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 /*
303 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
304 * get cleared
305 */
306 ext4_set_inode_flags(inode);
307 get_bh(is.iloc.bh);
308 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
309
310out:
311 brelse(is.iloc.bh);
312 return error;
313}
314
315static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
316 unsigned int len)
317{
318 int error;
319 void *value = NULL;
320 struct ext4_xattr_ibody_find is = {
321 .s = { .not_found = -ENODATA, },
322 };
323 struct ext4_xattr_info i = {
324 .name_index = EXT4_XATTR_INDEX_SYSTEM,
325 .name = EXT4_XATTR_SYSTEM_DATA,
326 };
327
328 /* If the old space is ok, write the data directly. */
329 if (len <= EXT4_I(inode)->i_inline_size)
330 return 0;
331
332 error = ext4_get_inode_loc(inode, &is.iloc);
333 if (error)
334 return error;
335
336 error = ext4_xattr_ibody_find(inode, &i, &is);
337 if (error)
338 goto out;
339
340 BUG_ON(is.s.not_found);
341
342 len -= EXT4_MIN_INLINE_DATA_SIZE;
343 value = kzalloc(len, GFP_NOFS);
344 if (!value) {
345 error = -ENOMEM;
346 goto out;
347 }
348
349 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
350 value, len);
351 if (error == -ENODATA)
352 goto out;
353
354 BUFFER_TRACE(is.iloc.bh, "get_write_access");
355 error = ext4_journal_get_write_access(handle, is.iloc.bh);
356 if (error)
357 goto out;
358
359 /* Update the xttr entry. */
360 i.value = value;
361 i.value_len = len;
362
363 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
364 if (error)
365 goto out;
366
367 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
368 (void *)ext4_raw_inode(&is.iloc));
369 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
370 le32_to_cpu(is.s.here->e_value_size);
371 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
372 get_bh(is.iloc.bh);
373 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
374
375out:
376 kfree(value);
377 brelse(is.iloc.bh);
378 return error;
379}
380
381static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
382 unsigned int len)
383{
384 int ret, size, no_expand;
385 struct ext4_inode_info *ei = EXT4_I(inode);
386
387 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
388 return -ENOSPC;
389
390 size = ext4_get_max_inline_size(inode);
391 if (size < len)
392 return -ENOSPC;
393
394 ext4_write_lock_xattr(inode, &no_expand);
395
396 if (ei->i_inline_off)
397 ret = ext4_update_inline_data(handle, inode, len);
398 else
399 ret = ext4_create_inline_data(handle, inode, len);
400
401 ext4_write_unlock_xattr(inode, &no_expand);
402 return ret;
403}
404
405static int ext4_destroy_inline_data_nolock(handle_t *handle,
406 struct inode *inode)
407{
408 struct ext4_inode_info *ei = EXT4_I(inode);
409 struct ext4_xattr_ibody_find is = {
410 .s = { .not_found = 0, },
411 };
412 struct ext4_xattr_info i = {
413 .name_index = EXT4_XATTR_INDEX_SYSTEM,
414 .name = EXT4_XATTR_SYSTEM_DATA,
415 .value = NULL,
416 .value_len = 0,
417 };
418 int error;
419
420 if (!ei->i_inline_off)
421 return 0;
422
423 error = ext4_get_inode_loc(inode, &is.iloc);
424 if (error)
425 return error;
426
427 error = ext4_xattr_ibody_find(inode, &i, &is);
428 if (error)
429 goto out;
430
431 BUFFER_TRACE(is.iloc.bh, "get_write_access");
432 error = ext4_journal_get_write_access(handle, is.iloc.bh);
433 if (error)
434 goto out;
435
436 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
437 if (error)
438 goto out;
439
440 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
441 0, EXT4_MIN_INLINE_DATA_SIZE);
442
443 if (ext4_has_feature_extents(inode->i_sb)) {
444 if (S_ISDIR(inode->i_mode) ||
445 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
446 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
447 ext4_ext_tree_init(handle, inode);
448 }
449 }
450 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
451 /*
452 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
453 * get set.
454 */
455 ext4_set_inode_flags(inode);
456
457 get_bh(is.iloc.bh);
458 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
459
460 EXT4_I(inode)->i_inline_off = 0;
461 EXT4_I(inode)->i_inline_size = 0;
462 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
463out:
464 brelse(is.iloc.bh);
465 if (error == -ENODATA)
466 error = 0;
467 return error;
468}
469
470static int ext4_read_inline_page(struct inode *inode, struct page *page)
471{
472 void *kaddr;
473 int ret = 0;
474 size_t len;
475 struct ext4_iloc iloc;
476
477 BUG_ON(!PageLocked(page));
478 BUG_ON(!ext4_has_inline_data(inode));
479 BUG_ON(page->index);
480
481 if (!EXT4_I(inode)->i_inline_off) {
482 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
483 inode->i_ino);
484 goto out;
485 }
486
487 ret = ext4_get_inode_loc(inode, &iloc);
488 if (ret)
489 goto out;
490
491 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
492 kaddr = kmap_atomic(page);
493 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
494 flush_dcache_page(page);
495 kunmap_atomic(kaddr);
496 zero_user_segment(page, len, PAGE_SIZE);
497 SetPageUptodate(page);
498 brelse(iloc.bh);
499
500out:
501 return ret;
502}
503
504int ext4_readpage_inline(struct inode *inode, struct page *page)
505{
506 int ret = 0;
507
508 down_read(&EXT4_I(inode)->xattr_sem);
509 if (!ext4_has_inline_data(inode)) {
510 up_read(&EXT4_I(inode)->xattr_sem);
511 return -EAGAIN;
512 }
513
514 /*
515 * Current inline data can only exist in the 1st page,
516 * So for all the other pages, just set them uptodate.
517 */
518 if (!page->index)
519 ret = ext4_read_inline_page(inode, page);
520 else if (!PageUptodate(page)) {
521 zero_user_segment(page, 0, PAGE_SIZE);
522 SetPageUptodate(page);
523 }
524
525 up_read(&EXT4_I(inode)->xattr_sem);
526
527 unlock_page(page);
528 return ret >= 0 ? 0 : ret;
529}
530
531static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
532 struct inode *inode,
533 unsigned flags)
534{
535 int ret, needed_blocks, no_expand;
536 handle_t *handle = NULL;
537 int retries = 0, sem_held = 0;
538 struct page *page = NULL;
539 unsigned from, to;
540 struct ext4_iloc iloc;
541
542 if (!ext4_has_inline_data(inode)) {
543 /*
544 * clear the flag so that no new write
545 * will trap here again.
546 */
547 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
548 return 0;
549 }
550
551 needed_blocks = ext4_writepage_trans_blocks(inode);
552
553 ret = ext4_get_inode_loc(inode, &iloc);
554 if (ret)
555 return ret;
556
557retry:
558 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
559 if (IS_ERR(handle)) {
560 ret = PTR_ERR(handle);
561 handle = NULL;
562 goto out;
563 }
564
565 /* We cannot recurse into the filesystem as the transaction is already
566 * started */
567 flags |= AOP_FLAG_NOFS;
568
569 page = grab_cache_page_write_begin(mapping, 0, flags);
570 if (!page) {
571 ret = -ENOMEM;
572 goto out;
573 }
574
575 ext4_write_lock_xattr(inode, &no_expand);
576 sem_held = 1;
577 /* If some one has already done this for us, just exit. */
578 if (!ext4_has_inline_data(inode)) {
579 ret = 0;
580 goto out;
581 }
582
583 from = 0;
584 to = ext4_get_inline_size(inode);
585 if (!PageUptodate(page)) {
586 ret = ext4_read_inline_page(inode, page);
587 if (ret < 0)
588 goto out;
589 }
590
591 ret = ext4_destroy_inline_data_nolock(handle, inode);
592 if (ret)
593 goto out;
594
595 if (ext4_should_dioread_nolock(inode)) {
596 ret = __block_write_begin(page, from, to,
597 ext4_get_block_unwritten);
598 } else
599 ret = __block_write_begin(page, from, to, ext4_get_block);
600
601 if (!ret && ext4_should_journal_data(inode)) {
602 ret = ext4_walk_page_buffers(handle, page_buffers(page),
603 from, to, NULL,
604 do_journal_get_write_access);
605 }
606
607 if (ret) {
608 unlock_page(page);
609 put_page(page);
610 page = NULL;
611 ext4_orphan_add(handle, inode);
612 ext4_write_unlock_xattr(inode, &no_expand);
613 sem_held = 0;
614 ext4_journal_stop(handle);
615 handle = NULL;
616 ext4_truncate_failed_write(inode);
617 /*
618 * If truncate failed early the inode might
619 * still be on the orphan list; we need to
620 * make sure the inode is removed from the
621 * orphan list in that case.
622 */
623 if (inode->i_nlink)
624 ext4_orphan_del(NULL, inode);
625 }
626
627 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
628 goto retry;
629
630 if (page)
631 block_commit_write(page, from, to);
632out:
633 if (page) {
634 unlock_page(page);
635 put_page(page);
636 }
637 if (sem_held)
638 ext4_write_unlock_xattr(inode, &no_expand);
639 if (handle)
640 ext4_journal_stop(handle);
641 brelse(iloc.bh);
642 return ret;
643}
644
645/*
646 * Try to write data in the inode.
647 * If the inode has inline data, check whether the new write can be
648 * in the inode also. If not, create the page the handle, move the data
649 * to the page make it update and let the later codes create extent for it.
650 */
651int ext4_try_to_write_inline_data(struct address_space *mapping,
652 struct inode *inode,
653 loff_t pos, unsigned len,
654 unsigned flags,
655 struct page **pagep)
656{
657 int ret;
658 handle_t *handle;
659 struct page *page;
660 struct ext4_iloc iloc;
661
662 if (pos + len > ext4_get_max_inline_size(inode))
663 goto convert;
664
665 ret = ext4_get_inode_loc(inode, &iloc);
666 if (ret)
667 return ret;
668
669 /*
670 * The possible write could happen in the inode,
671 * so try to reserve the space in inode first.
672 */
673 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
674 if (IS_ERR(handle)) {
675 ret = PTR_ERR(handle);
676 handle = NULL;
677 goto out;
678 }
679
680 ret = ext4_prepare_inline_data(handle, inode, pos + len);
681 if (ret && ret != -ENOSPC)
682 goto out;
683
684 /* We don't have space in inline inode, so convert it to extent. */
685 if (ret == -ENOSPC) {
686 ext4_journal_stop(handle);
687 brelse(iloc.bh);
688 goto convert;
689 }
690
691 flags |= AOP_FLAG_NOFS;
692
693 page = grab_cache_page_write_begin(mapping, 0, flags);
694 if (!page) {
695 ret = -ENOMEM;
696 goto out;
697 }
698
699 *pagep = page;
700 down_read(&EXT4_I(inode)->xattr_sem);
701 if (!ext4_has_inline_data(inode)) {
702 ret = 0;
703 unlock_page(page);
704 put_page(page);
705 goto out_up_read;
706 }
707
708 if (!PageUptodate(page)) {
709 ret = ext4_read_inline_page(inode, page);
710 if (ret < 0)
711 goto out_up_read;
712 }
713
714 ret = 1;
715 handle = NULL;
716out_up_read:
717 up_read(&EXT4_I(inode)->xattr_sem);
718out:
719 if (handle)
720 ext4_journal_stop(handle);
721 brelse(iloc.bh);
722 return ret;
723convert:
724 return ext4_convert_inline_data_to_extent(mapping,
725 inode, flags);
726}
727
728int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
729 unsigned copied, struct page *page)
730{
731 int ret, no_expand;
732 void *kaddr;
733 struct ext4_iloc iloc;
734
735 if (unlikely(copied < len)) {
736 if (!PageUptodate(page)) {
737 copied = 0;
738 goto out;
739 }
740 }
741
742 ret = ext4_get_inode_loc(inode, &iloc);
743 if (ret) {
744 ext4_std_error(inode->i_sb, ret);
745 copied = 0;
746 goto out;
747 }
748
749 ext4_write_lock_xattr(inode, &no_expand);
750 BUG_ON(!ext4_has_inline_data(inode));
751
752 kaddr = kmap_atomic(page);
753 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
754 kunmap_atomic(kaddr);
755 SetPageUptodate(page);
756 /* clear page dirty so that writepages wouldn't work for us. */
757 ClearPageDirty(page);
758
759 ext4_write_unlock_xattr(inode, &no_expand);
760 brelse(iloc.bh);
761out:
762 return copied;
763}
764
765struct buffer_head *
766ext4_journalled_write_inline_data(struct inode *inode,
767 unsigned len,
768 struct page *page)
769{
770 int ret, no_expand;
771 void *kaddr;
772 struct ext4_iloc iloc;
773
774 ret = ext4_get_inode_loc(inode, &iloc);
775 if (ret) {
776 ext4_std_error(inode->i_sb, ret);
777 return NULL;
778 }
779
780 ext4_write_lock_xattr(inode, &no_expand);
781 kaddr = kmap_atomic(page);
782 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
783 kunmap_atomic(kaddr);
784 ext4_write_unlock_xattr(inode, &no_expand);
785
786 return iloc.bh;
787}
788
789/*
790 * Try to make the page cache and handle ready for the inline data case.
791 * We can call this function in 2 cases:
792 * 1. The inode is created and the first write exceeds inline size. We can
793 * clear the inode state safely.
794 * 2. The inode has inline data, then we need to read the data, make it
795 * update and dirty so that ext4_da_writepages can handle it. We don't
796 * need to start the journal since the file's metatdata isn't changed now.
797 */
798static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
799 struct inode *inode,
800 unsigned flags,
801 void **fsdata)
802{
803 int ret = 0, inline_size;
804 struct page *page;
805
806 page = grab_cache_page_write_begin(mapping, 0, flags);
807 if (!page)
808 return -ENOMEM;
809
810 down_read(&EXT4_I(inode)->xattr_sem);
811 if (!ext4_has_inline_data(inode)) {
812 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
813 goto out;
814 }
815
816 inline_size = ext4_get_inline_size(inode);
817
818 if (!PageUptodate(page)) {
819 ret = ext4_read_inline_page(inode, page);
820 if (ret < 0)
821 goto out;
822 }
823
824 ret = __block_write_begin(page, 0, inline_size,
825 ext4_da_get_block_prep);
826 if (ret) {
827 up_read(&EXT4_I(inode)->xattr_sem);
828 unlock_page(page);
829 put_page(page);
830 ext4_truncate_failed_write(inode);
831 return ret;
832 }
833
834 SetPageDirty(page);
835 SetPageUptodate(page);
836 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
837 *fsdata = (void *)CONVERT_INLINE_DATA;
838
839out:
840 up_read(&EXT4_I(inode)->xattr_sem);
841 if (page) {
842 unlock_page(page);
843 put_page(page);
844 }
845 return ret;
846}
847
848/*
849 * Prepare the write for the inline data.
850 * If the the data can be written into the inode, we just read
851 * the page and make it uptodate, and start the journal.
852 * Otherwise read the page, makes it dirty so that it can be
853 * handle in writepages(the i_disksize update is left to the
854 * normal ext4_da_write_end).
855 */
856int ext4_da_write_inline_data_begin(struct address_space *mapping,
857 struct inode *inode,
858 loff_t pos, unsigned len,
859 unsigned flags,
860 struct page **pagep,
861 void **fsdata)
862{
863 int ret, inline_size;
864 handle_t *handle;
865 struct page *page;
866 struct ext4_iloc iloc;
867 int retries;
868
869 ret = ext4_get_inode_loc(inode, &iloc);
870 if (ret)
871 return ret;
872
873retry_journal:
874 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
875 if (IS_ERR(handle)) {
876 ret = PTR_ERR(handle);
877 goto out;
878 }
879
880 inline_size = ext4_get_max_inline_size(inode);
881
882 ret = -ENOSPC;
883 if (inline_size >= pos + len) {
884 ret = ext4_prepare_inline_data(handle, inode, pos + len);
885 if (ret && ret != -ENOSPC)
886 goto out_journal;
887 }
888
889 /*
890 * We cannot recurse into the filesystem as the transaction
891 * is already started.
892 */
893 flags |= AOP_FLAG_NOFS;
894
895 if (ret == -ENOSPC) {
896 ret = ext4_da_convert_inline_data_to_extent(mapping,
897 inode,
898 flags,
899 fsdata);
900 ext4_journal_stop(handle);
901 if (ret == -ENOSPC &&
902 ext4_should_retry_alloc(inode->i_sb, &retries))
903 goto retry_journal;
904 goto out;
905 }
906
907
908 page = grab_cache_page_write_begin(mapping, 0, flags);
909 if (!page) {
910 ret = -ENOMEM;
911 goto out_journal;
912 }
913
914 down_read(&EXT4_I(inode)->xattr_sem);
915 if (!ext4_has_inline_data(inode)) {
916 ret = 0;
917 goto out_release_page;
918 }
919
920 if (!PageUptodate(page)) {
921 ret = ext4_read_inline_page(inode, page);
922 if (ret < 0)
923 goto out_release_page;
924 }
925
926 up_read(&EXT4_I(inode)->xattr_sem);
927 *pagep = page;
928 brelse(iloc.bh);
929 return 1;
930out_release_page:
931 up_read(&EXT4_I(inode)->xattr_sem);
932 unlock_page(page);
933 put_page(page);
934out_journal:
935 ext4_journal_stop(handle);
936out:
937 brelse(iloc.bh);
938 return ret;
939}
940
941int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
942 unsigned len, unsigned copied,
943 struct page *page)
944{
945 int i_size_changed = 0;
946 int ret;
947
948 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
949 if (ret < 0) {
950 unlock_page(page);
951 put_page(page);
952 return ret;
953 }
954 copied = ret;
955
956 /*
957 * No need to use i_size_read() here, the i_size
958 * cannot change under us because we hold i_mutex.
959 *
960 * But it's important to update i_size while still holding page lock:
961 * page writeout could otherwise come in and zero beyond i_size.
962 */
963 if (pos+copied > inode->i_size) {
964 i_size_write(inode, pos+copied);
965 i_size_changed = 1;
966 }
967 unlock_page(page);
968 put_page(page);
969
970 /*
971 * Don't mark the inode dirty under page lock. First, it unnecessarily
972 * makes the holding time of page lock longer. Second, it forces lock
973 * ordering of page lock and transaction start for journaling
974 * filesystems.
975 */
976 if (i_size_changed)
977 mark_inode_dirty(inode);
978
979 return copied;
980}
981
982#ifdef INLINE_DIR_DEBUG
983void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
984 void *inline_start, int inline_size)
985{
986 int offset;
987 unsigned short de_len;
988 struct ext4_dir_entry_2 *de = inline_start;
989 void *dlimit = inline_start + inline_size;
990
991 trace_printk("inode %lu\n", dir->i_ino);
992 offset = 0;
993 while ((void *)de < dlimit) {
994 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
995 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
996 offset, de_len, de->name_len, de->name,
997 de->name_len, le32_to_cpu(de->inode));
998 if (ext4_check_dir_entry(dir, NULL, de, bh,
999 inline_start, inline_size, offset))
1000 BUG();
1001
1002 offset += de_len;
1003 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1004 }
1005}
1006#else
1007#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1008#endif
1009
1010/*
1011 * Add a new entry into a inline dir.
1012 * It will return -ENOSPC if no space is available, and -EIO
1013 * and -EEXIST if directory entry already exists.
1014 */
1015static int ext4_add_dirent_to_inline(handle_t *handle,
1016 struct ext4_filename *fname,
1017 struct inode *dir,
1018 struct inode *inode,
1019 struct ext4_iloc *iloc,
1020 void *inline_start, int inline_size)
1021{
1022 int err;
1023 struct ext4_dir_entry_2 *de;
1024
1025 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1026 inline_size, fname, &de);
1027 if (err)
1028 return err;
1029
1030 BUFFER_TRACE(iloc->bh, "get_write_access");
1031 err = ext4_journal_get_write_access(handle, iloc->bh);
1032 if (err)
1033 return err;
1034 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1035
1036 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1037
1038 /*
1039 * XXX shouldn't update any times until successful
1040 * completion of syscall, but too many callers depend
1041 * on this.
1042 *
1043 * XXX similarly, too many callers depend on
1044 * ext4_new_inode() setting the times, but error
1045 * recovery deletes the inode, so the worst that can
1046 * happen is that the times are slightly out of date
1047 * and/or different from the directory change time.
1048 */
1049 dir->i_mtime = dir->i_ctime = current_time(dir);
1050 ext4_update_dx_flag(dir);
1051 dir->i_version++;
1052 ext4_mark_inode_dirty(handle, dir);
1053 return 1;
1054}
1055
1056static void *ext4_get_inline_xattr_pos(struct inode *inode,
1057 struct ext4_iloc *iloc)
1058{
1059 struct ext4_xattr_entry *entry;
1060 struct ext4_xattr_ibody_header *header;
1061
1062 BUG_ON(!EXT4_I(inode)->i_inline_off);
1063
1064 header = IHDR(inode, ext4_raw_inode(iloc));
1065 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1066 EXT4_I(inode)->i_inline_off);
1067
1068 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1069}
1070
1071/* Set the final de to cover the whole block. */
1072static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1073{
1074 struct ext4_dir_entry_2 *de, *prev_de;
1075 void *limit;
1076 int de_len;
1077
1078 de = (struct ext4_dir_entry_2 *)de_buf;
1079 if (old_size) {
1080 limit = de_buf + old_size;
1081 do {
1082 prev_de = de;
1083 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1084 de_buf += de_len;
1085 de = (struct ext4_dir_entry_2 *)de_buf;
1086 } while (de_buf < limit);
1087
1088 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1089 old_size, new_size);
1090 } else {
1091 /* this is just created, so create an empty entry. */
1092 de->inode = 0;
1093 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1094 }
1095}
1096
1097static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1098 struct ext4_iloc *iloc)
1099{
1100 int ret;
1101 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1102 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1103
1104 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1105 return -ENOSPC;
1106
1107 ret = ext4_update_inline_data(handle, dir,
1108 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1109 if (ret)
1110 return ret;
1111
1112 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1113 EXT4_I(dir)->i_inline_size -
1114 EXT4_MIN_INLINE_DATA_SIZE);
1115 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1116 return 0;
1117}
1118
1119static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1120 struct ext4_iloc *iloc,
1121 void *buf, int inline_size)
1122{
1123 ext4_create_inline_data(handle, inode, inline_size);
1124 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1125 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1126}
1127
1128static int ext4_finish_convert_inline_dir(handle_t *handle,
1129 struct inode *inode,
1130 struct buffer_head *dir_block,
1131 void *buf,
1132 int inline_size)
1133{
1134 int err, csum_size = 0, header_size = 0;
1135 struct ext4_dir_entry_2 *de;
1136 struct ext4_dir_entry_tail *t;
1137 void *target = dir_block->b_data;
1138
1139 /*
1140 * First create "." and ".." and then copy the dir information
1141 * back to the block.
1142 */
1143 de = (struct ext4_dir_entry_2 *)target;
1144 de = ext4_init_dot_dotdot(inode, de,
1145 inode->i_sb->s_blocksize, csum_size,
1146 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1147 header_size = (void *)de - target;
1148
1149 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1150 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1151
1152 if (ext4_has_metadata_csum(inode->i_sb))
1153 csum_size = sizeof(struct ext4_dir_entry_tail);
1154
1155 inode->i_size = inode->i_sb->s_blocksize;
1156 i_size_write(inode, inode->i_sb->s_blocksize);
1157 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1158 ext4_update_final_de(dir_block->b_data,
1159 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1160 inode->i_sb->s_blocksize - csum_size);
1161
1162 if (csum_size) {
1163 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1164 inode->i_sb->s_blocksize);
1165 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1166 }
1167 set_buffer_uptodate(dir_block);
1168 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1169 if (err)
1170 return err;
1171 set_buffer_verified(dir_block);
1172 return ext4_mark_inode_dirty(handle, inode);
1173}
1174
1175static int ext4_convert_inline_data_nolock(handle_t *handle,
1176 struct inode *inode,
1177 struct ext4_iloc *iloc)
1178{
1179 int error;
1180 void *buf = NULL;
1181 struct buffer_head *data_bh = NULL;
1182 struct ext4_map_blocks map;
1183 int inline_size;
1184
1185 inline_size = ext4_get_inline_size(inode);
1186 buf = kmalloc(inline_size, GFP_NOFS);
1187 if (!buf) {
1188 error = -ENOMEM;
1189 goto out;
1190 }
1191
1192 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1193 if (error < 0)
1194 goto out;
1195
1196 /*
1197 * Make sure the inline directory entries pass checks before we try to
1198 * convert them, so that we avoid touching stuff that needs fsck.
1199 */
1200 if (S_ISDIR(inode->i_mode)) {
1201 error = ext4_check_all_de(inode, iloc->bh,
1202 buf + EXT4_INLINE_DOTDOT_SIZE,
1203 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1204 if (error)
1205 goto out;
1206 }
1207
1208 error = ext4_destroy_inline_data_nolock(handle, inode);
1209 if (error)
1210 goto out;
1211
1212 map.m_lblk = 0;
1213 map.m_len = 1;
1214 map.m_flags = 0;
1215 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1216 if (error < 0)
1217 goto out_restore;
1218 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1219 error = -EIO;
1220 goto out_restore;
1221 }
1222
1223 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1224 if (!data_bh) {
1225 error = -ENOMEM;
1226 goto out_restore;
1227 }
1228
1229 lock_buffer(data_bh);
1230 error = ext4_journal_get_create_access(handle, data_bh);
1231 if (error) {
1232 unlock_buffer(data_bh);
1233 error = -EIO;
1234 goto out_restore;
1235 }
1236 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1237
1238 if (!S_ISDIR(inode->i_mode)) {
1239 memcpy(data_bh->b_data, buf, inline_size);
1240 set_buffer_uptodate(data_bh);
1241 error = ext4_handle_dirty_metadata(handle,
1242 inode, data_bh);
1243 } else {
1244 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1245 buf, inline_size);
1246 }
1247
1248 unlock_buffer(data_bh);
1249out_restore:
1250 if (error)
1251 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1252
1253out:
1254 brelse(data_bh);
1255 kfree(buf);
1256 return error;
1257}
1258
1259/*
1260 * Try to add the new entry to the inline data.
1261 * If succeeds, return 0. If not, extended the inline dir and copied data to
1262 * the new created block.
1263 */
1264int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1265 struct inode *dir, struct inode *inode)
1266{
1267 int ret, inline_size, no_expand;
1268 void *inline_start;
1269 struct ext4_iloc iloc;
1270
1271 ret = ext4_get_inode_loc(dir, &iloc);
1272 if (ret)
1273 return ret;
1274
1275 ext4_write_lock_xattr(dir, &no_expand);
1276 if (!ext4_has_inline_data(dir))
1277 goto out;
1278
1279 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1280 EXT4_INLINE_DOTDOT_SIZE;
1281 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1282
1283 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1284 inline_start, inline_size);
1285 if (ret != -ENOSPC)
1286 goto out;
1287
1288 /* check whether it can be inserted to inline xattr space. */
1289 inline_size = EXT4_I(dir)->i_inline_size -
1290 EXT4_MIN_INLINE_DATA_SIZE;
1291 if (!inline_size) {
1292 /* Try to use the xattr space.*/
1293 ret = ext4_update_inline_dir(handle, dir, &iloc);
1294 if (ret && ret != -ENOSPC)
1295 goto out;
1296
1297 inline_size = EXT4_I(dir)->i_inline_size -
1298 EXT4_MIN_INLINE_DATA_SIZE;
1299 }
1300
1301 if (inline_size) {
1302 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1303
1304 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1305 inode, &iloc, inline_start,
1306 inline_size);
1307
1308 if (ret != -ENOSPC)
1309 goto out;
1310 }
1311
1312 /*
1313 * The inline space is filled up, so create a new block for it.
1314 * As the extent tree will be created, we have to save the inline
1315 * dir first.
1316 */
1317 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1318
1319out:
1320 ext4_mark_inode_dirty(handle, dir);
1321 ext4_write_unlock_xattr(dir, &no_expand);
1322 brelse(iloc.bh);
1323 return ret;
1324}
1325
1326/*
1327 * This function fills a red-black tree with information from an
1328 * inlined dir. It returns the number directory entries loaded
1329 * into the tree. If there is an error it is returned in err.
1330 */
1331int htree_inlinedir_to_tree(struct file *dir_file,
1332 struct inode *dir, ext4_lblk_t block,
1333 struct dx_hash_info *hinfo,
1334 __u32 start_hash, __u32 start_minor_hash,
1335 int *has_inline_data)
1336{
1337 int err = 0, count = 0;
1338 unsigned int parent_ino;
1339 int pos;
1340 struct ext4_dir_entry_2 *de;
1341 struct inode *inode = file_inode(dir_file);
1342 int ret, inline_size = 0;
1343 struct ext4_iloc iloc;
1344 void *dir_buf = NULL;
1345 struct ext4_dir_entry_2 fake;
1346 struct fscrypt_str tmp_str;
1347
1348 ret = ext4_get_inode_loc(inode, &iloc);
1349 if (ret)
1350 return ret;
1351
1352 down_read(&EXT4_I(inode)->xattr_sem);
1353 if (!ext4_has_inline_data(inode)) {
1354 up_read(&EXT4_I(inode)->xattr_sem);
1355 *has_inline_data = 0;
1356 goto out;
1357 }
1358
1359 inline_size = ext4_get_inline_size(inode);
1360 dir_buf = kmalloc(inline_size, GFP_NOFS);
1361 if (!dir_buf) {
1362 ret = -ENOMEM;
1363 up_read(&EXT4_I(inode)->xattr_sem);
1364 goto out;
1365 }
1366
1367 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1368 up_read(&EXT4_I(inode)->xattr_sem);
1369 if (ret < 0)
1370 goto out;
1371
1372 pos = 0;
1373 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1374 while (pos < inline_size) {
1375 /*
1376 * As inlined dir doesn't store any information about '.' and
1377 * only the inode number of '..' is stored, we have to handle
1378 * them differently.
1379 */
1380 if (pos == 0) {
1381 fake.inode = cpu_to_le32(inode->i_ino);
1382 fake.name_len = 1;
1383 strcpy(fake.name, ".");
1384 fake.rec_len = ext4_rec_len_to_disk(
1385 EXT4_DIR_REC_LEN(fake.name_len),
1386 inline_size);
1387 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1388 de = &fake;
1389 pos = EXT4_INLINE_DOTDOT_OFFSET;
1390 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1391 fake.inode = cpu_to_le32(parent_ino);
1392 fake.name_len = 2;
1393 strcpy(fake.name, "..");
1394 fake.rec_len = ext4_rec_len_to_disk(
1395 EXT4_DIR_REC_LEN(fake.name_len),
1396 inline_size);
1397 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1398 de = &fake;
1399 pos = EXT4_INLINE_DOTDOT_SIZE;
1400 } else {
1401 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1402 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1403 if (ext4_check_dir_entry(inode, dir_file, de,
1404 iloc.bh, dir_buf,
1405 inline_size, pos)) {
1406 ret = count;
1407 goto out;
1408 }
1409 }
1410
1411 ext4fs_dirhash(de->name, de->name_len, hinfo);
1412 if ((hinfo->hash < start_hash) ||
1413 ((hinfo->hash == start_hash) &&
1414 (hinfo->minor_hash < start_minor_hash)))
1415 continue;
1416 if (de->inode == 0)
1417 continue;
1418 tmp_str.name = de->name;
1419 tmp_str.len = de->name_len;
1420 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1421 hinfo->minor_hash, de, &tmp_str);
1422 if (err) {
1423 count = err;
1424 goto out;
1425 }
1426 count++;
1427 }
1428 ret = count;
1429out:
1430 kfree(dir_buf);
1431 brelse(iloc.bh);
1432 return ret;
1433}
1434
1435/*
1436 * So this function is called when the volume is mkfsed with
1437 * dir_index disabled. In order to keep f_pos persistent
1438 * after we convert from an inlined dir to a blocked based,
1439 * we just pretend that we are a normal dir and return the
1440 * offset as if '.' and '..' really take place.
1441 *
1442 */
1443int ext4_read_inline_dir(struct file *file,
1444 struct dir_context *ctx,
1445 int *has_inline_data)
1446{
1447 unsigned int offset, parent_ino;
1448 int i;
1449 struct ext4_dir_entry_2 *de;
1450 struct super_block *sb;
1451 struct inode *inode = file_inode(file);
1452 int ret, inline_size = 0;
1453 struct ext4_iloc iloc;
1454 void *dir_buf = NULL;
1455 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1456
1457 ret = ext4_get_inode_loc(inode, &iloc);
1458 if (ret)
1459 return ret;
1460
1461 down_read(&EXT4_I(inode)->xattr_sem);
1462 if (!ext4_has_inline_data(inode)) {
1463 up_read(&EXT4_I(inode)->xattr_sem);
1464 *has_inline_data = 0;
1465 goto out;
1466 }
1467
1468 inline_size = ext4_get_inline_size(inode);
1469 dir_buf = kmalloc(inline_size, GFP_NOFS);
1470 if (!dir_buf) {
1471 ret = -ENOMEM;
1472 up_read(&EXT4_I(inode)->xattr_sem);
1473 goto out;
1474 }
1475
1476 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1477 up_read(&EXT4_I(inode)->xattr_sem);
1478 if (ret < 0)
1479 goto out;
1480
1481 ret = 0;
1482 sb = inode->i_sb;
1483 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1484 offset = ctx->pos;
1485
1486 /*
1487 * dotdot_offset and dotdot_size is the real offset and
1488 * size for ".." and "." if the dir is block based while
1489 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1490 * So we will use extra_offset and extra_size to indicate them
1491 * during the inline dir iteration.
1492 */
1493 dotdot_offset = EXT4_DIR_REC_LEN(1);
1494 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1495 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1496 extra_size = extra_offset + inline_size;
1497
1498 /*
1499 * If the version has changed since the last call to
1500 * readdir(2), then we might be pointing to an invalid
1501 * dirent right now. Scan from the start of the inline
1502 * dir to make sure.
1503 */
1504 if (file->f_version != inode->i_version) {
1505 for (i = 0; i < extra_size && i < offset;) {
1506 /*
1507 * "." is with offset 0 and
1508 * ".." is dotdot_offset.
1509 */
1510 if (!i) {
1511 i = dotdot_offset;
1512 continue;
1513 } else if (i == dotdot_offset) {
1514 i = dotdot_size;
1515 continue;
1516 }
1517 /* for other entry, the real offset in
1518 * the buf has to be tuned accordingly.
1519 */
1520 de = (struct ext4_dir_entry_2 *)
1521 (dir_buf + i - extra_offset);
1522 /* It's too expensive to do a full
1523 * dirent test each time round this
1524 * loop, but we do have to test at
1525 * least that it is non-zero. A
1526 * failure will be detected in the
1527 * dirent test below. */
1528 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1529 < EXT4_DIR_REC_LEN(1))
1530 break;
1531 i += ext4_rec_len_from_disk(de->rec_len,
1532 extra_size);
1533 }
1534 offset = i;
1535 ctx->pos = offset;
1536 file->f_version = inode->i_version;
1537 }
1538
1539 while (ctx->pos < extra_size) {
1540 if (ctx->pos == 0) {
1541 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1542 goto out;
1543 ctx->pos = dotdot_offset;
1544 continue;
1545 }
1546
1547 if (ctx->pos == dotdot_offset) {
1548 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1549 goto out;
1550 ctx->pos = dotdot_size;
1551 continue;
1552 }
1553
1554 de = (struct ext4_dir_entry_2 *)
1555 (dir_buf + ctx->pos - extra_offset);
1556 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1557 extra_size, ctx->pos))
1558 goto out;
1559 if (le32_to_cpu(de->inode)) {
1560 if (!dir_emit(ctx, de->name, de->name_len,
1561 le32_to_cpu(de->inode),
1562 get_dtype(sb, de->file_type)))
1563 goto out;
1564 }
1565 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1566 }
1567out:
1568 kfree(dir_buf);
1569 brelse(iloc.bh);
1570 return ret;
1571}
1572
1573struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1574 struct ext4_dir_entry_2 **parent_de,
1575 int *retval)
1576{
1577 struct ext4_iloc iloc;
1578
1579 *retval = ext4_get_inode_loc(inode, &iloc);
1580 if (*retval)
1581 return NULL;
1582
1583 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1584
1585 return iloc.bh;
1586}
1587
1588/*
1589 * Try to create the inline data for the new dir.
1590 * If it succeeds, return 0, otherwise return the error.
1591 * In case of ENOSPC, the caller should create the normal disk layout dir.
1592 */
1593int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1594 struct inode *inode)
1595{
1596 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1597 struct ext4_iloc iloc;
1598 struct ext4_dir_entry_2 *de;
1599
1600 ret = ext4_get_inode_loc(inode, &iloc);
1601 if (ret)
1602 return ret;
1603
1604 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1605 if (ret)
1606 goto out;
1607
1608 /*
1609 * For inline dir, we only save the inode information for the ".."
1610 * and create a fake dentry to cover the left space.
1611 */
1612 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1613 de->inode = cpu_to_le32(parent->i_ino);
1614 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1615 de->inode = 0;
1616 de->rec_len = ext4_rec_len_to_disk(
1617 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1618 inline_size);
1619 set_nlink(inode, 2);
1620 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1621out:
1622 brelse(iloc.bh);
1623 return ret;
1624}
1625
1626struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1627 struct ext4_filename *fname,
1628 const struct qstr *d_name,
1629 struct ext4_dir_entry_2 **res_dir,
1630 int *has_inline_data)
1631{
1632 int ret;
1633 struct ext4_iloc iloc;
1634 void *inline_start;
1635 int inline_size;
1636
1637 if (ext4_get_inode_loc(dir, &iloc))
1638 return NULL;
1639
1640 down_read(&EXT4_I(dir)->xattr_sem);
1641 if (!ext4_has_inline_data(dir)) {
1642 *has_inline_data = 0;
1643 goto out;
1644 }
1645
1646 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1647 EXT4_INLINE_DOTDOT_SIZE;
1648 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1649 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1650 dir, fname, d_name, 0, res_dir);
1651 if (ret == 1)
1652 goto out_find;
1653 if (ret < 0)
1654 goto out;
1655
1656 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1657 goto out;
1658
1659 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1660 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1661
1662 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1663 dir, fname, d_name, 0, res_dir);
1664 if (ret == 1)
1665 goto out_find;
1666
1667out:
1668 brelse(iloc.bh);
1669 iloc.bh = NULL;
1670out_find:
1671 up_read(&EXT4_I(dir)->xattr_sem);
1672 return iloc.bh;
1673}
1674
1675int ext4_delete_inline_entry(handle_t *handle,
1676 struct inode *dir,
1677 struct ext4_dir_entry_2 *de_del,
1678 struct buffer_head *bh,
1679 int *has_inline_data)
1680{
1681 int err, inline_size, no_expand;
1682 struct ext4_iloc iloc;
1683 void *inline_start;
1684
1685 err = ext4_get_inode_loc(dir, &iloc);
1686 if (err)
1687 return err;
1688
1689 ext4_write_lock_xattr(dir, &no_expand);
1690 if (!ext4_has_inline_data(dir)) {
1691 *has_inline_data = 0;
1692 goto out;
1693 }
1694
1695 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1696 EXT4_MIN_INLINE_DATA_SIZE) {
1697 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1698 EXT4_INLINE_DOTDOT_SIZE;
1699 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1700 EXT4_INLINE_DOTDOT_SIZE;
1701 } else {
1702 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1703 inline_size = ext4_get_inline_size(dir) -
1704 EXT4_MIN_INLINE_DATA_SIZE;
1705 }
1706
1707 BUFFER_TRACE(bh, "get_write_access");
1708 err = ext4_journal_get_write_access(handle, bh);
1709 if (err)
1710 goto out;
1711
1712 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1713 inline_start, inline_size, 0);
1714 if (err)
1715 goto out;
1716
1717 err = ext4_mark_inode_dirty(handle, dir);
1718 if (unlikely(err))
1719 goto out;
1720
1721 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1722out:
1723 ext4_write_unlock_xattr(dir, &no_expand);
1724 brelse(iloc.bh);
1725 if (err != -ENOENT)
1726 ext4_std_error(dir->i_sb, err);
1727 return err;
1728}
1729
1730/*
1731 * Get the inline dentry at offset.
1732 */
1733static inline struct ext4_dir_entry_2 *
1734ext4_get_inline_entry(struct inode *inode,
1735 struct ext4_iloc *iloc,
1736 unsigned int offset,
1737 void **inline_start,
1738 int *inline_size)
1739{
1740 void *inline_pos;
1741
1742 BUG_ON(offset > ext4_get_inline_size(inode));
1743
1744 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1745 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1746 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1747 } else {
1748 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1749 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1750 *inline_size = ext4_get_inline_size(inode) -
1751 EXT4_MIN_INLINE_DATA_SIZE;
1752 }
1753
1754 if (inline_start)
1755 *inline_start = inline_pos;
1756 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1757}
1758
1759bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1760{
1761 int err, inline_size;
1762 struct ext4_iloc iloc;
1763 void *inline_pos;
1764 unsigned int offset;
1765 struct ext4_dir_entry_2 *de;
1766 bool ret = true;
1767
1768 err = ext4_get_inode_loc(dir, &iloc);
1769 if (err) {
1770 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1771 err, dir->i_ino);
1772 return true;
1773 }
1774
1775 down_read(&EXT4_I(dir)->xattr_sem);
1776 if (!ext4_has_inline_data(dir)) {
1777 *has_inline_data = 0;
1778 goto out;
1779 }
1780
1781 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1782 if (!le32_to_cpu(de->inode)) {
1783 ext4_warning(dir->i_sb,
1784 "bad inline directory (dir #%lu) - no `..'",
1785 dir->i_ino);
1786 ret = true;
1787 goto out;
1788 }
1789
1790 offset = EXT4_INLINE_DOTDOT_SIZE;
1791 while (offset < dir->i_size) {
1792 de = ext4_get_inline_entry(dir, &iloc, offset,
1793 &inline_pos, &inline_size);
1794 if (ext4_check_dir_entry(dir, NULL, de,
1795 iloc.bh, inline_pos,
1796 inline_size, offset)) {
1797 ext4_warning(dir->i_sb,
1798 "bad inline directory (dir #%lu) - "
1799 "inode %u, rec_len %u, name_len %d"
1800 "inline size %d",
1801 dir->i_ino, le32_to_cpu(de->inode),
1802 le16_to_cpu(de->rec_len), de->name_len,
1803 inline_size);
1804 ret = true;
1805 goto out;
1806 }
1807 if (le32_to_cpu(de->inode)) {
1808 ret = false;
1809 goto out;
1810 }
1811 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1812 }
1813
1814out:
1815 up_read(&EXT4_I(dir)->xattr_sem);
1816 brelse(iloc.bh);
1817 return ret;
1818}
1819
1820int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1821{
1822 int ret, no_expand;
1823
1824 ext4_write_lock_xattr(inode, &no_expand);
1825 ret = ext4_destroy_inline_data_nolock(handle, inode);
1826 ext4_write_unlock_xattr(inode, &no_expand);
1827
1828 return ret;
1829}
1830
1831int ext4_inline_data_fiemap(struct inode *inode,
1832 struct fiemap_extent_info *fieinfo,
1833 int *has_inline, __u64 start, __u64 len)
1834{
1835 __u64 physical = 0;
1836 __u64 inline_len;
1837 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1838 FIEMAP_EXTENT_LAST;
1839 int error = 0;
1840 struct ext4_iloc iloc;
1841
1842 down_read(&EXT4_I(inode)->xattr_sem);
1843 if (!ext4_has_inline_data(inode)) {
1844 *has_inline = 0;
1845 goto out;
1846 }
1847 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1848 i_size_read(inode));
1849 if (start >= inline_len)
1850 goto out;
1851 if (start + len < inline_len)
1852 inline_len = start + len;
1853 inline_len -= start;
1854
1855 error = ext4_get_inode_loc(inode, &iloc);
1856 if (error)
1857 goto out;
1858
1859 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1860 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1861 physical += offsetof(struct ext4_inode, i_block);
1862
1863 if (physical)
1864 error = fiemap_fill_next_extent(fieinfo, start, physical,
1865 inline_len, flags);
1866 brelse(iloc.bh);
1867out:
1868 up_read(&EXT4_I(inode)->xattr_sem);
1869 return (error < 0 ? error : 0);
1870}
1871
1872/*
1873 * Called during xattr set, and if we can sparse space 'needed',
1874 * just create the extent tree evict the data to the outer block.
1875 *
1876 * We use jbd2 instead of page cache to move data to the 1st block
1877 * so that the whole transaction can be committed as a whole and
1878 * the data isn't lost because of the delayed page cache write.
1879 */
1880int ext4_try_to_evict_inline_data(handle_t *handle,
1881 struct inode *inode,
1882 int needed)
1883{
1884 int error;
1885 struct ext4_xattr_entry *entry;
1886 struct ext4_inode *raw_inode;
1887 struct ext4_iloc iloc;
1888
1889 error = ext4_get_inode_loc(inode, &iloc);
1890 if (error)
1891 return error;
1892
1893 raw_inode = ext4_raw_inode(&iloc);
1894 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1895 EXT4_I(inode)->i_inline_off);
1896 if (EXT4_XATTR_LEN(entry->e_name_len) +
1897 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1898 error = -ENOSPC;
1899 goto out;
1900 }
1901
1902 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1903out:
1904 brelse(iloc.bh);
1905 return error;
1906}
1907
1908void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1909{
1910 handle_t *handle;
1911 int inline_size, value_len, needed_blocks, no_expand;
1912 size_t i_size;
1913 void *value = NULL;
1914 struct ext4_xattr_ibody_find is = {
1915 .s = { .not_found = -ENODATA, },
1916 };
1917 struct ext4_xattr_info i = {
1918 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1919 .name = EXT4_XATTR_SYSTEM_DATA,
1920 };
1921
1922
1923 needed_blocks = ext4_writepage_trans_blocks(inode);
1924 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1925 if (IS_ERR(handle))
1926 return;
1927
1928 ext4_write_lock_xattr(inode, &no_expand);
1929 if (!ext4_has_inline_data(inode)) {
1930 *has_inline = 0;
1931 ext4_journal_stop(handle);
1932 return;
1933 }
1934
1935 if (ext4_orphan_add(handle, inode))
1936 goto out;
1937
1938 if (ext4_get_inode_loc(inode, &is.iloc))
1939 goto out;
1940
1941 down_write(&EXT4_I(inode)->i_data_sem);
1942 i_size = inode->i_size;
1943 inline_size = ext4_get_inline_size(inode);
1944 EXT4_I(inode)->i_disksize = i_size;
1945
1946 if (i_size < inline_size) {
1947 /* Clear the content in the xattr space. */
1948 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1949 if (ext4_xattr_ibody_find(inode, &i, &is))
1950 goto out_error;
1951
1952 BUG_ON(is.s.not_found);
1953
1954 value_len = le32_to_cpu(is.s.here->e_value_size);
1955 value = kmalloc(value_len, GFP_NOFS);
1956 if (!value)
1957 goto out_error;
1958
1959 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1960 value, value_len))
1961 goto out_error;
1962
1963 i.value = value;
1964 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1965 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1966 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1967 goto out_error;
1968 }
1969
1970 /* Clear the content within i_blocks. */
1971 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1972 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1973 memset(p + i_size, 0,
1974 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1975 }
1976
1977 EXT4_I(inode)->i_inline_size = i_size <
1978 EXT4_MIN_INLINE_DATA_SIZE ?
1979 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1980 }
1981
1982out_error:
1983 up_write(&EXT4_I(inode)->i_data_sem);
1984out:
1985 brelse(is.iloc.bh);
1986 ext4_write_unlock_xattr(inode, &no_expand);
1987 kfree(value);
1988 if (inode->i_nlink)
1989 ext4_orphan_del(handle, inode);
1990
1991 inode->i_mtime = inode->i_ctime = current_time(inode);
1992 ext4_mark_inode_dirty(handle, inode);
1993 if (IS_SYNC(inode))
1994 ext4_handle_sync(handle);
1995
1996 ext4_journal_stop(handle);
1997 return;
1998}
1999
2000int ext4_convert_inline_data(struct inode *inode)
2001{
2002 int error, needed_blocks, no_expand;
2003 handle_t *handle;
2004 struct ext4_iloc iloc;
2005
2006 if (!ext4_has_inline_data(inode)) {
2007 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2008 return 0;
2009 }
2010
2011 needed_blocks = ext4_writepage_trans_blocks(inode);
2012
2013 iloc.bh = NULL;
2014 error = ext4_get_inode_loc(inode, &iloc);
2015 if (error)
2016 return error;
2017
2018 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2019 if (IS_ERR(handle)) {
2020 error = PTR_ERR(handle);
2021 goto out_free;
2022 }
2023
2024 ext4_write_lock_xattr(inode, &no_expand);
2025 if (ext4_has_inline_data(inode))
2026 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2027 ext4_write_unlock_xattr(inode, &no_expand);
2028 ext4_journal_stop(handle);
2029out_free:
2030 brelse(iloc.bh);
2031 return error;
2032}
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}