714
715 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
716                                unsigned copied, struct page *page)
717 {
718         int ret;
719         void *kaddr;
720         struct ext4_iloc iloc;
721
722         if (unlikely(copied < len)) {
723                 if (!PageUptodate(page)) {
724                         copied = 0;
725                         goto out;
726                 }
727         }
728
729         ret = ext4_get_inode_loc(inode, &iloc);
730         if (ret) {
731                 ext4_std_error(inode->i_sb, ret);
732                 copied = 0;
733                 goto out;
734         }
735
736         down_write(&EXT4_I(inode)->xattr_sem);
737         BUG_ON(!ext4_has_inline_data(inode));
738
739         kaddr = kmap_atomic(page);
740         ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
741         kunmap_atomic(kaddr);
742         SetPageUptodate(page);
743         /* clear page dirty so that writepages wouldn't work for us. */
744         ClearPageDirty(page);
745
746         up_write(&EXT4_I(inode)->xattr_sem);
747         brelse(iloc.bh);
748 out:
749         return copied;
750 }
751
752 struct buffer_head *
753 ext4_journalled_write_inline_data(struct inode *inode,
754                                   unsigned len,
755                                   struct page *page)
756 {
757         int ret;
758         void *kaddr;
759         struct ext4_iloc iloc;
760
761         ret = ext4_get_inode_loc(inode, &iloc);
762         if (ret) {
763                 ext4_std_error(inode->i_sb, ret);
764                 return NULL;
765         }
766
767         down_write(&EXT4_I(inode)->xattr_sem);
768         kaddr = kmap_atomic(page);
769         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
770         kunmap_atomic(kaddr);
771         up_write(&EXT4_I(inode)->xattr_sem);
772
773         return iloc.bh;
774 }
775
776 /*
777  * Try to make the page cache and handle ready for the inline data case.
778  * We can call this function in 2 cases:
779  * 1. The inode is created and the first write exceeds inline size. We can
780  *    clear the inode state safely.
781  * 2. The inode has inline data, then we need to read the data, make it
782  *    update and dirty so that ext4_da_writepages can handle it. We don't
783  *    need to start the journal since the file's metatdata isn't changed now.
784  */
785 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
786                                                  struct inode *inode,
787                                                  unsigned flags,
788                                                  void **fsdata)
789 {
790         int ret = 0, inline_size;
791         struct page *page;
792
793         page = grab_cache_page_write_begin(mapping, 0, flags);
794         if (!page)
795                 return -ENOMEM;
796
797         down_read(&EXT4_I(inode)->xattr_sem);
798         if (!ext4_has_inline_data(inode)) {
799                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
800                 goto out;
801         }
802
803         inline_size = ext4_get_inline_size(inode);
804
805         if (!PageUptodate(page)) {
806                 ret = ext4_read_inline_page(inode, page);
807                 if (ret < 0)
808                         goto out;
809         }
810
811         ret = __block_write_begin(page, 0, inline_size,
812                                   ext4_da_get_block_prep);
813         if (ret) {
814                 ext4_truncate_failed_write(inode);
815                 goto out;
816         }
817
818         SetPageDirty(page);
819         SetPageUptodate(page);
820         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
821         *fsdata = (void *)CONVERT_INLINE_DATA;
822
823 out:
824         up_read(&EXT4_I(inode)->xattr_sem);
825         if (page) {
826                 unlock_page(page);
827                 page_cache_release(page);
828         }
829         return ret;
830 }
831
832 /*
833  * Prepare the write for the inline data.
834  * If the the data can be written into the inode, we just read
835  * the page and make it uptodate, and start the journal.
836  * Otherwise read the page, makes it dirty so that it can be
837  * handle in writepages(the i_disksize update is left to the
838  * normal ext4_da_write_end).
839  */
840 int ext4_da_write_inline_data_begin(struct address_space *mapping,
841                                     struct inode *inode,
842                                     loff_t pos, unsigned len,
843                                     unsigned flags,
844                                     struct page **pagep,
845                                     void **fsdata)
846 {
847         int ret, inline_size;
848         handle_t *handle;
849         struct page *page;
850         struct ext4_iloc iloc;
851         int retries;
852
853         ret = ext4_get_inode_loc(inode, &iloc);
854         if (ret)
855                 return ret;
856
857 retry_journal:
858         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
859         if (IS_ERR(handle)) {
860                 ret = PTR_ERR(handle);
861                 goto out;
862         }
863
864         inline_size = ext4_get_max_inline_size(inode);
865
866         ret = -ENOSPC;
867         if (inline_size >= pos + len) {
868                 ret = ext4_prepare_inline_data(handle, inode, pos + len);
869                 if (ret && ret != -ENOSPC)
870                         goto out_journal;
871         }
872
873         if (ret == -ENOSPC) {
874                 ret = ext4_da_convert_inline_data_to_extent(mapping,
875                                                             inode,
876                                                             flags,
877                                                             fsdata);
878                 ext4_journal_stop(handle);
879                 if (ret == -ENOSPC &&
880                     ext4_should_retry_alloc(inode->i_sb, &retries))
881                         goto retry_journal;
882                 goto out;
883         }
884
885         /*
886          * We cannot recurse into the filesystem as the transaction
887          * is already started.
888          */
889         flags |= AOP_FLAG_NOFS;
890
891         page = grab_cache_page_write_begin(mapping, 0, flags);
892         if (!page) {
893                 ret = -ENOMEM;
894                 goto out_journal;
895         }
896
897         down_read(&EXT4_I(inode)->xattr_sem);
898         if (!ext4_has_inline_data(inode)) {
899                 ret = 0;
900                 goto out_release_page;
901         }
902
903         if (!PageUptodate(page)) {
904                 ret = ext4_read_inline_page(inode, page);
905                 if (ret < 0)
906                         goto out_release_page;
907         }
908
909         up_read(&EXT4_I(inode)->xattr_sem);
910         *pagep = page;
911         brelse(iloc.bh);
912         return 1;
913 out_release_page:
914         up_read(&EXT4_I(inode)->xattr_sem);
915         unlock_page(page);
916         page_cache_release(page);
917 out_journal:
918         ext4_journal_stop(handle);
919 out:
920         brelse(iloc.bh);
921         return ret;
922 }
923
924 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
925                                   unsigned len, unsigned copied,
926                                   struct page *page)
927 {
928         int i_size_changed = 0;
929
930         copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
931
932         /*
933          * No need to use i_size_read() here, the i_size
934          * cannot change under us because we hold i_mutex.
935          *
936          * But it's important to update i_size while still holding page lock:
937          * page writeout could otherwise come in and zero beyond i_size.
938          */
939         if (pos+copied > inode->i_size) {
940                 i_size_write(inode, pos+copied);
941                 i_size_changed = 1;
942         }
943         unlock_page(page);
944         page_cache_release(page);
945
946         /*
947          * Don't mark the inode dirty under page lock. First, it unnecessarily
948          * makes the holding time of page lock longer. Second, it forces lock
949          * ordering of page lock and transaction start for journaling
950          * filesystems.
951          */
952         if (i_size_changed)
953                 mark_inode_dirty(inode);
954
955         return copied;
956 }
957
958 #ifdef INLINE_DIR_DEBUG
959 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
960                           void *inline_start, int inline_size)
961 {
962         int offset;
963         unsigned short de_len;
964         struct ext4_dir_entry_2 *de = inline_start;
965         void *dlimit = inline_start + inline_size;
966
967         trace_printk("inode %lu\n", dir->i_ino);
968         offset = 0;
969         while ((void *)de < dlimit) {
970                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
971                 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
972                              offset, de_len, de->name_len, de->name,
973                              de->name_len, le32_to_cpu(de->inode));
974                 if (ext4_check_dir_entry(dir, NULL, de, bh,
975                                          inline_start, inline_size, offset))
976                         BUG();
977
978                 offset += de_len;
979                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
980         }
981 }
982 #else
983 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
984 #endif
985
986 /*
987  * Add a new entry into a inline dir.
988  * It will return -ENOSPC if no space is available, and -EIO
989  * and -EEXIST if directory entry already exists.
990  */
991 static int ext4_add_dirent_to_inline(handle_t *handle,
992                                      struct dentry *dentry,
993                                      struct inode *inode,
994                                      struct ext4_iloc *iloc,
995                                      void *inline_start, int inline_size)
996 {
997         struct inode    *dir = dentry->d_parent->d_inode;
998         const char      *name = dentry->d_name.name;
999         int             namelen = dentry->d_name.len;
1000         int             err;
1001         struct ext4_dir_entry_2 *de;
1002
1003         err = ext4_find_dest_de(dir, inode, iloc->bh,
1004                                 inline_start, inline_size,
1005                                 name, namelen, &de);
1006         if (err)
1007                 return err;
1008
1009         BUFFER_TRACE(iloc->bh, "get_write_access");
1010         err = ext4_journal_get_write_access(handle, iloc->bh);
1011         if (err)
1012                 return err;
1013         ext4_insert_dentry(inode, de, inline_size, name, namelen);
1014
1015         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1016
1017         /*
1018          * XXX shouldn't update any times until successful
1019          * completion of syscall, but too many callers depend
1020          * on this.
1021          *
1022          * XXX similarly, too many callers depend on
1023          * ext4_new_inode() setting the times, but error
1024          * recovery deletes the inode, so the worst that can
1025          * happen is that the times are slightly out of date
1026          * and/or different from the directory change time.
1027          */
1028         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1029         ext4_update_dx_flag(dir);
1030         dir->i_version++;
1031         ext4_mark_inode_dirty(handle, dir);
1032         return 1;
1033 }
1034
1035 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1036                                        struct ext4_iloc *iloc)
1037 {
1038         struct ext4_xattr_entry *entry;
1039         struct ext4_xattr_ibody_header *header;
1040
1041         BUG_ON(!EXT4_I(inode)->i_inline_off);
1042
1043         header = IHDR(inode, ext4_raw_inode(iloc));
1044         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1045                                             EXT4_I(inode)->i_inline_off);
1046
1047         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1048 }
1049
1050 /* Set the final de to cover the whole block. */
1051 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1052 {
1053         struct ext4_dir_entry_2 *de, *prev_de;
1054         void *limit;
1055         int de_len;
1056
1057         de = (struct ext4_dir_entry_2 *)de_buf;
1058         if (old_size) {
1059                 limit = de_buf + old_size;
1060                 do {
1061                         prev_de = de;
1062                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1063                         de_buf += de_len;
1064                         de = (struct ext4_dir_entry_2 *)de_buf;
1065                 } while (de_buf < limit);
1066
1067                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1068                                                         old_size, new_size);
1069         } else {
1070                 /* this is just created, so create an empty entry. */
1071                 de->inode = 0;
1072                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1073         }
1074 }
1075
1076 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1077                                   struct ext4_iloc *iloc)
1078 {
1079         int ret;
1080         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1081         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1082
1083         if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1084                 return -ENOSPC;
1085
1086         ret = ext4_update_inline_data(handle, dir,
1087                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1088         if (ret)
1089                 return ret;
1090
1091         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1092                              EXT4_I(dir)->i_inline_size -
1093                                                 EXT4_MIN_INLINE_DATA_SIZE);
1094         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1095         return 0;
1096 }
1097
1098 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1099                                      struct ext4_iloc *iloc,
1100                                      void *buf, int inline_size)
1101 {
1102         ext4_create_inline_data(handle, inode, inline_size);
1103         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1104         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1105 }
1106
1107 static int ext4_finish_convert_inline_dir(handle_t *handle,
1108                                           struct inode *inode,
1109                                           struct buffer_head *dir_block,
1110                                           void *buf,
1111                                           int inline_size)
1112 {
1113         int err, csum_size = 0, header_size = 0;
1114         struct ext4_dir_entry_2 *de;
1115         struct ext4_dir_entry_tail *t;
1116         void *target = dir_block->b_data;
1117
1118         /*
1119          * First create "." and ".." and then copy the dir information
1120          * back to the block.
1121          */
1122         de = (struct ext4_dir_entry_2 *)target;
1123         de = ext4_init_dot_dotdot(inode, de,
1124                 inode->i_sb->s_blocksize, csum_size,
1125                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1126         header_size = (void *)de - target;
1127
1128         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1129                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1130
1131         if (ext4_has_metadata_csum(inode->i_sb))
1132                 csum_size = sizeof(struct ext4_dir_entry_tail);
1133
1134         inode->i_size = inode->i_sb->s_blocksize;
1135         i_size_write(inode, inode->i_sb->s_blocksize);
1136         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1137         ext4_update_final_de(dir_block->b_data,
1138                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1139                         inode->i_sb->s_blocksize - csum_size);
1140
1141         if (csum_size) {
1142                 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1143                                      inode->i_sb->s_blocksize);
1144                 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1145         }
1146         set_buffer_uptodate(dir_block);
1147         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1148         if (err)
1149                 goto out;
1150         set_buffer_verified(dir_block);
1151 out:
1152         return err;
1153 }
1154
1155 static int ext4_convert_inline_data_nolock(handle_t *handle,
1156                                            struct inode *inode,
1157                                            struct ext4_iloc *iloc)
1158 {
1159         int error;
1160         void *buf = NULL;
1161         struct buffer_head *data_bh = NULL;
1162         struct ext4_map_blocks map;
1163         int inline_size;
1164
1165         inline_size = ext4_get_inline_size(inode);
1166         buf = kmalloc(inline_size, GFP_NOFS);
1167         if (!buf) {
1168                 error = -ENOMEM;
1169                 goto out;
1170         }
1171
1172         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1173         if (error < 0)
1174                 goto out;
1175
1176         /*
1177          * Make sure the inline directory entries pass checks before we try to
1178          * convert them, so that we avoid touching stuff that needs fsck.
1179          */
1180         if (S_ISDIR(inode->i_mode)) {
1181                 error = ext4_check_all_de(inode, iloc->bh,
1182                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1183                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1184                 if (error)
1185                         goto out;
1186         }
1187
1188         error = ext4_destroy_inline_data_nolock(handle, inode);
1189         if (error)
1190                 goto out;
1191
1192         map.m_lblk = 0;
1193         map.m_len = 1;
1194         map.m_flags = 0;
1195         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1196         if (error < 0)
1197                 goto out_restore;
1198         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1199                 error = -EIO;
1200                 goto out_restore;
1201         }
1202
1203         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1204         if (!data_bh) {
1205                 error = -ENOMEM;
1206                 goto out_restore;
1207         }
1208
1209         lock_buffer(data_bh);
1210         error = ext4_journal_get_create_access(handle, data_bh);
1211         if (error) {
1212                 unlock_buffer(data_bh);
1213                 error = -EIO;
1214                 goto out_restore;
1215         }
1216         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1217
1218         if (!S_ISDIR(inode->i_mode)) {
1219                 memcpy(data_bh->b_data, buf, inline_size);
1220                 set_buffer_uptodate(data_bh);
1221                 error = ext4_handle_dirty_metadata(handle,
1222                                                    inode, data_bh);
1223         } else {
1224                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1225                                                        buf, inline_size);
1226         }
1227
1228         unlock_buffer(data_bh);
1229 out_restore:
1230         if (error)
1231                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1232
1233 out:
1234         brelse(data_bh);
1235         kfree(buf);
1236         return error;
1237 }
1238
1239 /*
1240  * Try to add the new entry to the inline data.
1241  * If succeeds, return 0. If not, extended the inline dir and copied data to
1242  * the new created block.
1243  */
1244 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1245                               struct inode *inode)
1246 {
1247         int ret, inline_size;
1248         void *inline_start;
1249         struct ext4_iloc iloc;
1250         struct inode *dir = dentry->d_parent->d_inode;
1251
1252         ret = ext4_get_inode_loc(dir, &iloc);
1253         if (ret)
1254                 return ret;
1255
1256         down_write(&EXT4_I(dir)->xattr_sem);
1257         if (!ext4_has_inline_data(dir))
1258                 goto out;
1259
1260         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1261                                                  EXT4_INLINE_DOTDOT_SIZE;
1262         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1263
1264         ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1265                                         inline_start, inline_size);
1266         if (ret != -ENOSPC)
1267                 goto out;
1268
1269         /* check whether it can be inserted to inline xattr space. */
1270         inline_size = EXT4_I(dir)->i_inline_size -
1271                         EXT4_MIN_INLINE_DATA_SIZE;
1272         if (!inline_size) {
1273                 /* Try to use the xattr space.*/
1274                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1275                 if (ret && ret != -ENOSPC)
1276                         goto out;
1277
1278                 inline_size = EXT4_I(dir)->i_inline_size -
1279                                 EXT4_MIN_INLINE_DATA_SIZE;
1280         }
1281
1282         if (inline_size) {
1283                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1284
1285                 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1286                                                 inline_start, inline_size);
1287
1288                 if (ret != -ENOSPC)
1289                         goto out;
1290         }
1291
1292         /*
1293          * The inline space is filled up, so create a new block for it.
1294          * As the extent tree will be created, we have to save the inline
1295          * dir first.
1296          */
1297         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1298
1299 out:
1300         ext4_mark_inode_dirty(handle, dir);
1301         up_write(&EXT4_I(dir)->xattr_sem);
1302         brelse(iloc.bh);
1303         return ret;
1304 }
1305
1306 /*
1307  * This function fills a red-black tree with information from an
1308  * inlined dir.  It returns the number directory entries loaded
1309  * into the tree.  If there is an error it is returned in err.
1310  */
1311 int htree_inlinedir_to_tree(struct file *dir_file,
1312                             struct inode *dir, ext4_lblk_t block,
1313                             struct dx_hash_info *hinfo,
1314                             __u32 start_hash, __u32 start_minor_hash,
1315                             int *has_inline_data)
1316 {
1317         int err = 0, count = 0;
1318         unsigned int parent_ino;
1319         int pos;
1320         struct ext4_dir_entry_2 *de;
1321         struct inode *inode = file_inode(dir_file);
1322         int ret, inline_size = 0;
1323         struct ext4_iloc iloc;
1324         void *dir_buf = NULL;
1325         struct ext4_dir_entry_2 fake;
1326
1327         ret = ext4_get_inode_loc(inode, &iloc);
1328         if (ret)
1329                 return ret;
1330
1331         down_read(&EXT4_I(inode)->xattr_sem);
1332         if (!ext4_has_inline_data(inode)) {
1333                 up_read(&EXT4_I(inode)->xattr_sem);
1334                 *has_inline_data = 0;
1335                 goto out;
1336         }
1337
1338         inline_size = ext4_get_inline_size(inode);
1339         dir_buf = kmalloc(inline_size, GFP_NOFS);
1340         if (!dir_buf) {
1341                 ret = -ENOMEM;
1342                 up_read(&EXT4_I(inode)->xattr_sem);
1343                 goto out;
1344         }
1345
1346         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1347         up_read(&EXT4_I(inode)->xattr_sem);
1348         if (ret < 0)
1349                 goto out;
1350
1351         pos = 0;
1352         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1353         while (pos < inline_size) {
1354                 /*
1355                  * As inlined dir doesn't store any information about '.' and
1356                  * only the inode number of '..' is stored, we have to handle
1357                  * them differently.
1358                  */
1359                 if (pos == 0) {
1360                         fake.inode = cpu_to_le32(inode->i_ino);
1361                         fake.name_len = 1;
1362                         strcpy(fake.name, ".");
1363                         fake.rec_len = ext4_rec_len_to_disk(
1364                                                 EXT4_DIR_REC_LEN(fake.name_len),
1365                                                 inline_size);
1366                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1367                         de = &fake;
1368                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1369                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1370                         fake.inode = cpu_to_le32(parent_ino);
1371                         fake.name_len = 2;
1372                         strcpy(fake.name, "..");
1373                         fake.rec_len = ext4_rec_len_to_disk(
1374                                                 EXT4_DIR_REC_LEN(fake.name_len),
1375                                                 inline_size);
1376                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1377                         de = &fake;
1378                         pos = EXT4_INLINE_DOTDOT_SIZE;
1379                 } else {
1380                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1381                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1382                         if (ext4_check_dir_entry(inode, dir_file, de,
1383                                          iloc.bh, dir_buf,
1384                                          inline_size, pos)) {
1385                                 ret = count;
1386                                 goto out;
1387                         }
1388                 }
1389
1390                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1391                 if ((hinfo->hash < start_hash) ||
1392                     ((hinfo->hash == start_hash) &&
1393                      (hinfo->minor_hash < start_minor_hash)))
1394                         continue;
1395                 if (de->inode == 0)
1396                         continue;
1397                 err = ext4_htree_store_dirent(dir_file,
1398                                    hinfo->hash, hinfo->minor_hash, de);
1399                 if (err) {
1400                         count = err;
1401                         goto out;
1402                 }
1403                 count++;
1404         }
1405         ret = count;
1406 out:
1407         kfree(dir_buf);
1408         brelse(iloc.bh);
1409         return ret;
1410 }
1411
1412 /*
1413  * So this function is called when the volume is mkfsed with
1414  * dir_index disabled. In order to keep f_pos persistent
1415  * after we convert from an inlined dir to a blocked based,
1416  * we just pretend that we are a normal dir and return the
1417  * offset as if '.' and '..' really take place.
1418  *
1419  */
1420 int ext4_read_inline_dir(struct file *file,
1421                          struct dir_context *ctx,
1422                          int *has_inline_data)
1423 {
1424         unsigned int offset, parent_ino;
1425         int i;
1426         struct ext4_dir_entry_2 *de;
1427         struct super_block *sb;
1428         struct inode *inode = file_inode(file);
1429         int ret, inline_size = 0;
1430         struct ext4_iloc iloc;
1431         void *dir_buf = NULL;
1432         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1433
1434         ret = ext4_get_inode_loc(inode, &iloc);
1435         if (ret)
1436                 return ret;
1437
1438         down_read(&EXT4_I(inode)->xattr_sem);
1439         if (!ext4_has_inline_data(inode)) {
1440                 up_read(&EXT4_I(inode)->xattr_sem);
1441                 *has_inline_data = 0;
1442                 goto out;
1443         }
1444
1445         inline_size = ext4_get_inline_size(inode);
1446         dir_buf = kmalloc(inline_size, GFP_NOFS);
1447         if (!dir_buf) {
1448                 ret = -ENOMEM;
1449                 up_read(&EXT4_I(inode)->xattr_sem);
1450                 goto out;
1451         }
1452
1453         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1454         up_read(&EXT4_I(inode)->xattr_sem);
1455         if (ret < 0)
1456                 goto out;
1457
1458         ret = 0;
1459         sb = inode->i_sb;
1460         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1461         offset = ctx->pos;
1462
1463         /*
1464          * dotdot_offset and dotdot_size is the real offset and
1465          * size for ".." and "." if the dir is block based while
1466          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1467          * So we will use extra_offset and extra_size to indicate them
1468          * during the inline dir iteration.
1469          */
1470         dotdot_offset = EXT4_DIR_REC_LEN(1);
1471         dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1472         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1473         extra_size = extra_offset + inline_size;
1474
1475         /*
1476          * If the version has changed since the last call to
1477          * readdir(2), then we might be pointing to an invalid
1478          * dirent right now.  Scan from the start of the inline
1479          * dir to make sure.
1480          */
1481         if (file->f_version != inode->i_version) {
1482                 for (i = 0; i < extra_size && i < offset;) {
1483                         /*
1484                          * "." is with offset 0 and
1485                          * ".." is dotdot_offset.
1486                          */
1487                         if (!i) {
1488                                 i = dotdot_offset;
1489                                 continue;
1490                         } else if (i == dotdot_offset) {
1491                                 i = dotdot_size;
1492                                 continue;
1493                         }
1494                         /* for other entry, the real offset in
1495                          * the buf has to be tuned accordingly.
1496                          */
1497                         de = (struct ext4_dir_entry_2 *)
1498                                 (dir_buf + i - extra_offset);
1499                         /* It's too expensive to do a full
1500                          * dirent test each time round this
1501                          * loop, but we do have to test at
1502                          * least that it is non-zero.  A
1503                          * failure will be detected in the
1504                          * dirent test below. */
1505                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1506                                 < EXT4_DIR_REC_LEN(1))
1507                                 break;
1508                         i += ext4_rec_len_from_disk(de->rec_len,
1509                                                     extra_size);
1510                 }
1511                 offset = i;
1512                 ctx->pos = offset;
1513                 file->f_version = inode->i_version;
1514         }
1515
1516         while (ctx->pos < extra_size) {
1517                 if (ctx->pos == 0) {
1518                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1519                                 goto out;
1520                         ctx->pos = dotdot_offset;
1521                         continue;
1522                 }
1523
1524                 if (ctx->pos == dotdot_offset) {
1525                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1526                                 goto out;
1527                         ctx->pos = dotdot_size;
1528                         continue;
1529                 }
1530
1531                 de = (struct ext4_dir_entry_2 *)
1532                         (dir_buf + ctx->pos - extra_offset);
1533                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1534                                          extra_size, ctx->pos))
1535                         goto out;
1536                 if (le32_to_cpu(de->inode)) {
1537                         if (!dir_emit(ctx, de->name, de->name_len,
1538                                       le32_to_cpu(de->inode),
1539                                       get_dtype(sb, de->file_type)))
1540                                 goto out;
1541                 }
1542                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1543         }
1544 out:
1545         kfree(dir_buf);
1546         brelse(iloc.bh);
1547         return ret;
1548 }
1549
1550 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1551                                         struct ext4_dir_entry_2 **parent_de,
1552                                         int *retval)
1553 {
1554         struct ext4_iloc iloc;
1555
1556         *retval = ext4_get_inode_loc(inode, &iloc);
1557         if (*retval)
1558                 return NULL;
1559
1560         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1561
1562         return iloc.bh;
1563 }
1564
1565 /*
1566  * Try to create the inline data for the new dir.
1567  * If it succeeds, return 0, otherwise return the error.
1568  * In case of ENOSPC, the caller should create the normal disk layout dir.
1569  */
1570 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1571                                struct inode *inode)
1572 {
1573         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1574         struct ext4_iloc iloc;
1575         struct ext4_dir_entry_2 *de;
1576
1577         ret = ext4_get_inode_loc(inode, &iloc);
1578         if (ret)
1579                 return ret;
1580
1581         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1582         if (ret)
1583                 goto out;
1584
1585         /*
1586          * For inline dir, we only save the inode information for the ".."
1587          * and create a fake dentry to cover the left space.
1588          */
1589         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1590         de->inode = cpu_to_le32(parent->i_ino);
1591         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1592         de->inode = 0;
1593         de->rec_len = ext4_rec_len_to_disk(
1594                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1595                                 inline_size);
1596         set_nlink(inode, 2);
1597         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1598 out:
1599         brelse(iloc.bh);
1600         return ret;
1601 }
1602
1603 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1604                                         const struct qstr *d_name,
1605                                         struct ext4_dir_entry_2 **res_dir,
1606                                         int *has_inline_data)
1607 {
1608         int ret;
1609         struct ext4_iloc iloc;
1610         void *inline_start;
1611         int inline_size;
1612
1613         if (ext4_get_inode_loc(dir, &iloc))
1614                 return NULL;
1615
1616         down_read(&EXT4_I(dir)->xattr_sem);
1617         if (!ext4_has_inline_data(dir)) {
1618                 *has_inline_data = 0;
1619                 goto out;
1620         }
1621
1622         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1623                                                 EXT4_INLINE_DOTDOT_SIZE;
1624         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1625         ret = search_dir(iloc.bh, inline_start, inline_size,
1626                          dir, d_name, 0, res_dir);
1627         if (ret == 1)
1628                 goto out_find;
1629         if (ret < 0)
1630                 goto out;
1631
1632         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1633                 goto out;
1634
1635         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1636         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1637
1638         ret = search_dir(iloc.bh, inline_start, inline_size,
1639                          dir, d_name, 0, res_dir);
1640         if (ret == 1)
1641                 goto out_find;
1642
1643 out:
1644         brelse(iloc.bh);
1645         iloc.bh = NULL;
1646 out_find:
1647         up_read(&EXT4_I(dir)->xattr_sem);
1648         return iloc.bh;
1649 }
1650
1651 int ext4_delete_inline_entry(handle_t *handle,
1652                              struct inode *dir,
1653                              struct ext4_dir_entry_2 *de_del,
1654                              struct buffer_head *bh,
1655                              int *has_inline_data)
1656 {
1657         int err, inline_size;
1658         struct ext4_iloc iloc;
1659         void *inline_start;
1660
1661         err = ext4_get_inode_loc(dir, &iloc);
1662         if (err)
1663                 return err;
1664
1665         down_write(&EXT4_I(dir)->xattr_sem);
1666         if (!ext4_has_inline_data(dir)) {
1667                 *has_inline_data = 0;
1668                 goto out;
1669         }
1670
1671         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1672                 EXT4_MIN_INLINE_DATA_SIZE) {
1673                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1674                                         EXT4_INLINE_DOTDOT_SIZE;
1675                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1676                                 EXT4_INLINE_DOTDOT_SIZE;
1677         } else {
1678                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1679                 inline_size = ext4_get_inline_size(dir) -
1680                                 EXT4_MIN_INLINE_DATA_SIZE;
1681         }
1682
1683         BUFFER_TRACE(bh, "get_write_access");
1684         err = ext4_journal_get_write_access(handle, bh);
1685         if (err)
1686                 goto out;
1687
1688         err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1689                                         inline_start, inline_size, 0);
1690         if (err)
1691                 goto out;
1692
1693         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1694         err = ext4_mark_inode_dirty(handle, dir);
1695         if (unlikely(err))
1696                 goto out;
1697
1698         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1699 out:
1700         up_write(&EXT4_I(dir)->xattr_sem);
1701         brelse(iloc.bh);
1702         if (err != -ENOENT)
1703                 ext4_std_error(dir->i_sb, err);
1704         return err;
1705 }
1706
1707 /*
1708  * Get the inline dentry at offset.
1709  */
1710 static inline struct ext4_dir_entry_2 *
1711 ext4_get_inline_entry(struct inode *inode,
1712                       struct ext4_iloc *iloc,
1713                       unsigned int offset,
1714                       void **inline_start,
1715                       int *inline_size)
1716 {
1717         void *inline_pos;
1718
1719         BUG_ON(offset > ext4_get_inline_size(inode));
1720
1721         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1722                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1723                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1724         } else {
1725                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1726                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1727                 *inline_size = ext4_get_inline_size(inode) -
1728                                 EXT4_MIN_INLINE_DATA_SIZE;
1729         }
1730
1731         if (inline_start)
1732                 *inline_start = inline_pos;
1733         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1734 }
1735
1736 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1737 {
1738         int err, inline_size;
1739         struct ext4_iloc iloc;
1740         void *inline_pos;
1741         unsigned int offset;
1742         struct ext4_dir_entry_2 *de;
1743         int ret = 1;
1744
1745         err = ext4_get_inode_loc(dir, &iloc);
1746         if (err) {
1747                 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1748                                  err, dir->i_ino);
1749                 return 1;
1750         }
1751
1752         down_read(&EXT4_I(dir)->xattr_sem);
1753         if (!ext4_has_inline_data(dir)) {
1754                 *has_inline_data = 0;
1755                 goto out;
1756         }
1757
1758         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1759         if (!le32_to_cpu(de->inode)) {
1760                 ext4_warning(dir->i_sb,
1761                              "bad inline directory (dir #%lu) - no `..'",
1762                              dir->i_ino);
1763                 ret = 1;
1764                 goto out;
1765         }
1766
1767         offset = EXT4_INLINE_DOTDOT_SIZE;
1768         while (offset < dir->i_size) {
1769                 de = ext4_get_inline_entry(dir, &iloc, offset,
1770                                            &inline_pos, &inline_size);
1771                 if (ext4_check_dir_entry(dir, NULL, de,
1772                                          iloc.bh, inline_pos,
1773                                          inline_size, offset)) {
1774                         ext4_warning(dir->i_sb,
1775                                      "bad inline directory (dir #%lu) - "
1776                                      "inode %u, rec_len %u, name_len %d"
1777                                      "inline size %d\n",
1778                                      dir->i_ino, le32_to_cpu(de->inode),
1779                                      le16_to_cpu(de->rec_len), de->name_len,
1780                                      inline_size);
1781                         ret = 1;
1782                         goto out;
1783                 }
1784                 if (le32_to_cpu(de->inode)) {
1785                         ret = 0;
1786                         goto out;
1787                 }
1788                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1789         }
1790
1791 out:
1792         up_read(&EXT4_I(dir)->xattr_sem);
1793         brelse(iloc.bh);
1794         return ret;
1795 }
1796
1797 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1798 {
1799         int ret;
1800
1801         down_write(&EXT4_I(inode)->xattr_sem);
1802         ret = ext4_destroy_inline_data_nolock(handle, inode);
1803         up_write(&EXT4_I(inode)->xattr_sem);
1804
1805         return ret;
1806 }
1807
1808 int ext4_inline_data_fiemap(struct inode *inode,
1809                             struct fiemap_extent_info *fieinfo,
1810                             int *has_inline)
1811 {
1812         __u64 physical = 0;
1813         __u64 length;
1814         __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1815         int error = 0;
1816         struct ext4_iloc iloc;
1817
1818         down_read(&EXT4_I(inode)->xattr_sem);
1819         if (!ext4_has_inline_data(inode)) {
1820                 *has_inline = 0;
1821                 goto out;
1822         }
1823
1824         error = ext4_get_inode_loc(inode, &iloc);
1825         if (error)
1826                 goto out;
1827
1828         physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1829         physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1830         physical += offsetof(struct ext4_inode, i_block);
1831         length = i_size_read(inode);
1832
1833         if (physical)
1834                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1835                                                 length, flags);
1836         brelse(iloc.bh);
1837 out:
1838         up_read(&EXT4_I(inode)->xattr_sem);
1839         return (error < 0 ? error : 0);
1840 }
1841
1842 /*
1843  * Called during xattr set, and if we can sparse space 'needed',
1844  * just create the extent tree evict the data to the outer block.
1845  *
1846  * We use jbd2 instead of page cache to move data to the 1st block
1847  * so that the whole transaction can be committed as a whole and
1848  * the data isn't lost because of the delayed page cache write.
1849  */
1850 int ext4_try_to_evict_inline_data(handle_t *handle,
1851                                   struct inode *inode,
1852                                   int needed)
1853 {
1854         int error;
1855         struct ext4_xattr_entry *entry;
1856         struct ext4_inode *raw_inode;
1857         struct ext4_iloc iloc;
1858
1859         error = ext4_get_inode_loc(inode, &iloc);
1860         if (error)
1861                 return error;
1862
1863         raw_inode = ext4_raw_inode(&iloc);
1864         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1865                                             EXT4_I(inode)->i_inline_off);
1866         if (EXT4_XATTR_LEN(entry->e_name_len) +
1867             EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1868                 error = -ENOSPC;
1869                 goto out;
1870         }
1871
1872         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1873 out:
1874         brelse(iloc.bh);
1875         return error;
1876 }
1877
1878 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1879 {
1880         handle_t *handle;
1881         int inline_size, value_len, needed_blocks;
1882         size_t i_size;
1883         void *value = NULL;
1884         struct ext4_xattr_ibody_find is = {
1885                 .s = { .not_found = -ENODATA, },
1886         };
1887         struct ext4_xattr_info i = {
1888                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1889                 .name = EXT4_XATTR_SYSTEM_DATA,
1890         };
1891
1892
1893         needed_blocks = ext4_writepage_trans_blocks(inode);
1894         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1895         if (IS_ERR(handle))
1896                 return;
1897
1898         down_write(&EXT4_I(inode)->xattr_sem);
1899         if (!ext4_has_inline_data(inode)) {
1900                 *has_inline = 0;
1901                 ext4_journal_stop(handle);
1902                 return;
1903         }
1904
1905         if (ext4_orphan_add(handle, inode))
1906                 goto out;
1907
1908         if (ext4_get_inode_loc(inode, &is.iloc))
1909                 goto out;
1910
1911         down_write(&EXT4_I(inode)->i_data_sem);
1912         i_size = inode->i_size;
1913         inline_size = ext4_get_inline_size(inode);
1914         EXT4_I(inode)->i_disksize = i_size;
1915
1916         if (i_size < inline_size) {
1917                 /* Clear the content in the xattr space. */
1918                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1919                         if (ext4_xattr_ibody_find(inode, &i, &is))
1920                                 goto out_error;
1921
1922                         BUG_ON(is.s.not_found);
1923
1924                         value_len = le32_to_cpu(is.s.here->e_value_size);
1925                         value = kmalloc(value_len, GFP_NOFS);
1926                         if (!value)
1927                                 goto out_error;
1928
1929                         if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1930                                                 value, value_len))
1931                                 goto out_error;
1932
1933                         i.value = value;
1934                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1935                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1936                         if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1937                                 goto out_error;
1938                 }
1939
1940                 /* Clear the content within i_blocks. */
1941                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1942                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1943                         memset(p + i_size, 0,
1944                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1945                 }
1946
1947                 EXT4_I(inode)->i_inline_size = i_size <
1948                                         EXT4_MIN_INLINE_DATA_SIZE ?
1949                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
1950         }
1951
1952 out_error:
1953         up_write(&EXT4_I(inode)->i_data_sem);
1954 out:
1955         brelse(is.iloc.bh);
1956         up_write(&EXT4_I(inode)->xattr_sem);
1957         kfree(value);
1958         if (inode->i_nlink)
1959                 ext4_orphan_del(handle, inode);
1960
1961         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1962         ext4_mark_inode_dirty(handle, inode);
1963         if (IS_SYNC(inode))
1964                 ext4_handle_sync(handle);
1965
1966         ext4_journal_stop(handle);
1967         return;
1968 }
1969
1970 int ext4_convert_inline_data(struct inode *inode)
1971 {
1972         int error, needed_blocks;
1973         handle_t *handle;
1974         struct ext4_iloc iloc;
1975
1976         if (!ext4_has_inline_data(inode)) {
1977                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1978                 return 0;
1979         }
1980
1981         needed_blocks = ext4_writepage_trans_blocks(inode);
1982
1983         iloc.bh = NULL;
1984         error = ext4_get_inode_loc(inode, &iloc);
1985         if (error)
1986                 return error;
1987
1988         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1989         if (IS_ERR(handle)) {
1990                 error = PTR_ERR(handle);
1991                 goto out_free;
1992         }
1993
1994         down_write(&EXT4_I(inode)->xattr_sem);
1995         if (!ext4_has_inline_data(inode)) {
1996                 up_write(&EXT4_I(inode)->xattr_sem);
1997                 goto out;
1998         }
1999
2000         error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2001         up_write(&EXT4_I(inode)->xattr_sem);
2002 out:
2003         ext4_journal_stop(handle);
2004 out_free:
2005         brelse(iloc.bh);
2006         return error;
2007 }