Loading...
Note: File does not exist in v3.5.6.
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 = __block_write_begin(&folio->page, from, to,
605 ext4_get_block_unwritten);
606 } else
607 ret = __block_write_begin(&folio->page, from, to, ext4_get_block);
608
609 if (!ret && ext4_should_journal_data(inode)) {
610 ret = ext4_walk_page_buffers(handle, inode,
611 folio_buffers(folio), from, to,
612 NULL, do_journal_get_write_access);
613 }
614
615 if (ret) {
616 folio_unlock(folio);
617 folio_put(folio);
618 folio = NULL;
619 ext4_orphan_add(handle, inode);
620 ext4_write_unlock_xattr(inode, &no_expand);
621 sem_held = 0;
622 ext4_journal_stop(handle);
623 handle = NULL;
624 ext4_truncate_failed_write(inode);
625 /*
626 * If truncate failed early the inode might
627 * still be on the orphan list; we need to
628 * make sure the inode is removed from the
629 * orphan list in that case.
630 */
631 if (inode->i_nlink)
632 ext4_orphan_del(NULL, inode);
633 }
634
635 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
636 goto retry;
637
638 if (folio)
639 block_commit_write(&folio->page, from, to);
640out:
641 if (folio) {
642 folio_unlock(folio);
643 folio_put(folio);
644 }
645out_nofolio:
646 if (sem_held)
647 ext4_write_unlock_xattr(inode, &no_expand);
648 if (handle)
649 ext4_journal_stop(handle);
650 brelse(iloc.bh);
651 return ret;
652}
653
654/*
655 * Try to write data in the inode.
656 * If the inode has inline data, check whether the new write can be
657 * in the inode also. If not, create the page the handle, move the data
658 * to the page make it update and let the later codes create extent for it.
659 */
660int ext4_try_to_write_inline_data(struct address_space *mapping,
661 struct inode *inode,
662 loff_t pos, unsigned len,
663 struct page **pagep)
664{
665 int ret;
666 handle_t *handle;
667 struct folio *folio;
668 struct ext4_iloc iloc;
669
670 if (pos + len > ext4_get_max_inline_size(inode))
671 goto convert;
672
673 ret = ext4_get_inode_loc(inode, &iloc);
674 if (ret)
675 return ret;
676
677 /*
678 * The possible write could happen in the inode,
679 * so try to reserve the space in inode first.
680 */
681 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
682 if (IS_ERR(handle)) {
683 ret = PTR_ERR(handle);
684 handle = NULL;
685 goto out;
686 }
687
688 ret = ext4_prepare_inline_data(handle, inode, pos + len);
689 if (ret && ret != -ENOSPC)
690 goto out;
691
692 /* We don't have space in inline inode, so convert it to extent. */
693 if (ret == -ENOSPC) {
694 ext4_journal_stop(handle);
695 brelse(iloc.bh);
696 goto convert;
697 }
698
699 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
700 EXT4_JTR_NONE);
701 if (ret)
702 goto out;
703
704 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
705 mapping_gfp_mask(mapping));
706 if (IS_ERR(folio)) {
707 ret = PTR_ERR(folio);
708 goto out;
709 }
710
711 *pagep = &folio->page;
712 down_read(&EXT4_I(inode)->xattr_sem);
713 if (!ext4_has_inline_data(inode)) {
714 ret = 0;
715 folio_unlock(folio);
716 folio_put(folio);
717 goto out_up_read;
718 }
719
720 if (!folio_test_uptodate(folio)) {
721 ret = ext4_read_inline_folio(inode, folio);
722 if (ret < 0) {
723 folio_unlock(folio);
724 folio_put(folio);
725 goto out_up_read;
726 }
727 }
728
729 ret = 1;
730 handle = NULL;
731out_up_read:
732 up_read(&EXT4_I(inode)->xattr_sem);
733out:
734 if (handle && (ret != 1))
735 ext4_journal_stop(handle);
736 brelse(iloc.bh);
737 return ret;
738convert:
739 return ext4_convert_inline_data_to_extent(mapping, inode);
740}
741
742int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
743 unsigned copied, struct folio *folio)
744{
745 handle_t *handle = ext4_journal_current_handle();
746 int no_expand;
747 void *kaddr;
748 struct ext4_iloc iloc;
749 int ret = 0, ret2;
750
751 if (unlikely(copied < len) && !folio_test_uptodate(folio))
752 copied = 0;
753
754 if (likely(copied)) {
755 ret = ext4_get_inode_loc(inode, &iloc);
756 if (ret) {
757 folio_unlock(folio);
758 folio_put(folio);
759 ext4_std_error(inode->i_sb, ret);
760 goto out;
761 }
762 ext4_write_lock_xattr(inode, &no_expand);
763 BUG_ON(!ext4_has_inline_data(inode));
764
765 /*
766 * ei->i_inline_off may have changed since
767 * ext4_write_begin() called
768 * ext4_try_to_write_inline_data()
769 */
770 (void) ext4_find_inline_data_nolock(inode);
771
772 kaddr = kmap_local_folio(folio, 0);
773 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
774 kunmap_local(kaddr);
775 folio_mark_uptodate(folio);
776 /* clear dirty flag so that writepages wouldn't work for us. */
777 folio_clear_dirty(folio);
778
779 ext4_write_unlock_xattr(inode, &no_expand);
780 brelse(iloc.bh);
781
782 /*
783 * It's important to update i_size while still holding folio
784 * lock: page writeout could otherwise come in and zero
785 * beyond i_size.
786 */
787 ext4_update_inode_size(inode, pos + copied);
788 }
789 folio_unlock(folio);
790 folio_put(folio);
791
792 /*
793 * Don't mark the inode dirty under folio lock. First, it unnecessarily
794 * makes the holding time of folio lock longer. Second, it forces lock
795 * ordering of folio lock and transaction start for journaling
796 * filesystems.
797 */
798 if (likely(copied))
799 mark_inode_dirty(inode);
800out:
801 /*
802 * If we didn't copy as much data as expected, we need to trim back
803 * size of xattr containing inline data.
804 */
805 if (pos + len > inode->i_size && ext4_can_truncate(inode))
806 ext4_orphan_add(handle, inode);
807
808 ret2 = ext4_journal_stop(handle);
809 if (!ret)
810 ret = ret2;
811 if (pos + len > inode->i_size) {
812 ext4_truncate_failed_write(inode);
813 /*
814 * If truncate failed early the inode might still be
815 * on the orphan list; we need to make sure the inode
816 * is removed from the orphan list in that case.
817 */
818 if (inode->i_nlink)
819 ext4_orphan_del(NULL, inode);
820 }
821 return ret ? ret : copied;
822}
823
824/*
825 * Try to make the page cache and handle ready for the inline data case.
826 * We can call this function in 2 cases:
827 * 1. The inode is created and the first write exceeds inline size. We can
828 * clear the inode state safely.
829 * 2. The inode has inline data, then we need to read the data, make it
830 * update and dirty so that ext4_da_writepages can handle it. We don't
831 * need to start the journal since the file's metadata isn't changed now.
832 */
833static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
834 struct inode *inode,
835 void **fsdata)
836{
837 int ret = 0, inline_size;
838 struct folio *folio;
839
840 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN,
841 mapping_gfp_mask(mapping));
842 if (IS_ERR(folio))
843 return PTR_ERR(folio);
844
845 down_read(&EXT4_I(inode)->xattr_sem);
846 if (!ext4_has_inline_data(inode)) {
847 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
848 goto out;
849 }
850
851 inline_size = ext4_get_inline_size(inode);
852
853 if (!folio_test_uptodate(folio)) {
854 ret = ext4_read_inline_folio(inode, folio);
855 if (ret < 0)
856 goto out;
857 }
858
859 ret = __block_write_begin(&folio->page, 0, inline_size,
860 ext4_da_get_block_prep);
861 if (ret) {
862 up_read(&EXT4_I(inode)->xattr_sem);
863 folio_unlock(folio);
864 folio_put(folio);
865 ext4_truncate_failed_write(inode);
866 return ret;
867 }
868
869 folio_mark_dirty(folio);
870 folio_mark_uptodate(folio);
871 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
872 *fsdata = (void *)CONVERT_INLINE_DATA;
873
874out:
875 up_read(&EXT4_I(inode)->xattr_sem);
876 if (folio) {
877 folio_unlock(folio);
878 folio_put(folio);
879 }
880 return ret;
881}
882
883/*
884 * Prepare the write for the inline data.
885 * If the data can be written into the inode, we just read
886 * the page and make it uptodate, and start the journal.
887 * Otherwise read the page, makes it dirty so that it can be
888 * handle in writepages(the i_disksize update is left to the
889 * normal ext4_da_write_end).
890 */
891int ext4_da_write_inline_data_begin(struct address_space *mapping,
892 struct inode *inode,
893 loff_t pos, unsigned len,
894 struct page **pagep,
895 void **fsdata)
896{
897 int ret;
898 handle_t *handle;
899 struct folio *folio;
900 struct ext4_iloc iloc;
901 int retries = 0;
902
903 ret = ext4_get_inode_loc(inode, &iloc);
904 if (ret)
905 return ret;
906
907retry_journal:
908 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
909 if (IS_ERR(handle)) {
910 ret = PTR_ERR(handle);
911 goto out;
912 }
913
914 ret = ext4_prepare_inline_data(handle, inode, pos + len);
915 if (ret && ret != -ENOSPC)
916 goto out_journal;
917
918 if (ret == -ENOSPC) {
919 ext4_journal_stop(handle);
920 ret = ext4_da_convert_inline_data_to_extent(mapping,
921 inode,
922 fsdata);
923 if (ret == -ENOSPC &&
924 ext4_should_retry_alloc(inode->i_sb, &retries))
925 goto retry_journal;
926 goto out;
927 }
928
929 /*
930 * We cannot recurse into the filesystem as the transaction
931 * is already started.
932 */
933 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
934 mapping_gfp_mask(mapping));
935 if (IS_ERR(folio)) {
936 ret = PTR_ERR(folio);
937 goto out_journal;
938 }
939
940 down_read(&EXT4_I(inode)->xattr_sem);
941 if (!ext4_has_inline_data(inode)) {
942 ret = 0;
943 goto out_release_page;
944 }
945
946 if (!folio_test_uptodate(folio)) {
947 ret = ext4_read_inline_folio(inode, folio);
948 if (ret < 0)
949 goto out_release_page;
950 }
951 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
952 EXT4_JTR_NONE);
953 if (ret)
954 goto out_release_page;
955
956 up_read(&EXT4_I(inode)->xattr_sem);
957 *pagep = &folio->page;
958 brelse(iloc.bh);
959 return 1;
960out_release_page:
961 up_read(&EXT4_I(inode)->xattr_sem);
962 folio_unlock(folio);
963 folio_put(folio);
964out_journal:
965 ext4_journal_stop(handle);
966out:
967 brelse(iloc.bh);
968 return ret;
969}
970
971#ifdef INLINE_DIR_DEBUG
972void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
973 void *inline_start, int inline_size)
974{
975 int offset;
976 unsigned short de_len;
977 struct ext4_dir_entry_2 *de = inline_start;
978 void *dlimit = inline_start + inline_size;
979
980 trace_printk("inode %lu\n", dir->i_ino);
981 offset = 0;
982 while ((void *)de < dlimit) {
983 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
984 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
985 offset, de_len, de->name_len, de->name,
986 de->name_len, le32_to_cpu(de->inode));
987 if (ext4_check_dir_entry(dir, NULL, de, bh,
988 inline_start, inline_size, offset))
989 BUG();
990
991 offset += de_len;
992 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
993 }
994}
995#else
996#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
997#endif
998
999/*
1000 * Add a new entry into a inline dir.
1001 * It will return -ENOSPC if no space is available, and -EIO
1002 * and -EEXIST if directory entry already exists.
1003 */
1004static int ext4_add_dirent_to_inline(handle_t *handle,
1005 struct ext4_filename *fname,
1006 struct inode *dir,
1007 struct inode *inode,
1008 struct ext4_iloc *iloc,
1009 void *inline_start, int inline_size)
1010{
1011 int err;
1012 struct ext4_dir_entry_2 *de;
1013
1014 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1015 inline_size, fname, &de);
1016 if (err)
1017 return err;
1018
1019 BUFFER_TRACE(iloc->bh, "get_write_access");
1020 err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1021 EXT4_JTR_NONE);
1022 if (err)
1023 return err;
1024 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1025
1026 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1027
1028 /*
1029 * XXX shouldn't update any times until successful
1030 * completion of syscall, but too many callers depend
1031 * on this.
1032 *
1033 * XXX similarly, too many callers depend on
1034 * ext4_new_inode() setting the times, but error
1035 * recovery deletes the inode, so the worst that can
1036 * happen is that the times are slightly out of date
1037 * and/or different from the directory change time.
1038 */
1039 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
1040 ext4_update_dx_flag(dir);
1041 inode_inc_iversion(dir);
1042 return 1;
1043}
1044
1045static void *ext4_get_inline_xattr_pos(struct inode *inode,
1046 struct ext4_iloc *iloc)
1047{
1048 struct ext4_xattr_entry *entry;
1049 struct ext4_xattr_ibody_header *header;
1050
1051 BUG_ON(!EXT4_I(inode)->i_inline_off);
1052
1053 header = IHDR(inode, ext4_raw_inode(iloc));
1054 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1055 EXT4_I(inode)->i_inline_off);
1056
1057 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1058}
1059
1060/* Set the final de to cover the whole block. */
1061static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1062{
1063 struct ext4_dir_entry_2 *de, *prev_de;
1064 void *limit;
1065 int de_len;
1066
1067 de = de_buf;
1068 if (old_size) {
1069 limit = de_buf + old_size;
1070 do {
1071 prev_de = de;
1072 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1073 de_buf += de_len;
1074 de = de_buf;
1075 } while (de_buf < limit);
1076
1077 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1078 old_size, new_size);
1079 } else {
1080 /* this is just created, so create an empty entry. */
1081 de->inode = 0;
1082 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1083 }
1084}
1085
1086static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1087 struct ext4_iloc *iloc)
1088{
1089 int ret;
1090 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1091 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1092
1093 if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1094 return -ENOSPC;
1095
1096 ret = ext4_update_inline_data(handle, dir,
1097 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1098 if (ret)
1099 return ret;
1100
1101 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1102 EXT4_I(dir)->i_inline_size -
1103 EXT4_MIN_INLINE_DATA_SIZE);
1104 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1105 return 0;
1106}
1107
1108static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1109 struct ext4_iloc *iloc,
1110 void *buf, int inline_size)
1111{
1112 int ret;
1113
1114 ret = ext4_create_inline_data(handle, inode, inline_size);
1115 if (ret) {
1116 ext4_msg(inode->i_sb, KERN_EMERG,
1117 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1118 inode->i_ino, ret);
1119 return;
1120 }
1121 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1122 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1123}
1124
1125static int ext4_finish_convert_inline_dir(handle_t *handle,
1126 struct inode *inode,
1127 struct buffer_head *dir_block,
1128 void *buf,
1129 int inline_size)
1130{
1131 int err, csum_size = 0, header_size = 0;
1132 struct ext4_dir_entry_2 *de;
1133 void *target = dir_block->b_data;
1134
1135 /*
1136 * First create "." and ".." and then copy the dir information
1137 * back to the block.
1138 */
1139 de = target;
1140 de = ext4_init_dot_dotdot(inode, de,
1141 inode->i_sb->s_blocksize, csum_size,
1142 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1143 header_size = (void *)de - target;
1144
1145 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1146 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1147
1148 if (ext4_has_metadata_csum(inode->i_sb))
1149 csum_size = sizeof(struct ext4_dir_entry_tail);
1150
1151 inode->i_size = inode->i_sb->s_blocksize;
1152 i_size_write(inode, inode->i_sb->s_blocksize);
1153 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1154 ext4_update_final_de(dir_block->b_data,
1155 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1156 inode->i_sb->s_blocksize - csum_size);
1157
1158 if (csum_size)
1159 ext4_initialize_dirent_tail(dir_block,
1160 inode->i_sb->s_blocksize);
1161 set_buffer_uptodate(dir_block);
1162 unlock_buffer(dir_block);
1163 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1164 if (err)
1165 return err;
1166 set_buffer_verified(dir_block);
1167 return ext4_mark_inode_dirty(handle, inode);
1168}
1169
1170static int ext4_convert_inline_data_nolock(handle_t *handle,
1171 struct inode *inode,
1172 struct ext4_iloc *iloc)
1173{
1174 int error;
1175 void *buf = NULL;
1176 struct buffer_head *data_bh = NULL;
1177 struct ext4_map_blocks map;
1178 int inline_size;
1179
1180 inline_size = ext4_get_inline_size(inode);
1181 buf = kmalloc(inline_size, GFP_NOFS);
1182 if (!buf) {
1183 error = -ENOMEM;
1184 goto out;
1185 }
1186
1187 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1188 if (error < 0)
1189 goto out;
1190
1191 /*
1192 * Make sure the inline directory entries pass checks before we try to
1193 * convert them, so that we avoid touching stuff that needs fsck.
1194 */
1195 if (S_ISDIR(inode->i_mode)) {
1196 error = ext4_check_all_de(inode, iloc->bh,
1197 buf + EXT4_INLINE_DOTDOT_SIZE,
1198 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1199 if (error)
1200 goto out;
1201 }
1202
1203 error = ext4_destroy_inline_data_nolock(handle, inode);
1204 if (error)
1205 goto out;
1206
1207 map.m_lblk = 0;
1208 map.m_len = 1;
1209 map.m_flags = 0;
1210 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1211 if (error < 0)
1212 goto out_restore;
1213 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1214 error = -EIO;
1215 goto out_restore;
1216 }
1217
1218 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1219 if (!data_bh) {
1220 error = -ENOMEM;
1221 goto out_restore;
1222 }
1223
1224 lock_buffer(data_bh);
1225 error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1226 EXT4_JTR_NONE);
1227 if (error) {
1228 unlock_buffer(data_bh);
1229 error = -EIO;
1230 goto out_restore;
1231 }
1232 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1233
1234 if (!S_ISDIR(inode->i_mode)) {
1235 memcpy(data_bh->b_data, buf, inline_size);
1236 set_buffer_uptodate(data_bh);
1237 unlock_buffer(data_bh);
1238 error = ext4_handle_dirty_metadata(handle,
1239 inode, data_bh);
1240 } else {
1241 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1242 buf, inline_size);
1243 }
1244
1245out_restore:
1246 if (error)
1247 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1248
1249out:
1250 brelse(data_bh);
1251 kfree(buf);
1252 return error;
1253}
1254
1255/*
1256 * Try to add the new entry to the inline data.
1257 * If succeeds, return 0. If not, extended the inline dir and copied data to
1258 * the new created block.
1259 */
1260int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1261 struct inode *dir, struct inode *inode)
1262{
1263 int ret, ret2, inline_size, no_expand;
1264 void *inline_start;
1265 struct ext4_iloc iloc;
1266
1267 ret = ext4_get_inode_loc(dir, &iloc);
1268 if (ret)
1269 return ret;
1270
1271 ext4_write_lock_xattr(dir, &no_expand);
1272 if (!ext4_has_inline_data(dir))
1273 goto out;
1274
1275 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1276 EXT4_INLINE_DOTDOT_SIZE;
1277 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1278
1279 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1280 inline_start, inline_size);
1281 if (ret != -ENOSPC)
1282 goto out;
1283
1284 /* check whether it can be inserted to inline xattr space. */
1285 inline_size = EXT4_I(dir)->i_inline_size -
1286 EXT4_MIN_INLINE_DATA_SIZE;
1287 if (!inline_size) {
1288 /* Try to use the xattr space.*/
1289 ret = ext4_update_inline_dir(handle, dir, &iloc);
1290 if (ret && ret != -ENOSPC)
1291 goto out;
1292
1293 inline_size = EXT4_I(dir)->i_inline_size -
1294 EXT4_MIN_INLINE_DATA_SIZE;
1295 }
1296
1297 if (inline_size) {
1298 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1299
1300 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1301 inode, &iloc, inline_start,
1302 inline_size);
1303
1304 if (ret != -ENOSPC)
1305 goto out;
1306 }
1307
1308 /*
1309 * The inline space is filled up, so create a new block for it.
1310 * As the extent tree will be created, we have to save the inline
1311 * dir first.
1312 */
1313 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1314
1315out:
1316 ext4_write_unlock_xattr(dir, &no_expand);
1317 ret2 = ext4_mark_inode_dirty(handle, dir);
1318 if (unlikely(ret2 && !ret))
1319 ret = ret2;
1320 brelse(iloc.bh);
1321 return ret;
1322}
1323
1324/*
1325 * This function fills a red-black tree with information from an
1326 * inlined dir. It returns the number directory entries loaded
1327 * into the tree. If there is an error it is returned in err.
1328 */
1329int ext4_inlinedir_to_tree(struct file *dir_file,
1330 struct inode *dir, ext4_lblk_t block,
1331 struct dx_hash_info *hinfo,
1332 __u32 start_hash, __u32 start_minor_hash,
1333 int *has_inline_data)
1334{
1335 int err = 0, count = 0;
1336 unsigned int parent_ino;
1337 int pos;
1338 struct ext4_dir_entry_2 *de;
1339 struct inode *inode = file_inode(dir_file);
1340 int ret, inline_size = 0;
1341 struct ext4_iloc iloc;
1342 void *dir_buf = NULL;
1343 struct ext4_dir_entry_2 fake;
1344 struct fscrypt_str tmp_str;
1345
1346 ret = ext4_get_inode_loc(inode, &iloc);
1347 if (ret)
1348 return ret;
1349
1350 down_read(&EXT4_I(inode)->xattr_sem);
1351 if (!ext4_has_inline_data(inode)) {
1352 up_read(&EXT4_I(inode)->xattr_sem);
1353 *has_inline_data = 0;
1354 goto out;
1355 }
1356
1357 inline_size = ext4_get_inline_size(inode);
1358 dir_buf = kmalloc(inline_size, GFP_NOFS);
1359 if (!dir_buf) {
1360 ret = -ENOMEM;
1361 up_read(&EXT4_I(inode)->xattr_sem);
1362 goto out;
1363 }
1364
1365 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1366 up_read(&EXT4_I(inode)->xattr_sem);
1367 if (ret < 0)
1368 goto out;
1369
1370 pos = 0;
1371 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1372 while (pos < inline_size) {
1373 /*
1374 * As inlined dir doesn't store any information about '.' and
1375 * only the inode number of '..' is stored, we have to handle
1376 * them differently.
1377 */
1378 if (pos == 0) {
1379 fake.inode = cpu_to_le32(inode->i_ino);
1380 fake.name_len = 1;
1381 strcpy(fake.name, ".");
1382 fake.rec_len = ext4_rec_len_to_disk(
1383 ext4_dir_rec_len(fake.name_len, NULL),
1384 inline_size);
1385 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1386 de = &fake;
1387 pos = EXT4_INLINE_DOTDOT_OFFSET;
1388 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1389 fake.inode = cpu_to_le32(parent_ino);
1390 fake.name_len = 2;
1391 strcpy(fake.name, "..");
1392 fake.rec_len = ext4_rec_len_to_disk(
1393 ext4_dir_rec_len(fake.name_len, NULL),
1394 inline_size);
1395 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1396 de = &fake;
1397 pos = EXT4_INLINE_DOTDOT_SIZE;
1398 } else {
1399 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1400 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1401 if (ext4_check_dir_entry(inode, dir_file, de,
1402 iloc.bh, dir_buf,
1403 inline_size, pos)) {
1404 ret = count;
1405 goto out;
1406 }
1407 }
1408
1409 if (ext4_hash_in_dirent(dir)) {
1410 hinfo->hash = EXT4_DIRENT_HASH(de);
1411 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1412 } else {
1413 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1414 }
1415 if ((hinfo->hash < start_hash) ||
1416 ((hinfo->hash == start_hash) &&
1417 (hinfo->minor_hash < start_minor_hash)))
1418 continue;
1419 if (de->inode == 0)
1420 continue;
1421 tmp_str.name = de->name;
1422 tmp_str.len = de->name_len;
1423 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1424 hinfo->minor_hash, de, &tmp_str);
1425 if (err) {
1426 ret = err;
1427 goto out;
1428 }
1429 count++;
1430 }
1431 ret = count;
1432out:
1433 kfree(dir_buf);
1434 brelse(iloc.bh);
1435 return ret;
1436}
1437
1438/*
1439 * So this function is called when the volume is mkfsed with
1440 * dir_index disabled. In order to keep f_pos persistent
1441 * after we convert from an inlined dir to a blocked based,
1442 * we just pretend that we are a normal dir and return the
1443 * offset as if '.' and '..' really take place.
1444 *
1445 */
1446int ext4_read_inline_dir(struct file *file,
1447 struct dir_context *ctx,
1448 int *has_inline_data)
1449{
1450 unsigned int offset, parent_ino;
1451 int i;
1452 struct ext4_dir_entry_2 *de;
1453 struct super_block *sb;
1454 struct inode *inode = file_inode(file);
1455 int ret, inline_size = 0;
1456 struct ext4_iloc iloc;
1457 void *dir_buf = NULL;
1458 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1459
1460 ret = ext4_get_inode_loc(inode, &iloc);
1461 if (ret)
1462 return ret;
1463
1464 down_read(&EXT4_I(inode)->xattr_sem);
1465 if (!ext4_has_inline_data(inode)) {
1466 up_read(&EXT4_I(inode)->xattr_sem);
1467 *has_inline_data = 0;
1468 goto out;
1469 }
1470
1471 inline_size = ext4_get_inline_size(inode);
1472 dir_buf = kmalloc(inline_size, GFP_NOFS);
1473 if (!dir_buf) {
1474 ret = -ENOMEM;
1475 up_read(&EXT4_I(inode)->xattr_sem);
1476 goto out;
1477 }
1478
1479 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1480 up_read(&EXT4_I(inode)->xattr_sem);
1481 if (ret < 0)
1482 goto out;
1483
1484 ret = 0;
1485 sb = inode->i_sb;
1486 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1487 offset = ctx->pos;
1488
1489 /*
1490 * dotdot_offset and dotdot_size is the real offset and
1491 * size for ".." and "." if the dir is block based while
1492 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1493 * So we will use extra_offset and extra_size to indicate them
1494 * during the inline dir iteration.
1495 */
1496 dotdot_offset = ext4_dir_rec_len(1, NULL);
1497 dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1498 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1499 extra_size = extra_offset + inline_size;
1500
1501 /*
1502 * If the version has changed since the last call to
1503 * readdir(2), then we might be pointing to an invalid
1504 * dirent right now. Scan from the start of the inline
1505 * dir to make sure.
1506 */
1507 if (!inode_eq_iversion(inode, file->f_version)) {
1508 for (i = 0; i < extra_size && i < offset;) {
1509 /*
1510 * "." is with offset 0 and
1511 * ".." is dotdot_offset.
1512 */
1513 if (!i) {
1514 i = dotdot_offset;
1515 continue;
1516 } else if (i == dotdot_offset) {
1517 i = dotdot_size;
1518 continue;
1519 }
1520 /* for other entry, the real offset in
1521 * the buf has to be tuned accordingly.
1522 */
1523 de = (struct ext4_dir_entry_2 *)
1524 (dir_buf + i - extra_offset);
1525 /* It's too expensive to do a full
1526 * dirent test each time round this
1527 * loop, but we do have to test at
1528 * least that it is non-zero. A
1529 * failure will be detected in the
1530 * dirent test below. */
1531 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1532 < ext4_dir_rec_len(1, NULL))
1533 break;
1534 i += ext4_rec_len_from_disk(de->rec_len,
1535 extra_size);
1536 }
1537 offset = i;
1538 ctx->pos = offset;
1539 file->f_version = inode_query_iversion(inode);
1540 }
1541
1542 while (ctx->pos < extra_size) {
1543 if (ctx->pos == 0) {
1544 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1545 goto out;
1546 ctx->pos = dotdot_offset;
1547 continue;
1548 }
1549
1550 if (ctx->pos == dotdot_offset) {
1551 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1552 goto out;
1553 ctx->pos = dotdot_size;
1554 continue;
1555 }
1556
1557 de = (struct ext4_dir_entry_2 *)
1558 (dir_buf + ctx->pos - extra_offset);
1559 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1560 extra_size, ctx->pos))
1561 goto out;
1562 if (le32_to_cpu(de->inode)) {
1563 if (!dir_emit(ctx, de->name, de->name_len,
1564 le32_to_cpu(de->inode),
1565 get_dtype(sb, de->file_type)))
1566 goto out;
1567 }
1568 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1569 }
1570out:
1571 kfree(dir_buf);
1572 brelse(iloc.bh);
1573 return ret;
1574}
1575
1576void *ext4_read_inline_link(struct inode *inode)
1577{
1578 struct ext4_iloc iloc;
1579 int ret, inline_size;
1580 void *link;
1581
1582 ret = ext4_get_inode_loc(inode, &iloc);
1583 if (ret)
1584 return ERR_PTR(ret);
1585
1586 ret = -ENOMEM;
1587 inline_size = ext4_get_inline_size(inode);
1588 link = kmalloc(inline_size + 1, GFP_NOFS);
1589 if (!link)
1590 goto out;
1591
1592 ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1593 if (ret < 0) {
1594 kfree(link);
1595 goto out;
1596 }
1597 nd_terminate_link(link, inode->i_size, ret);
1598out:
1599 if (ret < 0)
1600 link = ERR_PTR(ret);
1601 brelse(iloc.bh);
1602 return link;
1603}
1604
1605struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1606 struct ext4_dir_entry_2 **parent_de,
1607 int *retval)
1608{
1609 struct ext4_iloc iloc;
1610
1611 *retval = ext4_get_inode_loc(inode, &iloc);
1612 if (*retval)
1613 return NULL;
1614
1615 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1616
1617 return iloc.bh;
1618}
1619
1620/*
1621 * Try to create the inline data for the new dir.
1622 * If it succeeds, return 0, otherwise return the error.
1623 * In case of ENOSPC, the caller should create the normal disk layout dir.
1624 */
1625int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1626 struct inode *inode)
1627{
1628 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1629 struct ext4_iloc iloc;
1630 struct ext4_dir_entry_2 *de;
1631
1632 ret = ext4_get_inode_loc(inode, &iloc);
1633 if (ret)
1634 return ret;
1635
1636 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1637 if (ret)
1638 goto out;
1639
1640 /*
1641 * For inline dir, we only save the inode information for the ".."
1642 * and create a fake dentry to cover the left space.
1643 */
1644 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1645 de->inode = cpu_to_le32(parent->i_ino);
1646 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1647 de->inode = 0;
1648 de->rec_len = ext4_rec_len_to_disk(
1649 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1650 inline_size);
1651 set_nlink(inode, 2);
1652 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1653out:
1654 brelse(iloc.bh);
1655 return ret;
1656}
1657
1658struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1659 struct ext4_filename *fname,
1660 struct ext4_dir_entry_2 **res_dir,
1661 int *has_inline_data)
1662{
1663 int ret;
1664 struct ext4_iloc iloc;
1665 void *inline_start;
1666 int inline_size;
1667
1668 if (ext4_get_inode_loc(dir, &iloc))
1669 return NULL;
1670
1671 down_read(&EXT4_I(dir)->xattr_sem);
1672 if (!ext4_has_inline_data(dir)) {
1673 *has_inline_data = 0;
1674 goto out;
1675 }
1676
1677 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1678 EXT4_INLINE_DOTDOT_SIZE;
1679 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1680 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1681 dir, fname, 0, res_dir);
1682 if (ret == 1)
1683 goto out_find;
1684 if (ret < 0)
1685 goto out;
1686
1687 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1688 goto out;
1689
1690 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1691 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1692
1693 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1694 dir, fname, 0, res_dir);
1695 if (ret == 1)
1696 goto out_find;
1697
1698out:
1699 brelse(iloc.bh);
1700 iloc.bh = NULL;
1701out_find:
1702 up_read(&EXT4_I(dir)->xattr_sem);
1703 return iloc.bh;
1704}
1705
1706int ext4_delete_inline_entry(handle_t *handle,
1707 struct inode *dir,
1708 struct ext4_dir_entry_2 *de_del,
1709 struct buffer_head *bh,
1710 int *has_inline_data)
1711{
1712 int err, inline_size, no_expand;
1713 struct ext4_iloc iloc;
1714 void *inline_start;
1715
1716 err = ext4_get_inode_loc(dir, &iloc);
1717 if (err)
1718 return err;
1719
1720 ext4_write_lock_xattr(dir, &no_expand);
1721 if (!ext4_has_inline_data(dir)) {
1722 *has_inline_data = 0;
1723 goto out;
1724 }
1725
1726 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1727 EXT4_MIN_INLINE_DATA_SIZE) {
1728 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1729 EXT4_INLINE_DOTDOT_SIZE;
1730 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1731 EXT4_INLINE_DOTDOT_SIZE;
1732 } else {
1733 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1734 inline_size = ext4_get_inline_size(dir) -
1735 EXT4_MIN_INLINE_DATA_SIZE;
1736 }
1737
1738 BUFFER_TRACE(bh, "get_write_access");
1739 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1740 EXT4_JTR_NONE);
1741 if (err)
1742 goto out;
1743
1744 err = ext4_generic_delete_entry(dir, de_del, bh,
1745 inline_start, inline_size, 0);
1746 if (err)
1747 goto out;
1748
1749 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1750out:
1751 ext4_write_unlock_xattr(dir, &no_expand);
1752 if (likely(err == 0))
1753 err = ext4_mark_inode_dirty(handle, dir);
1754 brelse(iloc.bh);
1755 if (err != -ENOENT)
1756 ext4_std_error(dir->i_sb, err);
1757 return err;
1758}
1759
1760/*
1761 * Get the inline dentry at offset.
1762 */
1763static inline struct ext4_dir_entry_2 *
1764ext4_get_inline_entry(struct inode *inode,
1765 struct ext4_iloc *iloc,
1766 unsigned int offset,
1767 void **inline_start,
1768 int *inline_size)
1769{
1770 void *inline_pos;
1771
1772 BUG_ON(offset > ext4_get_inline_size(inode));
1773
1774 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1775 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1776 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1777 } else {
1778 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1779 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1780 *inline_size = ext4_get_inline_size(inode) -
1781 EXT4_MIN_INLINE_DATA_SIZE;
1782 }
1783
1784 if (inline_start)
1785 *inline_start = inline_pos;
1786 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1787}
1788
1789bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1790{
1791 int err, inline_size;
1792 struct ext4_iloc iloc;
1793 size_t inline_len;
1794 void *inline_pos;
1795 unsigned int offset;
1796 struct ext4_dir_entry_2 *de;
1797 bool ret = false;
1798
1799 err = ext4_get_inode_loc(dir, &iloc);
1800 if (err) {
1801 EXT4_ERROR_INODE_ERR(dir, -err,
1802 "error %d getting inode %lu block",
1803 err, dir->i_ino);
1804 return false;
1805 }
1806
1807 down_read(&EXT4_I(dir)->xattr_sem);
1808 if (!ext4_has_inline_data(dir)) {
1809 *has_inline_data = 0;
1810 ret = true;
1811 goto out;
1812 }
1813
1814 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1815 if (!le32_to_cpu(de->inode)) {
1816 ext4_warning(dir->i_sb,
1817 "bad inline directory (dir #%lu) - no `..'",
1818 dir->i_ino);
1819 goto out;
1820 }
1821
1822 inline_len = ext4_get_inline_size(dir);
1823 offset = EXT4_INLINE_DOTDOT_SIZE;
1824 while (offset < inline_len) {
1825 de = ext4_get_inline_entry(dir, &iloc, offset,
1826 &inline_pos, &inline_size);
1827 if (ext4_check_dir_entry(dir, NULL, de,
1828 iloc.bh, inline_pos,
1829 inline_size, offset)) {
1830 ext4_warning(dir->i_sb,
1831 "bad inline directory (dir #%lu) - "
1832 "inode %u, rec_len %u, name_len %d"
1833 "inline size %d",
1834 dir->i_ino, le32_to_cpu(de->inode),
1835 le16_to_cpu(de->rec_len), de->name_len,
1836 inline_size);
1837 goto out;
1838 }
1839 if (le32_to_cpu(de->inode)) {
1840 goto out;
1841 }
1842 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1843 }
1844
1845 ret = true;
1846out:
1847 up_read(&EXT4_I(dir)->xattr_sem);
1848 brelse(iloc.bh);
1849 return ret;
1850}
1851
1852int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1853{
1854 int ret, no_expand;
1855
1856 ext4_write_lock_xattr(inode, &no_expand);
1857 ret = ext4_destroy_inline_data_nolock(handle, inode);
1858 ext4_write_unlock_xattr(inode, &no_expand);
1859
1860 return ret;
1861}
1862
1863int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1864{
1865 __u64 addr;
1866 int error = -EAGAIN;
1867 struct ext4_iloc iloc;
1868
1869 down_read(&EXT4_I(inode)->xattr_sem);
1870 if (!ext4_has_inline_data(inode))
1871 goto out;
1872
1873 error = ext4_get_inode_loc(inode, &iloc);
1874 if (error)
1875 goto out;
1876
1877 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1878 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1879 addr += offsetof(struct ext4_inode, i_block);
1880
1881 brelse(iloc.bh);
1882
1883 iomap->addr = addr;
1884 iomap->offset = 0;
1885 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1886 i_size_read(inode));
1887 iomap->type = IOMAP_INLINE;
1888 iomap->flags = 0;
1889
1890out:
1891 up_read(&EXT4_I(inode)->xattr_sem);
1892 return error;
1893}
1894
1895int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1896{
1897 handle_t *handle;
1898 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1899 size_t i_size;
1900 void *value = NULL;
1901 struct ext4_xattr_ibody_find is = {
1902 .s = { .not_found = -ENODATA, },
1903 };
1904 struct ext4_xattr_info i = {
1905 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1906 .name = EXT4_XATTR_SYSTEM_DATA,
1907 };
1908
1909
1910 needed_blocks = ext4_writepage_trans_blocks(inode);
1911 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1912 if (IS_ERR(handle))
1913 return PTR_ERR(handle);
1914
1915 ext4_write_lock_xattr(inode, &no_expand);
1916 if (!ext4_has_inline_data(inode)) {
1917 ext4_write_unlock_xattr(inode, &no_expand);
1918 *has_inline = 0;
1919 ext4_journal_stop(handle);
1920 return 0;
1921 }
1922
1923 if ((err = ext4_orphan_add(handle, inode)) != 0)
1924 goto out;
1925
1926 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1927 goto out;
1928
1929 down_write(&EXT4_I(inode)->i_data_sem);
1930 i_size = inode->i_size;
1931 inline_size = ext4_get_inline_size(inode);
1932 EXT4_I(inode)->i_disksize = i_size;
1933
1934 if (i_size < inline_size) {
1935 /*
1936 * if there's inline data to truncate and this file was
1937 * converted to extents after that inline data was written,
1938 * the extent status cache must be cleared to avoid leaving
1939 * behind stale delayed allocated extent entries
1940 */
1941 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1942 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1943
1944 /* Clear the content in the xattr space. */
1945 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1946 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1947 goto out_error;
1948
1949 BUG_ON(is.s.not_found);
1950
1951 value_len = le32_to_cpu(is.s.here->e_value_size);
1952 value = kmalloc(value_len, GFP_NOFS);
1953 if (!value) {
1954 err = -ENOMEM;
1955 goto out_error;
1956 }
1957
1958 err = ext4_xattr_ibody_get(inode, i.name_index,
1959 i.name, value, value_len);
1960 if (err <= 0)
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 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1967 if (err)
1968 goto out_error;
1969 }
1970
1971 /* Clear the content within i_blocks. */
1972 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1973 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1974 memset(p + i_size, 0,
1975 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1976 }
1977
1978 EXT4_I(inode)->i_inline_size = i_size <
1979 EXT4_MIN_INLINE_DATA_SIZE ?
1980 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1981 }
1982
1983out_error:
1984 up_write(&EXT4_I(inode)->i_data_sem);
1985out:
1986 brelse(is.iloc.bh);
1987 ext4_write_unlock_xattr(inode, &no_expand);
1988 kfree(value);
1989 if (inode->i_nlink)
1990 ext4_orphan_del(handle, inode);
1991
1992 if (err == 0) {
1993 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1994 err = ext4_mark_inode_dirty(handle, inode);
1995 if (IS_SYNC(inode))
1996 ext4_handle_sync(handle);
1997 }
1998 ext4_journal_stop(handle);
1999 return err;
2000}
2001
2002int ext4_convert_inline_data(struct inode *inode)
2003{
2004 int error, needed_blocks, no_expand;
2005 handle_t *handle;
2006 struct ext4_iloc iloc;
2007
2008 if (!ext4_has_inline_data(inode)) {
2009 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2010 return 0;
2011 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2012 /*
2013 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2014 * cleared. This means we are in the middle of moving of
2015 * inline data to delay allocated block. Just force writeout
2016 * here to finish conversion.
2017 */
2018 error = filemap_flush(inode->i_mapping);
2019 if (error)
2020 return error;
2021 if (!ext4_has_inline_data(inode))
2022 return 0;
2023 }
2024
2025 needed_blocks = ext4_writepage_trans_blocks(inode);
2026
2027 iloc.bh = NULL;
2028 error = ext4_get_inode_loc(inode, &iloc);
2029 if (error)
2030 return error;
2031
2032 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2033 if (IS_ERR(handle)) {
2034 error = PTR_ERR(handle);
2035 goto out_free;
2036 }
2037
2038 ext4_write_lock_xattr(inode, &no_expand);
2039 if (ext4_has_inline_data(inode))
2040 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2041 ext4_write_unlock_xattr(inode, &no_expand);
2042 ext4_journal_stop(handle);
2043out_free:
2044 brelse(iloc.bh);
2045 return error;
2046}