Loading...
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
27static 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
35static 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
88out:
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 */
97int 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 */
130int 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 }
159out:
160 brelse(is.iloc.bh);
161 return error;
162}
163
164static 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
200out:
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 */
210static 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
246static 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
305out:
306 brelse(is.iloc.bh);
307 return error;
308}
309
310static 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 goto out;
341
342 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
343 value, len);
344 if (error == -ENODATA)
345 goto out;
346
347 BUFFER_TRACE(is.iloc.bh, "get_write_access");
348 error = ext4_journal_get_write_access(handle, is.iloc.bh);
349 if (error)
350 goto out;
351
352 /* Update the xttr entry. */
353 i.value = value;
354 i.value_len = len;
355
356 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
357 if (error)
358 goto out;
359
360 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
361 (void *)ext4_raw_inode(&is.iloc));
362 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
363 le32_to_cpu(is.s.here->e_value_size);
364 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
365 get_bh(is.iloc.bh);
366 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
367
368out:
369 kfree(value);
370 brelse(is.iloc.bh);
371 return error;
372}
373
374static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
375 unsigned int len)
376{
377 int ret, size;
378 struct ext4_inode_info *ei = EXT4_I(inode);
379
380 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
381 return -ENOSPC;
382
383 size = ext4_get_max_inline_size(inode);
384 if (size < len)
385 return -ENOSPC;
386
387 down_write(&EXT4_I(inode)->xattr_sem);
388
389 if (ei->i_inline_off)
390 ret = ext4_update_inline_data(handle, inode, len);
391 else
392 ret = ext4_create_inline_data(handle, inode, len);
393
394 up_write(&EXT4_I(inode)->xattr_sem);
395
396 return ret;
397}
398
399static int ext4_destroy_inline_data_nolock(handle_t *handle,
400 struct inode *inode)
401{
402 struct ext4_inode_info *ei = EXT4_I(inode);
403 struct ext4_xattr_ibody_find is = {
404 .s = { .not_found = 0, },
405 };
406 struct ext4_xattr_info i = {
407 .name_index = EXT4_XATTR_INDEX_SYSTEM,
408 .name = EXT4_XATTR_SYSTEM_DATA,
409 .value = NULL,
410 .value_len = 0,
411 };
412 int error;
413
414 if (!ei->i_inline_off)
415 return 0;
416
417 error = ext4_get_inode_loc(inode, &is.iloc);
418 if (error)
419 return error;
420
421 error = ext4_xattr_ibody_find(inode, &i, &is);
422 if (error)
423 goto out;
424
425 BUFFER_TRACE(is.iloc.bh, "get_write_access");
426 error = ext4_journal_get_write_access(handle, is.iloc.bh);
427 if (error)
428 goto out;
429
430 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
431 if (error)
432 goto out;
433
434 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
435 0, EXT4_MIN_INLINE_DATA_SIZE);
436
437 if (ext4_has_feature_extents(inode->i_sb)) {
438 if (S_ISDIR(inode->i_mode) ||
439 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
440 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
441 ext4_ext_tree_init(handle, inode);
442 }
443 }
444 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
445
446 get_bh(is.iloc.bh);
447 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
448
449 EXT4_I(inode)->i_inline_off = 0;
450 EXT4_I(inode)->i_inline_size = 0;
451 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
452out:
453 brelse(is.iloc.bh);
454 if (error == -ENODATA)
455 error = 0;
456 return error;
457}
458
459static int ext4_read_inline_page(struct inode *inode, struct page *page)
460{
461 void *kaddr;
462 int ret = 0;
463 size_t len;
464 struct ext4_iloc iloc;
465
466 BUG_ON(!PageLocked(page));
467 BUG_ON(!ext4_has_inline_data(inode));
468 BUG_ON(page->index);
469
470 if (!EXT4_I(inode)->i_inline_off) {
471 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
472 inode->i_ino);
473 goto out;
474 }
475
476 ret = ext4_get_inode_loc(inode, &iloc);
477 if (ret)
478 goto out;
479
480 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
481 kaddr = kmap_atomic(page);
482 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
483 flush_dcache_page(page);
484 kunmap_atomic(kaddr);
485 zero_user_segment(page, len, PAGE_SIZE);
486 SetPageUptodate(page);
487 brelse(iloc.bh);
488
489out:
490 return ret;
491}
492
493int ext4_readpage_inline(struct inode *inode, struct page *page)
494{
495 int ret = 0;
496
497 down_read(&EXT4_I(inode)->xattr_sem);
498 if (!ext4_has_inline_data(inode)) {
499 up_read(&EXT4_I(inode)->xattr_sem);
500 return -EAGAIN;
501 }
502
503 /*
504 * Current inline data can only exist in the 1st page,
505 * So for all the other pages, just set them uptodate.
506 */
507 if (!page->index)
508 ret = ext4_read_inline_page(inode, page);
509 else if (!PageUptodate(page)) {
510 zero_user_segment(page, 0, PAGE_SIZE);
511 SetPageUptodate(page);
512 }
513
514 up_read(&EXT4_I(inode)->xattr_sem);
515
516 unlock_page(page);
517 return ret >= 0 ? 0 : ret;
518}
519
520static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
521 struct inode *inode,
522 unsigned flags)
523{
524 int ret, needed_blocks;
525 handle_t *handle = NULL;
526 int retries = 0, sem_held = 0;
527 struct page *page = NULL;
528 unsigned from, to;
529 struct ext4_iloc iloc;
530
531 if (!ext4_has_inline_data(inode)) {
532 /*
533 * clear the flag so that no new write
534 * will trap here again.
535 */
536 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
537 return 0;
538 }
539
540 needed_blocks = ext4_writepage_trans_blocks(inode);
541
542 ret = ext4_get_inode_loc(inode, &iloc);
543 if (ret)
544 return ret;
545
546retry:
547 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
548 if (IS_ERR(handle)) {
549 ret = PTR_ERR(handle);
550 handle = NULL;
551 goto out;
552 }
553
554 /* We cannot recurse into the filesystem as the transaction is already
555 * started */
556 flags |= AOP_FLAG_NOFS;
557
558 page = grab_cache_page_write_begin(mapping, 0, flags);
559 if (!page) {
560 ret = -ENOMEM;
561 goto out;
562 }
563
564 down_write(&EXT4_I(inode)->xattr_sem);
565 sem_held = 1;
566 /* If some one has already done this for us, just exit. */
567 if (!ext4_has_inline_data(inode)) {
568 ret = 0;
569 goto out;
570 }
571
572 from = 0;
573 to = ext4_get_inline_size(inode);
574 if (!PageUptodate(page)) {
575 ret = ext4_read_inline_page(inode, page);
576 if (ret < 0)
577 goto out;
578 }
579
580 ret = ext4_destroy_inline_data_nolock(handle, inode);
581 if (ret)
582 goto out;
583
584 if (ext4_should_dioread_nolock(inode)) {
585 ret = __block_write_begin(page, from, to,
586 ext4_get_block_unwritten);
587 } else
588 ret = __block_write_begin(page, from, to, ext4_get_block);
589
590 if (!ret && ext4_should_journal_data(inode)) {
591 ret = ext4_walk_page_buffers(handle, page_buffers(page),
592 from, to, NULL,
593 do_journal_get_write_access);
594 }
595
596 if (ret) {
597 unlock_page(page);
598 put_page(page);
599 page = NULL;
600 ext4_orphan_add(handle, inode);
601 up_write(&EXT4_I(inode)->xattr_sem);
602 sem_held = 0;
603 ext4_journal_stop(handle);
604 handle = NULL;
605 ext4_truncate_failed_write(inode);
606 /*
607 * If truncate failed early the inode might
608 * still be on the orphan list; we need to
609 * make sure the inode is removed from the
610 * orphan list in that case.
611 */
612 if (inode->i_nlink)
613 ext4_orphan_del(NULL, inode);
614 }
615
616 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617 goto retry;
618
619 if (page)
620 block_commit_write(page, from, to);
621out:
622 if (page) {
623 unlock_page(page);
624 put_page(page);
625 }
626 if (sem_held)
627 up_write(&EXT4_I(inode)->xattr_sem);
628 if (handle)
629 ext4_journal_stop(handle);
630 brelse(iloc.bh);
631 return ret;
632}
633
634/*
635 * Try to write data in the inode.
636 * If the inode has inline data, check whether the new write can be
637 * in the inode also. If not, create the page the handle, move the data
638 * to the page make it update and let the later codes create extent for it.
639 */
640int ext4_try_to_write_inline_data(struct address_space *mapping,
641 struct inode *inode,
642 loff_t pos, unsigned len,
643 unsigned flags,
644 struct page **pagep)
645{
646 int ret;
647 handle_t *handle;
648 struct page *page;
649 struct ext4_iloc iloc;
650
651 if (pos + len > ext4_get_max_inline_size(inode))
652 goto convert;
653
654 ret = ext4_get_inode_loc(inode, &iloc);
655 if (ret)
656 return ret;
657
658 /*
659 * The possible write could happen in the inode,
660 * so try to reserve the space in inode first.
661 */
662 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
663 if (IS_ERR(handle)) {
664 ret = PTR_ERR(handle);
665 handle = NULL;
666 goto out;
667 }
668
669 ret = ext4_prepare_inline_data(handle, inode, pos + len);
670 if (ret && ret != -ENOSPC)
671 goto out;
672
673 /* We don't have space in inline inode, so convert it to extent. */
674 if (ret == -ENOSPC) {
675 ext4_journal_stop(handle);
676 brelse(iloc.bh);
677 goto convert;
678 }
679
680 flags |= AOP_FLAG_NOFS;
681
682 page = grab_cache_page_write_begin(mapping, 0, flags);
683 if (!page) {
684 ret = -ENOMEM;
685 goto out;
686 }
687
688 *pagep = page;
689 down_read(&EXT4_I(inode)->xattr_sem);
690 if (!ext4_has_inline_data(inode)) {
691 ret = 0;
692 unlock_page(page);
693 put_page(page);
694 goto out_up_read;
695 }
696
697 if (!PageUptodate(page)) {
698 ret = ext4_read_inline_page(inode, page);
699 if (ret < 0)
700 goto out_up_read;
701 }
702
703 ret = 1;
704 handle = NULL;
705out_up_read:
706 up_read(&EXT4_I(inode)->xattr_sem);
707out:
708 if (handle)
709 ext4_journal_stop(handle);
710 brelse(iloc.bh);
711 return ret;
712convert:
713 return ext4_convert_inline_data_to_extent(mapping,
714 inode, flags);
715}
716
717int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
718 unsigned copied, struct page *page)
719{
720 int ret;
721 void *kaddr;
722 struct ext4_iloc iloc;
723
724 if (unlikely(copied < len)) {
725 if (!PageUptodate(page)) {
726 copied = 0;
727 goto out;
728 }
729 }
730
731 ret = ext4_get_inode_loc(inode, &iloc);
732 if (ret) {
733 ext4_std_error(inode->i_sb, ret);
734 copied = 0;
735 goto out;
736 }
737
738 down_write(&EXT4_I(inode)->xattr_sem);
739 BUG_ON(!ext4_has_inline_data(inode));
740
741 kaddr = kmap_atomic(page);
742 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
743 kunmap_atomic(kaddr);
744 SetPageUptodate(page);
745 /* clear page dirty so that writepages wouldn't work for us. */
746 ClearPageDirty(page);
747
748 up_write(&EXT4_I(inode)->xattr_sem);
749 brelse(iloc.bh);
750out:
751 return copied;
752}
753
754struct buffer_head *
755ext4_journalled_write_inline_data(struct inode *inode,
756 unsigned len,
757 struct page *page)
758{
759 int ret;
760 void *kaddr;
761 struct ext4_iloc iloc;
762
763 ret = ext4_get_inode_loc(inode, &iloc);
764 if (ret) {
765 ext4_std_error(inode->i_sb, ret);
766 return NULL;
767 }
768
769 down_write(&EXT4_I(inode)->xattr_sem);
770 kaddr = kmap_atomic(page);
771 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
772 kunmap_atomic(kaddr);
773 up_write(&EXT4_I(inode)->xattr_sem);
774
775 return iloc.bh;
776}
777
778/*
779 * Try to make the page cache and handle ready for the inline data case.
780 * We can call this function in 2 cases:
781 * 1. The inode is created and the first write exceeds inline size. We can
782 * clear the inode state safely.
783 * 2. The inode has inline data, then we need to read the data, make it
784 * update and dirty so that ext4_da_writepages can handle it. We don't
785 * need to start the journal since the file's metatdata isn't changed now.
786 */
787static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
788 struct inode *inode,
789 unsigned flags,
790 void **fsdata)
791{
792 int ret = 0, inline_size;
793 struct page *page;
794
795 page = grab_cache_page_write_begin(mapping, 0, flags);
796 if (!page)
797 return -ENOMEM;
798
799 down_read(&EXT4_I(inode)->xattr_sem);
800 if (!ext4_has_inline_data(inode)) {
801 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
802 goto out;
803 }
804
805 inline_size = ext4_get_inline_size(inode);
806
807 if (!PageUptodate(page)) {
808 ret = ext4_read_inline_page(inode, page);
809 if (ret < 0)
810 goto out;
811 }
812
813 ret = __block_write_begin(page, 0, inline_size,
814 ext4_da_get_block_prep);
815 if (ret) {
816 up_read(&EXT4_I(inode)->xattr_sem);
817 unlock_page(page);
818 put_page(page);
819 ext4_truncate_failed_write(inode);
820 return ret;
821 }
822
823 SetPageDirty(page);
824 SetPageUptodate(page);
825 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
826 *fsdata = (void *)CONVERT_INLINE_DATA;
827
828out:
829 up_read(&EXT4_I(inode)->xattr_sem);
830 if (page) {
831 unlock_page(page);
832 put_page(page);
833 }
834 return ret;
835}
836
837/*
838 * Prepare the write for the inline data.
839 * If the the data can be written into the inode, we just read
840 * the page and make it uptodate, and start the journal.
841 * Otherwise read the page, makes it dirty so that it can be
842 * handle in writepages(the i_disksize update is left to the
843 * normal ext4_da_write_end).
844 */
845int ext4_da_write_inline_data_begin(struct address_space *mapping,
846 struct inode *inode,
847 loff_t pos, unsigned len,
848 unsigned flags,
849 struct page **pagep,
850 void **fsdata)
851{
852 int ret, inline_size;
853 handle_t *handle;
854 struct page *page;
855 struct ext4_iloc iloc;
856 int retries;
857
858 ret = ext4_get_inode_loc(inode, &iloc);
859 if (ret)
860 return ret;
861
862retry_journal:
863 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
864 if (IS_ERR(handle)) {
865 ret = PTR_ERR(handle);
866 goto out;
867 }
868
869 inline_size = ext4_get_max_inline_size(inode);
870
871 ret = -ENOSPC;
872 if (inline_size >= pos + len) {
873 ret = ext4_prepare_inline_data(handle, inode, pos + len);
874 if (ret && ret != -ENOSPC)
875 goto out_journal;
876 }
877
878 /*
879 * We cannot recurse into the filesystem as the transaction
880 * is already started.
881 */
882 flags |= AOP_FLAG_NOFS;
883
884 if (ret == -ENOSPC) {
885 ret = ext4_da_convert_inline_data_to_extent(mapping,
886 inode,
887 flags,
888 fsdata);
889 ext4_journal_stop(handle);
890 if (ret == -ENOSPC &&
891 ext4_should_retry_alloc(inode->i_sb, &retries))
892 goto retry_journal;
893 goto out;
894 }
895
896
897 page = grab_cache_page_write_begin(mapping, 0, flags);
898 if (!page) {
899 ret = -ENOMEM;
900 goto out_journal;
901 }
902
903 down_read(&EXT4_I(inode)->xattr_sem);
904 if (!ext4_has_inline_data(inode)) {
905 ret = 0;
906 goto out_release_page;
907 }
908
909 if (!PageUptodate(page)) {
910 ret = ext4_read_inline_page(inode, page);
911 if (ret < 0)
912 goto out_release_page;
913 }
914
915 up_read(&EXT4_I(inode)->xattr_sem);
916 *pagep = page;
917 brelse(iloc.bh);
918 return 1;
919out_release_page:
920 up_read(&EXT4_I(inode)->xattr_sem);
921 unlock_page(page);
922 put_page(page);
923out_journal:
924 ext4_journal_stop(handle);
925out:
926 brelse(iloc.bh);
927 return ret;
928}
929
930int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
931 unsigned len, unsigned copied,
932 struct page *page)
933{
934 int i_size_changed = 0;
935
936 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
937
938 /*
939 * No need to use i_size_read() here, the i_size
940 * cannot change under us because we hold i_mutex.
941 *
942 * But it's important to update i_size while still holding page lock:
943 * page writeout could otherwise come in and zero beyond i_size.
944 */
945 if (pos+copied > inode->i_size) {
946 i_size_write(inode, pos+copied);
947 i_size_changed = 1;
948 }
949 unlock_page(page);
950 put_page(page);
951
952 /*
953 * Don't mark the inode dirty under page lock. First, it unnecessarily
954 * makes the holding time of page lock longer. Second, it forces lock
955 * ordering of page lock and transaction start for journaling
956 * filesystems.
957 */
958 if (i_size_changed)
959 mark_inode_dirty(inode);
960
961 return copied;
962}
963
964#ifdef INLINE_DIR_DEBUG
965void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
966 void *inline_start, int inline_size)
967{
968 int offset;
969 unsigned short de_len;
970 struct ext4_dir_entry_2 *de = inline_start;
971 void *dlimit = inline_start + inline_size;
972
973 trace_printk("inode %lu\n", dir->i_ino);
974 offset = 0;
975 while ((void *)de < dlimit) {
976 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
977 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
978 offset, de_len, de->name_len, de->name,
979 de->name_len, le32_to_cpu(de->inode));
980 if (ext4_check_dir_entry(dir, NULL, de, bh,
981 inline_start, inline_size, offset))
982 BUG();
983
984 offset += de_len;
985 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
986 }
987}
988#else
989#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
990#endif
991
992/*
993 * Add a new entry into a inline dir.
994 * It will return -ENOSPC if no space is available, and -EIO
995 * and -EEXIST if directory entry already exists.
996 */
997static int ext4_add_dirent_to_inline(handle_t *handle,
998 struct ext4_filename *fname,
999 struct inode *dir,
1000 struct inode *inode,
1001 struct ext4_iloc *iloc,
1002 void *inline_start, int inline_size)
1003{
1004 int err;
1005 struct ext4_dir_entry_2 *de;
1006
1007 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1008 inline_size, fname, &de);
1009 if (err)
1010 return err;
1011
1012 BUFFER_TRACE(iloc->bh, "get_write_access");
1013 err = ext4_journal_get_write_access(handle, iloc->bh);
1014 if (err)
1015 return err;
1016 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1017
1018 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1019
1020 /*
1021 * XXX shouldn't update any times until successful
1022 * completion of syscall, but too many callers depend
1023 * on this.
1024 *
1025 * XXX similarly, too many callers depend on
1026 * ext4_new_inode() setting the times, but error
1027 * recovery deletes the inode, so the worst that can
1028 * happen is that the times are slightly out of date
1029 * and/or different from the directory change time.
1030 */
1031 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1032 ext4_update_dx_flag(dir);
1033 dir->i_version++;
1034 ext4_mark_inode_dirty(handle, dir);
1035 return 1;
1036}
1037
1038static void *ext4_get_inline_xattr_pos(struct inode *inode,
1039 struct ext4_iloc *iloc)
1040{
1041 struct ext4_xattr_entry *entry;
1042 struct ext4_xattr_ibody_header *header;
1043
1044 BUG_ON(!EXT4_I(inode)->i_inline_off);
1045
1046 header = IHDR(inode, ext4_raw_inode(iloc));
1047 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1048 EXT4_I(inode)->i_inline_off);
1049
1050 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1051}
1052
1053/* Set the final de to cover the whole block. */
1054static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1055{
1056 struct ext4_dir_entry_2 *de, *prev_de;
1057 void *limit;
1058 int de_len;
1059
1060 de = (struct ext4_dir_entry_2 *)de_buf;
1061 if (old_size) {
1062 limit = de_buf + old_size;
1063 do {
1064 prev_de = de;
1065 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1066 de_buf += de_len;
1067 de = (struct ext4_dir_entry_2 *)de_buf;
1068 } while (de_buf < limit);
1069
1070 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1071 old_size, new_size);
1072 } else {
1073 /* this is just created, so create an empty entry. */
1074 de->inode = 0;
1075 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1076 }
1077}
1078
1079static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1080 struct ext4_iloc *iloc)
1081{
1082 int ret;
1083 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1084 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1085
1086 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1087 return -ENOSPC;
1088
1089 ret = ext4_update_inline_data(handle, dir,
1090 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1091 if (ret)
1092 return ret;
1093
1094 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1095 EXT4_I(dir)->i_inline_size -
1096 EXT4_MIN_INLINE_DATA_SIZE);
1097 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1098 return 0;
1099}
1100
1101static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1102 struct ext4_iloc *iloc,
1103 void *buf, int inline_size)
1104{
1105 ext4_create_inline_data(handle, inode, inline_size);
1106 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1107 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1108}
1109
1110static int ext4_finish_convert_inline_dir(handle_t *handle,
1111 struct inode *inode,
1112 struct buffer_head *dir_block,
1113 void *buf,
1114 int inline_size)
1115{
1116 int err, csum_size = 0, header_size = 0;
1117 struct ext4_dir_entry_2 *de;
1118 struct ext4_dir_entry_tail *t;
1119 void *target = dir_block->b_data;
1120
1121 /*
1122 * First create "." and ".." and then copy the dir information
1123 * back to the block.
1124 */
1125 de = (struct ext4_dir_entry_2 *)target;
1126 de = ext4_init_dot_dotdot(inode, de,
1127 inode->i_sb->s_blocksize, csum_size,
1128 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1129 header_size = (void *)de - target;
1130
1131 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1132 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1133
1134 if (ext4_has_metadata_csum(inode->i_sb))
1135 csum_size = sizeof(struct ext4_dir_entry_tail);
1136
1137 inode->i_size = inode->i_sb->s_blocksize;
1138 i_size_write(inode, inode->i_sb->s_blocksize);
1139 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1140 ext4_update_final_de(dir_block->b_data,
1141 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1142 inode->i_sb->s_blocksize - csum_size);
1143
1144 if (csum_size) {
1145 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1146 inode->i_sb->s_blocksize);
1147 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1148 }
1149 set_buffer_uptodate(dir_block);
1150 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1151 if (err)
1152 goto out;
1153 set_buffer_verified(dir_block);
1154out:
1155 return err;
1156}
1157
1158static int ext4_convert_inline_data_nolock(handle_t *handle,
1159 struct inode *inode,
1160 struct ext4_iloc *iloc)
1161{
1162 int error;
1163 void *buf = NULL;
1164 struct buffer_head *data_bh = NULL;
1165 struct ext4_map_blocks map;
1166 int inline_size;
1167
1168 inline_size = ext4_get_inline_size(inode);
1169 buf = kmalloc(inline_size, GFP_NOFS);
1170 if (!buf) {
1171 error = -ENOMEM;
1172 goto out;
1173 }
1174
1175 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1176 if (error < 0)
1177 goto out;
1178
1179 /*
1180 * Make sure the inline directory entries pass checks before we try to
1181 * convert them, so that we avoid touching stuff that needs fsck.
1182 */
1183 if (S_ISDIR(inode->i_mode)) {
1184 error = ext4_check_all_de(inode, iloc->bh,
1185 buf + EXT4_INLINE_DOTDOT_SIZE,
1186 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1187 if (error)
1188 goto out;
1189 }
1190
1191 error = ext4_destroy_inline_data_nolock(handle, inode);
1192 if (error)
1193 goto out;
1194
1195 map.m_lblk = 0;
1196 map.m_len = 1;
1197 map.m_flags = 0;
1198 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1199 if (error < 0)
1200 goto out_restore;
1201 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1202 error = -EIO;
1203 goto out_restore;
1204 }
1205
1206 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1207 if (!data_bh) {
1208 error = -ENOMEM;
1209 goto out_restore;
1210 }
1211
1212 lock_buffer(data_bh);
1213 error = ext4_journal_get_create_access(handle, data_bh);
1214 if (error) {
1215 unlock_buffer(data_bh);
1216 error = -EIO;
1217 goto out_restore;
1218 }
1219 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1220
1221 if (!S_ISDIR(inode->i_mode)) {
1222 memcpy(data_bh->b_data, buf, inline_size);
1223 set_buffer_uptodate(data_bh);
1224 error = ext4_handle_dirty_metadata(handle,
1225 inode, data_bh);
1226 } else {
1227 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1228 buf, inline_size);
1229 }
1230
1231 unlock_buffer(data_bh);
1232out_restore:
1233 if (error)
1234 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1235
1236out:
1237 brelse(data_bh);
1238 kfree(buf);
1239 return error;
1240}
1241
1242/*
1243 * Try to add the new entry to the inline data.
1244 * If succeeds, return 0. If not, extended the inline dir and copied data to
1245 * the new created block.
1246 */
1247int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1248 struct inode *dir, struct inode *inode)
1249{
1250 int ret, inline_size;
1251 void *inline_start;
1252 struct ext4_iloc iloc;
1253
1254 ret = ext4_get_inode_loc(dir, &iloc);
1255 if (ret)
1256 return ret;
1257
1258 down_write(&EXT4_I(dir)->xattr_sem);
1259 if (!ext4_has_inline_data(dir))
1260 goto out;
1261
1262 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1263 EXT4_INLINE_DOTDOT_SIZE;
1264 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1265
1266 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1267 inline_start, inline_size);
1268 if (ret != -ENOSPC)
1269 goto out;
1270
1271 /* check whether it can be inserted to inline xattr space. */
1272 inline_size = EXT4_I(dir)->i_inline_size -
1273 EXT4_MIN_INLINE_DATA_SIZE;
1274 if (!inline_size) {
1275 /* Try to use the xattr space.*/
1276 ret = ext4_update_inline_dir(handle, dir, &iloc);
1277 if (ret && ret != -ENOSPC)
1278 goto out;
1279
1280 inline_size = EXT4_I(dir)->i_inline_size -
1281 EXT4_MIN_INLINE_DATA_SIZE;
1282 }
1283
1284 if (inline_size) {
1285 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1286
1287 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1288 inode, &iloc, inline_start,
1289 inline_size);
1290
1291 if (ret != -ENOSPC)
1292 goto out;
1293 }
1294
1295 /*
1296 * The inline space is filled up, so create a new block for it.
1297 * As the extent tree will be created, we have to save the inline
1298 * dir first.
1299 */
1300 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1301
1302out:
1303 ext4_mark_inode_dirty(handle, dir);
1304 up_write(&EXT4_I(dir)->xattr_sem);
1305 brelse(iloc.bh);
1306 return ret;
1307}
1308
1309/*
1310 * This function fills a red-black tree with information from an
1311 * inlined dir. It returns the number directory entries loaded
1312 * into the tree. If there is an error it is returned in err.
1313 */
1314int htree_inlinedir_to_tree(struct file *dir_file,
1315 struct inode *dir, ext4_lblk_t block,
1316 struct dx_hash_info *hinfo,
1317 __u32 start_hash, __u32 start_minor_hash,
1318 int *has_inline_data)
1319{
1320 int err = 0, count = 0;
1321 unsigned int parent_ino;
1322 int pos;
1323 struct ext4_dir_entry_2 *de;
1324 struct inode *inode = file_inode(dir_file);
1325 int ret, inline_size = 0;
1326 struct ext4_iloc iloc;
1327 void *dir_buf = NULL;
1328 struct ext4_dir_entry_2 fake;
1329 struct ext4_str tmp_str;
1330
1331 ret = ext4_get_inode_loc(inode, &iloc);
1332 if (ret)
1333 return ret;
1334
1335 down_read(&EXT4_I(inode)->xattr_sem);
1336 if (!ext4_has_inline_data(inode)) {
1337 up_read(&EXT4_I(inode)->xattr_sem);
1338 *has_inline_data = 0;
1339 goto out;
1340 }
1341
1342 inline_size = ext4_get_inline_size(inode);
1343 dir_buf = kmalloc(inline_size, GFP_NOFS);
1344 if (!dir_buf) {
1345 ret = -ENOMEM;
1346 up_read(&EXT4_I(inode)->xattr_sem);
1347 goto out;
1348 }
1349
1350 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1351 up_read(&EXT4_I(inode)->xattr_sem);
1352 if (ret < 0)
1353 goto out;
1354
1355 pos = 0;
1356 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1357 while (pos < inline_size) {
1358 /*
1359 * As inlined dir doesn't store any information about '.' and
1360 * only the inode number of '..' is stored, we have to handle
1361 * them differently.
1362 */
1363 if (pos == 0) {
1364 fake.inode = cpu_to_le32(inode->i_ino);
1365 fake.name_len = 1;
1366 strcpy(fake.name, ".");
1367 fake.rec_len = ext4_rec_len_to_disk(
1368 EXT4_DIR_REC_LEN(fake.name_len),
1369 inline_size);
1370 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1371 de = &fake;
1372 pos = EXT4_INLINE_DOTDOT_OFFSET;
1373 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1374 fake.inode = cpu_to_le32(parent_ino);
1375 fake.name_len = 2;
1376 strcpy(fake.name, "..");
1377 fake.rec_len = ext4_rec_len_to_disk(
1378 EXT4_DIR_REC_LEN(fake.name_len),
1379 inline_size);
1380 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1381 de = &fake;
1382 pos = EXT4_INLINE_DOTDOT_SIZE;
1383 } else {
1384 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1385 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1386 if (ext4_check_dir_entry(inode, dir_file, de,
1387 iloc.bh, dir_buf,
1388 inline_size, pos)) {
1389 ret = count;
1390 goto out;
1391 }
1392 }
1393
1394 ext4fs_dirhash(de->name, de->name_len, hinfo);
1395 if ((hinfo->hash < start_hash) ||
1396 ((hinfo->hash == start_hash) &&
1397 (hinfo->minor_hash < start_minor_hash)))
1398 continue;
1399 if (de->inode == 0)
1400 continue;
1401 tmp_str.name = de->name;
1402 tmp_str.len = de->name_len;
1403 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1404 hinfo->minor_hash, de, &tmp_str);
1405 if (err) {
1406 count = err;
1407 goto out;
1408 }
1409 count++;
1410 }
1411 ret = count;
1412out:
1413 kfree(dir_buf);
1414 brelse(iloc.bh);
1415 return ret;
1416}
1417
1418/*
1419 * So this function is called when the volume is mkfsed with
1420 * dir_index disabled. In order to keep f_pos persistent
1421 * after we convert from an inlined dir to a blocked based,
1422 * we just pretend that we are a normal dir and return the
1423 * offset as if '.' and '..' really take place.
1424 *
1425 */
1426int ext4_read_inline_dir(struct file *file,
1427 struct dir_context *ctx,
1428 int *has_inline_data)
1429{
1430 unsigned int offset, parent_ino;
1431 int i;
1432 struct ext4_dir_entry_2 *de;
1433 struct super_block *sb;
1434 struct inode *inode = file_inode(file);
1435 int ret, inline_size = 0;
1436 struct ext4_iloc iloc;
1437 void *dir_buf = NULL;
1438 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1439
1440 ret = ext4_get_inode_loc(inode, &iloc);
1441 if (ret)
1442 return ret;
1443
1444 down_read(&EXT4_I(inode)->xattr_sem);
1445 if (!ext4_has_inline_data(inode)) {
1446 up_read(&EXT4_I(inode)->xattr_sem);
1447 *has_inline_data = 0;
1448 goto out;
1449 }
1450
1451 inline_size = ext4_get_inline_size(inode);
1452 dir_buf = kmalloc(inline_size, GFP_NOFS);
1453 if (!dir_buf) {
1454 ret = -ENOMEM;
1455 up_read(&EXT4_I(inode)->xattr_sem);
1456 goto out;
1457 }
1458
1459 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1460 up_read(&EXT4_I(inode)->xattr_sem);
1461 if (ret < 0)
1462 goto out;
1463
1464 ret = 0;
1465 sb = inode->i_sb;
1466 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1467 offset = ctx->pos;
1468
1469 /*
1470 * dotdot_offset and dotdot_size is the real offset and
1471 * size for ".." and "." if the dir is block based while
1472 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1473 * So we will use extra_offset and extra_size to indicate them
1474 * during the inline dir iteration.
1475 */
1476 dotdot_offset = EXT4_DIR_REC_LEN(1);
1477 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1478 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1479 extra_size = extra_offset + inline_size;
1480
1481 /*
1482 * If the version has changed since the last call to
1483 * readdir(2), then we might be pointing to an invalid
1484 * dirent right now. Scan from the start of the inline
1485 * dir to make sure.
1486 */
1487 if (file->f_version != inode->i_version) {
1488 for (i = 0; i < extra_size && i < offset;) {
1489 /*
1490 * "." is with offset 0 and
1491 * ".." is dotdot_offset.
1492 */
1493 if (!i) {
1494 i = dotdot_offset;
1495 continue;
1496 } else if (i == dotdot_offset) {
1497 i = dotdot_size;
1498 continue;
1499 }
1500 /* for other entry, the real offset in
1501 * the buf has to be tuned accordingly.
1502 */
1503 de = (struct ext4_dir_entry_2 *)
1504 (dir_buf + i - extra_offset);
1505 /* It's too expensive to do a full
1506 * dirent test each time round this
1507 * loop, but we do have to test at
1508 * least that it is non-zero. A
1509 * failure will be detected in the
1510 * dirent test below. */
1511 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1512 < EXT4_DIR_REC_LEN(1))
1513 break;
1514 i += ext4_rec_len_from_disk(de->rec_len,
1515 extra_size);
1516 }
1517 offset = i;
1518 ctx->pos = offset;
1519 file->f_version = inode->i_version;
1520 }
1521
1522 while (ctx->pos < extra_size) {
1523 if (ctx->pos == 0) {
1524 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1525 goto out;
1526 ctx->pos = dotdot_offset;
1527 continue;
1528 }
1529
1530 if (ctx->pos == dotdot_offset) {
1531 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1532 goto out;
1533 ctx->pos = dotdot_size;
1534 continue;
1535 }
1536
1537 de = (struct ext4_dir_entry_2 *)
1538 (dir_buf + ctx->pos - extra_offset);
1539 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1540 extra_size, ctx->pos))
1541 goto out;
1542 if (le32_to_cpu(de->inode)) {
1543 if (!dir_emit(ctx, de->name, de->name_len,
1544 le32_to_cpu(de->inode),
1545 get_dtype(sb, de->file_type)))
1546 goto out;
1547 }
1548 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1549 }
1550out:
1551 kfree(dir_buf);
1552 brelse(iloc.bh);
1553 return ret;
1554}
1555
1556struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1557 struct ext4_dir_entry_2 **parent_de,
1558 int *retval)
1559{
1560 struct ext4_iloc iloc;
1561
1562 *retval = ext4_get_inode_loc(inode, &iloc);
1563 if (*retval)
1564 return NULL;
1565
1566 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1567
1568 return iloc.bh;
1569}
1570
1571/*
1572 * Try to create the inline data for the new dir.
1573 * If it succeeds, return 0, otherwise return the error.
1574 * In case of ENOSPC, the caller should create the normal disk layout dir.
1575 */
1576int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1577 struct inode *inode)
1578{
1579 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1580 struct ext4_iloc iloc;
1581 struct ext4_dir_entry_2 *de;
1582
1583 ret = ext4_get_inode_loc(inode, &iloc);
1584 if (ret)
1585 return ret;
1586
1587 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1588 if (ret)
1589 goto out;
1590
1591 /*
1592 * For inline dir, we only save the inode information for the ".."
1593 * and create a fake dentry to cover the left space.
1594 */
1595 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1596 de->inode = cpu_to_le32(parent->i_ino);
1597 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1598 de->inode = 0;
1599 de->rec_len = ext4_rec_len_to_disk(
1600 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1601 inline_size);
1602 set_nlink(inode, 2);
1603 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1604out:
1605 brelse(iloc.bh);
1606 return ret;
1607}
1608
1609struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1610 struct ext4_filename *fname,
1611 const struct qstr *d_name,
1612 struct ext4_dir_entry_2 **res_dir,
1613 int *has_inline_data)
1614{
1615 int ret;
1616 struct ext4_iloc iloc;
1617 void *inline_start;
1618 int inline_size;
1619
1620 if (ext4_get_inode_loc(dir, &iloc))
1621 return NULL;
1622
1623 down_read(&EXT4_I(dir)->xattr_sem);
1624 if (!ext4_has_inline_data(dir)) {
1625 *has_inline_data = 0;
1626 goto out;
1627 }
1628
1629 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1630 EXT4_INLINE_DOTDOT_SIZE;
1631 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1632 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1633 dir, fname, d_name, 0, res_dir);
1634 if (ret == 1)
1635 goto out_find;
1636 if (ret < 0)
1637 goto out;
1638
1639 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1640 goto out;
1641
1642 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1643 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1644
1645 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1646 dir, fname, d_name, 0, res_dir);
1647 if (ret == 1)
1648 goto out_find;
1649
1650out:
1651 brelse(iloc.bh);
1652 iloc.bh = NULL;
1653out_find:
1654 up_read(&EXT4_I(dir)->xattr_sem);
1655 return iloc.bh;
1656}
1657
1658int ext4_delete_inline_entry(handle_t *handle,
1659 struct inode *dir,
1660 struct ext4_dir_entry_2 *de_del,
1661 struct buffer_head *bh,
1662 int *has_inline_data)
1663{
1664 int err, inline_size;
1665 struct ext4_iloc iloc;
1666 void *inline_start;
1667
1668 err = ext4_get_inode_loc(dir, &iloc);
1669 if (err)
1670 return err;
1671
1672 down_write(&EXT4_I(dir)->xattr_sem);
1673 if (!ext4_has_inline_data(dir)) {
1674 *has_inline_data = 0;
1675 goto out;
1676 }
1677
1678 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1679 EXT4_MIN_INLINE_DATA_SIZE) {
1680 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1681 EXT4_INLINE_DOTDOT_SIZE;
1682 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1683 EXT4_INLINE_DOTDOT_SIZE;
1684 } else {
1685 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1686 inline_size = ext4_get_inline_size(dir) -
1687 EXT4_MIN_INLINE_DATA_SIZE;
1688 }
1689
1690 BUFFER_TRACE(bh, "get_write_access");
1691 err = ext4_journal_get_write_access(handle, bh);
1692 if (err)
1693 goto out;
1694
1695 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1696 inline_start, inline_size, 0);
1697 if (err)
1698 goto out;
1699
1700 err = ext4_mark_inode_dirty(handle, dir);
1701 if (unlikely(err))
1702 goto out;
1703
1704 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1705out:
1706 up_write(&EXT4_I(dir)->xattr_sem);
1707 brelse(iloc.bh);
1708 if (err != -ENOENT)
1709 ext4_std_error(dir->i_sb, err);
1710 return err;
1711}
1712
1713/*
1714 * Get the inline dentry at offset.
1715 */
1716static inline struct ext4_dir_entry_2 *
1717ext4_get_inline_entry(struct inode *inode,
1718 struct ext4_iloc *iloc,
1719 unsigned int offset,
1720 void **inline_start,
1721 int *inline_size)
1722{
1723 void *inline_pos;
1724
1725 BUG_ON(offset > ext4_get_inline_size(inode));
1726
1727 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1728 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1729 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1730 } else {
1731 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1732 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1733 *inline_size = ext4_get_inline_size(inode) -
1734 EXT4_MIN_INLINE_DATA_SIZE;
1735 }
1736
1737 if (inline_start)
1738 *inline_start = inline_pos;
1739 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1740}
1741
1742int empty_inline_dir(struct inode *dir, int *has_inline_data)
1743{
1744 int err, inline_size;
1745 struct ext4_iloc iloc;
1746 void *inline_pos;
1747 unsigned int offset;
1748 struct ext4_dir_entry_2 *de;
1749 int ret = 1;
1750
1751 err = ext4_get_inode_loc(dir, &iloc);
1752 if (err) {
1753 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1754 err, dir->i_ino);
1755 return 1;
1756 }
1757
1758 down_read(&EXT4_I(dir)->xattr_sem);
1759 if (!ext4_has_inline_data(dir)) {
1760 *has_inline_data = 0;
1761 goto out;
1762 }
1763
1764 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1765 if (!le32_to_cpu(de->inode)) {
1766 ext4_warning(dir->i_sb,
1767 "bad inline directory (dir #%lu) - no `..'",
1768 dir->i_ino);
1769 ret = 1;
1770 goto out;
1771 }
1772
1773 offset = EXT4_INLINE_DOTDOT_SIZE;
1774 while (offset < dir->i_size) {
1775 de = ext4_get_inline_entry(dir, &iloc, offset,
1776 &inline_pos, &inline_size);
1777 if (ext4_check_dir_entry(dir, NULL, de,
1778 iloc.bh, inline_pos,
1779 inline_size, offset)) {
1780 ext4_warning(dir->i_sb,
1781 "bad inline directory (dir #%lu) - "
1782 "inode %u, rec_len %u, name_len %d"
1783 "inline size %d\n",
1784 dir->i_ino, le32_to_cpu(de->inode),
1785 le16_to_cpu(de->rec_len), de->name_len,
1786 inline_size);
1787 ret = 1;
1788 goto out;
1789 }
1790 if (le32_to_cpu(de->inode)) {
1791 ret = 0;
1792 goto out;
1793 }
1794 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1795 }
1796
1797out:
1798 up_read(&EXT4_I(dir)->xattr_sem);
1799 brelse(iloc.bh);
1800 return ret;
1801}
1802
1803int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1804{
1805 int ret;
1806
1807 down_write(&EXT4_I(inode)->xattr_sem);
1808 ret = ext4_destroy_inline_data_nolock(handle, inode);
1809 up_write(&EXT4_I(inode)->xattr_sem);
1810
1811 return ret;
1812}
1813
1814int ext4_inline_data_fiemap(struct inode *inode,
1815 struct fiemap_extent_info *fieinfo,
1816 int *has_inline, __u64 start, __u64 len)
1817{
1818 __u64 physical = 0;
1819 __u64 inline_len;
1820 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1821 FIEMAP_EXTENT_LAST;
1822 int error = 0;
1823 struct ext4_iloc iloc;
1824
1825 down_read(&EXT4_I(inode)->xattr_sem);
1826 if (!ext4_has_inline_data(inode)) {
1827 *has_inline = 0;
1828 goto out;
1829 }
1830 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1831 i_size_read(inode));
1832 if (start >= inline_len)
1833 goto out;
1834 if (start + len < inline_len)
1835 inline_len = start + len;
1836 inline_len -= start;
1837
1838 error = ext4_get_inode_loc(inode, &iloc);
1839 if (error)
1840 goto out;
1841
1842 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1843 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1844 physical += offsetof(struct ext4_inode, i_block);
1845
1846 if (physical)
1847 error = fiemap_fill_next_extent(fieinfo, start, physical,
1848 inline_len, flags);
1849 brelse(iloc.bh);
1850out:
1851 up_read(&EXT4_I(inode)->xattr_sem);
1852 return (error < 0 ? error : 0);
1853}
1854
1855/*
1856 * Called during xattr set, and if we can sparse space 'needed',
1857 * just create the extent tree evict the data to the outer block.
1858 *
1859 * We use jbd2 instead of page cache to move data to the 1st block
1860 * so that the whole transaction can be committed as a whole and
1861 * the data isn't lost because of the delayed page cache write.
1862 */
1863int ext4_try_to_evict_inline_data(handle_t *handle,
1864 struct inode *inode,
1865 int needed)
1866{
1867 int error;
1868 struct ext4_xattr_entry *entry;
1869 struct ext4_inode *raw_inode;
1870 struct ext4_iloc iloc;
1871
1872 error = ext4_get_inode_loc(inode, &iloc);
1873 if (error)
1874 return error;
1875
1876 raw_inode = ext4_raw_inode(&iloc);
1877 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1878 EXT4_I(inode)->i_inline_off);
1879 if (EXT4_XATTR_LEN(entry->e_name_len) +
1880 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1881 error = -ENOSPC;
1882 goto out;
1883 }
1884
1885 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1886out:
1887 brelse(iloc.bh);
1888 return error;
1889}
1890
1891void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1892{
1893 handle_t *handle;
1894 int inline_size, value_len, needed_blocks;
1895 size_t i_size;
1896 void *value = NULL;
1897 struct ext4_xattr_ibody_find is = {
1898 .s = { .not_found = -ENODATA, },
1899 };
1900 struct ext4_xattr_info i = {
1901 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1902 .name = EXT4_XATTR_SYSTEM_DATA,
1903 };
1904
1905
1906 needed_blocks = ext4_writepage_trans_blocks(inode);
1907 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1908 if (IS_ERR(handle))
1909 return;
1910
1911 down_write(&EXT4_I(inode)->xattr_sem);
1912 if (!ext4_has_inline_data(inode)) {
1913 *has_inline = 0;
1914 ext4_journal_stop(handle);
1915 return;
1916 }
1917
1918 if (ext4_orphan_add(handle, inode))
1919 goto out;
1920
1921 if (ext4_get_inode_loc(inode, &is.iloc))
1922 goto out;
1923
1924 down_write(&EXT4_I(inode)->i_data_sem);
1925 i_size = inode->i_size;
1926 inline_size = ext4_get_inline_size(inode);
1927 EXT4_I(inode)->i_disksize = i_size;
1928
1929 if (i_size < inline_size) {
1930 /* Clear the content in the xattr space. */
1931 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1932 if (ext4_xattr_ibody_find(inode, &i, &is))
1933 goto out_error;
1934
1935 BUG_ON(is.s.not_found);
1936
1937 value_len = le32_to_cpu(is.s.here->e_value_size);
1938 value = kmalloc(value_len, GFP_NOFS);
1939 if (!value)
1940 goto out_error;
1941
1942 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1943 value, value_len))
1944 goto out_error;
1945
1946 i.value = value;
1947 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1948 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1949 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1950 goto out_error;
1951 }
1952
1953 /* Clear the content within i_blocks. */
1954 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1955 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1956 memset(p + i_size, 0,
1957 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1958 }
1959
1960 EXT4_I(inode)->i_inline_size = i_size <
1961 EXT4_MIN_INLINE_DATA_SIZE ?
1962 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1963 }
1964
1965out_error:
1966 up_write(&EXT4_I(inode)->i_data_sem);
1967out:
1968 brelse(is.iloc.bh);
1969 up_write(&EXT4_I(inode)->xattr_sem);
1970 kfree(value);
1971 if (inode->i_nlink)
1972 ext4_orphan_del(handle, inode);
1973
1974 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1975 ext4_mark_inode_dirty(handle, inode);
1976 if (IS_SYNC(inode))
1977 ext4_handle_sync(handle);
1978
1979 ext4_journal_stop(handle);
1980 return;
1981}
1982
1983int ext4_convert_inline_data(struct inode *inode)
1984{
1985 int error, needed_blocks;
1986 handle_t *handle;
1987 struct ext4_iloc iloc;
1988
1989 if (!ext4_has_inline_data(inode)) {
1990 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1991 return 0;
1992 }
1993
1994 needed_blocks = ext4_writepage_trans_blocks(inode);
1995
1996 iloc.bh = NULL;
1997 error = ext4_get_inode_loc(inode, &iloc);
1998 if (error)
1999 return error;
2000
2001 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2002 if (IS_ERR(handle)) {
2003 error = PTR_ERR(handle);
2004 goto out_free;
2005 }
2006
2007 down_write(&EXT4_I(inode)->xattr_sem);
2008 if (!ext4_has_inline_data(inode)) {
2009 up_write(&EXT4_I(inode)->xattr_sem);
2010 goto out;
2011 }
2012
2013 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2014 up_write(&EXT4_I(inode)->xattr_sem);
2015out:
2016 ext4_journal_stop(handle);
2017out_free:
2018 brelse(iloc.bh);
2019 return error;
2020}
1// SPDX-License-Identifier: LGPL-2.1
2/*
3 * Copyright (c) 2012 Taobao.
4 * Written by Tao Ma <boyu.mt@taobao.com>
5 */
6
7#include <linux/iomap.h>
8#include <linux/fiemap.h>
9#include <linux/namei.h>
10#include <linux/iversion.h>
11#include <linux/sched/mm.h>
12
13#include "ext4_jbd2.h"
14#include "ext4.h"
15#include "xattr.h"
16#include "truncate.h"
17
18#define EXT4_XATTR_SYSTEM_DATA "data"
19#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
20#define EXT4_INLINE_DOTDOT_OFFSET 2
21#define EXT4_INLINE_DOTDOT_SIZE 4
22
23static int ext4_get_inline_size(struct inode *inode)
24{
25 if (EXT4_I(inode)->i_inline_off)
26 return EXT4_I(inode)->i_inline_size;
27
28 return 0;
29}
30
31static int get_max_inline_xattr_value_size(struct inode *inode,
32 struct ext4_iloc *iloc)
33{
34 struct ext4_xattr_ibody_header *header;
35 struct ext4_xattr_entry *entry;
36 struct ext4_inode *raw_inode;
37 void *end;
38 int free, min_offs;
39
40 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
41 return 0;
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 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
62
63 /* Compute min_offs. */
64 while (!IS_LAST_ENTRY(entry)) {
65 void *next = EXT4_XATTR_NEXT(entry);
66
67 if (next >= end) {
68 EXT4_ERROR_INODE(inode,
69 "corrupt xattr in inline inode");
70 return 0;
71 }
72 if (!entry->e_value_inum && entry->e_value_size) {
73 size_t offs = le16_to_cpu(entry->e_value_offs);
74 if (offs < min_offs)
75 min_offs = offs;
76 }
77 entry = next;
78 }
79 free = min_offs -
80 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
81
82 if (EXT4_I(inode)->i_inline_off) {
83 entry = (struct ext4_xattr_entry *)
84 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
85
86 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
87 goto out;
88 }
89
90 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
91
92 if (free > EXT4_XATTR_ROUND)
93 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
94 else
95 free = 0;
96
97out:
98 return free;
99}
100
101/*
102 * Get the maximum size we now can store in an inode.
103 * If we can't find the space for a xattr entry, don't use the space
104 * of the extents since we have no space to indicate the inline data.
105 */
106int ext4_get_max_inline_size(struct inode *inode)
107{
108 int error, max_inline_size;
109 struct ext4_iloc iloc;
110
111 if (EXT4_I(inode)->i_extra_isize == 0)
112 return 0;
113
114 error = ext4_get_inode_loc(inode, &iloc);
115 if (error) {
116 ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
117 "can't get inode location %lu",
118 inode->i_ino);
119 return 0;
120 }
121
122 down_read(&EXT4_I(inode)->xattr_sem);
123 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
124 up_read(&EXT4_I(inode)->xattr_sem);
125
126 brelse(iloc.bh);
127
128 if (!max_inline_size)
129 return 0;
130
131 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
132}
133
134/*
135 * this function does not take xattr_sem, which is OK because it is
136 * currently only used in a code path coming form ext4_iget, before
137 * the new inode has been unlocked
138 */
139int ext4_find_inline_data_nolock(struct inode *inode)
140{
141 struct ext4_xattr_ibody_find is = {
142 .s = { .not_found = -ENODATA, },
143 };
144 struct ext4_xattr_info i = {
145 .name_index = EXT4_XATTR_INDEX_SYSTEM,
146 .name = EXT4_XATTR_SYSTEM_DATA,
147 };
148 int error;
149
150 if (EXT4_I(inode)->i_extra_isize == 0)
151 return 0;
152
153 error = ext4_get_inode_loc(inode, &is.iloc);
154 if (error)
155 return error;
156
157 error = ext4_xattr_ibody_find(inode, &i, &is);
158 if (error)
159 goto out;
160
161 if (!is.s.not_found) {
162 if (is.s.here->e_value_inum) {
163 EXT4_ERROR_INODE(inode, "inline data xattr refers "
164 "to an external xattr inode");
165 error = -EFSCORRUPTED;
166 goto out;
167 }
168 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
169 (void *)ext4_raw_inode(&is.iloc));
170 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
171 le32_to_cpu(is.s.here->e_value_size);
172 }
173out:
174 brelse(is.iloc.bh);
175 return error;
176}
177
178static int ext4_read_inline_data(struct inode *inode, void *buffer,
179 unsigned int len,
180 struct ext4_iloc *iloc)
181{
182 struct ext4_xattr_entry *entry;
183 struct ext4_xattr_ibody_header *header;
184 int cp_len = 0;
185 struct ext4_inode *raw_inode;
186
187 if (!len)
188 return 0;
189
190 BUG_ON(len > EXT4_I(inode)->i_inline_size);
191
192 cp_len = min_t(unsigned int, len, EXT4_MIN_INLINE_DATA_SIZE);
193
194 raw_inode = ext4_raw_inode(iloc);
195 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
196
197 len -= cp_len;
198 buffer += cp_len;
199
200 if (!len)
201 goto out;
202
203 header = IHDR(inode, raw_inode);
204 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
205 EXT4_I(inode)->i_inline_off);
206 len = min_t(unsigned int, len,
207 (unsigned int)le32_to_cpu(entry->e_value_size));
208
209 memcpy(buffer,
210 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
211 cp_len += len;
212
213out:
214 return cp_len;
215}
216
217/*
218 * write the buffer to the inline inode.
219 * If 'create' is set, we don't need to do the extra copy in the xattr
220 * value since it is already handled by ext4_xattr_ibody_set.
221 * That saves us one memcpy.
222 */
223static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
224 void *buffer, loff_t pos, unsigned int len)
225{
226 struct ext4_xattr_entry *entry;
227 struct ext4_xattr_ibody_header *header;
228 struct ext4_inode *raw_inode;
229 int cp_len = 0;
230
231 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
232 return;
233
234 BUG_ON(!EXT4_I(inode)->i_inline_off);
235 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
236
237 raw_inode = ext4_raw_inode(iloc);
238 buffer += pos;
239
240 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
241 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
242 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
243 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
244
245 len -= cp_len;
246 buffer += cp_len;
247 pos += cp_len;
248 }
249
250 if (!len)
251 return;
252
253 pos -= EXT4_MIN_INLINE_DATA_SIZE;
254 header = IHDR(inode, raw_inode);
255 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
256 EXT4_I(inode)->i_inline_off);
257
258 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
259 buffer, len);
260}
261
262static int ext4_create_inline_data(handle_t *handle,
263 struct inode *inode, unsigned len)
264{
265 int error;
266 void *value = NULL;
267 struct ext4_xattr_ibody_find is = {
268 .s = { .not_found = -ENODATA, },
269 };
270 struct ext4_xattr_info i = {
271 .name_index = EXT4_XATTR_INDEX_SYSTEM,
272 .name = EXT4_XATTR_SYSTEM_DATA,
273 };
274
275 error = ext4_get_inode_loc(inode, &is.iloc);
276 if (error)
277 return error;
278
279 BUFFER_TRACE(is.iloc.bh, "get_write_access");
280 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
281 EXT4_JTR_NONE);
282 if (error)
283 goto out;
284
285 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
286 value = EXT4_ZERO_XATTR_VALUE;
287 len -= EXT4_MIN_INLINE_DATA_SIZE;
288 } else {
289 value = "";
290 len = 0;
291 }
292
293 /* Insert the xttr entry. */
294 i.value = value;
295 i.value_len = len;
296
297 error = ext4_xattr_ibody_find(inode, &i, &is);
298 if (error)
299 goto out;
300
301 BUG_ON(!is.s.not_found);
302
303 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
304 if (error) {
305 if (error == -ENOSPC)
306 ext4_clear_inode_state(inode,
307 EXT4_STATE_MAY_INLINE_DATA);
308 goto out;
309 }
310
311 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
312 0, EXT4_MIN_INLINE_DATA_SIZE);
313
314 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
315 (void *)ext4_raw_inode(&is.iloc));
316 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
317 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
318 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
319 get_bh(is.iloc.bh);
320 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
321
322out:
323 brelse(is.iloc.bh);
324 return error;
325}
326
327static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
328 unsigned int len)
329{
330 int error;
331 void *value = NULL;
332 struct ext4_xattr_ibody_find is = {
333 .s = { .not_found = -ENODATA, },
334 };
335 struct ext4_xattr_info i = {
336 .name_index = EXT4_XATTR_INDEX_SYSTEM,
337 .name = EXT4_XATTR_SYSTEM_DATA,
338 };
339
340 /* If the old space is ok, write the data directly. */
341 if (len <= EXT4_I(inode)->i_inline_size)
342 return 0;
343
344 error = ext4_get_inode_loc(inode, &is.iloc);
345 if (error)
346 return error;
347
348 error = ext4_xattr_ibody_find(inode, &i, &is);
349 if (error)
350 goto out;
351
352 BUG_ON(is.s.not_found);
353
354 len -= EXT4_MIN_INLINE_DATA_SIZE;
355 value = kzalloc(len, GFP_NOFS);
356 if (!value) {
357 error = -ENOMEM;
358 goto out;
359 }
360
361 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
362 value, len);
363 if (error < 0)
364 goto out;
365
366 BUFFER_TRACE(is.iloc.bh, "get_write_access");
367 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
368 EXT4_JTR_NONE);
369 if (error)
370 goto out;
371
372 /* Update the xattr entry. */
373 i.value = value;
374 i.value_len = len;
375
376 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
377 if (error)
378 goto out;
379
380 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
381 (void *)ext4_raw_inode(&is.iloc));
382 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
383 le32_to_cpu(is.s.here->e_value_size);
384 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
385 get_bh(is.iloc.bh);
386 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
387
388out:
389 kfree(value);
390 brelse(is.iloc.bh);
391 return error;
392}
393
394static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
395 unsigned int len)
396{
397 int ret, size, no_expand;
398 struct ext4_inode_info *ei = EXT4_I(inode);
399
400 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
401 return -ENOSPC;
402
403 size = ext4_get_max_inline_size(inode);
404 if (size < len)
405 return -ENOSPC;
406
407 ext4_write_lock_xattr(inode, &no_expand);
408
409 if (ei->i_inline_off)
410 ret = ext4_update_inline_data(handle, inode, len);
411 else
412 ret = ext4_create_inline_data(handle, inode, len);
413
414 ext4_write_unlock_xattr(inode, &no_expand);
415 return ret;
416}
417
418static int ext4_destroy_inline_data_nolock(handle_t *handle,
419 struct inode *inode)
420{
421 struct ext4_inode_info *ei = EXT4_I(inode);
422 struct ext4_xattr_ibody_find is = {
423 .s = { .not_found = 0, },
424 };
425 struct ext4_xattr_info i = {
426 .name_index = EXT4_XATTR_INDEX_SYSTEM,
427 .name = EXT4_XATTR_SYSTEM_DATA,
428 .value = NULL,
429 .value_len = 0,
430 };
431 int error;
432
433 if (!ei->i_inline_off)
434 return 0;
435
436 error = ext4_get_inode_loc(inode, &is.iloc);
437 if (error)
438 return error;
439
440 error = ext4_xattr_ibody_find(inode, &i, &is);
441 if (error)
442 goto out;
443
444 BUFFER_TRACE(is.iloc.bh, "get_write_access");
445 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
446 EXT4_JTR_NONE);
447 if (error)
448 goto out;
449
450 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
451 if (error)
452 goto out;
453
454 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
455 0, EXT4_MIN_INLINE_DATA_SIZE);
456 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
457
458 if (ext4_has_feature_extents(inode->i_sb)) {
459 if (S_ISDIR(inode->i_mode) ||
460 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
461 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
462 ext4_ext_tree_init(handle, inode);
463 }
464 }
465 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
466
467 get_bh(is.iloc.bh);
468 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
469
470 EXT4_I(inode)->i_inline_off = 0;
471 EXT4_I(inode)->i_inline_size = 0;
472 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
473out:
474 brelse(is.iloc.bh);
475 if (error == -ENODATA)
476 error = 0;
477 return error;
478}
479
480static int ext4_read_inline_folio(struct inode *inode, struct folio *folio)
481{
482 void *kaddr;
483 int ret = 0;
484 size_t len;
485 struct ext4_iloc iloc;
486
487 BUG_ON(!folio_test_locked(folio));
488 BUG_ON(!ext4_has_inline_data(inode));
489 BUG_ON(folio->index);
490
491 if (!EXT4_I(inode)->i_inline_off) {
492 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
493 inode->i_ino);
494 goto out;
495 }
496
497 ret = ext4_get_inode_loc(inode, &iloc);
498 if (ret)
499 goto out;
500
501 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
502 BUG_ON(len > PAGE_SIZE);
503 kaddr = kmap_local_folio(folio, 0);
504 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
505 kaddr = folio_zero_tail(folio, len, kaddr + len);
506 kunmap_local(kaddr);
507 folio_mark_uptodate(folio);
508 brelse(iloc.bh);
509
510out:
511 return ret;
512}
513
514int ext4_readpage_inline(struct inode *inode, struct folio *folio)
515{
516 int ret = 0;
517
518 down_read(&EXT4_I(inode)->xattr_sem);
519 if (!ext4_has_inline_data(inode)) {
520 up_read(&EXT4_I(inode)->xattr_sem);
521 return -EAGAIN;
522 }
523
524 /*
525 * Current inline data can only exist in the 1st page,
526 * So for all the other pages, just set them uptodate.
527 */
528 if (!folio->index)
529 ret = ext4_read_inline_folio(inode, folio);
530 else if (!folio_test_uptodate(folio)) {
531 folio_zero_segment(folio, 0, folio_size(folio));
532 folio_mark_uptodate(folio);
533 }
534
535 up_read(&EXT4_I(inode)->xattr_sem);
536
537 folio_unlock(folio);
538 return ret >= 0 ? 0 : ret;
539}
540
541static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
542 struct inode *inode)
543{
544 int ret, needed_blocks, no_expand;
545 handle_t *handle = NULL;
546 int retries = 0, sem_held = 0;
547 struct folio *folio = NULL;
548 unsigned from, to;
549 struct ext4_iloc iloc;
550
551 if (!ext4_has_inline_data(inode)) {
552 /*
553 * clear the flag so that no new write
554 * will trap here again.
555 */
556 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
557 return 0;
558 }
559
560 needed_blocks = ext4_writepage_trans_blocks(inode);
561
562 ret = ext4_get_inode_loc(inode, &iloc);
563 if (ret)
564 return ret;
565
566retry:
567 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
568 if (IS_ERR(handle)) {
569 ret = PTR_ERR(handle);
570 handle = NULL;
571 goto out;
572 }
573
574 /* We cannot recurse into the filesystem as the transaction is already
575 * started */
576 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
577 mapping_gfp_mask(mapping));
578 if (IS_ERR(folio)) {
579 ret = PTR_ERR(folio);
580 goto out_nofolio;
581 }
582
583 ext4_write_lock_xattr(inode, &no_expand);
584 sem_held = 1;
585 /* If some one has already done this for us, just exit. */
586 if (!ext4_has_inline_data(inode)) {
587 ret = 0;
588 goto out;
589 }
590
591 from = 0;
592 to = ext4_get_inline_size(inode);
593 if (!folio_test_uptodate(folio)) {
594 ret = ext4_read_inline_folio(inode, folio);
595 if (ret < 0)
596 goto out;
597 }
598
599 ret = ext4_destroy_inline_data_nolock(handle, inode);
600 if (ret)
601 goto out;
602
603 if (ext4_should_dioread_nolock(inode)) {
604 ret = ext4_block_write_begin(handle, folio, from, to,
605 ext4_get_block_unwritten);
606 } else
607 ret = ext4_block_write_begin(handle, folio, from, to,
608 ext4_get_block);
609
610 if (!ret && ext4_should_journal_data(inode)) {
611 ret = ext4_walk_page_buffers(handle, inode,
612 folio_buffers(folio), from, to,
613 NULL, do_journal_get_write_access);
614 }
615
616 if (ret) {
617 folio_unlock(folio);
618 folio_put(folio);
619 folio = NULL;
620 ext4_orphan_add(handle, inode);
621 ext4_write_unlock_xattr(inode, &no_expand);
622 sem_held = 0;
623 ext4_journal_stop(handle);
624 handle = NULL;
625 ext4_truncate_failed_write(inode);
626 /*
627 * If truncate failed early the inode might
628 * still be on the orphan list; we need to
629 * make sure the inode is removed from the
630 * orphan list in that case.
631 */
632 if (inode->i_nlink)
633 ext4_orphan_del(NULL, inode);
634 }
635
636 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
637 goto retry;
638
639 if (folio)
640 block_commit_write(&folio->page, from, to);
641out:
642 if (folio) {
643 folio_unlock(folio);
644 folio_put(folio);
645 }
646out_nofolio:
647 if (sem_held)
648 ext4_write_unlock_xattr(inode, &no_expand);
649 if (handle)
650 ext4_journal_stop(handle);
651 brelse(iloc.bh);
652 return ret;
653}
654
655/*
656 * Try to write data in the inode.
657 * If the inode has inline data, check whether the new write can be
658 * in the inode also. If not, create the page the handle, move the data
659 * to the page make it update and let the later codes create extent for it.
660 */
661int ext4_try_to_write_inline_data(struct address_space *mapping,
662 struct inode *inode,
663 loff_t pos, unsigned len,
664 struct folio **foliop)
665{
666 int ret;
667 handle_t *handle;
668 struct folio *folio;
669 struct ext4_iloc iloc;
670
671 if (pos + len > ext4_get_max_inline_size(inode))
672 goto convert;
673
674 ret = ext4_get_inode_loc(inode, &iloc);
675 if (ret)
676 return ret;
677
678 /*
679 * The possible write could happen in the inode,
680 * so try to reserve the space in inode first.
681 */
682 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
683 if (IS_ERR(handle)) {
684 ret = PTR_ERR(handle);
685 handle = NULL;
686 goto out;
687 }
688
689 ret = ext4_prepare_inline_data(handle, inode, pos + len);
690 if (ret && ret != -ENOSPC)
691 goto out;
692
693 /* We don't have space in inline inode, so convert it to extent. */
694 if (ret == -ENOSPC) {
695 ext4_journal_stop(handle);
696 brelse(iloc.bh);
697 goto convert;
698 }
699
700 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
701 EXT4_JTR_NONE);
702 if (ret)
703 goto out;
704
705 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
706 mapping_gfp_mask(mapping));
707 if (IS_ERR(folio)) {
708 ret = PTR_ERR(folio);
709 goto out;
710 }
711
712 *foliop = folio;
713 down_read(&EXT4_I(inode)->xattr_sem);
714 if (!ext4_has_inline_data(inode)) {
715 ret = 0;
716 folio_unlock(folio);
717 folio_put(folio);
718 goto out_up_read;
719 }
720
721 if (!folio_test_uptodate(folio)) {
722 ret = ext4_read_inline_folio(inode, folio);
723 if (ret < 0) {
724 folio_unlock(folio);
725 folio_put(folio);
726 goto out_up_read;
727 }
728 }
729
730 ret = 1;
731 handle = NULL;
732out_up_read:
733 up_read(&EXT4_I(inode)->xattr_sem);
734out:
735 if (handle && (ret != 1))
736 ext4_journal_stop(handle);
737 brelse(iloc.bh);
738 return ret;
739convert:
740 return ext4_convert_inline_data_to_extent(mapping, inode);
741}
742
743int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
744 unsigned copied, struct folio *folio)
745{
746 handle_t *handle = ext4_journal_current_handle();
747 int no_expand;
748 void *kaddr;
749 struct ext4_iloc iloc;
750 int ret = 0, ret2;
751
752 if (unlikely(copied < len) && !folio_test_uptodate(folio))
753 copied = 0;
754
755 if (likely(copied)) {
756 ret = ext4_get_inode_loc(inode, &iloc);
757 if (ret) {
758 folio_unlock(folio);
759 folio_put(folio);
760 ext4_std_error(inode->i_sb, ret);
761 goto out;
762 }
763 ext4_write_lock_xattr(inode, &no_expand);
764 BUG_ON(!ext4_has_inline_data(inode));
765
766 /*
767 * ei->i_inline_off may have changed since
768 * ext4_write_begin() called
769 * ext4_try_to_write_inline_data()
770 */
771 (void) ext4_find_inline_data_nolock(inode);
772
773 kaddr = kmap_local_folio(folio, 0);
774 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
775 kunmap_local(kaddr);
776 folio_mark_uptodate(folio);
777 /* clear dirty flag so that writepages wouldn't work for us. */
778 folio_clear_dirty(folio);
779
780 ext4_write_unlock_xattr(inode, &no_expand);
781 brelse(iloc.bh);
782
783 /*
784 * It's important to update i_size while still holding folio
785 * lock: page writeout could otherwise come in and zero
786 * beyond i_size.
787 */
788 ext4_update_inode_size(inode, pos + copied);
789 }
790 folio_unlock(folio);
791 folio_put(folio);
792
793 /*
794 * Don't mark the inode dirty under folio lock. First, it unnecessarily
795 * makes the holding time of folio lock longer. Second, it forces lock
796 * ordering of folio lock and transaction start for journaling
797 * filesystems.
798 */
799 if (likely(copied))
800 mark_inode_dirty(inode);
801out:
802 /*
803 * If we didn't copy as much data as expected, we need to trim back
804 * size of xattr containing inline data.
805 */
806 if (pos + len > inode->i_size && ext4_can_truncate(inode))
807 ext4_orphan_add(handle, inode);
808
809 ret2 = ext4_journal_stop(handle);
810 if (!ret)
811 ret = ret2;
812 if (pos + len > inode->i_size) {
813 ext4_truncate_failed_write(inode);
814 /*
815 * If truncate failed early the inode might still be
816 * on the orphan list; we need to make sure the inode
817 * is removed from the orphan list in that case.
818 */
819 if (inode->i_nlink)
820 ext4_orphan_del(NULL, inode);
821 }
822 return ret ? ret : copied;
823}
824
825/*
826 * Try to make the page cache and handle ready for the inline data case.
827 * We can call this function in 2 cases:
828 * 1. The inode is created and the first write exceeds inline size. We can
829 * clear the inode state safely.
830 * 2. The inode has inline data, then we need to read the data, make it
831 * update and dirty so that ext4_da_writepages can handle it. We don't
832 * need to start the journal since the file's metadata isn't changed now.
833 */
834static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
835 struct inode *inode,
836 void **fsdata)
837{
838 int ret = 0, inline_size;
839 struct folio *folio;
840
841 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN,
842 mapping_gfp_mask(mapping));
843 if (IS_ERR(folio))
844 return PTR_ERR(folio);
845
846 down_read(&EXT4_I(inode)->xattr_sem);
847 if (!ext4_has_inline_data(inode)) {
848 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
849 goto out;
850 }
851
852 inline_size = ext4_get_inline_size(inode);
853
854 if (!folio_test_uptodate(folio)) {
855 ret = ext4_read_inline_folio(inode, folio);
856 if (ret < 0)
857 goto out;
858 }
859
860 ret = ext4_block_write_begin(NULL, folio, 0, inline_size,
861 ext4_da_get_block_prep);
862 if (ret) {
863 up_read(&EXT4_I(inode)->xattr_sem);
864 folio_unlock(folio);
865 folio_put(folio);
866 ext4_truncate_failed_write(inode);
867 return ret;
868 }
869
870 folio_mark_dirty(folio);
871 folio_mark_uptodate(folio);
872 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
873 *fsdata = (void *)CONVERT_INLINE_DATA;
874
875out:
876 up_read(&EXT4_I(inode)->xattr_sem);
877 if (folio) {
878 folio_unlock(folio);
879 folio_put(folio);
880 }
881 return ret;
882}
883
884/*
885 * Prepare the write for the inline data.
886 * If the data can be written into the inode, we just read
887 * the page and make it uptodate, and start the journal.
888 * Otherwise read the page, makes it dirty so that it can be
889 * handle in writepages(the i_disksize update is left to the
890 * normal ext4_da_write_end).
891 */
892int ext4_da_write_inline_data_begin(struct address_space *mapping,
893 struct inode *inode,
894 loff_t pos, unsigned len,
895 struct folio **foliop,
896 void **fsdata)
897{
898 int ret;
899 handle_t *handle;
900 struct folio *folio;
901 struct ext4_iloc iloc;
902 int retries = 0;
903
904 ret = ext4_get_inode_loc(inode, &iloc);
905 if (ret)
906 return ret;
907
908retry_journal:
909 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
910 if (IS_ERR(handle)) {
911 ret = PTR_ERR(handle);
912 goto out;
913 }
914
915 ret = ext4_prepare_inline_data(handle, inode, pos + len);
916 if (ret && ret != -ENOSPC)
917 goto out_journal;
918
919 if (ret == -ENOSPC) {
920 ext4_journal_stop(handle);
921 ret = ext4_da_convert_inline_data_to_extent(mapping,
922 inode,
923 fsdata);
924 if (ret == -ENOSPC &&
925 ext4_should_retry_alloc(inode->i_sb, &retries))
926 goto retry_journal;
927 goto out;
928 }
929
930 /*
931 * We cannot recurse into the filesystem as the transaction
932 * is already started.
933 */
934 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
935 mapping_gfp_mask(mapping));
936 if (IS_ERR(folio)) {
937 ret = PTR_ERR(folio);
938 goto out_journal;
939 }
940
941 down_read(&EXT4_I(inode)->xattr_sem);
942 if (!ext4_has_inline_data(inode)) {
943 ret = 0;
944 goto out_release_page;
945 }
946
947 if (!folio_test_uptodate(folio)) {
948 ret = ext4_read_inline_folio(inode, folio);
949 if (ret < 0)
950 goto out_release_page;
951 }
952 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
953 EXT4_JTR_NONE);
954 if (ret)
955 goto out_release_page;
956
957 up_read(&EXT4_I(inode)->xattr_sem);
958 *foliop = folio;
959 brelse(iloc.bh);
960 return 1;
961out_release_page:
962 up_read(&EXT4_I(inode)->xattr_sem);
963 folio_unlock(folio);
964 folio_put(folio);
965out_journal:
966 ext4_journal_stop(handle);
967out:
968 brelse(iloc.bh);
969 return ret;
970}
971
972#ifdef INLINE_DIR_DEBUG
973void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
974 void *inline_start, int inline_size)
975{
976 int offset;
977 unsigned short de_len;
978 struct ext4_dir_entry_2 *de = inline_start;
979 void *dlimit = inline_start + inline_size;
980
981 trace_printk("inode %lu\n", dir->i_ino);
982 offset = 0;
983 while ((void *)de < dlimit) {
984 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
985 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
986 offset, de_len, de->name_len, de->name,
987 de->name_len, le32_to_cpu(de->inode));
988 if (ext4_check_dir_entry(dir, NULL, de, bh,
989 inline_start, inline_size, offset))
990 BUG();
991
992 offset += de_len;
993 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
994 }
995}
996#else
997#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
998#endif
999
1000/*
1001 * Add a new entry into a inline dir.
1002 * It will return -ENOSPC if no space is available, and -EIO
1003 * and -EEXIST if directory entry already exists.
1004 */
1005static int ext4_add_dirent_to_inline(handle_t *handle,
1006 struct ext4_filename *fname,
1007 struct inode *dir,
1008 struct inode *inode,
1009 struct ext4_iloc *iloc,
1010 void *inline_start, int inline_size)
1011{
1012 int err;
1013 struct ext4_dir_entry_2 *de;
1014
1015 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1016 inline_size, fname, &de);
1017 if (err)
1018 return err;
1019
1020 BUFFER_TRACE(iloc->bh, "get_write_access");
1021 err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1022 EXT4_JTR_NONE);
1023 if (err)
1024 return err;
1025 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1026
1027 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1028
1029 /*
1030 * XXX shouldn't update any times until successful
1031 * completion of syscall, but too many callers depend
1032 * on this.
1033 *
1034 * XXX similarly, too many callers depend on
1035 * ext4_new_inode() setting the times, but error
1036 * recovery deletes the inode, so the worst that can
1037 * happen is that the times are slightly out of date
1038 * and/or different from the directory change time.
1039 */
1040 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
1041 ext4_update_dx_flag(dir);
1042 inode_inc_iversion(dir);
1043 return 1;
1044}
1045
1046static void *ext4_get_inline_xattr_pos(struct inode *inode,
1047 struct ext4_iloc *iloc)
1048{
1049 struct ext4_xattr_entry *entry;
1050 struct ext4_xattr_ibody_header *header;
1051
1052 BUG_ON(!EXT4_I(inode)->i_inline_off);
1053
1054 header = IHDR(inode, ext4_raw_inode(iloc));
1055 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1056 EXT4_I(inode)->i_inline_off);
1057
1058 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1059}
1060
1061/* Set the final de to cover the whole block. */
1062static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1063{
1064 struct ext4_dir_entry_2 *de, *prev_de;
1065 void *limit;
1066 int de_len;
1067
1068 de = de_buf;
1069 if (old_size) {
1070 limit = de_buf + old_size;
1071 do {
1072 prev_de = de;
1073 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1074 de_buf += de_len;
1075 de = de_buf;
1076 } while (de_buf < limit);
1077
1078 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1079 old_size, new_size);
1080 } else {
1081 /* this is just created, so create an empty entry. */
1082 de->inode = 0;
1083 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1084 }
1085}
1086
1087static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1088 struct ext4_iloc *iloc)
1089{
1090 int ret;
1091 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1092 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1093
1094 if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1095 return -ENOSPC;
1096
1097 ret = ext4_update_inline_data(handle, dir,
1098 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1099 if (ret)
1100 return ret;
1101
1102 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1103 EXT4_I(dir)->i_inline_size -
1104 EXT4_MIN_INLINE_DATA_SIZE);
1105 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1106 return 0;
1107}
1108
1109static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1110 struct ext4_iloc *iloc,
1111 void *buf, int inline_size)
1112{
1113 int ret;
1114
1115 ret = ext4_create_inline_data(handle, inode, inline_size);
1116 if (ret) {
1117 ext4_msg(inode->i_sb, KERN_EMERG,
1118 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1119 inode->i_ino, ret);
1120 return;
1121 }
1122 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1123 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1124}
1125
1126static int ext4_finish_convert_inline_dir(handle_t *handle,
1127 struct inode *inode,
1128 struct buffer_head *dir_block,
1129 void *buf,
1130 int inline_size)
1131{
1132 int err, csum_size = 0, header_size = 0;
1133 struct ext4_dir_entry_2 *de;
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 = 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 ext4_initialize_dirent_tail(dir_block,
1161 inode->i_sb->s_blocksize);
1162 set_buffer_uptodate(dir_block);
1163 unlock_buffer(dir_block);
1164 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1165 if (err)
1166 return err;
1167 set_buffer_verified(dir_block);
1168 return ext4_mark_inode_dirty(handle, inode);
1169}
1170
1171static int ext4_convert_inline_data_nolock(handle_t *handle,
1172 struct inode *inode,
1173 struct ext4_iloc *iloc)
1174{
1175 int error;
1176 void *buf = NULL;
1177 struct buffer_head *data_bh = NULL;
1178 struct ext4_map_blocks map;
1179 int inline_size;
1180
1181 inline_size = ext4_get_inline_size(inode);
1182 buf = kmalloc(inline_size, GFP_NOFS);
1183 if (!buf) {
1184 error = -ENOMEM;
1185 goto out;
1186 }
1187
1188 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1189 if (error < 0)
1190 goto out;
1191
1192 /*
1193 * Make sure the inline directory entries pass checks before we try to
1194 * convert them, so that we avoid touching stuff that needs fsck.
1195 */
1196 if (S_ISDIR(inode->i_mode)) {
1197 error = ext4_check_all_de(inode, iloc->bh,
1198 buf + EXT4_INLINE_DOTDOT_SIZE,
1199 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1200 if (error)
1201 goto out;
1202 }
1203
1204 error = ext4_destroy_inline_data_nolock(handle, inode);
1205 if (error)
1206 goto out;
1207
1208 map.m_lblk = 0;
1209 map.m_len = 1;
1210 map.m_flags = 0;
1211 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1212 if (error < 0)
1213 goto out_restore;
1214 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1215 error = -EIO;
1216 goto out_restore;
1217 }
1218
1219 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1220 if (!data_bh) {
1221 error = -ENOMEM;
1222 goto out_restore;
1223 }
1224
1225 lock_buffer(data_bh);
1226 error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1227 EXT4_JTR_NONE);
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 unlock_buffer(data_bh);
1239 error = ext4_handle_dirty_metadata(handle,
1240 inode, data_bh);
1241 } else {
1242 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1243 buf, inline_size);
1244 }
1245
1246out_restore:
1247 if (error)
1248 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1249
1250out:
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 */
1261int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1262 struct inode *dir, struct inode *inode)
1263{
1264 int ret, ret2, inline_size, no_expand;
1265 void *inline_start;
1266 struct ext4_iloc iloc;
1267
1268 ret = ext4_get_inode_loc(dir, &iloc);
1269 if (ret)
1270 return ret;
1271
1272 ext4_write_lock_xattr(dir, &no_expand);
1273 if (!ext4_has_inline_data(dir))
1274 goto out;
1275
1276 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1277 EXT4_INLINE_DOTDOT_SIZE;
1278 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1279
1280 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1281 inline_start, inline_size);
1282 if (ret != -ENOSPC)
1283 goto out;
1284
1285 /* check whether it can be inserted to inline xattr space. */
1286 inline_size = EXT4_I(dir)->i_inline_size -
1287 EXT4_MIN_INLINE_DATA_SIZE;
1288 if (!inline_size) {
1289 /* Try to use the xattr space.*/
1290 ret = ext4_update_inline_dir(handle, dir, &iloc);
1291 if (ret && ret != -ENOSPC)
1292 goto out;
1293
1294 inline_size = EXT4_I(dir)->i_inline_size -
1295 EXT4_MIN_INLINE_DATA_SIZE;
1296 }
1297
1298 if (inline_size) {
1299 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1300
1301 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1302 inode, &iloc, inline_start,
1303 inline_size);
1304
1305 if (ret != -ENOSPC)
1306 goto out;
1307 }
1308
1309 /*
1310 * The inline space is filled up, so create a new block for it.
1311 * As the extent tree will be created, we have to save the inline
1312 * dir first.
1313 */
1314 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1315
1316out:
1317 ext4_write_unlock_xattr(dir, &no_expand);
1318 ret2 = ext4_mark_inode_dirty(handle, dir);
1319 if (unlikely(ret2 && !ret))
1320 ret = ret2;
1321 brelse(iloc.bh);
1322 return ret;
1323}
1324
1325/*
1326 * This function fills a red-black tree with information from an
1327 * inlined dir. It returns the number directory entries loaded
1328 * into the tree. If there is an error it is returned in err.
1329 */
1330int ext4_inlinedir_to_tree(struct file *dir_file,
1331 struct inode *dir, ext4_lblk_t block,
1332 struct dx_hash_info *hinfo,
1333 __u32 start_hash, __u32 start_minor_hash,
1334 int *has_inline_data)
1335{
1336 int err = 0, count = 0;
1337 unsigned int parent_ino;
1338 int pos;
1339 struct ext4_dir_entry_2 *de;
1340 struct inode *inode = file_inode(dir_file);
1341 int ret, inline_size = 0;
1342 struct ext4_iloc iloc;
1343 void *dir_buf = NULL;
1344 struct ext4_dir_entry_2 fake;
1345 struct fscrypt_str tmp_str;
1346
1347 ret = ext4_get_inode_loc(inode, &iloc);
1348 if (ret)
1349 return ret;
1350
1351 down_read(&EXT4_I(inode)->xattr_sem);
1352 if (!ext4_has_inline_data(inode)) {
1353 up_read(&EXT4_I(inode)->xattr_sem);
1354 *has_inline_data = 0;
1355 goto out;
1356 }
1357
1358 inline_size = ext4_get_inline_size(inode);
1359 dir_buf = kmalloc(inline_size, GFP_NOFS);
1360 if (!dir_buf) {
1361 ret = -ENOMEM;
1362 up_read(&EXT4_I(inode)->xattr_sem);
1363 goto out;
1364 }
1365
1366 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1367 up_read(&EXT4_I(inode)->xattr_sem);
1368 if (ret < 0)
1369 goto out;
1370
1371 pos = 0;
1372 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1373 while (pos < inline_size) {
1374 /*
1375 * As inlined dir doesn't store any information about '.' and
1376 * only the inode number of '..' is stored, we have to handle
1377 * them differently.
1378 */
1379 if (pos == 0) {
1380 fake.inode = cpu_to_le32(inode->i_ino);
1381 fake.name_len = 1;
1382 strcpy(fake.name, ".");
1383 fake.rec_len = ext4_rec_len_to_disk(
1384 ext4_dir_rec_len(fake.name_len, NULL),
1385 inline_size);
1386 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1387 de = &fake;
1388 pos = EXT4_INLINE_DOTDOT_OFFSET;
1389 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1390 fake.inode = cpu_to_le32(parent_ino);
1391 fake.name_len = 2;
1392 strcpy(fake.name, "..");
1393 fake.rec_len = ext4_rec_len_to_disk(
1394 ext4_dir_rec_len(fake.name_len, NULL),
1395 inline_size);
1396 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1397 de = &fake;
1398 pos = EXT4_INLINE_DOTDOT_SIZE;
1399 } else {
1400 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1401 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1402 if (ext4_check_dir_entry(inode, dir_file, de,
1403 iloc.bh, dir_buf,
1404 inline_size, pos)) {
1405 ret = count;
1406 goto out;
1407 }
1408 }
1409
1410 if (ext4_hash_in_dirent(dir)) {
1411 hinfo->hash = EXT4_DIRENT_HASH(de);
1412 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1413 } else {
1414 err = ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1415 if (err) {
1416 ret = err;
1417 goto out;
1418 }
1419 }
1420 if ((hinfo->hash < start_hash) ||
1421 ((hinfo->hash == start_hash) &&
1422 (hinfo->minor_hash < start_minor_hash)))
1423 continue;
1424 if (de->inode == 0)
1425 continue;
1426 tmp_str.name = de->name;
1427 tmp_str.len = de->name_len;
1428 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1429 hinfo->minor_hash, de, &tmp_str);
1430 if (err) {
1431 ret = err;
1432 goto out;
1433 }
1434 count++;
1435 }
1436 ret = count;
1437out:
1438 kfree(dir_buf);
1439 brelse(iloc.bh);
1440 return ret;
1441}
1442
1443/*
1444 * So this function is called when the volume is mkfsed with
1445 * dir_index disabled. In order to keep f_pos persistent
1446 * after we convert from an inlined dir to a blocked based,
1447 * we just pretend that we are a normal dir and return the
1448 * offset as if '.' and '..' really take place.
1449 *
1450 */
1451int ext4_read_inline_dir(struct file *file,
1452 struct dir_context *ctx,
1453 int *has_inline_data)
1454{
1455 unsigned int offset, parent_ino;
1456 int i;
1457 struct ext4_dir_entry_2 *de;
1458 struct super_block *sb;
1459 struct inode *inode = file_inode(file);
1460 int ret, inline_size = 0;
1461 struct ext4_iloc iloc;
1462 void *dir_buf = NULL;
1463 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1464 struct dir_private_info *info = file->private_data;
1465
1466 ret = ext4_get_inode_loc(inode, &iloc);
1467 if (ret)
1468 return ret;
1469
1470 down_read(&EXT4_I(inode)->xattr_sem);
1471 if (!ext4_has_inline_data(inode)) {
1472 up_read(&EXT4_I(inode)->xattr_sem);
1473 *has_inline_data = 0;
1474 goto out;
1475 }
1476
1477 inline_size = ext4_get_inline_size(inode);
1478 dir_buf = kmalloc(inline_size, GFP_NOFS);
1479 if (!dir_buf) {
1480 ret = -ENOMEM;
1481 up_read(&EXT4_I(inode)->xattr_sem);
1482 goto out;
1483 }
1484
1485 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1486 up_read(&EXT4_I(inode)->xattr_sem);
1487 if (ret < 0)
1488 goto out;
1489
1490 ret = 0;
1491 sb = inode->i_sb;
1492 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1493 offset = ctx->pos;
1494
1495 /*
1496 * dotdot_offset and dotdot_size is the real offset and
1497 * size for ".." and "." if the dir is block based while
1498 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1499 * So we will use extra_offset and extra_size to indicate them
1500 * during the inline dir iteration.
1501 */
1502 dotdot_offset = ext4_dir_rec_len(1, NULL);
1503 dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1504 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1505 extra_size = extra_offset + inline_size;
1506
1507 /*
1508 * If the cookie has changed since the last call to
1509 * readdir(2), then we might be pointing to an invalid
1510 * dirent right now. Scan from the start of the inline
1511 * dir to make sure.
1512 */
1513 if (!inode_eq_iversion(inode, info->cookie)) {
1514 for (i = 0; i < extra_size && i < offset;) {
1515 /*
1516 * "." is with offset 0 and
1517 * ".." is dotdot_offset.
1518 */
1519 if (!i) {
1520 i = dotdot_offset;
1521 continue;
1522 } else if (i == dotdot_offset) {
1523 i = dotdot_size;
1524 continue;
1525 }
1526 /* for other entry, the real offset in
1527 * the buf has to be tuned accordingly.
1528 */
1529 de = (struct ext4_dir_entry_2 *)
1530 (dir_buf + i - extra_offset);
1531 /* It's too expensive to do a full
1532 * dirent test each time round this
1533 * loop, but we do have to test at
1534 * least that it is non-zero. A
1535 * failure will be detected in the
1536 * dirent test below. */
1537 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1538 < ext4_dir_rec_len(1, NULL))
1539 break;
1540 i += ext4_rec_len_from_disk(de->rec_len,
1541 extra_size);
1542 }
1543 offset = i;
1544 ctx->pos = offset;
1545 info->cookie = inode_query_iversion(inode);
1546 }
1547
1548 while (ctx->pos < extra_size) {
1549 if (ctx->pos == 0) {
1550 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1551 goto out;
1552 ctx->pos = dotdot_offset;
1553 continue;
1554 }
1555
1556 if (ctx->pos == dotdot_offset) {
1557 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1558 goto out;
1559 ctx->pos = dotdot_size;
1560 continue;
1561 }
1562
1563 de = (struct ext4_dir_entry_2 *)
1564 (dir_buf + ctx->pos - extra_offset);
1565 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1566 extra_size, ctx->pos))
1567 goto out;
1568 if (le32_to_cpu(de->inode)) {
1569 if (!dir_emit(ctx, de->name, de->name_len,
1570 le32_to_cpu(de->inode),
1571 get_dtype(sb, de->file_type)))
1572 goto out;
1573 }
1574 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1575 }
1576out:
1577 kfree(dir_buf);
1578 brelse(iloc.bh);
1579 return ret;
1580}
1581
1582void *ext4_read_inline_link(struct inode *inode)
1583{
1584 struct ext4_iloc iloc;
1585 int ret, inline_size;
1586 void *link;
1587
1588 ret = ext4_get_inode_loc(inode, &iloc);
1589 if (ret)
1590 return ERR_PTR(ret);
1591
1592 ret = -ENOMEM;
1593 inline_size = ext4_get_inline_size(inode);
1594 link = kmalloc(inline_size + 1, GFP_NOFS);
1595 if (!link)
1596 goto out;
1597
1598 ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1599 if (ret < 0) {
1600 kfree(link);
1601 goto out;
1602 }
1603 nd_terminate_link(link, inode->i_size, ret);
1604out:
1605 if (ret < 0)
1606 link = ERR_PTR(ret);
1607 brelse(iloc.bh);
1608 return link;
1609}
1610
1611struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1612 struct ext4_dir_entry_2 **parent_de,
1613 int *retval)
1614{
1615 struct ext4_iloc iloc;
1616
1617 *retval = ext4_get_inode_loc(inode, &iloc);
1618 if (*retval)
1619 return NULL;
1620
1621 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1622
1623 return iloc.bh;
1624}
1625
1626/*
1627 * Try to create the inline data for the new dir.
1628 * If it succeeds, return 0, otherwise return the error.
1629 * In case of ENOSPC, the caller should create the normal disk layout dir.
1630 */
1631int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1632 struct inode *inode)
1633{
1634 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1635 struct ext4_iloc iloc;
1636 struct ext4_dir_entry_2 *de;
1637
1638 ret = ext4_get_inode_loc(inode, &iloc);
1639 if (ret)
1640 return ret;
1641
1642 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1643 if (ret)
1644 goto out;
1645
1646 /*
1647 * For inline dir, we only save the inode information for the ".."
1648 * and create a fake dentry to cover the left space.
1649 */
1650 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1651 de->inode = cpu_to_le32(parent->i_ino);
1652 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1653 de->inode = 0;
1654 de->rec_len = ext4_rec_len_to_disk(
1655 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1656 inline_size);
1657 set_nlink(inode, 2);
1658 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1659out:
1660 brelse(iloc.bh);
1661 return ret;
1662}
1663
1664struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1665 struct ext4_filename *fname,
1666 struct ext4_dir_entry_2 **res_dir,
1667 int *has_inline_data)
1668{
1669 struct ext4_xattr_ibody_find is = {
1670 .s = { .not_found = -ENODATA, },
1671 };
1672 struct ext4_xattr_info i = {
1673 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1674 .name = EXT4_XATTR_SYSTEM_DATA,
1675 };
1676 int ret;
1677 void *inline_start;
1678 int inline_size;
1679
1680 ret = ext4_get_inode_loc(dir, &is.iloc);
1681 if (ret)
1682 return ERR_PTR(ret);
1683
1684 down_read(&EXT4_I(dir)->xattr_sem);
1685
1686 ret = ext4_xattr_ibody_find(dir, &i, &is);
1687 if (ret)
1688 goto out;
1689
1690 if (!ext4_has_inline_data(dir)) {
1691 *has_inline_data = 0;
1692 goto out;
1693 }
1694
1695 inline_start = (void *)ext4_raw_inode(&is.iloc)->i_block +
1696 EXT4_INLINE_DOTDOT_SIZE;
1697 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1698 ret = ext4_search_dir(is.iloc.bh, inline_start, inline_size,
1699 dir, fname, 0, res_dir);
1700 if (ret == 1)
1701 goto out_find;
1702 if (ret < 0)
1703 goto out;
1704
1705 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1706 goto out;
1707
1708 inline_start = ext4_get_inline_xattr_pos(dir, &is.iloc);
1709 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1710
1711 ret = ext4_search_dir(is.iloc.bh, inline_start, inline_size,
1712 dir, fname, 0, res_dir);
1713 if (ret == 1)
1714 goto out_find;
1715
1716out:
1717 brelse(is.iloc.bh);
1718 if (ret < 0)
1719 is.iloc.bh = ERR_PTR(ret);
1720 else
1721 is.iloc.bh = NULL;
1722out_find:
1723 up_read(&EXT4_I(dir)->xattr_sem);
1724 return is.iloc.bh;
1725}
1726
1727int ext4_delete_inline_entry(handle_t *handle,
1728 struct inode *dir,
1729 struct ext4_dir_entry_2 *de_del,
1730 struct buffer_head *bh,
1731 int *has_inline_data)
1732{
1733 int err, inline_size, no_expand;
1734 struct ext4_iloc iloc;
1735 void *inline_start;
1736
1737 err = ext4_get_inode_loc(dir, &iloc);
1738 if (err)
1739 return err;
1740
1741 ext4_write_lock_xattr(dir, &no_expand);
1742 if (!ext4_has_inline_data(dir)) {
1743 *has_inline_data = 0;
1744 goto out;
1745 }
1746
1747 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1748 EXT4_MIN_INLINE_DATA_SIZE) {
1749 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1750 EXT4_INLINE_DOTDOT_SIZE;
1751 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1752 EXT4_INLINE_DOTDOT_SIZE;
1753 } else {
1754 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1755 inline_size = ext4_get_inline_size(dir) -
1756 EXT4_MIN_INLINE_DATA_SIZE;
1757 }
1758
1759 BUFFER_TRACE(bh, "get_write_access");
1760 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1761 EXT4_JTR_NONE);
1762 if (err)
1763 goto out;
1764
1765 err = ext4_generic_delete_entry(dir, de_del, bh,
1766 inline_start, inline_size, 0);
1767 if (err)
1768 goto out;
1769
1770 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1771out:
1772 ext4_write_unlock_xattr(dir, &no_expand);
1773 if (likely(err == 0))
1774 err = ext4_mark_inode_dirty(handle, dir);
1775 brelse(iloc.bh);
1776 if (err != -ENOENT)
1777 ext4_std_error(dir->i_sb, err);
1778 return err;
1779}
1780
1781/*
1782 * Get the inline dentry at offset.
1783 */
1784static inline struct ext4_dir_entry_2 *
1785ext4_get_inline_entry(struct inode *inode,
1786 struct ext4_iloc *iloc,
1787 unsigned int offset,
1788 void **inline_start,
1789 int *inline_size)
1790{
1791 void *inline_pos;
1792
1793 BUG_ON(offset > ext4_get_inline_size(inode));
1794
1795 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1796 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1797 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1798 } else {
1799 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1800 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1801 *inline_size = ext4_get_inline_size(inode) -
1802 EXT4_MIN_INLINE_DATA_SIZE;
1803 }
1804
1805 if (inline_start)
1806 *inline_start = inline_pos;
1807 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1808}
1809
1810bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1811{
1812 int err, inline_size;
1813 struct ext4_iloc iloc;
1814 size_t inline_len;
1815 void *inline_pos;
1816 unsigned int offset;
1817 struct ext4_dir_entry_2 *de;
1818 bool ret = false;
1819
1820 err = ext4_get_inode_loc(dir, &iloc);
1821 if (err) {
1822 EXT4_ERROR_INODE_ERR(dir, -err,
1823 "error %d getting inode %lu block",
1824 err, dir->i_ino);
1825 return false;
1826 }
1827
1828 down_read(&EXT4_I(dir)->xattr_sem);
1829 if (!ext4_has_inline_data(dir)) {
1830 *has_inline_data = 0;
1831 ret = true;
1832 goto out;
1833 }
1834
1835 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1836 if (!le32_to_cpu(de->inode)) {
1837 ext4_warning(dir->i_sb,
1838 "bad inline directory (dir #%lu) - no `..'",
1839 dir->i_ino);
1840 goto out;
1841 }
1842
1843 inline_len = ext4_get_inline_size(dir);
1844 offset = EXT4_INLINE_DOTDOT_SIZE;
1845 while (offset < inline_len) {
1846 de = ext4_get_inline_entry(dir, &iloc, offset,
1847 &inline_pos, &inline_size);
1848 if (ext4_check_dir_entry(dir, NULL, de,
1849 iloc.bh, inline_pos,
1850 inline_size, offset)) {
1851 ext4_warning(dir->i_sb,
1852 "bad inline directory (dir #%lu) - "
1853 "inode %u, rec_len %u, name_len %d"
1854 "inline size %d",
1855 dir->i_ino, le32_to_cpu(de->inode),
1856 le16_to_cpu(de->rec_len), de->name_len,
1857 inline_size);
1858 goto out;
1859 }
1860 if (le32_to_cpu(de->inode)) {
1861 goto out;
1862 }
1863 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1864 }
1865
1866 ret = true;
1867out:
1868 up_read(&EXT4_I(dir)->xattr_sem);
1869 brelse(iloc.bh);
1870 return ret;
1871}
1872
1873int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1874{
1875 int ret, no_expand;
1876
1877 ext4_write_lock_xattr(inode, &no_expand);
1878 ret = ext4_destroy_inline_data_nolock(handle, inode);
1879 ext4_write_unlock_xattr(inode, &no_expand);
1880
1881 return ret;
1882}
1883
1884int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1885{
1886 __u64 addr;
1887 int error = -EAGAIN;
1888 struct ext4_iloc iloc;
1889
1890 down_read(&EXT4_I(inode)->xattr_sem);
1891 if (!ext4_has_inline_data(inode))
1892 goto out;
1893
1894 error = ext4_get_inode_loc(inode, &iloc);
1895 if (error)
1896 goto out;
1897
1898 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1899 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1900 addr += offsetof(struct ext4_inode, i_block);
1901
1902 brelse(iloc.bh);
1903
1904 iomap->addr = addr;
1905 iomap->offset = 0;
1906 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1907 i_size_read(inode));
1908 iomap->type = IOMAP_INLINE;
1909 iomap->flags = 0;
1910
1911out:
1912 up_read(&EXT4_I(inode)->xattr_sem);
1913 return error;
1914}
1915
1916int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1917{
1918 handle_t *handle;
1919 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1920 size_t i_size;
1921 void *value = NULL;
1922 struct ext4_xattr_ibody_find is = {
1923 .s = { .not_found = -ENODATA, },
1924 };
1925 struct ext4_xattr_info i = {
1926 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1927 .name = EXT4_XATTR_SYSTEM_DATA,
1928 };
1929
1930
1931 needed_blocks = ext4_writepage_trans_blocks(inode);
1932 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1933 if (IS_ERR(handle))
1934 return PTR_ERR(handle);
1935
1936 ext4_write_lock_xattr(inode, &no_expand);
1937 if (!ext4_has_inline_data(inode)) {
1938 ext4_write_unlock_xattr(inode, &no_expand);
1939 *has_inline = 0;
1940 ext4_journal_stop(handle);
1941 return 0;
1942 }
1943
1944 if ((err = ext4_orphan_add(handle, inode)) != 0)
1945 goto out;
1946
1947 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1948 goto out;
1949
1950 down_write(&EXT4_I(inode)->i_data_sem);
1951 i_size = inode->i_size;
1952 inline_size = ext4_get_inline_size(inode);
1953 EXT4_I(inode)->i_disksize = i_size;
1954
1955 if (i_size < inline_size) {
1956 /*
1957 * if there's inline data to truncate and this file was
1958 * converted to extents after that inline data was written,
1959 * the extent status cache must be cleared to avoid leaving
1960 * behind stale delayed allocated extent entries
1961 */
1962 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1963 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1964
1965 /* Clear the content in the xattr space. */
1966 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1967 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1968 goto out_error;
1969
1970 BUG_ON(is.s.not_found);
1971
1972 value_len = le32_to_cpu(is.s.here->e_value_size);
1973 value = kmalloc(value_len, GFP_NOFS);
1974 if (!value) {
1975 err = -ENOMEM;
1976 goto out_error;
1977 }
1978
1979 err = ext4_xattr_ibody_get(inode, i.name_index,
1980 i.name, value, value_len);
1981 if (err <= 0)
1982 goto out_error;
1983
1984 i.value = value;
1985 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1986 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1987 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1988 if (err)
1989 goto out_error;
1990 }
1991
1992 /* Clear the content within i_blocks. */
1993 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1994 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1995 memset(p + i_size, 0,
1996 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1997 }
1998
1999 EXT4_I(inode)->i_inline_size = i_size <
2000 EXT4_MIN_INLINE_DATA_SIZE ?
2001 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2002 }
2003
2004out_error:
2005 up_write(&EXT4_I(inode)->i_data_sem);
2006out:
2007 brelse(is.iloc.bh);
2008 ext4_write_unlock_xattr(inode, &no_expand);
2009 kfree(value);
2010 if (inode->i_nlink)
2011 ext4_orphan_del(handle, inode);
2012
2013 if (err == 0) {
2014 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2015 err = ext4_mark_inode_dirty(handle, inode);
2016 if (IS_SYNC(inode))
2017 ext4_handle_sync(handle);
2018 }
2019 ext4_journal_stop(handle);
2020 return err;
2021}
2022
2023int ext4_convert_inline_data(struct inode *inode)
2024{
2025 int error, needed_blocks, no_expand;
2026 handle_t *handle;
2027 struct ext4_iloc iloc;
2028
2029 if (!ext4_has_inline_data(inode)) {
2030 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2031 return 0;
2032 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2033 /*
2034 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2035 * cleared. This means we are in the middle of moving of
2036 * inline data to delay allocated block. Just force writeout
2037 * here to finish conversion.
2038 */
2039 error = filemap_flush(inode->i_mapping);
2040 if (error)
2041 return error;
2042 if (!ext4_has_inline_data(inode))
2043 return 0;
2044 }
2045
2046 needed_blocks = ext4_writepage_trans_blocks(inode);
2047
2048 iloc.bh = NULL;
2049 error = ext4_get_inode_loc(inode, &iloc);
2050 if (error)
2051 return error;
2052
2053 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2054 if (IS_ERR(handle)) {
2055 error = PTR_ERR(handle);
2056 goto out_free;
2057 }
2058
2059 ext4_write_lock_xattr(inode, &no_expand);
2060 if (ext4_has_inline_data(inode))
2061 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2062 ext4_write_unlock_xattr(inode, &no_expand);
2063 ext4_journal_stop(handle);
2064out_free:
2065 brelse(iloc.bh);
2066 return error;
2067}