?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
powerpc/rfi-flush: Make it possible to call setup_rfi_flush() again
[linux/fpc-iii.git] / fs / ext4 / inline.c
blob0dcd33f626376a39ce9ecd262f61635c95a7971f
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_CACHE_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_CACHE_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, ext4_get_block_write);
588 else
589 ret = __block_write_begin(page, from, to, ext4_get_block);
590
591 if (!ret && ext4_should_journal_data(inode)) {
592 ret = ext4_walk_page_buffers(handle, page_buffers(page),
593 from, to, NULL,
594 do_journal_get_write_access);
595 }
596
597 if (ret) {
598 unlock_page(page);
599 page_cache_release(page);
600 page = NULL;
601 ext4_orphan_add(handle, inode);
602 ext4_write_unlock_xattr(inode, &no_expand);
603 sem_held = 0;
604 ext4_journal_stop(handle);
605 handle = NULL;
606 ext4_truncate_failed_write(inode);
607 /*
608 * If truncate failed early the inode might
609 * still be on the orphan list; we need to
610 * make sure the inode is removed from the
611 * orphan list in that case.
612 */
613 if (inode->i_nlink)
614 ext4_orphan_del(NULL, inode);
615 }
616
617 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
618 goto retry;
619
620 if (page)
621 block_commit_write(page, from, to);
622 out:
623 if (page) {
624 unlock_page(page);
625 page_cache_release(page);
626 }
627 if (sem_held)
628 ext4_write_unlock_xattr(inode, &no_expand);
629 if (handle)
630 ext4_journal_stop(handle);
631 brelse(iloc.bh);
632 return ret;
633 }
634
635 /*
636 * Try to write data in the inode.
637 * If the inode has inline data, check whether the new write can be
638 * in the inode also. If not, create the page the handle, move the data
639 * to the page make it update and let the later codes create extent for it.
640 */
641 int ext4_try_to_write_inline_data(struct address_space *mapping,
642 struct inode *inode,
643 loff_t pos, unsigned len,
644 unsigned flags,
645 struct page **pagep)
646 {
647 int ret;
648 handle_t *handle;
649 struct page *page;
650 struct ext4_iloc iloc;
651
652 if (pos + len > ext4_get_max_inline_size(inode))
653 goto convert;
654
655 ret = ext4_get_inode_loc(inode, &iloc);
656 if (ret)
657 return ret;
658
659 /*
660 * The possible write could happen in the inode,
661 * so try to reserve the space in inode first.
662 */
663 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
664 if (IS_ERR(handle)) {
665 ret = PTR_ERR(handle);
666 handle = NULL;
667 goto out;
668 }
669
670 ret = ext4_prepare_inline_data(handle, inode, pos + len);
671 if (ret && ret != -ENOSPC)
672 goto out;
673
674 /* We don't have space in inline inode, so convert it to extent. */
675 if (ret == -ENOSPC) {
676 ext4_journal_stop(handle);
677 brelse(iloc.bh);
678 goto convert;
679 }
680
681 ret = ext4_journal_get_write_access(handle, iloc.bh);
682 if (ret)
683 goto out;
684
685 flags |= AOP_FLAG_NOFS;
686
687 page = grab_cache_page_write_begin(mapping, 0, flags);
688 if (!page) {
689 ret = -ENOMEM;
690 goto out;
691 }
692
693 *pagep = page;
694 down_read(&EXT4_I(inode)->xattr_sem);
695 if (!ext4_has_inline_data(inode)) {
696 ret = 0;
697 unlock_page(page);
698 page_cache_release(page);
699 goto out_up_read;
700 }
701
702 if (!PageUptodate(page)) {
703 ret = ext4_read_inline_page(inode, page);
704 if (ret < 0) {
705 unlock_page(page);
706 put_page(page);
707 goto out_up_read;
708 }
709 }
710
711 ret = 1;
712 handle = NULL;
713 out_up_read:
714 up_read(&EXT4_I(inode)->xattr_sem);
715 out:
716 if (handle && (ret != 1))
717 ext4_journal_stop(handle);
718 brelse(iloc.bh);
719 return ret;
720 convert:
721 return ext4_convert_inline_data_to_extent(mapping,
722 inode, flags);
723 }
724
725 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
726 unsigned copied, struct page *page)
727 {
728 int ret, no_expand;
729 void *kaddr;
730 struct ext4_iloc iloc;
731
732 if (unlikely(copied < len)) {
733 if (!PageUptodate(page)) {
734 copied = 0;
735 goto out;
736 }
737 }
738
739 ret = ext4_get_inode_loc(inode, &iloc);
740 if (ret) {
741 ext4_std_error(inode->i_sb, ret);
742 copied = 0;
743 goto out;
744 }
745
746 ext4_write_lock_xattr(inode, &no_expand);
747 BUG_ON(!ext4_has_inline_data(inode));
748
749 kaddr = kmap_atomic(page);
750 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
751 kunmap_atomic(kaddr);
752 SetPageUptodate(page);
753 /* clear page dirty so that writepages wouldn't work for us. */
754 ClearPageDirty(page);
755
756 ext4_write_unlock_xattr(inode, &no_expand);
757 brelse(iloc.bh);
758 mark_inode_dirty(inode);
759 out:
760 return copied;
761 }
762
763 struct buffer_head *
764 ext4_journalled_write_inline_data(struct inode *inode,
765 unsigned len,
766 struct page *page)
767 {
768 int ret, no_expand;
769 void *kaddr;
770 struct ext4_iloc iloc;
771
772 ret = ext4_get_inode_loc(inode, &iloc);
773 if (ret) {
774 ext4_std_error(inode->i_sb, ret);
775 return NULL;
776 }
777
778 ext4_write_lock_xattr(inode, &no_expand);
779 kaddr = kmap_atomic(page);
780 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
781 kunmap_atomic(kaddr);
782 ext4_write_unlock_xattr(inode, &no_expand);
783
784 return iloc.bh;
785 }
786
787 /*
788 * Try to make the page cache and handle ready for the inline data case.
789 * We can call this function in 2 cases:
790 * 1. The inode is created and the first write exceeds inline size. We can
791 * clear the inode state safely.
792 * 2. The inode has inline data, then we need to read the data, make it
793 * update and dirty so that ext4_da_writepages can handle it. We don't
794 * need to start the journal since the file's metatdata isn't changed now.
795 */
796 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
797 struct inode *inode,
798 unsigned flags,
799 void **fsdata)
800 {
801 int ret = 0, inline_size;
802 struct page *page;
803
804 page = grab_cache_page_write_begin(mapping, 0, flags);
805 if (!page)
806 return -ENOMEM;
807
808 down_read(&EXT4_I(inode)->xattr_sem);
809 if (!ext4_has_inline_data(inode)) {
810 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
811 goto out;
812 }
813
814 inline_size = ext4_get_inline_size(inode);
815
816 if (!PageUptodate(page)) {
817 ret = ext4_read_inline_page(inode, page);
818 if (ret < 0)
819 goto out;
820 }
821
822 ret = __block_write_begin(page, 0, inline_size,
823 ext4_da_get_block_prep);
824 if (ret) {
825 up_read(&EXT4_I(inode)->xattr_sem);
826 unlock_page(page);
827 page_cache_release(page);
828 ext4_truncate_failed_write(inode);
829 return ret;
830 }
831
832 SetPageDirty(page);
833 SetPageUptodate(page);
834 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
835 *fsdata = (void *)CONVERT_INLINE_DATA;
836
837 out:
838 up_read(&EXT4_I(inode)->xattr_sem);
839 if (page) {
840 unlock_page(page);
841 page_cache_release(page);
842 }
843 return ret;
844 }
845
846 /*
847 * Prepare the write for the inline data.
848 * If the the data can be written into the inode, we just read
849 * the page and make it uptodate, and start the journal.
850 * Otherwise read the page, makes it dirty so that it can be
851 * handle in writepages(the i_disksize update is left to the
852 * normal ext4_da_write_end).
853 */
854 int ext4_da_write_inline_data_begin(struct address_space *mapping,
855 struct inode *inode,
856 loff_t pos, unsigned len,
857 unsigned flags,
858 struct page **pagep,
859 void **fsdata)
860 {
861 int ret, inline_size;
862 handle_t *handle;
863 struct page *page;
864 struct ext4_iloc iloc;
865 int retries = 0;
866
867 ret = ext4_get_inode_loc(inode, &iloc);
868 if (ret)
869 return ret;
870
871 retry_journal:
872 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
873 if (IS_ERR(handle)) {
874 ret = PTR_ERR(handle);
875 goto out;
876 }
877
878 inline_size = ext4_get_max_inline_size(inode);
879
880 ret = -ENOSPC;
881 if (inline_size >= pos + len) {
882 ret = ext4_prepare_inline_data(handle, inode, pos + len);
883 if (ret && ret != -ENOSPC)
884 goto out_journal;
885 }
886
887 /*
888 * We cannot recurse into the filesystem as the transaction
889 * is already started.
890 */
891 flags |= AOP_FLAG_NOFS;
892
893 if (ret == -ENOSPC) {
894 ext4_journal_stop(handle);
895 ret = ext4_da_convert_inline_data_to_extent(mapping,
896 inode,
897 flags,
898 fsdata);
899 if (ret == -ENOSPC &&
900 ext4_should_retry_alloc(inode->i_sb, &retries))
901 goto retry_journal;
902 goto out;
903 }
904
905 page = grab_cache_page_write_begin(mapping, 0, flags);
906 if (!page) {
907 ret = -ENOMEM;
908 goto out_journal;
909 }
910
911 down_read(&EXT4_I(inode)->xattr_sem);
912 if (!ext4_has_inline_data(inode)) {
913 ret = 0;
914 goto out_release_page;
915 }
916
917 if (!PageUptodate(page)) {
918 ret = ext4_read_inline_page(inode, page);
919 if (ret < 0)
920 goto out_release_page;
921 }
922 ret = ext4_journal_get_write_access(handle, iloc.bh);
923 if (ret)
924 goto out_release_page;
925
926 up_read(&EXT4_I(inode)->xattr_sem);
927 *pagep = page;
928 brelse(iloc.bh);
929 return 1;
930 out_release_page:
931 up_read(&EXT4_I(inode)->xattr_sem);
932 unlock_page(page);
933 page_cache_release(page);
934 out_journal:
935 ext4_journal_stop(handle);
936 out:
937 brelse(iloc.bh);
938 return ret;
939 }
940
941 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
942 unsigned len, unsigned copied,
943 struct page *page)
944 {
945 int ret;
946
947 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
948 if (ret < 0) {
949 unlock_page(page);
950 put_page(page);
951 return ret;
952 }
953 copied = ret;
954
955 /*
956 * No need to use i_size_read() here, the i_size
957 * cannot change under us because we hold i_mutex.
958 *
959 * But it's important to update i_size while still holding page lock:
960 * page writeout could otherwise come in and zero beyond i_size.
961 */
962 if (pos+copied > inode->i_size)
963 i_size_write(inode, pos+copied);
964 unlock_page(page);
965 page_cache_release(page);
966
967 /*
968 * Don't mark the inode dirty under page lock. First, it unnecessarily
969 * makes the holding time of page lock longer. Second, it forces lock
970 * ordering of page lock and transaction start for journaling
971 * filesystems.
972 */
973 mark_inode_dirty(inode);
974
975 return copied;
976 }
977
978 #ifdef INLINE_DIR_DEBUG
979 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
980 void *inline_start, int inline_size)
981 {
982 int offset;
983 unsigned short de_len;
984 struct ext4_dir_entry_2 *de = inline_start;
985 void *dlimit = inline_start + inline_size;
986
987 trace_printk("inode %lu\n", dir->i_ino);
988 offset = 0;
989 while ((void *)de < dlimit) {
990 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
991 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
992 offset, de_len, de->name_len, de->name,
993 de->name_len, le32_to_cpu(de->inode));
994 if (ext4_check_dir_entry(dir, NULL, de, bh,
995 inline_start, inline_size, offset))
996 BUG();
997
998 offset += de_len;
999 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1000 }
1001 }
1002 #else
1003 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1004 #endif
1005
1006 /*
1007 * Add a new entry into a inline dir.
1008 * It will return -ENOSPC if no space is available, and -EIO
1009 * and -EEXIST if directory entry already exists.
1010 */
1011 static int ext4_add_dirent_to_inline(handle_t *handle,
1012 struct ext4_filename *fname,
1013 struct dentry *dentry,
1014 struct inode *inode,
1015 struct ext4_iloc *iloc,
1016 void *inline_start, int inline_size)
1017 {
1018 struct inode *dir = d_inode(dentry->d_parent);
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 dentry *dentry, struct inode *inode)
1263 {
1264 int ret, inline_size, no_expand;
1265 void *inline_start;
1266 struct ext4_iloc iloc;
1267 struct inode *dir = d_inode(dentry->d_parent);
1268
1269 ret = ext4_get_inode_loc(dir, &iloc);
1270 if (ret)
1271 return ret;
1272
1273 ext4_write_lock_xattr(dir, &no_expand);
1274 if (!ext4_has_inline_data(dir))
1275 goto out;
1276
1277 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1278 EXT4_INLINE_DOTDOT_SIZE;
1279 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1280
1281 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1282 inline_start, inline_size);
1283 if (ret != -ENOSPC)
1284 goto out;
1285
1286 /* check whether it can be inserted to inline xattr space. */
1287 inline_size = EXT4_I(dir)->i_inline_size -
1288 EXT4_MIN_INLINE_DATA_SIZE;
1289 if (!inline_size) {
1290 /* Try to use the xattr space.*/
1291 ret = ext4_update_inline_dir(handle, dir, &iloc);
1292 if (ret && ret != -ENOSPC)
1293 goto out;
1294
1295 inline_size = EXT4_I(dir)->i_inline_size -
1296 EXT4_MIN_INLINE_DATA_SIZE;
1297 }
1298
1299 if (inline_size) {
1300 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1301
1302 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1303 inode, &iloc, inline_start,
1304 inline_size);
1305
1306 if (ret != -ENOSPC)
1307 goto out;
1308 }
1309
1310 /*
1311 * The inline space is filled up, so create a new block for it.
1312 * As the extent tree will be created, we have to save the inline
1313 * dir first.
1314 */
1315 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1316
1317 out:
1318 ext4_mark_inode_dirty(handle, dir);
1319 ext4_write_unlock_xattr(dir, &no_expand);
1320 brelse(iloc.bh);
1321 return ret;
1322 }
1323
1324 /*
1325 * This function fills a red-black tree with information from an
1326 * inlined dir. It returns the number directory entries loaded
1327 * into the tree. If there is an error it is returned in err.
1328 */
1329 int htree_inlinedir_to_tree(struct file *dir_file,
1330 struct inode *dir, ext4_lblk_t block,
1331 struct dx_hash_info *hinfo,
1332 __u32 start_hash, __u32 start_minor_hash,
1333 int *has_inline_data)
1334 {
1335 int err = 0, count = 0;
1336 unsigned int parent_ino;
1337 int pos;
1338 struct ext4_dir_entry_2 *de;
1339 struct inode *inode = file_inode(dir_file);
1340 int ret, inline_size = 0;
1341 struct ext4_iloc iloc;
1342 void *dir_buf = NULL;
1343 struct ext4_dir_entry_2 fake;
1344 struct ext4_str tmp_str;
1345
1346 ret = ext4_get_inode_loc(inode, &iloc);
1347 if (ret)
1348 return ret;
1349
1350 down_read(&EXT4_I(inode)->xattr_sem);
1351 if (!ext4_has_inline_data(inode)) {
1352 up_read(&EXT4_I(inode)->xattr_sem);
1353 *has_inline_data = 0;
1354 goto out;
1355 }
1356
1357 inline_size = ext4_get_inline_size(inode);
1358 dir_buf = kmalloc(inline_size, GFP_NOFS);
1359 if (!dir_buf) {
1360 ret = -ENOMEM;
1361 up_read(&EXT4_I(inode)->xattr_sem);
1362 goto out;
1363 }
1364
1365 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1366 up_read(&EXT4_I(inode)->xattr_sem);
1367 if (ret < 0)
1368 goto out;
1369
1370 pos = 0;
1371 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1372 while (pos < inline_size) {
1373 /*
1374 * As inlined dir doesn't store any information about '.' and
1375 * only the inode number of '..' is stored, we have to handle
1376 * them differently.
1377 */
1378 if (pos == 0) {
1379 fake.inode = cpu_to_le32(inode->i_ino);
1380 fake.name_len = 1;
1381 strcpy(fake.name, ".");
1382 fake.rec_len = ext4_rec_len_to_disk(
1383 EXT4_DIR_REC_LEN(fake.name_len),
1384 inline_size);
1385 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1386 de = &fake;
1387 pos = EXT4_INLINE_DOTDOT_OFFSET;
1388 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1389 fake.inode = cpu_to_le32(parent_ino);
1390 fake.name_len = 2;
1391 strcpy(fake.name, "..");
1392 fake.rec_len = ext4_rec_len_to_disk(
1393 EXT4_DIR_REC_LEN(fake.name_len),
1394 inline_size);
1395 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1396 de = &fake;
1397 pos = EXT4_INLINE_DOTDOT_SIZE;
1398 } else {
1399 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1400 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1401 if (ext4_check_dir_entry(inode, dir_file, de,
1402 iloc.bh, dir_buf,
1403 inline_size, pos)) {
1404 ret = count;
1405 goto out;
1406 }
1407 }
1408
1409 ext4fs_dirhash(de->name, de->name_len, hinfo);
1410 if ((hinfo->hash < start_hash) ||
1411 ((hinfo->hash == start_hash) &&
1412 (hinfo->minor_hash < start_minor_hash)))
1413 continue;
1414 if (de->inode == 0)
1415 continue;
1416 tmp_str.name = de->name;
1417 tmp_str.len = de->name_len;
1418 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1419 hinfo->minor_hash, de, &tmp_str);
1420 if (err) {
1421 count = err;
1422 goto out;
1423 }
1424 count++;
1425 }
1426 ret = count;
1427 out:
1428 kfree(dir_buf);
1429 brelse(iloc.bh);
1430 return ret;
1431 }
1432
1433 /*
1434 * So this function is called when the volume is mkfsed with
1435 * dir_index disabled. In order to keep f_pos persistent
1436 * after we convert from an inlined dir to a blocked based,
1437 * we just pretend that we are a normal dir and return the
1438 * offset as if '.' and '..' really take place.
1439 *
1440 */
1441 int ext4_read_inline_dir(struct file *file,
1442 struct dir_context *ctx,
1443 int *has_inline_data)
1444 {
1445 unsigned int offset, parent_ino;
1446 int i;
1447 struct ext4_dir_entry_2 *de;
1448 struct super_block *sb;
1449 struct inode *inode = file_inode(file);
1450 int ret, inline_size = 0;
1451 struct ext4_iloc iloc;
1452 void *dir_buf = NULL;
1453 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1454
1455 ret = ext4_get_inode_loc(inode, &iloc);
1456 if (ret)
1457 return ret;
1458
1459 down_read(&EXT4_I(inode)->xattr_sem);
1460 if (!ext4_has_inline_data(inode)) {
1461 up_read(&EXT4_I(inode)->xattr_sem);
1462 *has_inline_data = 0;
1463 goto out;
1464 }
1465
1466 inline_size = ext4_get_inline_size(inode);
1467 dir_buf = kmalloc(inline_size, GFP_NOFS);
1468 if (!dir_buf) {
1469 ret = -ENOMEM;
1470 up_read(&EXT4_I(inode)->xattr_sem);
1471 goto out;
1472 }
1473
1474 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1475 up_read(&EXT4_I(inode)->xattr_sem);
1476 if (ret < 0)
1477 goto out;
1478
1479 ret = 0;
1480 sb = inode->i_sb;
1481 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1482 offset = ctx->pos;
1483
1484 /*
1485 * dotdot_offset and dotdot_size is the real offset and
1486 * size for ".." and "." if the dir is block based while
1487 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1488 * So we will use extra_offset and extra_size to indicate them
1489 * during the inline dir iteration.
1490 */
1491 dotdot_offset = EXT4_DIR_REC_LEN(1);
1492 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1493 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1494 extra_size = extra_offset + inline_size;
1495
1496 /*
1497 * If the version has changed since the last call to
1498 * readdir(2), then we might be pointing to an invalid
1499 * dirent right now. Scan from the start of the inline
1500 * dir to make sure.
1501 */
1502 if (file->f_version != inode->i_version) {
1503 for (i = 0; i < extra_size && i < offset;) {
1504 /*
1505 * "." is with offset 0 and
1506 * ".." is dotdot_offset.
1507 */
1508 if (!i) {
1509 i = dotdot_offset;
1510 continue;
1511 } else if (i == dotdot_offset) {
1512 i = dotdot_size;
1513 continue;
1514 }
1515 /* for other entry, the real offset in
1516 * the buf has to be tuned accordingly.
1517 */
1518 de = (struct ext4_dir_entry_2 *)
1519 (dir_buf + i - extra_offset);
1520 /* It's too expensive to do a full
1521 * dirent test each time round this
1522 * loop, but we do have to test at
1523 * least that it is non-zero. A
1524 * failure will be detected in the
1525 * dirent test below. */
1526 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1527 < EXT4_DIR_REC_LEN(1))
1528 break;
1529 i += ext4_rec_len_from_disk(de->rec_len,
1530 extra_size);
1531 }
1532 offset = i;
1533 ctx->pos = offset;
1534 file->f_version = inode->i_version;
1535 }
1536
1537 while (ctx->pos < extra_size) {
1538 if (ctx->pos == 0) {
1539 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1540 goto out;
1541 ctx->pos = dotdot_offset;
1542 continue;
1543 }
1544
1545 if (ctx->pos == dotdot_offset) {
1546 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1547 goto out;
1548 ctx->pos = dotdot_size;
1549 continue;
1550 }
1551
1552 de = (struct ext4_dir_entry_2 *)
1553 (dir_buf + ctx->pos - extra_offset);
1554 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1555 extra_size, ctx->pos))
1556 goto out;
1557 if (le32_to_cpu(de->inode)) {
1558 if (!dir_emit(ctx, de->name, de->name_len,
1559 le32_to_cpu(de->inode),
1560 get_dtype(sb, de->file_type)))
1561 goto out;
1562 }
1563 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1564 }
1565 out:
1566 kfree(dir_buf);
1567 brelse(iloc.bh);
1568 return ret;
1569 }
1570
1571 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1572 struct ext4_dir_entry_2 **parent_de,
1573 int *retval)
1574 {
1575 struct ext4_iloc iloc;
1576
1577 *retval = ext4_get_inode_loc(inode, &iloc);
1578 if (*retval)
1579 return NULL;
1580
1581 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1582
1583 return iloc.bh;
1584 }
1585
1586 /*
1587 * Try to create the inline data for the new dir.
1588 * If it succeeds, return 0, otherwise return the error.
1589 * In case of ENOSPC, the caller should create the normal disk layout dir.
1590 */
1591 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1592 struct inode *inode)
1593 {
1594 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1595 struct ext4_iloc iloc;
1596 struct ext4_dir_entry_2 *de;
1597
1598 ret = ext4_get_inode_loc(inode, &iloc);
1599 if (ret)
1600 return ret;
1601
1602 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1603 if (ret)
1604 goto out;
1605
1606 /*
1607 * For inline dir, we only save the inode information for the ".."
1608 * and create a fake dentry to cover the left space.
1609 */
1610 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1611 de->inode = cpu_to_le32(parent->i_ino);
1612 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1613 de->inode = 0;
1614 de->rec_len = ext4_rec_len_to_disk(
1615 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1616 inline_size);
1617 set_nlink(inode, 2);
1618 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1619 out:
1620 brelse(iloc.bh);
1621 return ret;
1622 }
1623
1624 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1625 struct ext4_filename *fname,
1626 const struct qstr *d_name,
1627 struct ext4_dir_entry_2 **res_dir,
1628 int *has_inline_data)
1629 {
1630 int ret;
1631 struct ext4_iloc iloc;
1632 void *inline_start;
1633 int inline_size;
1634
1635 if (ext4_get_inode_loc(dir, &iloc))
1636 return NULL;
1637
1638 down_read(&EXT4_I(dir)->xattr_sem);
1639 if (!ext4_has_inline_data(dir)) {
1640 *has_inline_data = 0;
1641 goto out;
1642 }
1643
1644 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1645 EXT4_INLINE_DOTDOT_SIZE;
1646 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1647 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1648 dir, fname, d_name, 0, res_dir);
1649 if (ret == 1)
1650 goto out_find;
1651 if (ret < 0)
1652 goto out;
1653
1654 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1655 goto out;
1656
1657 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1658 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1659
1660 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1661 dir, fname, d_name, 0, res_dir);
1662 if (ret == 1)
1663 goto out_find;
1664
1665 out:
1666 brelse(iloc.bh);
1667 iloc.bh = NULL;
1668 out_find:
1669 up_read(&EXT4_I(dir)->xattr_sem);
1670 return iloc.bh;
1671 }
1672
1673 int ext4_delete_inline_entry(handle_t *handle,
1674 struct inode *dir,
1675 struct ext4_dir_entry_2 *de_del,
1676 struct buffer_head *bh,
1677 int *has_inline_data)
1678 {
1679 int err, inline_size, no_expand;
1680 struct ext4_iloc iloc;
1681 void *inline_start;
1682
1683 err = ext4_get_inode_loc(dir, &iloc);
1684 if (err)
1685 return err;
1686
1687 ext4_write_lock_xattr(dir, &no_expand);
1688 if (!ext4_has_inline_data(dir)) {
1689 *has_inline_data = 0;
1690 goto out;
1691 }
1692
1693 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1694 EXT4_MIN_INLINE_DATA_SIZE) {
1695 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1696 EXT4_INLINE_DOTDOT_SIZE;
1697 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1698 EXT4_INLINE_DOTDOT_SIZE;
1699 } else {
1700 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1701 inline_size = ext4_get_inline_size(dir) -
1702 EXT4_MIN_INLINE_DATA_SIZE;
1703 }
1704
1705 BUFFER_TRACE(bh, "get_write_access");
1706 err = ext4_journal_get_write_access(handle, bh);
1707 if (err)
1708 goto out;
1709
1710 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1711 inline_start, inline_size, 0);
1712 if (err)
1713 goto out;
1714
1715 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1716 err = ext4_mark_inode_dirty(handle, dir);
1717 if (unlikely(err))
1718 goto out;
1719
1720 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1721 out:
1722 ext4_write_unlock_xattr(dir, &no_expand);
1723 brelse(iloc.bh);
1724 if (err != -ENOENT)
1725 ext4_std_error(dir->i_sb, err);
1726 return err;
1727 }
1728
1729 /*
1730 * Get the inline dentry at offset.
1731 */
1732 static inline struct ext4_dir_entry_2 *
1733 ext4_get_inline_entry(struct inode *inode,
1734 struct ext4_iloc *iloc,
1735 unsigned int offset,
1736 void **inline_start,
1737 int *inline_size)
1738 {
1739 void *inline_pos;
1740
1741 BUG_ON(offset > ext4_get_inline_size(inode));
1742
1743 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1744 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1745 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1746 } else {
1747 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1748 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1749 *inline_size = ext4_get_inline_size(inode) -
1750 EXT4_MIN_INLINE_DATA_SIZE;
1751 }
1752
1753 if (inline_start)
1754 *inline_start = inline_pos;
1755 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1756 }
1757
1758 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1759 {
1760 int err, inline_size;
1761 struct ext4_iloc iloc;
1762 size_t inline_len;
1763 void *inline_pos;
1764 unsigned int offset;
1765 struct ext4_dir_entry_2 *de;
1766 int ret = 1;
1767
1768 err = ext4_get_inode_loc(dir, &iloc);
1769 if (err) {
1770 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1771 err, dir->i_ino);
1772 return 1;
1773 }
1774
1775 down_read(&EXT4_I(dir)->xattr_sem);
1776 if (!ext4_has_inline_data(dir)) {
1777 *has_inline_data = 0;
1778 goto out;
1779 }
1780
1781 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1782 if (!le32_to_cpu(de->inode)) {
1783 ext4_warning(dir->i_sb,
1784 "bad inline directory (dir #%lu) - no `..'",
1785 dir->i_ino);
1786 ret = 1;
1787 goto out;
1788 }
1789
1790 inline_len = ext4_get_inline_size(dir);
1791 offset = EXT4_INLINE_DOTDOT_SIZE;
1792 while (offset < inline_len) {
1793 de = ext4_get_inline_entry(dir, &iloc, offset,
1794 &inline_pos, &inline_size);
1795 if (ext4_check_dir_entry(dir, NULL, de,
1796 iloc.bh, inline_pos,
1797 inline_size, offset)) {
1798 ext4_warning(dir->i_sb,
1799 "bad inline directory (dir #%lu) - "
1800 "inode %u, rec_len %u, name_len %d"
1801 "inline size %d\n",
1802 dir->i_ino, le32_to_cpu(de->inode),
1803 le16_to_cpu(de->rec_len), de->name_len,
1804 inline_size);
1805 ret = 1;
1806 goto out;
1807 }
1808 if (le32_to_cpu(de->inode)) {
1809 ret = 0;
1810 goto out;
1811 }
1812 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1813 }
1814
1815 out:
1816 up_read(&EXT4_I(dir)->xattr_sem);
1817 brelse(iloc.bh);
1818 return ret;
1819 }
1820
1821 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1822 {
1823 int ret, no_expand;
1824
1825 ext4_write_lock_xattr(inode, &no_expand);
1826 ret = ext4_destroy_inline_data_nolock(handle, inode);
1827 ext4_write_unlock_xattr(inode, &no_expand);
1828
1829 return ret;
1830 }
1831
1832 int ext4_inline_data_fiemap(struct inode *inode,
1833 struct fiemap_extent_info *fieinfo,
1834 int *has_inline, __u64 start, __u64 len)
1835 {
1836 __u64 physical = 0;
1837 __u64 inline_len;
1838 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1839 FIEMAP_EXTENT_LAST;
1840 int error = 0;
1841 struct ext4_iloc iloc;
1842
1843 down_read(&EXT4_I(inode)->xattr_sem);
1844 if (!ext4_has_inline_data(inode)) {
1845 *has_inline = 0;
1846 goto out;
1847 }
1848 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1849 i_size_read(inode));
1850 if (start >= inline_len)
1851 goto out;
1852 if (start + len < inline_len)
1853 inline_len = start + len;
1854 inline_len -= start;
1855
1856 error = ext4_get_inode_loc(inode, &iloc);
1857 if (error)
1858 goto out;
1859
1860 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1861 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1862 physical += offsetof(struct ext4_inode, i_block);
1863
1864 brelse(iloc.bh);
1865 out:
1866 up_read(&EXT4_I(inode)->xattr_sem);
1867 if (physical)
1868 error = fiemap_fill_next_extent(fieinfo, start, physical,
1869 inline_len, flags);
1870 return (error < 0 ? error : 0);
1871 }
1872
1873 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1874 {
1875 handle_t *handle;
1876 int inline_size, value_len, needed_blocks, no_expand;
1877 size_t i_size;
1878 void *value = NULL;
1879 struct ext4_xattr_ibody_find is = {
1880 .s = { .not_found = -ENODATA, },
1881 };
1882 struct ext4_xattr_info i = {
1883 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1884 .name = EXT4_XATTR_SYSTEM_DATA,
1885 };
1886
1887
1888 needed_blocks = ext4_writepage_trans_blocks(inode);
1889 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1890 if (IS_ERR(handle))
1891 return;
1892
1893 ext4_write_lock_xattr(inode, &no_expand);
1894 if (!ext4_has_inline_data(inode)) {
1895 *has_inline = 0;
1896 ext4_journal_stop(handle);
1897 return;
1898 }
1899
1900 if (ext4_orphan_add(handle, inode))
1901 goto out;
1902
1903 if (ext4_get_inode_loc(inode, &is.iloc))
1904 goto out;
1905
1906 down_write(&EXT4_I(inode)->i_data_sem);
1907 i_size = inode->i_size;
1908 inline_size = ext4_get_inline_size(inode);
1909 EXT4_I(inode)->i_disksize = i_size;
1910
1911 if (i_size < inline_size) {
1912 /* Clear the content in the xattr space. */
1913 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1914 if (ext4_xattr_ibody_find(inode, &i, &is))
1915 goto out_error;
1916
1917 BUG_ON(is.s.not_found);
1918
1919 value_len = le32_to_cpu(is.s.here->e_value_size);
1920 value = kmalloc(value_len, GFP_NOFS);
1921 if (!value)
1922 goto out_error;
1923
1924 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1925 value, value_len))
1926 goto out_error;
1927
1928 i.value = value;
1929 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1930 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1931 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1932 goto out_error;
1933 }
1934
1935 /* Clear the content within i_blocks. */
1936 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1937 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1938 memset(p + i_size, 0,
1939 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1940 }
1941
1942 EXT4_I(inode)->i_inline_size = i_size <
1943 EXT4_MIN_INLINE_DATA_SIZE ?
1944 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1945 }
1946
1947 out_error:
1948 up_write(&EXT4_I(inode)->i_data_sem);
1949 out:
1950 brelse(is.iloc.bh);
1951 ext4_write_unlock_xattr(inode, &no_expand);
1952 kfree(value);
1953 if (inode->i_nlink)
1954 ext4_orphan_del(handle, inode);
1955
1956 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1957 ext4_mark_inode_dirty(handle, inode);
1958 if (IS_SYNC(inode))
1959 ext4_handle_sync(handle);
1960
1961 ext4_journal_stop(handle);
1962 return;
1963 }
1964
1965 int ext4_convert_inline_data(struct inode *inode)
1966 {
1967 int error, needed_blocks, no_expand;
1968 handle_t *handle;
1969 struct ext4_iloc iloc;
1970
1971 if (!ext4_has_inline_data(inode)) {
1972 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1973 return 0;
1974 }
1975
1976 needed_blocks = ext4_writepage_trans_blocks(inode);
1977
1978 iloc.bh = NULL;
1979 error = ext4_get_inode_loc(inode, &iloc);
1980 if (error)
1981 return error;
1982
1983 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1984 if (IS_ERR(handle)) {
1985 error = PTR_ERR(handle);
1986 goto out_free;
1987 }
1988
1989 ext4_write_lock_xattr(inode, &no_expand);
1990 if (ext4_has_inline_data(inode))
1991 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1992 ext4_write_unlock_xattr(inode, &no_expand);
1993 ext4_journal_stop(handle);
1994 out_free:
1995 brelse(iloc.bh);
1996 return error;
1997 }
Linux with added FPC-III support