]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/ext4/inline.c
Merge branch 'work.misc' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[mirror_ubuntu-jammy-kernel.git] / fs / ext4 / inline.c
1 /*
2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 */
14
15 #include <linux/iomap.h>
16 #include <linux/fiemap.h>
17 #include <linux/iversion.h>
18
19 #include "ext4_jbd2.h"
20 #include "ext4.h"
21 #include "xattr.h"
22 #include "truncate.h"
23
24 #define EXT4_XATTR_SYSTEM_DATA "data"
25 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
26 #define EXT4_INLINE_DOTDOT_OFFSET 2
27 #define EXT4_INLINE_DOTDOT_SIZE 4
28
29 static int ext4_get_inline_size(struct inode *inode)
30 {
31 if (EXT4_I(inode)->i_inline_off)
32 return EXT4_I(inode)->i_inline_size;
33
34 return 0;
35 }
36
37 static int get_max_inline_xattr_value_size(struct inode *inode,
38 struct ext4_iloc *iloc)
39 {
40 struct ext4_xattr_ibody_header *header;
41 struct ext4_xattr_entry *entry;
42 struct ext4_inode *raw_inode;
43 int free, min_offs;
44
45 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
46 EXT4_GOOD_OLD_INODE_SIZE -
47 EXT4_I(inode)->i_extra_isize -
48 sizeof(struct ext4_xattr_ibody_header);
49
50 /*
51 * We need to subtract another sizeof(__u32) since an in-inode xattr
52 * needs an empty 4 bytes to indicate the gap between the xattr entry
53 * and the name/value pair.
54 */
55 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
56 return EXT4_XATTR_SIZE(min_offs -
57 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
58 EXT4_XATTR_ROUND - sizeof(__u32));
59
60 raw_inode = ext4_raw_inode(iloc);
61 header = IHDR(inode, raw_inode);
62 entry = IFIRST(header);
63
64 /* Compute min_offs. */
65 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
66 if (!entry->e_value_inum && entry->e_value_size) {
67 size_t offs = le16_to_cpu(entry->e_value_offs);
68 if (offs < min_offs)
69 min_offs = offs;
70 }
71 }
72 free = min_offs -
73 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
74
75 if (EXT4_I(inode)->i_inline_off) {
76 entry = (struct ext4_xattr_entry *)
77 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
78
79 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
80 goto out;
81 }
82
83 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
84
85 if (free > EXT4_XATTR_ROUND)
86 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
87 else
88 free = 0;
89
90 out:
91 return free;
92 }
93
94 /*
95 * Get the maximum size we now can store in an inode.
96 * If we can't find the space for a xattr entry, don't use the space
97 * of the extents since we have no space to indicate the inline data.
98 */
99 int ext4_get_max_inline_size(struct inode *inode)
100 {
101 int error, max_inline_size;
102 struct ext4_iloc iloc;
103
104 if (EXT4_I(inode)->i_extra_isize == 0)
105 return 0;
106
107 error = ext4_get_inode_loc(inode, &iloc);
108 if (error) {
109 ext4_error_inode(inode, __func__, __LINE__, 0,
110 "can't get inode location %lu",
111 inode->i_ino);
112 return 0;
113 }
114
115 down_read(&EXT4_I(inode)->xattr_sem);
116 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
117 up_read(&EXT4_I(inode)->xattr_sem);
118
119 brelse(iloc.bh);
120
121 if (!max_inline_size)
122 return 0;
123
124 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
125 }
126
127 /*
128 * this function does not take xattr_sem, which is OK because it is
129 * currently only used in a code path coming form ext4_iget, before
130 * the new inode has been unlocked
131 */
132 int ext4_find_inline_data_nolock(struct inode *inode)
133 {
134 struct ext4_xattr_ibody_find is = {
135 .s = { .not_found = -ENODATA, },
136 };
137 struct ext4_xattr_info i = {
138 .name_index = EXT4_XATTR_INDEX_SYSTEM,
139 .name = EXT4_XATTR_SYSTEM_DATA,
140 };
141 int error;
142
143 if (EXT4_I(inode)->i_extra_isize == 0)
144 return 0;
145
146 error = ext4_get_inode_loc(inode, &is.iloc);
147 if (error)
148 return error;
149
150 error = ext4_xattr_ibody_find(inode, &i, &is);
151 if (error)
152 goto out;
153
154 if (!is.s.not_found) {
155 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
156 (void *)ext4_raw_inode(&is.iloc));
157 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
158 le32_to_cpu(is.s.here->e_value_size);
159 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
160 }
161 out:
162 brelse(is.iloc.bh);
163 return error;
164 }
165
166 static int ext4_read_inline_data(struct inode *inode, void *buffer,
167 unsigned int len,
168 struct ext4_iloc *iloc)
169 {
170 struct ext4_xattr_entry *entry;
171 struct ext4_xattr_ibody_header *header;
172 int cp_len = 0;
173 struct ext4_inode *raw_inode;
174
175 if (!len)
176 return 0;
177
178 BUG_ON(len > EXT4_I(inode)->i_inline_size);
179
180 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
181 len : EXT4_MIN_INLINE_DATA_SIZE;
182
183 raw_inode = ext4_raw_inode(iloc);
184 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
185
186 len -= cp_len;
187 buffer += cp_len;
188
189 if (!len)
190 goto out;
191
192 header = IHDR(inode, raw_inode);
193 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
194 EXT4_I(inode)->i_inline_off);
195 len = min_t(unsigned int, len,
196 (unsigned int)le32_to_cpu(entry->e_value_size));
197
198 memcpy(buffer,
199 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
200 cp_len += len;
201
202 out:
203 return cp_len;
204 }
205
206 /*
207 * write the buffer to the inline inode.
208 * If 'create' is set, we don't need to do the extra copy in the xattr
209 * value since it is already handled by ext4_xattr_ibody_inline_set.
210 * That saves us one memcpy.
211 */
212 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
213 void *buffer, loff_t pos, unsigned int len)
214 {
215 struct ext4_xattr_entry *entry;
216 struct ext4_xattr_ibody_header *header;
217 struct ext4_inode *raw_inode;
218 int cp_len = 0;
219
220 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
221 return;
222
223 BUG_ON(!EXT4_I(inode)->i_inline_off);
224 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
225
226 raw_inode = ext4_raw_inode(iloc);
227 buffer += pos;
228
229 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
230 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
231 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
232 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
233
234 len -= cp_len;
235 buffer += cp_len;
236 pos += cp_len;
237 }
238
239 if (!len)
240 return;
241
242 pos -= EXT4_MIN_INLINE_DATA_SIZE;
243 header = IHDR(inode, raw_inode);
244 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
245 EXT4_I(inode)->i_inline_off);
246
247 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
248 buffer, len);
249 }
250
251 static int ext4_create_inline_data(handle_t *handle,
252 struct inode *inode, unsigned len)
253 {
254 int error;
255 void *value = NULL;
256 struct ext4_xattr_ibody_find is = {
257 .s = { .not_found = -ENODATA, },
258 };
259 struct ext4_xattr_info i = {
260 .name_index = EXT4_XATTR_INDEX_SYSTEM,
261 .name = EXT4_XATTR_SYSTEM_DATA,
262 };
263
264 error = ext4_get_inode_loc(inode, &is.iloc);
265 if (error)
266 return error;
267
268 BUFFER_TRACE(is.iloc.bh, "get_write_access");
269 error = ext4_journal_get_write_access(handle, is.iloc.bh);
270 if (error)
271 goto out;
272
273 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
274 value = EXT4_ZERO_XATTR_VALUE;
275 len -= EXT4_MIN_INLINE_DATA_SIZE;
276 } else {
277 value = "";
278 len = 0;
279 }
280
281 /* Insert the the xttr entry. */
282 i.value = value;
283 i.value_len = len;
284
285 error = ext4_xattr_ibody_find(inode, &i, &is);
286 if (error)
287 goto out;
288
289 BUG_ON(!is.s.not_found);
290
291 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
292 if (error) {
293 if (error == -ENOSPC)
294 ext4_clear_inode_state(inode,
295 EXT4_STATE_MAY_INLINE_DATA);
296 goto out;
297 }
298
299 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
300 0, EXT4_MIN_INLINE_DATA_SIZE);
301
302 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
303 (void *)ext4_raw_inode(&is.iloc));
304 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
305 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
306 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
307 get_bh(is.iloc.bh);
308 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
309
310 out:
311 brelse(is.iloc.bh);
312 return error;
313 }
314
315 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
316 unsigned int len)
317 {
318 int error;
319 void *value = NULL;
320 struct ext4_xattr_ibody_find is = {
321 .s = { .not_found = -ENODATA, },
322 };
323 struct ext4_xattr_info i = {
324 .name_index = EXT4_XATTR_INDEX_SYSTEM,
325 .name = EXT4_XATTR_SYSTEM_DATA,
326 };
327
328 /* If the old space is ok, write the data directly. */
329 if (len <= EXT4_I(inode)->i_inline_size)
330 return 0;
331
332 error = ext4_get_inode_loc(inode, &is.iloc);
333 if (error)
334 return error;
335
336 error = ext4_xattr_ibody_find(inode, &i, &is);
337 if (error)
338 goto out;
339
340 BUG_ON(is.s.not_found);
341
342 len -= EXT4_MIN_INLINE_DATA_SIZE;
343 value = kzalloc(len, GFP_NOFS);
344 if (!value) {
345 error = -ENOMEM;
346 goto out;
347 }
348
349 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
350 value, len);
351 if (error == -ENODATA)
352 goto out;
353
354 BUFFER_TRACE(is.iloc.bh, "get_write_access");
355 error = ext4_journal_get_write_access(handle, is.iloc.bh);
356 if (error)
357 goto out;
358
359 /* Update the xttr entry. */
360 i.value = value;
361 i.value_len = len;
362
363 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
364 if (error)
365 goto out;
366
367 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
368 (void *)ext4_raw_inode(&is.iloc));
369 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
370 le32_to_cpu(is.s.here->e_value_size);
371 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
372 get_bh(is.iloc.bh);
373 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
374
375 out:
376 kfree(value);
377 brelse(is.iloc.bh);
378 return error;
379 }
380
381 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
382 unsigned int len)
383 {
384 int ret, size, no_expand;
385 struct ext4_inode_info *ei = EXT4_I(inode);
386
387 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
388 return -ENOSPC;
389
390 size = ext4_get_max_inline_size(inode);
391 if (size < len)
392 return -ENOSPC;
393
394 ext4_write_lock_xattr(inode, &no_expand);
395
396 if (ei->i_inline_off)
397 ret = ext4_update_inline_data(handle, inode, len);
398 else
399 ret = ext4_create_inline_data(handle, inode, len);
400
401 ext4_write_unlock_xattr(inode, &no_expand);
402 return ret;
403 }
404
405 static int ext4_destroy_inline_data_nolock(handle_t *handle,
406 struct inode *inode)
407 {
408 struct ext4_inode_info *ei = EXT4_I(inode);
409 struct ext4_xattr_ibody_find is = {
410 .s = { .not_found = 0, },
411 };
412 struct ext4_xattr_info i = {
413 .name_index = EXT4_XATTR_INDEX_SYSTEM,
414 .name = EXT4_XATTR_SYSTEM_DATA,
415 .value = NULL,
416 .value_len = 0,
417 };
418 int error;
419
420 if (!ei->i_inline_off)
421 return 0;
422
423 error = ext4_get_inode_loc(inode, &is.iloc);
424 if (error)
425 return error;
426
427 error = ext4_xattr_ibody_find(inode, &i, &is);
428 if (error)
429 goto out;
430
431 BUFFER_TRACE(is.iloc.bh, "get_write_access");
432 error = ext4_journal_get_write_access(handle, is.iloc.bh);
433 if (error)
434 goto out;
435
436 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
437 if (error)
438 goto out;
439
440 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
441 0, EXT4_MIN_INLINE_DATA_SIZE);
442
443 if (ext4_has_feature_extents(inode->i_sb)) {
444 if (S_ISDIR(inode->i_mode) ||
445 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
446 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
447 ext4_ext_tree_init(handle, inode);
448 }
449 }
450 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
451
452 get_bh(is.iloc.bh);
453 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
454
455 EXT4_I(inode)->i_inline_off = 0;
456 EXT4_I(inode)->i_inline_size = 0;
457 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
458 out:
459 brelse(is.iloc.bh);
460 if (error == -ENODATA)
461 error = 0;
462 return error;
463 }
464
465 static int ext4_read_inline_page(struct inode *inode, struct page *page)
466 {
467 void *kaddr;
468 int ret = 0;
469 size_t len;
470 struct ext4_iloc iloc;
471
472 BUG_ON(!PageLocked(page));
473 BUG_ON(!ext4_has_inline_data(inode));
474 BUG_ON(page->index);
475
476 if (!EXT4_I(inode)->i_inline_off) {
477 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
478 inode->i_ino);
479 goto out;
480 }
481
482 ret = ext4_get_inode_loc(inode, &iloc);
483 if (ret)
484 goto out;
485
486 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
487 kaddr = kmap_atomic(page);
488 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
489 flush_dcache_page(page);
490 kunmap_atomic(kaddr);
491 zero_user_segment(page, len, PAGE_SIZE);
492 SetPageUptodate(page);
493 brelse(iloc.bh);
494
495 out:
496 return ret;
497 }
498
499 int ext4_readpage_inline(struct inode *inode, struct page *page)
500 {
501 int ret = 0;
502
503 down_read(&EXT4_I(inode)->xattr_sem);
504 if (!ext4_has_inline_data(inode)) {
505 up_read(&EXT4_I(inode)->xattr_sem);
506 return -EAGAIN;
507 }
508
509 /*
510 * Current inline data can only exist in the 1st page,
511 * So for all the other pages, just set them uptodate.
512 */
513 if (!page->index)
514 ret = ext4_read_inline_page(inode, page);
515 else if (!PageUptodate(page)) {
516 zero_user_segment(page, 0, PAGE_SIZE);
517 SetPageUptodate(page);
518 }
519
520 up_read(&EXT4_I(inode)->xattr_sem);
521
522 unlock_page(page);
523 return ret >= 0 ? 0 : ret;
524 }
525
526 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
527 struct inode *inode,
528 unsigned flags)
529 {
530 int ret, needed_blocks, no_expand;
531 handle_t *handle = NULL;
532 int retries = 0, sem_held = 0;
533 struct page *page = NULL;
534 unsigned from, to;
535 struct ext4_iloc iloc;
536
537 if (!ext4_has_inline_data(inode)) {
538 /*
539 * clear the flag so that no new write
540 * will trap here again.
541 */
542 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
543 return 0;
544 }
545
546 needed_blocks = ext4_writepage_trans_blocks(inode);
547
548 ret = ext4_get_inode_loc(inode, &iloc);
549 if (ret)
550 return ret;
551
552 retry:
553 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
554 if (IS_ERR(handle)) {
555 ret = PTR_ERR(handle);
556 handle = NULL;
557 goto out;
558 }
559
560 /* We cannot recurse into the filesystem as the transaction is already
561 * started */
562 flags |= AOP_FLAG_NOFS;
563
564 page = grab_cache_page_write_begin(mapping, 0, flags);
565 if (!page) {
566 ret = -ENOMEM;
567 goto out;
568 }
569
570 ext4_write_lock_xattr(inode, &no_expand);
571 sem_held = 1;
572 /* If some one has already done this for us, just exit. */
573 if (!ext4_has_inline_data(inode)) {
574 ret = 0;
575 goto out;
576 }
577
578 from = 0;
579 to = ext4_get_inline_size(inode);
580 if (!PageUptodate(page)) {
581 ret = ext4_read_inline_page(inode, page);
582 if (ret < 0)
583 goto out;
584 }
585
586 ret = ext4_destroy_inline_data_nolock(handle, inode);
587 if (ret)
588 goto out;
589
590 if (ext4_should_dioread_nolock(inode)) {
591 ret = __block_write_begin(page, from, to,
592 ext4_get_block_unwritten);
593 } else
594 ret = __block_write_begin(page, from, to, ext4_get_block);
595
596 if (!ret && ext4_should_journal_data(inode)) {
597 ret = ext4_walk_page_buffers(handle, page_buffers(page),
598 from, to, NULL,
599 do_journal_get_write_access);
600 }
601
602 if (ret) {
603 unlock_page(page);
604 put_page(page);
605 page = NULL;
606 ext4_orphan_add(handle, inode);
607 ext4_write_unlock_xattr(inode, &no_expand);
608 sem_held = 0;
609 ext4_journal_stop(handle);
610 handle = NULL;
611 ext4_truncate_failed_write(inode);
612 /*
613 * If truncate failed early the inode might
614 * still be on the orphan list; we need to
615 * make sure the inode is removed from the
616 * orphan list in that case.
617 */
618 if (inode->i_nlink)
619 ext4_orphan_del(NULL, inode);
620 }
621
622 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
623 goto retry;
624
625 if (page)
626 block_commit_write(page, from, to);
627 out:
628 if (page) {
629 unlock_page(page);
630 put_page(page);
631 }
632 if (sem_held)
633 ext4_write_unlock_xattr(inode, &no_expand);
634 if (handle)
635 ext4_journal_stop(handle);
636 brelse(iloc.bh);
637 return ret;
638 }
639
640 /*
641 * Try to write data in the inode.
642 * If the inode has inline data, check whether the new write can be
643 * in the inode also. If not, create the page the handle, move the data
644 * to the page make it update and let the later codes create extent for it.
645 */
646 int ext4_try_to_write_inline_data(struct address_space *mapping,
647 struct inode *inode,
648 loff_t pos, unsigned len,
649 unsigned flags,
650 struct page **pagep)
651 {
652 int ret;
653 handle_t *handle;
654 struct page *page;
655 struct ext4_iloc iloc;
656
657 if (pos + len > ext4_get_max_inline_size(inode))
658 goto convert;
659
660 ret = ext4_get_inode_loc(inode, &iloc);
661 if (ret)
662 return ret;
663
664 /*
665 * The possible write could happen in the inode,
666 * so try to reserve the space in inode first.
667 */
668 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
669 if (IS_ERR(handle)) {
670 ret = PTR_ERR(handle);
671 handle = NULL;
672 goto out;
673 }
674
675 ret = ext4_prepare_inline_data(handle, inode, pos + len);
676 if (ret && ret != -ENOSPC)
677 goto out;
678
679 /* We don't have space in inline inode, so convert it to extent. */
680 if (ret == -ENOSPC) {
681 ext4_journal_stop(handle);
682 brelse(iloc.bh);
683 goto convert;
684 }
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 goto out_up_read;
707 }
708
709 ret = 1;
710 handle = NULL;
711 out_up_read:
712 up_read(&EXT4_I(inode)->xattr_sem);
713 out:
714 if (handle)
715 ext4_journal_stop(handle);
716 brelse(iloc.bh);
717 return ret;
718 convert:
719 return ext4_convert_inline_data_to_extent(mapping,
720 inode, flags);
721 }
722
723 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
724 unsigned copied, struct page *page)
725 {
726 int ret, no_expand;
727 void *kaddr;
728 struct ext4_iloc iloc;
729
730 if (unlikely(copied < len)) {
731 if (!PageUptodate(page)) {
732 copied = 0;
733 goto out;
734 }
735 }
736
737 ret = ext4_get_inode_loc(inode, &iloc);
738 if (ret) {
739 ext4_std_error(inode->i_sb, ret);
740 copied = 0;
741 goto out;
742 }
743
744 ext4_write_lock_xattr(inode, &no_expand);
745 BUG_ON(!ext4_has_inline_data(inode));
746
747 kaddr = kmap_atomic(page);
748 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
749 kunmap_atomic(kaddr);
750 SetPageUptodate(page);
751 /* clear page dirty so that writepages wouldn't work for us. */
752 ClearPageDirty(page);
753
754 ext4_write_unlock_xattr(inode, &no_expand);
755 brelse(iloc.bh);
756 out:
757 return copied;
758 }
759
760 struct buffer_head *
761 ext4_journalled_write_inline_data(struct inode *inode,
762 unsigned len,
763 struct page *page)
764 {
765 int ret, no_expand;
766 void *kaddr;
767 struct ext4_iloc iloc;
768
769 ret = ext4_get_inode_loc(inode, &iloc);
770 if (ret) {
771 ext4_std_error(inode->i_sb, ret);
772 return NULL;
773 }
774
775 ext4_write_lock_xattr(inode, &no_expand);
776 kaddr = kmap_atomic(page);
777 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
778 kunmap_atomic(kaddr);
779 ext4_write_unlock_xattr(inode, &no_expand);
780
781 return iloc.bh;
782 }
783
784 /*
785 * Try to make the page cache and handle ready for the inline data case.
786 * We can call this function in 2 cases:
787 * 1. The inode is created and the first write exceeds inline size. We can
788 * clear the inode state safely.
789 * 2. The inode has inline data, then we need to read the data, make it
790 * update and dirty so that ext4_da_writepages can handle it. We don't
791 * need to start the journal since the file's metatdata isn't changed now.
792 */
793 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
794 struct inode *inode,
795 unsigned flags,
796 void **fsdata)
797 {
798 int ret = 0, inline_size;
799 struct page *page;
800
801 page = grab_cache_page_write_begin(mapping, 0, flags);
802 if (!page)
803 return -ENOMEM;
804
805 down_read(&EXT4_I(inode)->xattr_sem);
806 if (!ext4_has_inline_data(inode)) {
807 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
808 goto out;
809 }
810
811 inline_size = ext4_get_inline_size(inode);
812
813 if (!PageUptodate(page)) {
814 ret = ext4_read_inline_page(inode, page);
815 if (ret < 0)
816 goto out;
817 }
818
819 ret = __block_write_begin(page, 0, inline_size,
820 ext4_da_get_block_prep);
821 if (ret) {
822 up_read(&EXT4_I(inode)->xattr_sem);
823 unlock_page(page);
824 put_page(page);
825 ext4_truncate_failed_write(inode);
826 return ret;
827 }
828
829 SetPageDirty(page);
830 SetPageUptodate(page);
831 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
832 *fsdata = (void *)CONVERT_INLINE_DATA;
833
834 out:
835 up_read(&EXT4_I(inode)->xattr_sem);
836 if (page) {
837 unlock_page(page);
838 put_page(page);
839 }
840 return ret;
841 }
842
843 /*
844 * Prepare the write for the inline data.
845 * If the the data can be written into the inode, we just read
846 * the page and make it uptodate, and start the journal.
847 * Otherwise read the page, makes it dirty so that it can be
848 * handle in writepages(the i_disksize update is left to the
849 * normal ext4_da_write_end).
850 */
851 int ext4_da_write_inline_data_begin(struct address_space *mapping,
852 struct inode *inode,
853 loff_t pos, unsigned len,
854 unsigned flags,
855 struct page **pagep,
856 void **fsdata)
857 {
858 int ret, inline_size;
859 handle_t *handle;
860 struct page *page;
861 struct ext4_iloc iloc;
862 int retries;
863
864 ret = ext4_get_inode_loc(inode, &iloc);
865 if (ret)
866 return ret;
867
868 retry_journal:
869 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
870 if (IS_ERR(handle)) {
871 ret = PTR_ERR(handle);
872 goto out;
873 }
874
875 inline_size = ext4_get_max_inline_size(inode);
876
877 ret = -ENOSPC;
878 if (inline_size >= pos + len) {
879 ret = ext4_prepare_inline_data(handle, inode, pos + len);
880 if (ret && ret != -ENOSPC)
881 goto out_journal;
882 }
883
884 /*
885 * We cannot recurse into the filesystem as the transaction
886 * is already started.
887 */
888 flags |= AOP_FLAG_NOFS;
889
890 if (ret == -ENOSPC) {
891 ret = ext4_da_convert_inline_data_to_extent(mapping,
892 inode,
893 flags,
894 fsdata);
895 ext4_journal_stop(handle);
896 if (ret == -ENOSPC &&
897 ext4_should_retry_alloc(inode->i_sb, &retries))
898 goto retry_journal;
899 goto out;
900 }
901
902
903 page = grab_cache_page_write_begin(mapping, 0, flags);
904 if (!page) {
905 ret = -ENOMEM;
906 goto out_journal;
907 }
908
909 down_read(&EXT4_I(inode)->xattr_sem);
910 if (!ext4_has_inline_data(inode)) {
911 ret = 0;
912 goto out_release_page;
913 }
914
915 if (!PageUptodate(page)) {
916 ret = ext4_read_inline_page(inode, page);
917 if (ret < 0)
918 goto out_release_page;
919 }
920
921 up_read(&EXT4_I(inode)->xattr_sem);
922 *pagep = page;
923 brelse(iloc.bh);
924 return 1;
925 out_release_page:
926 up_read(&EXT4_I(inode)->xattr_sem);
927 unlock_page(page);
928 put_page(page);
929 out_journal:
930 ext4_journal_stop(handle);
931 out:
932 brelse(iloc.bh);
933 return ret;
934 }
935
936 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
937 unsigned len, unsigned copied,
938 struct page *page)
939 {
940 int i_size_changed = 0;
941 int ret;
942
943 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
944 if (ret < 0) {
945 unlock_page(page);
946 put_page(page);
947 return ret;
948 }
949 copied = ret;
950
951 /*
952 * No need to use i_size_read() here, the i_size
953 * cannot change under us because we hold i_mutex.
954 *
955 * But it's important to update i_size while still holding page lock:
956 * page writeout could otherwise come in and zero beyond i_size.
957 */
958 if (pos+copied > inode->i_size) {
959 i_size_write(inode, pos+copied);
960 i_size_changed = 1;
961 }
962 unlock_page(page);
963 put_page(page);
964
965 /*
966 * Don't mark the inode dirty under page lock. First, it unnecessarily
967 * makes the holding time of page lock longer. Second, it forces lock
968 * ordering of page lock and transaction start for journaling
969 * filesystems.
970 */
971 if (i_size_changed)
972 mark_inode_dirty(inode);
973
974 return copied;
975 }
976
977 #ifdef INLINE_DIR_DEBUG
978 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
979 void *inline_start, int inline_size)
980 {
981 int offset;
982 unsigned short de_len;
983 struct ext4_dir_entry_2 *de = inline_start;
984 void *dlimit = inline_start + inline_size;
985
986 trace_printk("inode %lu\n", dir->i_ino);
987 offset = 0;
988 while ((void *)de < dlimit) {
989 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
990 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
991 offset, de_len, de->name_len, de->name,
992 de->name_len, le32_to_cpu(de->inode));
993 if (ext4_check_dir_entry(dir, NULL, de, bh,
994 inline_start, inline_size, offset))
995 BUG();
996
997 offset += de_len;
998 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
999 }
1000 }
1001 #else
1002 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1003 #endif
1004
1005 /*
1006 * Add a new entry into a inline dir.
1007 * It will return -ENOSPC if no space is available, and -EIO
1008 * and -EEXIST if directory entry already exists.
1009 */
1010 static int ext4_add_dirent_to_inline(handle_t *handle,
1011 struct ext4_filename *fname,
1012 struct inode *dir,
1013 struct inode *inode,
1014 struct ext4_iloc *iloc,
1015 void *inline_start, int inline_size)
1016 {
1017 int err;
1018 struct ext4_dir_entry_2 *de;
1019
1020 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1021 inline_size, fname, &de);
1022 if (err)
1023 return err;
1024
1025 BUFFER_TRACE(iloc->bh, "get_write_access");
1026 err = ext4_journal_get_write_access(handle, iloc->bh);
1027 if (err)
1028 return err;
1029 ext4_insert_dentry(inode, de, inline_size, fname);
1030
1031 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1032
1033 /*
1034 * XXX shouldn't update any times until successful
1035 * completion of syscall, but too many callers depend
1036 * on this.
1037 *
1038 * XXX similarly, too many callers depend on
1039 * ext4_new_inode() setting the times, but error
1040 * recovery deletes the inode, so the worst that can
1041 * happen is that the times are slightly out of date
1042 * and/or different from the directory change time.
1043 */
1044 dir->i_mtime = dir->i_ctime = current_time(dir);
1045 ext4_update_dx_flag(dir);
1046 inode_inc_iversion(dir);
1047 return 1;
1048 }
1049
1050 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1051 struct ext4_iloc *iloc)
1052 {
1053 struct ext4_xattr_entry *entry;
1054 struct ext4_xattr_ibody_header *header;
1055
1056 BUG_ON(!EXT4_I(inode)->i_inline_off);
1057
1058 header = IHDR(inode, ext4_raw_inode(iloc));
1059 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1060 EXT4_I(inode)->i_inline_off);
1061
1062 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1063 }
1064
1065 /* Set the final de to cover the whole block. */
1066 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1067 {
1068 struct ext4_dir_entry_2 *de, *prev_de;
1069 void *limit;
1070 int de_len;
1071
1072 de = (struct ext4_dir_entry_2 *)de_buf;
1073 if (old_size) {
1074 limit = de_buf + old_size;
1075 do {
1076 prev_de = de;
1077 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1078 de_buf += de_len;
1079 de = (struct ext4_dir_entry_2 *)de_buf;
1080 } while (de_buf < limit);
1081
1082 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1083 old_size, new_size);
1084 } else {
1085 /* this is just created, so create an empty entry. */
1086 de->inode = 0;
1087 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1088 }
1089 }
1090
1091 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1092 struct ext4_iloc *iloc)
1093 {
1094 int ret;
1095 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1096 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1097
1098 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1099 return -ENOSPC;
1100
1101 ret = ext4_update_inline_data(handle, dir,
1102 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1103 if (ret)
1104 return ret;
1105
1106 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1107 EXT4_I(dir)->i_inline_size -
1108 EXT4_MIN_INLINE_DATA_SIZE);
1109 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1110 return 0;
1111 }
1112
1113 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1114 struct ext4_iloc *iloc,
1115 void *buf, int inline_size)
1116 {
1117 ext4_create_inline_data(handle, inode, inline_size);
1118 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1119 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1120 }
1121
1122 static int ext4_finish_convert_inline_dir(handle_t *handle,
1123 struct inode *inode,
1124 struct buffer_head *dir_block,
1125 void *buf,
1126 int inline_size)
1127 {
1128 int err, csum_size = 0, header_size = 0;
1129 struct ext4_dir_entry_2 *de;
1130 struct ext4_dir_entry_tail *t;
1131 void *target = dir_block->b_data;
1132
1133 /*
1134 * First create "." and ".." and then copy the dir information
1135 * back to the block.
1136 */
1137 de = (struct ext4_dir_entry_2 *)target;
1138 de = ext4_init_dot_dotdot(inode, de,
1139 inode->i_sb->s_blocksize, csum_size,
1140 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1141 header_size = (void *)de - target;
1142
1143 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1144 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1145
1146 if (ext4_has_metadata_csum(inode->i_sb))
1147 csum_size = sizeof(struct ext4_dir_entry_tail);
1148
1149 inode->i_size = inode->i_sb->s_blocksize;
1150 i_size_write(inode, inode->i_sb->s_blocksize);
1151 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1152 ext4_update_final_de(dir_block->b_data,
1153 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1154 inode->i_sb->s_blocksize - csum_size);
1155
1156 if (csum_size) {
1157 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1158 inode->i_sb->s_blocksize);
1159 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1160 }
1161 set_buffer_uptodate(dir_block);
1162 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1163 if (err)
1164 return err;
1165 set_buffer_verified(dir_block);
1166 return ext4_mark_inode_dirty(handle, inode);
1167 }
1168
1169 static int ext4_convert_inline_data_nolock(handle_t *handle,
1170 struct inode *inode,
1171 struct ext4_iloc *iloc)
1172 {
1173 int error;
1174 void *buf = NULL;
1175 struct buffer_head *data_bh = NULL;
1176 struct ext4_map_blocks map;
1177 int inline_size;
1178
1179 inline_size = ext4_get_inline_size(inode);
1180 buf = kmalloc(inline_size, GFP_NOFS);
1181 if (!buf) {
1182 error = -ENOMEM;
1183 goto out;
1184 }
1185
1186 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1187 if (error < 0)
1188 goto out;
1189
1190 /*
1191 * Make sure the inline directory entries pass checks before we try to
1192 * convert them, so that we avoid touching stuff that needs fsck.
1193 */
1194 if (S_ISDIR(inode->i_mode)) {
1195 error = ext4_check_all_de(inode, iloc->bh,
1196 buf + EXT4_INLINE_DOTDOT_SIZE,
1197 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1198 if (error)
1199 goto out;
1200 }
1201
1202 error = ext4_destroy_inline_data_nolock(handle, inode);
1203 if (error)
1204 goto out;
1205
1206 map.m_lblk = 0;
1207 map.m_len = 1;
1208 map.m_flags = 0;
1209 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1210 if (error < 0)
1211 goto out_restore;
1212 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1213 error = -EIO;
1214 goto out_restore;
1215 }
1216
1217 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1218 if (!data_bh) {
1219 error = -ENOMEM;
1220 goto out_restore;
1221 }
1222
1223 lock_buffer(data_bh);
1224 error = ext4_journal_get_create_access(handle, data_bh);
1225 if (error) {
1226 unlock_buffer(data_bh);
1227 error = -EIO;
1228 goto out_restore;
1229 }
1230 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1231
1232 if (!S_ISDIR(inode->i_mode)) {
1233 memcpy(data_bh->b_data, buf, inline_size);
1234 set_buffer_uptodate(data_bh);
1235 error = ext4_handle_dirty_metadata(handle,
1236 inode, data_bh);
1237 } else {
1238 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1239 buf, inline_size);
1240 }
1241
1242 unlock_buffer(data_bh);
1243 out_restore:
1244 if (error)
1245 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1246
1247 out:
1248 brelse(data_bh);
1249 kfree(buf);
1250 return error;
1251 }
1252
1253 /*
1254 * Try to add the new entry to the inline data.
1255 * If succeeds, return 0. If not, extended the inline dir and copied data to
1256 * the new created block.
1257 */
1258 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1259 struct inode *dir, struct inode *inode)
1260 {
1261 int ret, inline_size, no_expand;
1262 void *inline_start;
1263 struct ext4_iloc iloc;
1264
1265 ret = ext4_get_inode_loc(dir, &iloc);
1266 if (ret)
1267 return ret;
1268
1269 ext4_write_lock_xattr(dir, &no_expand);
1270 if (!ext4_has_inline_data(dir))
1271 goto out;
1272
1273 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1274 EXT4_INLINE_DOTDOT_SIZE;
1275 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1276
1277 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1278 inline_start, inline_size);
1279 if (ret != -ENOSPC)
1280 goto out;
1281
1282 /* check whether it can be inserted to inline xattr space. */
1283 inline_size = EXT4_I(dir)->i_inline_size -
1284 EXT4_MIN_INLINE_DATA_SIZE;
1285 if (!inline_size) {
1286 /* Try to use the xattr space.*/
1287 ret = ext4_update_inline_dir(handle, dir, &iloc);
1288 if (ret && ret != -ENOSPC)
1289 goto out;
1290
1291 inline_size = EXT4_I(dir)->i_inline_size -
1292 EXT4_MIN_INLINE_DATA_SIZE;
1293 }
1294
1295 if (inline_size) {
1296 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1297
1298 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1299 inode, &iloc, inline_start,
1300 inline_size);
1301
1302 if (ret != -ENOSPC)
1303 goto out;
1304 }
1305
1306 /*
1307 * The inline space is filled up, so create a new block for it.
1308 * As the extent tree will be created, we have to save the inline
1309 * dir first.
1310 */
1311 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1312
1313 out:
1314 ext4_write_unlock_xattr(dir, &no_expand);
1315 ext4_mark_inode_dirty(handle, dir);
1316 brelse(iloc.bh);
1317 return ret;
1318 }
1319
1320 /*
1321 * This function fills a red-black tree with information from an
1322 * inlined dir. It returns the number directory entries loaded
1323 * into the tree. If there is an error it is returned in err.
1324 */
1325 int htree_inlinedir_to_tree(struct file *dir_file,
1326 struct inode *dir, ext4_lblk_t block,
1327 struct dx_hash_info *hinfo,
1328 __u32 start_hash, __u32 start_minor_hash,
1329 int *has_inline_data)
1330 {
1331 int err = 0, count = 0;
1332 unsigned int parent_ino;
1333 int pos;
1334 struct ext4_dir_entry_2 *de;
1335 struct inode *inode = file_inode(dir_file);
1336 int ret, inline_size = 0;
1337 struct ext4_iloc iloc;
1338 void *dir_buf = NULL;
1339 struct ext4_dir_entry_2 fake;
1340 struct fscrypt_str tmp_str;
1341
1342 ret = ext4_get_inode_loc(inode, &iloc);
1343 if (ret)
1344 return ret;
1345
1346 down_read(&EXT4_I(inode)->xattr_sem);
1347 if (!ext4_has_inline_data(inode)) {
1348 up_read(&EXT4_I(inode)->xattr_sem);
1349 *has_inline_data = 0;
1350 goto out;
1351 }
1352
1353 inline_size = ext4_get_inline_size(inode);
1354 dir_buf = kmalloc(inline_size, GFP_NOFS);
1355 if (!dir_buf) {
1356 ret = -ENOMEM;
1357 up_read(&EXT4_I(inode)->xattr_sem);
1358 goto out;
1359 }
1360
1361 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1362 up_read(&EXT4_I(inode)->xattr_sem);
1363 if (ret < 0)
1364 goto out;
1365
1366 pos = 0;
1367 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1368 while (pos < inline_size) {
1369 /*
1370 * As inlined dir doesn't store any information about '.' and
1371 * only the inode number of '..' is stored, we have to handle
1372 * them differently.
1373 */
1374 if (pos == 0) {
1375 fake.inode = cpu_to_le32(inode->i_ino);
1376 fake.name_len = 1;
1377 strcpy(fake.name, ".");
1378 fake.rec_len = ext4_rec_len_to_disk(
1379 EXT4_DIR_REC_LEN(fake.name_len),
1380 inline_size);
1381 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1382 de = &fake;
1383 pos = EXT4_INLINE_DOTDOT_OFFSET;
1384 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1385 fake.inode = cpu_to_le32(parent_ino);
1386 fake.name_len = 2;
1387 strcpy(fake.name, "..");
1388 fake.rec_len = ext4_rec_len_to_disk(
1389 EXT4_DIR_REC_LEN(fake.name_len),
1390 inline_size);
1391 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1392 de = &fake;
1393 pos = EXT4_INLINE_DOTDOT_SIZE;
1394 } else {
1395 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1396 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1397 if (ext4_check_dir_entry(inode, dir_file, de,
1398 iloc.bh, dir_buf,
1399 inline_size, pos)) {
1400 ret = count;
1401 goto out;
1402 }
1403 }
1404
1405 ext4fs_dirhash(de->name, de->name_len, hinfo);
1406 if ((hinfo->hash < start_hash) ||
1407 ((hinfo->hash == start_hash) &&
1408 (hinfo->minor_hash < start_minor_hash)))
1409 continue;
1410 if (de->inode == 0)
1411 continue;
1412 tmp_str.name = de->name;
1413 tmp_str.len = de->name_len;
1414 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1415 hinfo->minor_hash, de, &tmp_str);
1416 if (err) {
1417 count = err;
1418 goto out;
1419 }
1420 count++;
1421 }
1422 ret = count;
1423 out:
1424 kfree(dir_buf);
1425 brelse(iloc.bh);
1426 return ret;
1427 }
1428
1429 /*
1430 * So this function is called when the volume is mkfsed with
1431 * dir_index disabled. In order to keep f_pos persistent
1432 * after we convert from an inlined dir to a blocked based,
1433 * we just pretend that we are a normal dir and return the
1434 * offset as if '.' and '..' really take place.
1435 *
1436 */
1437 int ext4_read_inline_dir(struct file *file,
1438 struct dir_context *ctx,
1439 int *has_inline_data)
1440 {
1441 unsigned int offset, parent_ino;
1442 int i;
1443 struct ext4_dir_entry_2 *de;
1444 struct super_block *sb;
1445 struct inode *inode = file_inode(file);
1446 int ret, inline_size = 0;
1447 struct ext4_iloc iloc;
1448 void *dir_buf = NULL;
1449 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1450
1451 ret = ext4_get_inode_loc(inode, &iloc);
1452 if (ret)
1453 return ret;
1454
1455 down_read(&EXT4_I(inode)->xattr_sem);
1456 if (!ext4_has_inline_data(inode)) {
1457 up_read(&EXT4_I(inode)->xattr_sem);
1458 *has_inline_data = 0;
1459 goto out;
1460 }
1461
1462 inline_size = ext4_get_inline_size(inode);
1463 dir_buf = kmalloc(inline_size, GFP_NOFS);
1464 if (!dir_buf) {
1465 ret = -ENOMEM;
1466 up_read(&EXT4_I(inode)->xattr_sem);
1467 goto out;
1468 }
1469
1470 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1471 up_read(&EXT4_I(inode)->xattr_sem);
1472 if (ret < 0)
1473 goto out;
1474
1475 ret = 0;
1476 sb = inode->i_sb;
1477 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1478 offset = ctx->pos;
1479
1480 /*
1481 * dotdot_offset and dotdot_size is the real offset and
1482 * size for ".." and "." if the dir is block based while
1483 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1484 * So we will use extra_offset and extra_size to indicate them
1485 * during the inline dir iteration.
1486 */
1487 dotdot_offset = EXT4_DIR_REC_LEN(1);
1488 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1489 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1490 extra_size = extra_offset + inline_size;
1491
1492 /*
1493 * If the version has changed since the last call to
1494 * readdir(2), then we might be pointing to an invalid
1495 * dirent right now. Scan from the start of the inline
1496 * dir to make sure.
1497 */
1498 if (inode_cmp_iversion(inode, file->f_version)) {
1499 for (i = 0; i < extra_size && i < offset;) {
1500 /*
1501 * "." is with offset 0 and
1502 * ".." is dotdot_offset.
1503 */
1504 if (!i) {
1505 i = dotdot_offset;
1506 continue;
1507 } else if (i == dotdot_offset) {
1508 i = dotdot_size;
1509 continue;
1510 }
1511 /* for other entry, the real offset in
1512 * the buf has to be tuned accordingly.
1513 */
1514 de = (struct ext4_dir_entry_2 *)
1515 (dir_buf + i - extra_offset);
1516 /* It's too expensive to do a full
1517 * dirent test each time round this
1518 * loop, but we do have to test at
1519 * least that it is non-zero. A
1520 * failure will be detected in the
1521 * dirent test below. */
1522 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1523 < EXT4_DIR_REC_LEN(1))
1524 break;
1525 i += ext4_rec_len_from_disk(de->rec_len,
1526 extra_size);
1527 }
1528 offset = i;
1529 ctx->pos = offset;
1530 file->f_version = inode_query_iversion(inode);
1531 }
1532
1533 while (ctx->pos < extra_size) {
1534 if (ctx->pos == 0) {
1535 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1536 goto out;
1537 ctx->pos = dotdot_offset;
1538 continue;
1539 }
1540
1541 if (ctx->pos == dotdot_offset) {
1542 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1543 goto out;
1544 ctx->pos = dotdot_size;
1545 continue;
1546 }
1547
1548 de = (struct ext4_dir_entry_2 *)
1549 (dir_buf + ctx->pos - extra_offset);
1550 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1551 extra_size, ctx->pos))
1552 goto out;
1553 if (le32_to_cpu(de->inode)) {
1554 if (!dir_emit(ctx, de->name, de->name_len,
1555 le32_to_cpu(de->inode),
1556 get_dtype(sb, de->file_type)))
1557 goto out;
1558 }
1559 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1560 }
1561 out:
1562 kfree(dir_buf);
1563 brelse(iloc.bh);
1564 return ret;
1565 }
1566
1567 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1568 struct ext4_dir_entry_2 **parent_de,
1569 int *retval)
1570 {
1571 struct ext4_iloc iloc;
1572
1573 *retval = ext4_get_inode_loc(inode, &iloc);
1574 if (*retval)
1575 return NULL;
1576
1577 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1578
1579 return iloc.bh;
1580 }
1581
1582 /*
1583 * Try to create the inline data for the new dir.
1584 * If it succeeds, return 0, otherwise return the error.
1585 * In case of ENOSPC, the caller should create the normal disk layout dir.
1586 */
1587 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1588 struct inode *inode)
1589 {
1590 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1591 struct ext4_iloc iloc;
1592 struct ext4_dir_entry_2 *de;
1593
1594 ret = ext4_get_inode_loc(inode, &iloc);
1595 if (ret)
1596 return ret;
1597
1598 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1599 if (ret)
1600 goto out;
1601
1602 /*
1603 * For inline dir, we only save the inode information for the ".."
1604 * and create a fake dentry to cover the left space.
1605 */
1606 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1607 de->inode = cpu_to_le32(parent->i_ino);
1608 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1609 de->inode = 0;
1610 de->rec_len = ext4_rec_len_to_disk(
1611 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1612 inline_size);
1613 set_nlink(inode, 2);
1614 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1615 out:
1616 brelse(iloc.bh);
1617 return ret;
1618 }
1619
1620 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1621 struct ext4_filename *fname,
1622 struct ext4_dir_entry_2 **res_dir,
1623 int *has_inline_data)
1624 {
1625 int ret;
1626 struct ext4_iloc iloc;
1627 void *inline_start;
1628 int inline_size;
1629
1630 if (ext4_get_inode_loc(dir, &iloc))
1631 return NULL;
1632
1633 down_read(&EXT4_I(dir)->xattr_sem);
1634 if (!ext4_has_inline_data(dir)) {
1635 *has_inline_data = 0;
1636 goto out;
1637 }
1638
1639 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1640 EXT4_INLINE_DOTDOT_SIZE;
1641 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1642 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1643 dir, fname, 0, res_dir);
1644 if (ret == 1)
1645 goto out_find;
1646 if (ret < 0)
1647 goto out;
1648
1649 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1650 goto out;
1651
1652 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1653 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1654
1655 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1656 dir, fname, 0, res_dir);
1657 if (ret == 1)
1658 goto out_find;
1659
1660 out:
1661 brelse(iloc.bh);
1662 iloc.bh = NULL;
1663 out_find:
1664 up_read(&EXT4_I(dir)->xattr_sem);
1665 return iloc.bh;
1666 }
1667
1668 int ext4_delete_inline_entry(handle_t *handle,
1669 struct inode *dir,
1670 struct ext4_dir_entry_2 *de_del,
1671 struct buffer_head *bh,
1672 int *has_inline_data)
1673 {
1674 int err, inline_size, no_expand;
1675 struct ext4_iloc iloc;
1676 void *inline_start;
1677
1678 err = ext4_get_inode_loc(dir, &iloc);
1679 if (err)
1680 return err;
1681
1682 ext4_write_lock_xattr(dir, &no_expand);
1683 if (!ext4_has_inline_data(dir)) {
1684 *has_inline_data = 0;
1685 goto out;
1686 }
1687
1688 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1689 EXT4_MIN_INLINE_DATA_SIZE) {
1690 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1691 EXT4_INLINE_DOTDOT_SIZE;
1692 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1693 EXT4_INLINE_DOTDOT_SIZE;
1694 } else {
1695 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1696 inline_size = ext4_get_inline_size(dir) -
1697 EXT4_MIN_INLINE_DATA_SIZE;
1698 }
1699
1700 BUFFER_TRACE(bh, "get_write_access");
1701 err = ext4_journal_get_write_access(handle, bh);
1702 if (err)
1703 goto out;
1704
1705 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1706 inline_start, inline_size, 0);
1707 if (err)
1708 goto out;
1709
1710 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1711 out:
1712 ext4_write_unlock_xattr(dir, &no_expand);
1713 if (likely(err == 0))
1714 err = ext4_mark_inode_dirty(handle, dir);
1715 brelse(iloc.bh);
1716 if (err != -ENOENT)
1717 ext4_std_error(dir->i_sb, err);
1718 return err;
1719 }
1720
1721 /*
1722 * Get the inline dentry at offset.
1723 */
1724 static inline struct ext4_dir_entry_2 *
1725 ext4_get_inline_entry(struct inode *inode,
1726 struct ext4_iloc *iloc,
1727 unsigned int offset,
1728 void **inline_start,
1729 int *inline_size)
1730 {
1731 void *inline_pos;
1732
1733 BUG_ON(offset > ext4_get_inline_size(inode));
1734
1735 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1736 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1737 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1738 } else {
1739 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1740 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1741 *inline_size = ext4_get_inline_size(inode) -
1742 EXT4_MIN_INLINE_DATA_SIZE;
1743 }
1744
1745 if (inline_start)
1746 *inline_start = inline_pos;
1747 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1748 }
1749
1750 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1751 {
1752 int err, inline_size;
1753 struct ext4_iloc iloc;
1754 void *inline_pos;
1755 unsigned int offset;
1756 struct ext4_dir_entry_2 *de;
1757 bool ret = true;
1758
1759 err = ext4_get_inode_loc(dir, &iloc);
1760 if (err) {
1761 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1762 err, dir->i_ino);
1763 return true;
1764 }
1765
1766 down_read(&EXT4_I(dir)->xattr_sem);
1767 if (!ext4_has_inline_data(dir)) {
1768 *has_inline_data = 0;
1769 goto out;
1770 }
1771
1772 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1773 if (!le32_to_cpu(de->inode)) {
1774 ext4_warning(dir->i_sb,
1775 "bad inline directory (dir #%lu) - no `..'",
1776 dir->i_ino);
1777 ret = true;
1778 goto out;
1779 }
1780
1781 offset = EXT4_INLINE_DOTDOT_SIZE;
1782 while (offset < dir->i_size) {
1783 de = ext4_get_inline_entry(dir, &iloc, offset,
1784 &inline_pos, &inline_size);
1785 if (ext4_check_dir_entry(dir, NULL, de,
1786 iloc.bh, inline_pos,
1787 inline_size, offset)) {
1788 ext4_warning(dir->i_sb,
1789 "bad inline directory (dir #%lu) - "
1790 "inode %u, rec_len %u, name_len %d"
1791 "inline size %d",
1792 dir->i_ino, le32_to_cpu(de->inode),
1793 le16_to_cpu(de->rec_len), de->name_len,
1794 inline_size);
1795 ret = true;
1796 goto out;
1797 }
1798 if (le32_to_cpu(de->inode)) {
1799 ret = false;
1800 goto out;
1801 }
1802 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1803 }
1804
1805 out:
1806 up_read(&EXT4_I(dir)->xattr_sem);
1807 brelse(iloc.bh);
1808 return ret;
1809 }
1810
1811 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1812 {
1813 int ret, no_expand;
1814
1815 ext4_write_lock_xattr(inode, &no_expand);
1816 ret = ext4_destroy_inline_data_nolock(handle, inode);
1817 ext4_write_unlock_xattr(inode, &no_expand);
1818
1819 return ret;
1820 }
1821
1822 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1823 {
1824 __u64 addr;
1825 int error = -EAGAIN;
1826 struct ext4_iloc iloc;
1827
1828 down_read(&EXT4_I(inode)->xattr_sem);
1829 if (!ext4_has_inline_data(inode))
1830 goto out;
1831
1832 error = ext4_get_inode_loc(inode, &iloc);
1833 if (error)
1834 goto out;
1835
1836 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1837 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1838 addr += offsetof(struct ext4_inode, i_block);
1839
1840 brelse(iloc.bh);
1841
1842 iomap->addr = addr;
1843 iomap->offset = 0;
1844 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1845 i_size_read(inode));
1846 iomap->type = 0;
1847 iomap->flags = IOMAP_F_DATA_INLINE;
1848
1849 out:
1850 up_read(&EXT4_I(inode)->xattr_sem);
1851 return error;
1852 }
1853
1854 int ext4_inline_data_fiemap(struct inode *inode,
1855 struct fiemap_extent_info *fieinfo,
1856 int *has_inline, __u64 start, __u64 len)
1857 {
1858 __u64 physical = 0;
1859 __u64 inline_len;
1860 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1861 FIEMAP_EXTENT_LAST;
1862 int error = 0;
1863 struct ext4_iloc iloc;
1864
1865 down_read(&EXT4_I(inode)->xattr_sem);
1866 if (!ext4_has_inline_data(inode)) {
1867 *has_inline = 0;
1868 goto out;
1869 }
1870 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1871 i_size_read(inode));
1872 if (start >= inline_len)
1873 goto out;
1874 if (start + len < inline_len)
1875 inline_len = start + len;
1876 inline_len -= start;
1877
1878 error = ext4_get_inode_loc(inode, &iloc);
1879 if (error)
1880 goto out;
1881
1882 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1883 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1884 physical += offsetof(struct ext4_inode, i_block);
1885
1886 if (physical)
1887 error = fiemap_fill_next_extent(fieinfo, start, physical,
1888 inline_len, flags);
1889 brelse(iloc.bh);
1890 out:
1891 up_read(&EXT4_I(inode)->xattr_sem);
1892 return (error < 0 ? error : 0);
1893 }
1894
1895 /*
1896 * Called during xattr set, and if we can sparse space 'needed',
1897 * just create the extent tree evict the data to the outer block.
1898 *
1899 * We use jbd2 instead of page cache to move data to the 1st block
1900 * so that the whole transaction can be committed as a whole and
1901 * the data isn't lost because of the delayed page cache write.
1902 */
1903 int ext4_try_to_evict_inline_data(handle_t *handle,
1904 struct inode *inode,
1905 int needed)
1906 {
1907 int error;
1908 struct ext4_xattr_entry *entry;
1909 struct ext4_inode *raw_inode;
1910 struct ext4_iloc iloc;
1911
1912 error = ext4_get_inode_loc(inode, &iloc);
1913 if (error)
1914 return error;
1915
1916 raw_inode = ext4_raw_inode(&iloc);
1917 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1918 EXT4_I(inode)->i_inline_off);
1919 if (EXT4_XATTR_LEN(entry->e_name_len) +
1920 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1921 error = -ENOSPC;
1922 goto out;
1923 }
1924
1925 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1926 out:
1927 brelse(iloc.bh);
1928 return error;
1929 }
1930
1931 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1932 {
1933 handle_t *handle;
1934 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1935 size_t i_size;
1936 void *value = NULL;
1937 struct ext4_xattr_ibody_find is = {
1938 .s = { .not_found = -ENODATA, },
1939 };
1940 struct ext4_xattr_info i = {
1941 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1942 .name = EXT4_XATTR_SYSTEM_DATA,
1943 };
1944
1945
1946 needed_blocks = ext4_writepage_trans_blocks(inode);
1947 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1948 if (IS_ERR(handle))
1949 return PTR_ERR(handle);
1950
1951 ext4_write_lock_xattr(inode, &no_expand);
1952 if (!ext4_has_inline_data(inode)) {
1953 *has_inline = 0;
1954 ext4_journal_stop(handle);
1955 return 0;
1956 }
1957
1958 if ((err = ext4_orphan_add(handle, inode)) != 0)
1959 goto out;
1960
1961 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1962 goto out;
1963
1964 down_write(&EXT4_I(inode)->i_data_sem);
1965 i_size = inode->i_size;
1966 inline_size = ext4_get_inline_size(inode);
1967 EXT4_I(inode)->i_disksize = i_size;
1968
1969 if (i_size < inline_size) {
1970 /* Clear the content in the xattr space. */
1971 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1972 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1973 goto out_error;
1974
1975 BUG_ON(is.s.not_found);
1976
1977 value_len = le32_to_cpu(is.s.here->e_value_size);
1978 value = kmalloc(value_len, GFP_NOFS);
1979 if (!value) {
1980 err = -ENOMEM;
1981 goto out_error;
1982 }
1983
1984 err = ext4_xattr_ibody_get(inode, i.name_index,
1985 i.name, value, value_len);
1986 if (err <= 0)
1987 goto out_error;
1988
1989 i.value = value;
1990 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1991 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1992 err = ext4_xattr_ibody_inline_set(handle, inode,
1993 &i, &is);
1994 if (err)
1995 goto out_error;
1996 }
1997
1998 /* Clear the content within i_blocks. */
1999 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
2000 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
2001 memset(p + i_size, 0,
2002 EXT4_MIN_INLINE_DATA_SIZE - i_size);
2003 }
2004
2005 EXT4_I(inode)->i_inline_size = i_size <
2006 EXT4_MIN_INLINE_DATA_SIZE ?
2007 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2008 }
2009
2010 out_error:
2011 up_write(&EXT4_I(inode)->i_data_sem);
2012 out:
2013 brelse(is.iloc.bh);
2014 ext4_write_unlock_xattr(inode, &no_expand);
2015 kfree(value);
2016 if (inode->i_nlink)
2017 ext4_orphan_del(handle, inode);
2018
2019 if (err == 0) {
2020 inode->i_mtime = inode->i_ctime = current_time(inode);
2021 err = ext4_mark_inode_dirty(handle, inode);
2022 if (IS_SYNC(inode))
2023 ext4_handle_sync(handle);
2024 }
2025 ext4_journal_stop(handle);
2026 return err;
2027 }
2028
2029 int ext4_convert_inline_data(struct inode *inode)
2030 {
2031 int error, needed_blocks, no_expand;
2032 handle_t *handle;
2033 struct ext4_iloc iloc;
2034
2035 if (!ext4_has_inline_data(inode)) {
2036 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2037 return 0;
2038 }
2039
2040 needed_blocks = ext4_writepage_trans_blocks(inode);
2041
2042 iloc.bh = NULL;
2043 error = ext4_get_inode_loc(inode, &iloc);
2044 if (error)
2045 return error;
2046
2047 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2048 if (IS_ERR(handle)) {
2049 error = PTR_ERR(handle);
2050 goto out_free;
2051 }
2052
2053 ext4_write_lock_xattr(inode, &no_expand);
2054 if (ext4_has_inline_data(inode))
2055 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2056 ext4_write_unlock_xattr(inode, &no_expand);
2057 ext4_journal_stop(handle);
2058 out_free:
2059 brelse(iloc.bh);
2060 return error;
2061 }