?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
clocksource/drivers/arm_arch_timer: Correct fault programming of CNTKCTL_EL1.EVNTI
[linux/fpc-iii.git] / fs / ext4 / inline.c
blob5bf62bdd84b8721b5ca7dee2122a9de8539f33ae
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
15 #include <linux/fiemap.h>
16
17 #include "ext4_jbd2.h"
18 #include "ext4.h"
19 #include "xattr.h"
20 #include "truncate.h"
21
22 #define EXT4_XATTR_SYSTEM_DATA "data"
23 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
24 #define EXT4_INLINE_DOTDOT_OFFSET 2
25 #define EXT4_INLINE_DOTDOT_SIZE 4
26
27 static int ext4_get_inline_size(struct inode *inode)
28 {
29 if (EXT4_I(inode)->i_inline_off)
30 return EXT4_I(inode)->i_inline_size;
31
32 return 0;
33 }
34
35 static int get_max_inline_xattr_value_size(struct inode *inode,
36 struct ext4_iloc *iloc)
37 {
38 struct ext4_xattr_ibody_header *header;
39 struct ext4_xattr_entry *entry;
40 struct ext4_inode *raw_inode;
41 int free, min_offs;
42
43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 EXT4_GOOD_OLD_INODE_SIZE -
45 EXT4_I(inode)->i_extra_isize -
46 sizeof(struct ext4_xattr_ibody_header);
47
48 /*
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
52 */
53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 return EXT4_XATTR_SIZE(min_offs -
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 EXT4_XATTR_ROUND - sizeof(__u32));
57
58 raw_inode = ext4_raw_inode(iloc);
59 header = IHDR(inode, raw_inode);
60 entry = IFIRST(header);
61
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64 if (!entry->e_value_block && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
66 if (offs < min_offs)
67 min_offs = offs;
68 }
69 }
70 free = min_offs -
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78 goto out;
79 }
80
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85 else
86 free = 0;
87
88 out:
89 return free;
90 }
91
92 /*
93 * Get the maximum size we now can store in an inode.
94 * If we can't find the space for a xattr entry, don't use the space
95 * of the extents since we have no space to indicate the inline data.
96 */
97 int ext4_get_max_inline_size(struct inode *inode)
98 {
99 int error, max_inline_size;
100 struct ext4_iloc iloc;
101
102 if (EXT4_I(inode)->i_extra_isize == 0)
103 return 0;
104
105 error = ext4_get_inode_loc(inode, &iloc);
106 if (error) {
107 ext4_error_inode(inode, __func__, __LINE__, 0,
108 "can't get inode location %lu",
109 inode->i_ino);
110 return 0;
111 }
112
113 down_read(&EXT4_I(inode)->xattr_sem);
114 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
115 up_read(&EXT4_I(inode)->xattr_sem);
116
117 brelse(iloc.bh);
118
119 if (!max_inline_size)
120 return 0;
121
122 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
123 }
124
125 /*
126 * this function does not take xattr_sem, which is OK because it is
127 * currently only used in a code path coming form ext4_iget, before
128 * the new inode has been unlocked
129 */
130 int ext4_find_inline_data_nolock(struct inode *inode)
131 {
132 struct ext4_xattr_ibody_find is = {
133 .s = { .not_found = -ENODATA, },
134 };
135 struct ext4_xattr_info i = {
136 .name_index = EXT4_XATTR_INDEX_SYSTEM,
137 .name = EXT4_XATTR_SYSTEM_DATA,
138 };
139 int error;
140
141 if (EXT4_I(inode)->i_extra_isize == 0)
142 return 0;
143
144 error = ext4_get_inode_loc(inode, &is.iloc);
145 if (error)
146 return error;
147
148 error = ext4_xattr_ibody_find(inode, &i, &is);
149 if (error)
150 goto out;
151
152 if (!is.s.not_found) {
153 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154 (void *)ext4_raw_inode(&is.iloc));
155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156 le32_to_cpu(is.s.here->e_value_size);
157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158 }
159 out:
160 brelse(is.iloc.bh);
161 return error;
162 }
163
164 static int ext4_read_inline_data(struct inode *inode, void *buffer,
165 unsigned int len,
166 struct ext4_iloc *iloc)
167 {
168 struct ext4_xattr_entry *entry;
169 struct ext4_xattr_ibody_header *header;
170 int cp_len = 0;
171 struct ext4_inode *raw_inode;
172
173 if (!len)
174 return 0;
175
176 BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179 len : EXT4_MIN_INLINE_DATA_SIZE;
180
181 raw_inode = ext4_raw_inode(iloc);
182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184 len -= cp_len;
185 buffer += cp_len;
186
187 if (!len)
188 goto out;
189
190 header = IHDR(inode, raw_inode);
191 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192 EXT4_I(inode)->i_inline_off);
193 len = min_t(unsigned int, len,
194 (unsigned int)le32_to_cpu(entry->e_value_size));
195
196 memcpy(buffer,
197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198 cp_len += len;
199
200 out:
201 return cp_len;
202 }
203
204 /*
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
209 */
210 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211 void *buffer, loff_t pos, unsigned int len)
212 {
213 struct ext4_xattr_entry *entry;
214 struct ext4_xattr_ibody_header *header;
215 struct ext4_inode *raw_inode;
216 int cp_len = 0;
217
218 BUG_ON(!EXT4_I(inode)->i_inline_off);
219 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
220
221 raw_inode = ext4_raw_inode(iloc);
222 buffer += pos;
223
224 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
225 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
226 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
227 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
228
229 len -= cp_len;
230 buffer += cp_len;
231 pos += cp_len;
232 }
233
234 if (!len)
235 return;
236
237 pos -= EXT4_MIN_INLINE_DATA_SIZE;
238 header = IHDR(inode, raw_inode);
239 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
240 EXT4_I(inode)->i_inline_off);
241
242 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
243 buffer, len);
244 }
245
246 static int ext4_create_inline_data(handle_t *handle,
247 struct inode *inode, unsigned len)
248 {
249 int error;
250 void *value = NULL;
251 struct ext4_xattr_ibody_find is = {
252 .s = { .not_found = -ENODATA, },
253 };
254 struct ext4_xattr_info i = {
255 .name_index = EXT4_XATTR_INDEX_SYSTEM,
256 .name = EXT4_XATTR_SYSTEM_DATA,
257 };
258
259 error = ext4_get_inode_loc(inode, &is.iloc);
260 if (error)
261 return error;
262
263 BUFFER_TRACE(is.iloc.bh, "get_write_access");
264 error = ext4_journal_get_write_access(handle, is.iloc.bh);
265 if (error)
266 goto out;
267
268 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
269 value = EXT4_ZERO_XATTR_VALUE;
270 len -= EXT4_MIN_INLINE_DATA_SIZE;
271 } else {
272 value = "";
273 len = 0;
274 }
275
276 /* Insert the the xttr entry. */
277 i.value = value;
278 i.value_len = len;
279
280 error = ext4_xattr_ibody_find(inode, &i, &is);
281 if (error)
282 goto out;
283
284 BUG_ON(!is.s.not_found);
285
286 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
287 if (error) {
288 if (error == -ENOSPC)
289 ext4_clear_inode_state(inode,
290 EXT4_STATE_MAY_INLINE_DATA);
291 goto out;
292 }
293
294 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
295 0, EXT4_MIN_INLINE_DATA_SIZE);
296
297 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
298 (void *)ext4_raw_inode(&is.iloc));
299 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
300 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
301 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
302 get_bh(is.iloc.bh);
303 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
304
305 out:
306 brelse(is.iloc.bh);
307 return error;
308 }
309
310 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
311 unsigned int len)
312 {
313 int error;
314 void *value = NULL;
315 struct ext4_xattr_ibody_find is = {
316 .s = { .not_found = -ENODATA, },
317 };
318 struct ext4_xattr_info i = {
319 .name_index = EXT4_XATTR_INDEX_SYSTEM,
320 .name = EXT4_XATTR_SYSTEM_DATA,
321 };
322
323 /* If the old space is ok, write the data directly. */
324 if (len <= EXT4_I(inode)->i_inline_size)
325 return 0;
326
327 error = ext4_get_inode_loc(inode, &is.iloc);
328 if (error)
329 return error;
330
331 error = ext4_xattr_ibody_find(inode, &i, &is);
332 if (error)
333 goto out;
334
335 BUG_ON(is.s.not_found);
336
337 len -= EXT4_MIN_INLINE_DATA_SIZE;
338 value = kzalloc(len, GFP_NOFS);
339 if (!value) {
340 error = -ENOMEM;
341 goto out;
342 }
343
344 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
345 value, len);
346 if (error == -ENODATA)
347 goto out;
348
349 BUFFER_TRACE(is.iloc.bh, "get_write_access");
350 error = ext4_journal_get_write_access(handle, is.iloc.bh);
351 if (error)
352 goto out;
353
354 /* Update the xttr entry. */
355 i.value = value;
356 i.value_len = len;
357
358 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
359 if (error)
360 goto out;
361
362 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
363 (void *)ext4_raw_inode(&is.iloc));
364 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
365 le32_to_cpu(is.s.here->e_value_size);
366 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
367 get_bh(is.iloc.bh);
368 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
369
370 out:
371 kfree(value);
372 brelse(is.iloc.bh);
373 return error;
374 }
375
376 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
377 unsigned int len)
378 {
379 int ret, size, no_expand;
380 struct ext4_inode_info *ei = EXT4_I(inode);
381
382 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
383 return -ENOSPC;
384
385 size = ext4_get_max_inline_size(inode);
386 if (size < len)
387 return -ENOSPC;
388
389 ext4_write_lock_xattr(inode, &no_expand);
390
391 if (ei->i_inline_off)
392 ret = ext4_update_inline_data(handle, inode, len);
393 else
394 ret = ext4_create_inline_data(handle, inode, len);
395
396 ext4_write_unlock_xattr(inode, &no_expand);
397 return ret;
398 }
399
400 static int ext4_destroy_inline_data_nolock(handle_t *handle,
401 struct inode *inode)
402 {
403 struct ext4_inode_info *ei = EXT4_I(inode);
404 struct ext4_xattr_ibody_find is = {
405 .s = { .not_found = 0, },
406 };
407 struct ext4_xattr_info i = {
408 .name_index = EXT4_XATTR_INDEX_SYSTEM,
409 .name = EXT4_XATTR_SYSTEM_DATA,
410 .value = NULL,
411 .value_len = 0,
412 };
413 int error;
414
415 if (!ei->i_inline_off)
416 return 0;
417
418 error = ext4_get_inode_loc(inode, &is.iloc);
419 if (error)
420 return error;
421
422 error = ext4_xattr_ibody_find(inode, &i, &is);
423 if (error)
424 goto out;
425
426 BUFFER_TRACE(is.iloc.bh, "get_write_access");
427 error = ext4_journal_get_write_access(handle, is.iloc.bh);
428 if (error)
429 goto out;
430
431 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
432 if (error)
433 goto out;
434
435 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
436 0, EXT4_MIN_INLINE_DATA_SIZE);
437 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
438
439 if (ext4_has_feature_extents(inode->i_sb)) {
440 if (S_ISDIR(inode->i_mode) ||
441 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
442 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
443 ext4_ext_tree_init(handle, inode);
444 }
445 }
446 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
447
448 get_bh(is.iloc.bh);
449 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
450
451 EXT4_I(inode)->i_inline_off = 0;
452 EXT4_I(inode)->i_inline_size = 0;
453 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
454 out:
455 brelse(is.iloc.bh);
456 if (error == -ENODATA)
457 error = 0;
458 return error;
459 }
460
461 static int ext4_read_inline_page(struct inode *inode, struct page *page)
462 {
463 void *kaddr;
464 int ret = 0;
465 size_t len;
466 struct ext4_iloc iloc;
467
468 BUG_ON(!PageLocked(page));
469 BUG_ON(!ext4_has_inline_data(inode));
470 BUG_ON(page->index);
471
472 if (!EXT4_I(inode)->i_inline_off) {
473 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
474 inode->i_ino);
475 goto out;
476 }
477
478 ret = ext4_get_inode_loc(inode, &iloc);
479 if (ret)
480 goto out;
481
482 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
483 kaddr = kmap_atomic(page);
484 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
485 flush_dcache_page(page);
486 kunmap_atomic(kaddr);
487 zero_user_segment(page, len, PAGE_SIZE);
488 SetPageUptodate(page);
489 brelse(iloc.bh);
490
491 out:
492 return ret;
493 }
494
495 int ext4_readpage_inline(struct inode *inode, struct page *page)
496 {
497 int ret = 0;
498
499 down_read(&EXT4_I(inode)->xattr_sem);
500 if (!ext4_has_inline_data(inode)) {
501 up_read(&EXT4_I(inode)->xattr_sem);
502 return -EAGAIN;
503 }
504
505 /*
506 * Current inline data can only exist in the 1st page,
507 * So for all the other pages, just set them uptodate.
508 */
509 if (!page->index)
510 ret = ext4_read_inline_page(inode, page);
511 else if (!PageUptodate(page)) {
512 zero_user_segment(page, 0, PAGE_SIZE);
513 SetPageUptodate(page);
514 }
515
516 up_read(&EXT4_I(inode)->xattr_sem);
517
518 unlock_page(page);
519 return ret >= 0 ? 0 : ret;
520 }
521
522 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
523 struct inode *inode,
524 unsigned flags)
525 {
526 int ret, needed_blocks, no_expand;
527 handle_t *handle = NULL;
528 int retries = 0, sem_held = 0;
529 struct page *page = NULL;
530 unsigned from, to;
531 struct ext4_iloc iloc;
532
533 if (!ext4_has_inline_data(inode)) {
534 /*
535 * clear the flag so that no new write
536 * will trap here again.
537 */
538 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
539 return 0;
540 }
541
542 needed_blocks = ext4_writepage_trans_blocks(inode);
543
544 ret = ext4_get_inode_loc(inode, &iloc);
545 if (ret)
546 return ret;
547
548 retry:
549 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
550 if (IS_ERR(handle)) {
551 ret = PTR_ERR(handle);
552 handle = NULL;
553 goto out;
554 }
555
556 /* We cannot recurse into the filesystem as the transaction is already
557 * started */
558 flags |= AOP_FLAG_NOFS;
559
560 page = grab_cache_page_write_begin(mapping, 0, flags);
561 if (!page) {
562 ret = -ENOMEM;
563 goto out;
564 }
565
566 ext4_write_lock_xattr(inode, &no_expand);
567 sem_held = 1;
568 /* If some one has already done this for us, just exit. */
569 if (!ext4_has_inline_data(inode)) {
570 ret = 0;
571 goto out;
572 }
573
574 from = 0;
575 to = ext4_get_inline_size(inode);
576 if (!PageUptodate(page)) {
577 ret = ext4_read_inline_page(inode, page);
578 if (ret < 0)
579 goto out;
580 }
581
582 ret = ext4_destroy_inline_data_nolock(handle, inode);
583 if (ret)
584 goto out;
585
586 if (ext4_should_dioread_nolock(inode)) {
587 ret = __block_write_begin(page, from, to,
588 ext4_get_block_unwritten);
589 } else
590 ret = __block_write_begin(page, from, to, ext4_get_block);
591
592 if (!ret && ext4_should_journal_data(inode)) {
593 ret = ext4_walk_page_buffers(handle, page_buffers(page),
594 from, to, NULL,
595 do_journal_get_write_access);
596 }
597
598 if (ret) {
599 unlock_page(page);
600 put_page(page);
601 page = NULL;
602 ext4_orphan_add(handle, inode);
603 ext4_write_unlock_xattr(inode, &no_expand);
604 sem_held = 0;
605 ext4_journal_stop(handle);
606 handle = NULL;
607 ext4_truncate_failed_write(inode);
608 /*
609 * If truncate failed early the inode might
610 * still be on the orphan list; we need to
611 * make sure the inode is removed from the
612 * orphan list in that case.
613 */
614 if (inode->i_nlink)
615 ext4_orphan_del(NULL, inode);
616 }
617
618 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
619 goto retry;
620
621 if (page)
622 block_commit_write(page, from, to);
623 out:
624 if (page) {
625 unlock_page(page);
626 put_page(page);
627 }
628 if (sem_held)
629 ext4_write_unlock_xattr(inode, &no_expand);
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 ret = ext4_journal_get_write_access(handle, iloc.bh);
683 if (ret)
684 goto out;
685
686 flags |= AOP_FLAG_NOFS;
687
688 page = grab_cache_page_write_begin(mapping, 0, flags);
689 if (!page) {
690 ret = -ENOMEM;
691 goto out;
692 }
693
694 *pagep = page;
695 down_read(&EXT4_I(inode)->xattr_sem);
696 if (!ext4_has_inline_data(inode)) {
697 ret = 0;
698 unlock_page(page);
699 put_page(page);
700 goto out_up_read;
701 }
702
703 if (!PageUptodate(page)) {
704 ret = ext4_read_inline_page(inode, page);
705 if (ret < 0) {
706 unlock_page(page);
707 put_page(page);
708 goto out_up_read;
709 }
710 }
711
712 ret = 1;
713 handle = NULL;
714 out_up_read:
715 up_read(&EXT4_I(inode)->xattr_sem);
716 out:
717 if (handle && (ret != 1))
718 ext4_journal_stop(handle);
719 brelse(iloc.bh);
720 return ret;
721 convert:
722 return ext4_convert_inline_data_to_extent(mapping,
723 inode, flags);
724 }
725
726 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
727 unsigned copied, struct page *page)
728 {
729 int ret, no_expand;
730 void *kaddr;
731 struct ext4_iloc iloc;
732
733 if (unlikely(copied < len)) {
734 if (!PageUptodate(page)) {
735 copied = 0;
736 goto out;
737 }
738 }
739
740 ret = ext4_get_inode_loc(inode, &iloc);
741 if (ret) {
742 ext4_std_error(inode->i_sb, ret);
743 copied = 0;
744 goto out;
745 }
746
747 ext4_write_lock_xattr(inode, &no_expand);
748 BUG_ON(!ext4_has_inline_data(inode));
749
750 kaddr = kmap_atomic(page);
751 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
752 kunmap_atomic(kaddr);
753 SetPageUptodate(page);
754 /* clear page dirty so that writepages wouldn't work for us. */
755 ClearPageDirty(page);
756
757 ext4_write_unlock_xattr(inode, &no_expand);
758 brelse(iloc.bh);
759 mark_inode_dirty(inode);
760 out:
761 return copied;
762 }
763
764 struct buffer_head *
765 ext4_journalled_write_inline_data(struct inode *inode,
766 unsigned len,
767 struct page *page)
768 {
769 int ret, no_expand;
770 void *kaddr;
771 struct ext4_iloc iloc;
772
773 ret = ext4_get_inode_loc(inode, &iloc);
774 if (ret) {
775 ext4_std_error(inode->i_sb, ret);
776 return NULL;
777 }
778
779 ext4_write_lock_xattr(inode, &no_expand);
780 kaddr = kmap_atomic(page);
781 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
782 kunmap_atomic(kaddr);
783 ext4_write_unlock_xattr(inode, &no_expand);
784
785 return iloc.bh;
786 }
787
788 /*
789 * Try to make the page cache and handle ready for the inline data case.
790 * We can call this function in 2 cases:
791 * 1. The inode is created and the first write exceeds inline size. We can
792 * clear the inode state safely.
793 * 2. The inode has inline data, then we need to read the data, make it
794 * update and dirty so that ext4_da_writepages can handle it. We don't
795 * need to start the journal since the file's metatdata isn't changed now.
796 */
797 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
798 struct inode *inode,
799 unsigned flags,
800 void **fsdata)
801 {
802 int ret = 0, inline_size;
803 struct page *page;
804
805 page = grab_cache_page_write_begin(mapping, 0, flags);
806 if (!page)
807 return -ENOMEM;
808
809 down_read(&EXT4_I(inode)->xattr_sem);
810 if (!ext4_has_inline_data(inode)) {
811 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
812 goto out;
813 }
814
815 inline_size = ext4_get_inline_size(inode);
816
817 if (!PageUptodate(page)) {
818 ret = ext4_read_inline_page(inode, page);
819 if (ret < 0)
820 goto out;
821 }
822
823 ret = __block_write_begin(page, 0, inline_size,
824 ext4_da_get_block_prep);
825 if (ret) {
826 up_read(&EXT4_I(inode)->xattr_sem);
827 unlock_page(page);
828 put_page(page);
829 ext4_truncate_failed_write(inode);
830 return ret;
831 }
832
833 SetPageDirty(page);
834 SetPageUptodate(page);
835 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
836 *fsdata = (void *)CONVERT_INLINE_DATA;
837
838 out:
839 up_read(&EXT4_I(inode)->xattr_sem);
840 if (page) {
841 unlock_page(page);
842 put_page(page);
843 }
844 return ret;
845 }
846
847 /*
848 * Prepare the write for the inline data.
849 * If the the data can be written into the inode, we just read
850 * the page and make it uptodate, and start the journal.
851 * Otherwise read the page, makes it dirty so that it can be
852 * handle in writepages(the i_disksize update is left to the
853 * normal ext4_da_write_end).
854 */
855 int ext4_da_write_inline_data_begin(struct address_space *mapping,
856 struct inode *inode,
857 loff_t pos, unsigned len,
858 unsigned flags,
859 struct page **pagep,
860 void **fsdata)
861 {
862 int ret, inline_size;
863 handle_t *handle;
864 struct page *page;
865 struct ext4_iloc iloc;
866 int retries = 0;
867
868 ret = ext4_get_inode_loc(inode, &iloc);
869 if (ret)
870 return ret;
871
872 retry_journal:
873 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
874 if (IS_ERR(handle)) {
875 ret = PTR_ERR(handle);
876 goto out;
877 }
878
879 inline_size = ext4_get_max_inline_size(inode);
880
881 ret = -ENOSPC;
882 if (inline_size >= pos + len) {
883 ret = ext4_prepare_inline_data(handle, inode, pos + len);
884 if (ret && ret != -ENOSPC)
885 goto out_journal;
886 }
887
888 /*
889 * We cannot recurse into the filesystem as the transaction
890 * is already started.
891 */
892 flags |= AOP_FLAG_NOFS;
893
894 if (ret == -ENOSPC) {
895 ext4_journal_stop(handle);
896 ret = ext4_da_convert_inline_data_to_extent(mapping,
897 inode,
898 flags,
899 fsdata);
900 if (ret == -ENOSPC &&
901 ext4_should_retry_alloc(inode->i_sb, &retries))
902 goto retry_journal;
903 goto out;
904 }
905
906 page = grab_cache_page_write_begin(mapping, 0, flags);
907 if (!page) {
908 ret = -ENOMEM;
909 goto out_journal;
910 }
911
912 down_read(&EXT4_I(inode)->xattr_sem);
913 if (!ext4_has_inline_data(inode)) {
914 ret = 0;
915 goto out_release_page;
916 }
917
918 if (!PageUptodate(page)) {
919 ret = ext4_read_inline_page(inode, page);
920 if (ret < 0)
921 goto out_release_page;
922 }
923 ret = ext4_journal_get_write_access(handle, iloc.bh);
924 if (ret)
925 goto out_release_page;
926
927 up_read(&EXT4_I(inode)->xattr_sem);
928 *pagep = page;
929 brelse(iloc.bh);
930 return 1;
931 out_release_page:
932 up_read(&EXT4_I(inode)->xattr_sem);
933 unlock_page(page);
934 put_page(page);
935 out_journal:
936 ext4_journal_stop(handle);
937 out:
938 brelse(iloc.bh);
939 return ret;
940 }
941
942 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
943 unsigned len, unsigned copied,
944 struct page *page)
945 {
946 int ret;
947
948 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
949 if (ret < 0) {
950 unlock_page(page);
951 put_page(page);
952 return ret;
953 }
954 copied = ret;
955
956 /*
957 * No need to use i_size_read() here, the i_size
958 * cannot change under us because we hold i_mutex.
959 *
960 * But it's important to update i_size while still holding page lock:
961 * page writeout could otherwise come in and zero beyond i_size.
962 */
963 if (pos+copied > inode->i_size)
964 i_size_write(inode, pos+copied);
965 unlock_page(page);
966 put_page(page);
967
968 /*
969 * Don't mark the inode dirty under page lock. First, it unnecessarily
970 * makes the holding time of page lock longer. Second, it forces lock
971 * ordering of page lock and transaction start for journaling
972 * filesystems.
973 */
974 mark_inode_dirty(inode);
975
976 return copied;
977 }
978
979 #ifdef INLINE_DIR_DEBUG
980 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
981 void *inline_start, int inline_size)
982 {
983 int offset;
984 unsigned short de_len;
985 struct ext4_dir_entry_2 *de = inline_start;
986 void *dlimit = inline_start + inline_size;
987
988 trace_printk("inode %lu\n", dir->i_ino);
989 offset = 0;
990 while ((void *)de < dlimit) {
991 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
992 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
993 offset, de_len, de->name_len, de->name,
994 de->name_len, le32_to_cpu(de->inode));
995 if (ext4_check_dir_entry(dir, NULL, de, bh,
996 inline_start, inline_size, offset))
997 BUG();
998
999 offset += de_len;
1000 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1001 }
1002 }
1003 #else
1004 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1005 #endif
1006
1007 /*
1008 * Add a new entry into a inline dir.
1009 * It will return -ENOSPC if no space is available, and -EIO
1010 * and -EEXIST if directory entry already exists.
1011 */
1012 static int ext4_add_dirent_to_inline(handle_t *handle,
1013 struct ext4_filename *fname,
1014 struct inode *dir,
1015 struct inode *inode,
1016 struct ext4_iloc *iloc,
1017 void *inline_start, int inline_size)
1018 {
1019 int err;
1020 struct ext4_dir_entry_2 *de;
1021
1022 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1023 inline_size, fname, &de);
1024 if (err)
1025 return err;
1026
1027 BUFFER_TRACE(iloc->bh, "get_write_access");
1028 err = ext4_journal_get_write_access(handle, iloc->bh);
1029 if (err)
1030 return err;
1031 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1032
1033 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1034
1035 /*
1036 * XXX shouldn't update any times until successful
1037 * completion of syscall, but too many callers depend
1038 * on this.
1039 *
1040 * XXX similarly, too many callers depend on
1041 * ext4_new_inode() setting the times, but error
1042 * recovery deletes the inode, so the worst that can
1043 * happen is that the times are slightly out of date
1044 * and/or different from the directory change time.
1045 */
1046 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1047 ext4_update_dx_flag(dir);
1048 dir->i_version++;
1049 ext4_mark_inode_dirty(handle, dir);
1050 return 1;
1051 }
1052
1053 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1054 struct ext4_iloc *iloc)
1055 {
1056 struct ext4_xattr_entry *entry;
1057 struct ext4_xattr_ibody_header *header;
1058
1059 BUG_ON(!EXT4_I(inode)->i_inline_off);
1060
1061 header = IHDR(inode, ext4_raw_inode(iloc));
1062 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1063 EXT4_I(inode)->i_inline_off);
1064
1065 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1066 }
1067
1068 /* Set the final de to cover the whole block. */
1069 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1070 {
1071 struct ext4_dir_entry_2 *de, *prev_de;
1072 void *limit;
1073 int de_len;
1074
1075 de = (struct ext4_dir_entry_2 *)de_buf;
1076 if (old_size) {
1077 limit = de_buf + old_size;
1078 do {
1079 prev_de = de;
1080 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1081 de_buf += de_len;
1082 de = (struct ext4_dir_entry_2 *)de_buf;
1083 } while (de_buf < limit);
1084
1085 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1086 old_size, new_size);
1087 } else {
1088 /* this is just created, so create an empty entry. */
1089 de->inode = 0;
1090 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1091 }
1092 }
1093
1094 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1095 struct ext4_iloc *iloc)
1096 {
1097 int ret;
1098 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1099 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1100
1101 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1102 return -ENOSPC;
1103
1104 ret = ext4_update_inline_data(handle, dir,
1105 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1106 if (ret)
1107 return ret;
1108
1109 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1110 EXT4_I(dir)->i_inline_size -
1111 EXT4_MIN_INLINE_DATA_SIZE);
1112 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1113 return 0;
1114 }
1115
1116 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1117 struct ext4_iloc *iloc,
1118 void *buf, int inline_size)
1119 {
1120 ext4_create_inline_data(handle, inode, inline_size);
1121 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1122 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1123 }
1124
1125 static int ext4_finish_convert_inline_dir(handle_t *handle,
1126 struct inode *inode,
1127 struct buffer_head *dir_block,
1128 void *buf,
1129 int inline_size)
1130 {
1131 int err, csum_size = 0, header_size = 0;
1132 struct ext4_dir_entry_2 *de;
1133 struct ext4_dir_entry_tail *t;
1134 void *target = dir_block->b_data;
1135
1136 /*
1137 * First create "." and ".." and then copy the dir information
1138 * back to the block.
1139 */
1140 de = (struct ext4_dir_entry_2 *)target;
1141 de = ext4_init_dot_dotdot(inode, de,
1142 inode->i_sb->s_blocksize, csum_size,
1143 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1144 header_size = (void *)de - target;
1145
1146 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1147 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1148
1149 if (ext4_has_metadata_csum(inode->i_sb))
1150 csum_size = sizeof(struct ext4_dir_entry_tail);
1151
1152 inode->i_size = inode->i_sb->s_blocksize;
1153 i_size_write(inode, inode->i_sb->s_blocksize);
1154 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1155 ext4_update_final_de(dir_block->b_data,
1156 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1157 inode->i_sb->s_blocksize - csum_size);
1158
1159 if (csum_size) {
1160 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1161 inode->i_sb->s_blocksize);
1162 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1163 }
1164 set_buffer_uptodate(dir_block);
1165 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1166 if (err)
1167 return err;
1168 set_buffer_verified(dir_block);
1169 return ext4_mark_inode_dirty(handle, inode);
1170 }
1171
1172 static int ext4_convert_inline_data_nolock(handle_t *handle,
1173 struct inode *inode,
1174 struct ext4_iloc *iloc)
1175 {
1176 int error;
1177 void *buf = NULL;
1178 struct buffer_head *data_bh = NULL;
1179 struct ext4_map_blocks map;
1180 int inline_size;
1181
1182 inline_size = ext4_get_inline_size(inode);
1183 buf = kmalloc(inline_size, GFP_NOFS);
1184 if (!buf) {
1185 error = -ENOMEM;
1186 goto out;
1187 }
1188
1189 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1190 if (error < 0)
1191 goto out;
1192
1193 /*
1194 * Make sure the inline directory entries pass checks before we try to
1195 * convert them, so that we avoid touching stuff that needs fsck.
1196 */
1197 if (S_ISDIR(inode->i_mode)) {
1198 error = ext4_check_all_de(inode, iloc->bh,
1199 buf + EXT4_INLINE_DOTDOT_SIZE,
1200 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1201 if (error)
1202 goto out;
1203 }
1204
1205 error = ext4_destroy_inline_data_nolock(handle, inode);
1206 if (error)
1207 goto out;
1208
1209 map.m_lblk = 0;
1210 map.m_len = 1;
1211 map.m_flags = 0;
1212 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1213 if (error < 0)
1214 goto out_restore;
1215 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1216 error = -EIO;
1217 goto out_restore;
1218 }
1219
1220 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1221 if (!data_bh) {
1222 error = -ENOMEM;
1223 goto out_restore;
1224 }
1225
1226 lock_buffer(data_bh);
1227 error = ext4_journal_get_create_access(handle, data_bh);
1228 if (error) {
1229 unlock_buffer(data_bh);
1230 error = -EIO;
1231 goto out_restore;
1232 }
1233 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1234
1235 if (!S_ISDIR(inode->i_mode)) {
1236 memcpy(data_bh->b_data, buf, inline_size);
1237 set_buffer_uptodate(data_bh);
1238 error = ext4_handle_dirty_metadata(handle,
1239 inode, data_bh);
1240 } else {
1241 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1242 buf, inline_size);
1243 }
1244
1245 unlock_buffer(data_bh);
1246 out_restore:
1247 if (error)
1248 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1249
1250 out:
1251 brelse(data_bh);
1252 kfree(buf);
1253 return error;
1254 }
1255
1256 /*
1257 * Try to add the new entry to the inline data.
1258 * If succeeds, return 0. If not, extended the inline dir and copied data to
1259 * the new created block.
1260 */
1261 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1262 struct inode *dir, struct inode *inode)
1263 {
1264 int ret, inline_size, no_expand;
1265 void *inline_start;
1266 struct ext4_iloc iloc;
1267
1268 ret = ext4_get_inode_loc(dir, &iloc);
1269 if (ret)
1270 return ret;
1271
1272 ext4_write_lock_xattr(dir, &no_expand);
1273 if (!ext4_has_inline_data(dir))
1274 goto out;
1275
1276 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1277 EXT4_INLINE_DOTDOT_SIZE;
1278 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1279
1280 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1281 inline_start, inline_size);
1282 if (ret != -ENOSPC)
1283 goto out;
1284
1285 /* check whether it can be inserted to inline xattr space. */
1286 inline_size = EXT4_I(dir)->i_inline_size -
1287 EXT4_MIN_INLINE_DATA_SIZE;
1288 if (!inline_size) {
1289 /* Try to use the xattr space.*/
1290 ret = ext4_update_inline_dir(handle, dir, &iloc);
1291 if (ret && ret != -ENOSPC)
1292 goto out;
1293
1294 inline_size = EXT4_I(dir)->i_inline_size -
1295 EXT4_MIN_INLINE_DATA_SIZE;
1296 }
1297
1298 if (inline_size) {
1299 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1300
1301 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1302 inode, &iloc, inline_start,
1303 inline_size);
1304
1305 if (ret != -ENOSPC)
1306 goto out;
1307 }
1308
1309 /*
1310 * The inline space is filled up, so create a new block for it.
1311 * As the extent tree will be created, we have to save the inline
1312 * dir first.
1313 */
1314 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1315
1316 out:
1317 ext4_mark_inode_dirty(handle, dir);
1318 ext4_write_unlock_xattr(dir, &no_expand);
1319 brelse(iloc.bh);
1320 return ret;
1321 }
1322
1323 /*
1324 * This function fills a red-black tree with information from an
1325 * inlined dir. It returns the number directory entries loaded
1326 * into the tree. If there is an error it is returned in err.
1327 */
1328 int htree_inlinedir_to_tree(struct file *dir_file,
1329 struct inode *dir, ext4_lblk_t block,
1330 struct dx_hash_info *hinfo,
1331 __u32 start_hash, __u32 start_minor_hash,
1332 int *has_inline_data)
1333 {
1334 int err = 0, count = 0;
1335 unsigned int parent_ino;
1336 int pos;
1337 struct ext4_dir_entry_2 *de;
1338 struct inode *inode = file_inode(dir_file);
1339 int ret, inline_size = 0;
1340 struct ext4_iloc iloc;
1341 void *dir_buf = NULL;
1342 struct ext4_dir_entry_2 fake;
1343 struct fscrypt_str tmp_str;
1344
1345 ret = ext4_get_inode_loc(inode, &iloc);
1346 if (ret)
1347 return ret;
1348
1349 down_read(&EXT4_I(inode)->xattr_sem);
1350 if (!ext4_has_inline_data(inode)) {
1351 up_read(&EXT4_I(inode)->xattr_sem);
1352 *has_inline_data = 0;
1353 goto out;
1354 }
1355
1356 inline_size = ext4_get_inline_size(inode);
1357 dir_buf = kmalloc(inline_size, GFP_NOFS);
1358 if (!dir_buf) {
1359 ret = -ENOMEM;
1360 up_read(&EXT4_I(inode)->xattr_sem);
1361 goto out;
1362 }
1363
1364 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1365 up_read(&EXT4_I(inode)->xattr_sem);
1366 if (ret < 0)
1367 goto out;
1368
1369 pos = 0;
1370 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1371 while (pos < inline_size) {
1372 /*
1373 * As inlined dir doesn't store any information about '.' and
1374 * only the inode number of '..' is stored, we have to handle
1375 * them differently.
1376 */
1377 if (pos == 0) {
1378 fake.inode = cpu_to_le32(inode->i_ino);
1379 fake.name_len = 1;
1380 strcpy(fake.name, ".");
1381 fake.rec_len = ext4_rec_len_to_disk(
1382 EXT4_DIR_REC_LEN(fake.name_len),
1383 inline_size);
1384 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1385 de = &fake;
1386 pos = EXT4_INLINE_DOTDOT_OFFSET;
1387 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1388 fake.inode = cpu_to_le32(parent_ino);
1389 fake.name_len = 2;
1390 strcpy(fake.name, "..");
1391 fake.rec_len = ext4_rec_len_to_disk(
1392 EXT4_DIR_REC_LEN(fake.name_len),
1393 inline_size);
1394 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1395 de = &fake;
1396 pos = EXT4_INLINE_DOTDOT_SIZE;
1397 } else {
1398 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1399 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1400 if (ext4_check_dir_entry(inode, dir_file, de,
1401 iloc.bh, dir_buf,
1402 inline_size, pos)) {
1403 ret = count;
1404 goto out;
1405 }
1406 }
1407
1408 ext4fs_dirhash(de->name, de->name_len, hinfo);
1409 if ((hinfo->hash < start_hash) ||
1410 ((hinfo->hash == start_hash) &&
1411 (hinfo->minor_hash < start_minor_hash)))
1412 continue;
1413 if (de->inode == 0)
1414 continue;
1415 tmp_str.name = de->name;
1416 tmp_str.len = de->name_len;
1417 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1418 hinfo->minor_hash, de, &tmp_str);
1419 if (err) {
1420 ret = err;
1421 goto out;
1422 }
1423 count++;
1424 }
1425 ret = count;
1426 out:
1427 kfree(dir_buf);
1428 brelse(iloc.bh);
1429 return ret;
1430 }
1431
1432 /*
1433 * So this function is called when the volume is mkfsed with
1434 * dir_index disabled. In order to keep f_pos persistent
1435 * after we convert from an inlined dir to a blocked based,
1436 * we just pretend that we are a normal dir and return the
1437 * offset as if '.' and '..' really take place.
1438 *
1439 */
1440 int ext4_read_inline_dir(struct file *file,
1441 struct dir_context *ctx,
1442 int *has_inline_data)
1443 {
1444 unsigned int offset, parent_ino;
1445 int i;
1446 struct ext4_dir_entry_2 *de;
1447 struct super_block *sb;
1448 struct inode *inode = file_inode(file);
1449 int ret, inline_size = 0;
1450 struct ext4_iloc iloc;
1451 void *dir_buf = NULL;
1452 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1453
1454 ret = ext4_get_inode_loc(inode, &iloc);
1455 if (ret)
1456 return ret;
1457
1458 down_read(&EXT4_I(inode)->xattr_sem);
1459 if (!ext4_has_inline_data(inode)) {
1460 up_read(&EXT4_I(inode)->xattr_sem);
1461 *has_inline_data = 0;
1462 goto out;
1463 }
1464
1465 inline_size = ext4_get_inline_size(inode);
1466 dir_buf = kmalloc(inline_size, GFP_NOFS);
1467 if (!dir_buf) {
1468 ret = -ENOMEM;
1469 up_read(&EXT4_I(inode)->xattr_sem);
1470 goto out;
1471 }
1472
1473 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1474 up_read(&EXT4_I(inode)->xattr_sem);
1475 if (ret < 0)
1476 goto out;
1477
1478 ret = 0;
1479 sb = inode->i_sb;
1480 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1481 offset = ctx->pos;
1482
1483 /*
1484 * dotdot_offset and dotdot_size is the real offset and
1485 * size for ".." and "." if the dir is block based while
1486 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1487 * So we will use extra_offset and extra_size to indicate them
1488 * during the inline dir iteration.
1489 */
1490 dotdot_offset = EXT4_DIR_REC_LEN(1);
1491 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1492 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1493 extra_size = extra_offset + inline_size;
1494
1495 /*
1496 * If the version has changed since the last call to
1497 * readdir(2), then we might be pointing to an invalid
1498 * dirent right now. Scan from the start of the inline
1499 * dir to make sure.
1500 */
1501 if (file->f_version != inode->i_version) {
1502 for (i = 0; i < extra_size && i < offset;) {
1503 /*
1504 * "." is with offset 0 and
1505 * ".." is dotdot_offset.
1506 */
1507 if (!i) {
1508 i = dotdot_offset;
1509 continue;
1510 } else if (i == dotdot_offset) {
1511 i = dotdot_size;
1512 continue;
1513 }
1514 /* for other entry, the real offset in
1515 * the buf has to be tuned accordingly.
1516 */
1517 de = (struct ext4_dir_entry_2 *)
1518 (dir_buf + i - extra_offset);
1519 /* It's too expensive to do a full
1520 * dirent test each time round this
1521 * loop, but we do have to test at
1522 * least that it is non-zero. A
1523 * failure will be detected in the
1524 * dirent test below. */
1525 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1526 < EXT4_DIR_REC_LEN(1))
1527 break;
1528 i += ext4_rec_len_from_disk(de->rec_len,
1529 extra_size);
1530 }
1531 offset = i;
1532 ctx->pos = offset;
1533 file->f_version = inode->i_version;
1534 }
1535
1536 while (ctx->pos < extra_size) {
1537 if (ctx->pos == 0) {
1538 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1539 goto out;
1540 ctx->pos = dotdot_offset;
1541 continue;
1542 }
1543
1544 if (ctx->pos == dotdot_offset) {
1545 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1546 goto out;
1547 ctx->pos = dotdot_size;
1548 continue;
1549 }
1550
1551 de = (struct ext4_dir_entry_2 *)
1552 (dir_buf + ctx->pos - extra_offset);
1553 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1554 extra_size, ctx->pos))
1555 goto out;
1556 if (le32_to_cpu(de->inode)) {
1557 if (!dir_emit(ctx, de->name, de->name_len,
1558 le32_to_cpu(de->inode),
1559 get_dtype(sb, de->file_type)))
1560 goto out;
1561 }
1562 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1563 }
1564 out:
1565 kfree(dir_buf);
1566 brelse(iloc.bh);
1567 return ret;
1568 }
1569
1570 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1571 struct ext4_dir_entry_2 **parent_de,
1572 int *retval)
1573 {
1574 struct ext4_iloc iloc;
1575
1576 *retval = ext4_get_inode_loc(inode, &iloc);
1577 if (*retval)
1578 return NULL;
1579
1580 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1581
1582 return iloc.bh;
1583 }
1584
1585 /*
1586 * Try to create the inline data for the new dir.
1587 * If it succeeds, return 0, otherwise return the error.
1588 * In case of ENOSPC, the caller should create the normal disk layout dir.
1589 */
1590 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1591 struct inode *inode)
1592 {
1593 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1594 struct ext4_iloc iloc;
1595 struct ext4_dir_entry_2 *de;
1596
1597 ret = ext4_get_inode_loc(inode, &iloc);
1598 if (ret)
1599 return ret;
1600
1601 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1602 if (ret)
1603 goto out;
1604
1605 /*
1606 * For inline dir, we only save the inode information for the ".."
1607 * and create a fake dentry to cover the left space.
1608 */
1609 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1610 de->inode = cpu_to_le32(parent->i_ino);
1611 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1612 de->inode = 0;
1613 de->rec_len = ext4_rec_len_to_disk(
1614 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1615 inline_size);
1616 set_nlink(inode, 2);
1617 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1618 out:
1619 brelse(iloc.bh);
1620 return ret;
1621 }
1622
1623 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1624 struct ext4_filename *fname,
1625 const struct qstr *d_name,
1626 struct ext4_dir_entry_2 **res_dir,
1627 int *has_inline_data)
1628 {
1629 int ret;
1630 struct ext4_iloc iloc;
1631 void *inline_start;
1632 int inline_size;
1633
1634 if (ext4_get_inode_loc(dir, &iloc))
1635 return NULL;
1636
1637 down_read(&EXT4_I(dir)->xattr_sem);
1638 if (!ext4_has_inline_data(dir)) {
1639 *has_inline_data = 0;
1640 goto out;
1641 }
1642
1643 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1644 EXT4_INLINE_DOTDOT_SIZE;
1645 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1646 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1647 dir, fname, d_name, 0, res_dir);
1648 if (ret == 1)
1649 goto out_find;
1650 if (ret < 0)
1651 goto out;
1652
1653 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1654 goto out;
1655
1656 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1657 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1658
1659 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1660 dir, fname, d_name, 0, res_dir);
1661 if (ret == 1)
1662 goto out_find;
1663
1664 out:
1665 brelse(iloc.bh);
1666 iloc.bh = NULL;
1667 out_find:
1668 up_read(&EXT4_I(dir)->xattr_sem);
1669 return iloc.bh;
1670 }
1671
1672 int ext4_delete_inline_entry(handle_t *handle,
1673 struct inode *dir,
1674 struct ext4_dir_entry_2 *de_del,
1675 struct buffer_head *bh,
1676 int *has_inline_data)
1677 {
1678 int err, inline_size, no_expand;
1679 struct ext4_iloc iloc;
1680 void *inline_start;
1681
1682 err = ext4_get_inode_loc(dir, &iloc);
1683 if (err)
1684 return err;
1685
1686 ext4_write_lock_xattr(dir, &no_expand);
1687 if (!ext4_has_inline_data(dir)) {
1688 *has_inline_data = 0;
1689 goto out;
1690 }
1691
1692 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1693 EXT4_MIN_INLINE_DATA_SIZE) {
1694 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1695 EXT4_INLINE_DOTDOT_SIZE;
1696 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1697 EXT4_INLINE_DOTDOT_SIZE;
1698 } else {
1699 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1700 inline_size = ext4_get_inline_size(dir) -
1701 EXT4_MIN_INLINE_DATA_SIZE;
1702 }
1703
1704 BUFFER_TRACE(bh, "get_write_access");
1705 err = ext4_journal_get_write_access(handle, bh);
1706 if (err)
1707 goto out;
1708
1709 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1710 inline_start, inline_size, 0);
1711 if (err)
1712 goto out;
1713
1714 err = ext4_mark_inode_dirty(handle, dir);
1715 if (unlikely(err))
1716 goto out;
1717
1718 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1719 out:
1720 ext4_write_unlock_xattr(dir, &no_expand);
1721 brelse(iloc.bh);
1722 if (err != -ENOENT)
1723 ext4_std_error(dir->i_sb, err);
1724 return err;
1725 }
1726
1727 /*
1728 * Get the inline dentry at offset.
1729 */
1730 static inline struct ext4_dir_entry_2 *
1731 ext4_get_inline_entry(struct inode *inode,
1732 struct ext4_iloc *iloc,
1733 unsigned int offset,
1734 void **inline_start,
1735 int *inline_size)
1736 {
1737 void *inline_pos;
1738
1739 BUG_ON(offset > ext4_get_inline_size(inode));
1740
1741 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1742 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1743 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1744 } else {
1745 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1746 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1747 *inline_size = ext4_get_inline_size(inode) -
1748 EXT4_MIN_INLINE_DATA_SIZE;
1749 }
1750
1751 if (inline_start)
1752 *inline_start = inline_pos;
1753 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1754 }
1755
1756 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1757 {
1758 int err, inline_size;
1759 struct ext4_iloc iloc;
1760 size_t inline_len;
1761 void *inline_pos;
1762 unsigned int offset;
1763 struct ext4_dir_entry_2 *de;
1764 bool ret = true;
1765
1766 err = ext4_get_inode_loc(dir, &iloc);
1767 if (err) {
1768 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1769 err, dir->i_ino);
1770 return true;
1771 }
1772
1773 down_read(&EXT4_I(dir)->xattr_sem);
1774 if (!ext4_has_inline_data(dir)) {
1775 *has_inline_data = 0;
1776 goto out;
1777 }
1778
1779 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1780 if (!le32_to_cpu(de->inode)) {
1781 ext4_warning(dir->i_sb,
1782 "bad inline directory (dir #%lu) - no `..'",
1783 dir->i_ino);
1784 ret = true;
1785 goto out;
1786 }
1787
1788 inline_len = ext4_get_inline_size(dir);
1789 offset = EXT4_INLINE_DOTDOT_SIZE;
1790 while (offset < inline_len) {
1791 de = ext4_get_inline_entry(dir, &iloc, offset,
1792 &inline_pos, &inline_size);
1793 if (ext4_check_dir_entry(dir, NULL, de,
1794 iloc.bh, inline_pos,
1795 inline_size, offset)) {
1796 ext4_warning(dir->i_sb,
1797 "bad inline directory (dir #%lu) - "
1798 "inode %u, rec_len %u, name_len %d"
1799 "inline size %d",
1800 dir->i_ino, le32_to_cpu(de->inode),
1801 le16_to_cpu(de->rec_len), de->name_len,
1802 inline_size);
1803 ret = true;
1804 goto out;
1805 }
1806 if (le32_to_cpu(de->inode)) {
1807 ret = false;
1808 goto out;
1809 }
1810 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1811 }
1812
1813 out:
1814 up_read(&EXT4_I(dir)->xattr_sem);
1815 brelse(iloc.bh);
1816 return ret;
1817 }
1818
1819 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1820 {
1821 int ret, no_expand;
1822
1823 ext4_write_lock_xattr(inode, &no_expand);
1824 ret = ext4_destroy_inline_data_nolock(handle, inode);
1825 ext4_write_unlock_xattr(inode, &no_expand);
1826
1827 return ret;
1828 }
1829
1830 int ext4_inline_data_fiemap(struct inode *inode,
1831 struct fiemap_extent_info *fieinfo,
1832 int *has_inline, __u64 start, __u64 len)
1833 {
1834 __u64 physical = 0;
1835 __u64 inline_len;
1836 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1837 FIEMAP_EXTENT_LAST;
1838 int error = 0;
1839 struct ext4_iloc iloc;
1840
1841 down_read(&EXT4_I(inode)->xattr_sem);
1842 if (!ext4_has_inline_data(inode)) {
1843 *has_inline = 0;
1844 goto out;
1845 }
1846 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1847 i_size_read(inode));
1848 if (start >= inline_len)
1849 goto out;
1850 if (start + len < inline_len)
1851 inline_len = start + len;
1852 inline_len -= start;
1853
1854 error = ext4_get_inode_loc(inode, &iloc);
1855 if (error)
1856 goto out;
1857
1858 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1859 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1860 physical += offsetof(struct ext4_inode, i_block);
1861
1862 brelse(iloc.bh);
1863 out:
1864 up_read(&EXT4_I(inode)->xattr_sem);
1865 if (physical)
1866 error = fiemap_fill_next_extent(fieinfo, start, physical,
1867 inline_len, flags);
1868 return (error < 0 ? error : 0);
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, no_expand;
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 ext4_write_lock_xattr(inode, &no_expand);
1892 if (!ext4_has_inline_data(inode)) {
1893 ext4_write_unlock_xattr(inode, &no_expand);
1894 *has_inline = 0;
1895 ext4_journal_stop(handle);
1896 return;
1897 }
1898
1899 if (ext4_orphan_add(handle, inode))
1900 goto out;
1901
1902 if (ext4_get_inode_loc(inode, &is.iloc))
1903 goto out;
1904
1905 down_write(&EXT4_I(inode)->i_data_sem);
1906 i_size = inode->i_size;
1907 inline_size = ext4_get_inline_size(inode);
1908 EXT4_I(inode)->i_disksize = i_size;
1909
1910 if (i_size < inline_size) {
1911 /* Clear the content in the xattr space. */
1912 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1913 if (ext4_xattr_ibody_find(inode, &i, &is))
1914 goto out_error;
1915
1916 BUG_ON(is.s.not_found);
1917
1918 value_len = le32_to_cpu(is.s.here->e_value_size);
1919 value = kmalloc(value_len, GFP_NOFS);
1920 if (!value)
1921 goto out_error;
1922
1923 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1924 value, value_len))
1925 goto out_error;
1926
1927 i.value = value;
1928 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1929 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1930 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1931 goto out_error;
1932 }
1933
1934 /* Clear the content within i_blocks. */
1935 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1936 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1937 memset(p + i_size, 0,
1938 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1939 }
1940
1941 EXT4_I(inode)->i_inline_size = i_size <
1942 EXT4_MIN_INLINE_DATA_SIZE ?
1943 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1944 }
1945
1946 out_error:
1947 up_write(&EXT4_I(inode)->i_data_sem);
1948 out:
1949 brelse(is.iloc.bh);
1950 ext4_write_unlock_xattr(inode, &no_expand);
1951 kfree(value);
1952 if (inode->i_nlink)
1953 ext4_orphan_del(handle, inode);
1954
1955 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1956 ext4_mark_inode_dirty(handle, inode);
1957 if (IS_SYNC(inode))
1958 ext4_handle_sync(handle);
1959
1960 ext4_journal_stop(handle);
1961 return;
1962 }
1963
1964 int ext4_convert_inline_data(struct inode *inode)
1965 {
1966 int error, needed_blocks, no_expand;
1967 handle_t *handle;
1968 struct ext4_iloc iloc;
1969
1970 if (!ext4_has_inline_data(inode)) {
1971 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1972 return 0;
1973 }
1974
1975 needed_blocks = ext4_writepage_trans_blocks(inode);
1976
1977 iloc.bh = NULL;
1978 error = ext4_get_inode_loc(inode, &iloc);
1979 if (error)
1980 return error;
1981
1982 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1983 if (IS_ERR(handle)) {
1984 error = PTR_ERR(handle);
1985 goto out_free;
1986 }
1987
1988 ext4_write_lock_xattr(inode, &no_expand);
1989 if (ext4_has_inline_data(inode))
1990 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1991 ext4_write_unlock_xattr(inode, &no_expand);
1992 ext4_journal_stop(handle);
1993 out_free:
1994 brelse(iloc.bh);
1995 return error;
1996 }
Linux with added FPC-III support