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/iversion.h>
10
11#include "ext4_jbd2.h"
12#include "ext4.h"
13#include "xattr.h"
14#include "truncate.h"
15
16#define EXT4_XATTR_SYSTEM_DATA "data"
17#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
18#define EXT4_INLINE_DOTDOT_OFFSET 2
19#define EXT4_INLINE_DOTDOT_SIZE 4
20
21static int ext4_get_inline_size(struct inode *inode)
22{
23 if (EXT4_I(inode)->i_inline_off)
24 return EXT4_I(inode)->i_inline_size;
25
26 return 0;
27}
28
29static int get_max_inline_xattr_value_size(struct inode *inode,
30 struct ext4_iloc *iloc)
31{
32 struct ext4_xattr_ibody_header *header;
33 struct ext4_xattr_entry *entry;
34 struct ext4_inode *raw_inode;
35 int free, min_offs;
36
37 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
38 EXT4_GOOD_OLD_INODE_SIZE -
39 EXT4_I(inode)->i_extra_isize -
40 sizeof(struct ext4_xattr_ibody_header);
41
42 /*
43 * We need to subtract another sizeof(__u32) since an in-inode xattr
44 * needs an empty 4 bytes to indicate the gap between the xattr entry
45 * and the name/value pair.
46 */
47 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
48 return EXT4_XATTR_SIZE(min_offs -
49 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
50 EXT4_XATTR_ROUND - sizeof(__u32));
51
52 raw_inode = ext4_raw_inode(iloc);
53 header = IHDR(inode, raw_inode);
54 entry = IFIRST(header);
55
56 /* Compute min_offs. */
57 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
58 if (!entry->e_value_inum && entry->e_value_size) {
59 size_t offs = le16_to_cpu(entry->e_value_offs);
60 if (offs < min_offs)
61 min_offs = offs;
62 }
63 }
64 free = min_offs -
65 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
66
67 if (EXT4_I(inode)->i_inline_off) {
68 entry = (struct ext4_xattr_entry *)
69 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
70
71 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
72 goto out;
73 }
74
75 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
76
77 if (free > EXT4_XATTR_ROUND)
78 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
79 else
80 free = 0;
81
82out:
83 return free;
84}
85
86/*
87 * Get the maximum size we now can store in an inode.
88 * If we can't find the space for a xattr entry, don't use the space
89 * of the extents since we have no space to indicate the inline data.
90 */
91int ext4_get_max_inline_size(struct inode *inode)
92{
93 int error, max_inline_size;
94 struct ext4_iloc iloc;
95
96 if (EXT4_I(inode)->i_extra_isize == 0)
97 return 0;
98
99 error = ext4_get_inode_loc(inode, &iloc);
100 if (error) {
101 ext4_error_inode(inode, __func__, __LINE__, 0,
102 "can't get inode location %lu",
103 inode->i_ino);
104 return 0;
105 }
106
107 down_read(&EXT4_I(inode)->xattr_sem);
108 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
109 up_read(&EXT4_I(inode)->xattr_sem);
110
111 brelse(iloc.bh);
112
113 if (!max_inline_size)
114 return 0;
115
116 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
117}
118
119/*
120 * this function does not take xattr_sem, which is OK because it is
121 * currently only used in a code path coming form ext4_iget, before
122 * the new inode has been unlocked
123 */
124int ext4_find_inline_data_nolock(struct inode *inode)
125{
126 struct ext4_xattr_ibody_find is = {
127 .s = { .not_found = -ENODATA, },
128 };
129 struct ext4_xattr_info i = {
130 .name_index = EXT4_XATTR_INDEX_SYSTEM,
131 .name = EXT4_XATTR_SYSTEM_DATA,
132 };
133 int error;
134
135 if (EXT4_I(inode)->i_extra_isize == 0)
136 return 0;
137
138 error = ext4_get_inode_loc(inode, &is.iloc);
139 if (error)
140 return error;
141
142 error = ext4_xattr_ibody_find(inode, &i, &is);
143 if (error)
144 goto out;
145
146 if (!is.s.not_found) {
147 if (is.s.here->e_value_inum) {
148 EXT4_ERROR_INODE(inode, "inline data xattr refers "
149 "to an external xattr inode");
150 error = -EFSCORRUPTED;
151 goto out;
152 }
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 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
219 return;
220
221 BUG_ON(!EXT4_I(inode)->i_inline_off);
222 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
223
224 raw_inode = ext4_raw_inode(iloc);
225 buffer += pos;
226
227 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
228 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
229 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
230 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
231
232 len -= cp_len;
233 buffer += cp_len;
234 pos += cp_len;
235 }
236
237 if (!len)
238 return;
239
240 pos -= EXT4_MIN_INLINE_DATA_SIZE;
241 header = IHDR(inode, raw_inode);
242 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
243 EXT4_I(inode)->i_inline_off);
244
245 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
246 buffer, len);
247}
248
249static int ext4_create_inline_data(handle_t *handle,
250 struct inode *inode, unsigned len)
251{
252 int error;
253 void *value = NULL;
254 struct ext4_xattr_ibody_find is = {
255 .s = { .not_found = -ENODATA, },
256 };
257 struct ext4_xattr_info i = {
258 .name_index = EXT4_XATTR_INDEX_SYSTEM,
259 .name = EXT4_XATTR_SYSTEM_DATA,
260 };
261
262 error = ext4_get_inode_loc(inode, &is.iloc);
263 if (error)
264 return error;
265
266 BUFFER_TRACE(is.iloc.bh, "get_write_access");
267 error = ext4_journal_get_write_access(handle, is.iloc.bh);
268 if (error)
269 goto out;
270
271 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272 value = EXT4_ZERO_XATTR_VALUE;
273 len -= EXT4_MIN_INLINE_DATA_SIZE;
274 } else {
275 value = "";
276 len = 0;
277 }
278
279 /* Insert the the xttr entry. */
280 i.value = value;
281 i.value_len = len;
282
283 error = ext4_xattr_ibody_find(inode, &i, &is);
284 if (error)
285 goto out;
286
287 BUG_ON(!is.s.not_found);
288
289 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290 if (error) {
291 if (error == -ENOSPC)
292 ext4_clear_inode_state(inode,
293 EXT4_STATE_MAY_INLINE_DATA);
294 goto out;
295 }
296
297 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301 (void *)ext4_raw_inode(&is.iloc));
302 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305 get_bh(is.iloc.bh);
306 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
307
308out:
309 brelse(is.iloc.bh);
310 return error;
311}
312
313static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
314 unsigned int len)
315{
316 int error;
317 void *value = NULL;
318 struct ext4_xattr_ibody_find is = {
319 .s = { .not_found = -ENODATA, },
320 };
321 struct ext4_xattr_info i = {
322 .name_index = EXT4_XATTR_INDEX_SYSTEM,
323 .name = EXT4_XATTR_SYSTEM_DATA,
324 };
325
326 /* If the old space is ok, write the data directly. */
327 if (len <= EXT4_I(inode)->i_inline_size)
328 return 0;
329
330 error = ext4_get_inode_loc(inode, &is.iloc);
331 if (error)
332 return error;
333
334 error = ext4_xattr_ibody_find(inode, &i, &is);
335 if (error)
336 goto out;
337
338 BUG_ON(is.s.not_found);
339
340 len -= EXT4_MIN_INLINE_DATA_SIZE;
341 value = kzalloc(len, GFP_NOFS);
342 if (!value) {
343 error = -ENOMEM;
344 goto out;
345 }
346
347 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
348 value, len);
349 if (error == -ENODATA)
350 goto out;
351
352 BUFFER_TRACE(is.iloc.bh, "get_write_access");
353 error = ext4_journal_get_write_access(handle, is.iloc.bh);
354 if (error)
355 goto out;
356
357 /* Update the xttr entry. */
358 i.value = value;
359 i.value_len = len;
360
361 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
362 if (error)
363 goto out;
364
365 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
366 (void *)ext4_raw_inode(&is.iloc));
367 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
368 le32_to_cpu(is.s.here->e_value_size);
369 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
370 get_bh(is.iloc.bh);
371 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
372
373out:
374 kfree(value);
375 brelse(is.iloc.bh);
376 return error;
377}
378
379static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
380 unsigned int len)
381{
382 int ret, size, no_expand;
383 struct ext4_inode_info *ei = EXT4_I(inode);
384
385 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
386 return -ENOSPC;
387
388 size = ext4_get_max_inline_size(inode);
389 if (size < len)
390 return -ENOSPC;
391
392 ext4_write_lock_xattr(inode, &no_expand);
393
394 if (ei->i_inline_off)
395 ret = ext4_update_inline_data(handle, inode, len);
396 else
397 ret = ext4_create_inline_data(handle, inode, len);
398
399 ext4_write_unlock_xattr(inode, &no_expand);
400 return ret;
401}
402
403static int ext4_destroy_inline_data_nolock(handle_t *handle,
404 struct inode *inode)
405{
406 struct ext4_inode_info *ei = EXT4_I(inode);
407 struct ext4_xattr_ibody_find is = {
408 .s = { .not_found = 0, },
409 };
410 struct ext4_xattr_info i = {
411 .name_index = EXT4_XATTR_INDEX_SYSTEM,
412 .name = EXT4_XATTR_SYSTEM_DATA,
413 .value = NULL,
414 .value_len = 0,
415 };
416 int error;
417
418 if (!ei->i_inline_off)
419 return 0;
420
421 error = ext4_get_inode_loc(inode, &is.iloc);
422 if (error)
423 return error;
424
425 error = ext4_xattr_ibody_find(inode, &i, &is);
426 if (error)
427 goto out;
428
429 BUFFER_TRACE(is.iloc.bh, "get_write_access");
430 error = ext4_journal_get_write_access(handle, is.iloc.bh);
431 if (error)
432 goto out;
433
434 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
435 if (error)
436 goto out;
437
438 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
439 0, EXT4_MIN_INLINE_DATA_SIZE);
440 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
441
442 if (ext4_has_feature_extents(inode->i_sb)) {
443 if (S_ISDIR(inode->i_mode) ||
444 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
445 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
446 ext4_ext_tree_init(handle, inode);
447 }
448 }
449 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
450
451 get_bh(is.iloc.bh);
452 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
453
454 EXT4_I(inode)->i_inline_off = 0;
455 EXT4_I(inode)->i_inline_size = 0;
456 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
457out:
458 brelse(is.iloc.bh);
459 if (error == -ENODATA)
460 error = 0;
461 return error;
462}
463
464static int ext4_read_inline_page(struct inode *inode, struct page *page)
465{
466 void *kaddr;
467 int ret = 0;
468 size_t len;
469 struct ext4_iloc iloc;
470
471 BUG_ON(!PageLocked(page));
472 BUG_ON(!ext4_has_inline_data(inode));
473 BUG_ON(page->index);
474
475 if (!EXT4_I(inode)->i_inline_off) {
476 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
477 inode->i_ino);
478 goto out;
479 }
480
481 ret = ext4_get_inode_loc(inode, &iloc);
482 if (ret)
483 goto out;
484
485 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
486 kaddr = kmap_atomic(page);
487 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
488 flush_dcache_page(page);
489 kunmap_atomic(kaddr);
490 zero_user_segment(page, len, PAGE_SIZE);
491 SetPageUptodate(page);
492 brelse(iloc.bh);
493
494out:
495 return ret;
496}
497
498int ext4_readpage_inline(struct inode *inode, struct page *page)
499{
500 int ret = 0;
501
502 down_read(&EXT4_I(inode)->xattr_sem);
503 if (!ext4_has_inline_data(inode)) {
504 up_read(&EXT4_I(inode)->xattr_sem);
505 return -EAGAIN;
506 }
507
508 /*
509 * Current inline data can only exist in the 1st page,
510 * So for all the other pages, just set them uptodate.
511 */
512 if (!page->index)
513 ret = ext4_read_inline_page(inode, page);
514 else if (!PageUptodate(page)) {
515 zero_user_segment(page, 0, PAGE_SIZE);
516 SetPageUptodate(page);
517 }
518
519 up_read(&EXT4_I(inode)->xattr_sem);
520
521 unlock_page(page);
522 return ret >= 0 ? 0 : ret;
523}
524
525static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
526 struct inode *inode,
527 unsigned flags)
528{
529 int ret, needed_blocks, no_expand;
530 handle_t *handle = NULL;
531 int retries = 0, sem_held = 0;
532 struct page *page = NULL;
533 unsigned from, to;
534 struct ext4_iloc iloc;
535
536 if (!ext4_has_inline_data(inode)) {
537 /*
538 * clear the flag so that no new write
539 * will trap here again.
540 */
541 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
542 return 0;
543 }
544
545 needed_blocks = ext4_writepage_trans_blocks(inode);
546
547 ret = ext4_get_inode_loc(inode, &iloc);
548 if (ret)
549 return ret;
550
551retry:
552 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
553 if (IS_ERR(handle)) {
554 ret = PTR_ERR(handle);
555 handle = NULL;
556 goto out;
557 }
558
559 /* We cannot recurse into the filesystem as the transaction is already
560 * started */
561 flags |= AOP_FLAG_NOFS;
562
563 page = grab_cache_page_write_begin(mapping, 0, flags);
564 if (!page) {
565 ret = -ENOMEM;
566 goto out;
567 }
568
569 ext4_write_lock_xattr(inode, &no_expand);
570 sem_held = 1;
571 /* If some one has already done this for us, just exit. */
572 if (!ext4_has_inline_data(inode)) {
573 ret = 0;
574 goto out;
575 }
576
577 from = 0;
578 to = ext4_get_inline_size(inode);
579 if (!PageUptodate(page)) {
580 ret = ext4_read_inline_page(inode, page);
581 if (ret < 0)
582 goto out;
583 }
584
585 ret = ext4_destroy_inline_data_nolock(handle, inode);
586 if (ret)
587 goto out;
588
589 if (ext4_should_dioread_nolock(inode)) {
590 ret = __block_write_begin(page, from, to,
591 ext4_get_block_unwritten);
592 } else
593 ret = __block_write_begin(page, from, to, ext4_get_block);
594
595 if (!ret && ext4_should_journal_data(inode)) {
596 ret = ext4_walk_page_buffers(handle, page_buffers(page),
597 from, to, NULL,
598 do_journal_get_write_access);
599 }
600
601 if (ret) {
602 unlock_page(page);
603 put_page(page);
604 page = NULL;
605 ext4_orphan_add(handle, inode);
606 ext4_write_unlock_xattr(inode, &no_expand);
607 sem_held = 0;
608 ext4_journal_stop(handle);
609 handle = NULL;
610 ext4_truncate_failed_write(inode);
611 /*
612 * If truncate failed early the inode might
613 * still be on the orphan list; we need to
614 * make sure the inode is removed from the
615 * orphan list in that case.
616 */
617 if (inode->i_nlink)
618 ext4_orphan_del(NULL, inode);
619 }
620
621 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
622 goto retry;
623
624 if (page)
625 block_commit_write(page, from, to);
626out:
627 if (page) {
628 unlock_page(page);
629 put_page(page);
630 }
631 if (sem_held)
632 ext4_write_unlock_xattr(inode, &no_expand);
633 if (handle)
634 ext4_journal_stop(handle);
635 brelse(iloc.bh);
636 return ret;
637}
638
639/*
640 * Try to write data in the inode.
641 * If the inode has inline data, check whether the new write can be
642 * in the inode also. If not, create the page the handle, move the data
643 * to the page make it update and let the later codes create extent for it.
644 */
645int ext4_try_to_write_inline_data(struct address_space *mapping,
646 struct inode *inode,
647 loff_t pos, unsigned len,
648 unsigned flags,
649 struct page **pagep)
650{
651 int ret;
652 handle_t *handle;
653 struct page *page;
654 struct ext4_iloc iloc;
655
656 if (pos + len > ext4_get_max_inline_size(inode))
657 goto convert;
658
659 ret = ext4_get_inode_loc(inode, &iloc);
660 if (ret)
661 return ret;
662
663 /*
664 * The possible write could happen in the inode,
665 * so try to reserve the space in inode first.
666 */
667 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
668 if (IS_ERR(handle)) {
669 ret = PTR_ERR(handle);
670 handle = NULL;
671 goto out;
672 }
673
674 ret = ext4_prepare_inline_data(handle, inode, pos + len);
675 if (ret && ret != -ENOSPC)
676 goto out;
677
678 /* We don't have space in inline inode, so convert it to extent. */
679 if (ret == -ENOSPC) {
680 ext4_journal_stop(handle);
681 brelse(iloc.bh);
682 goto convert;
683 }
684
685 ret = ext4_journal_get_write_access(handle, iloc.bh);
686 if (ret)
687 goto out;
688
689 flags |= AOP_FLAG_NOFS;
690
691 page = grab_cache_page_write_begin(mapping, 0, flags);
692 if (!page) {
693 ret = -ENOMEM;
694 goto out;
695 }
696
697 *pagep = page;
698 down_read(&EXT4_I(inode)->xattr_sem);
699 if (!ext4_has_inline_data(inode)) {
700 ret = 0;
701 unlock_page(page);
702 put_page(page);
703 goto out_up_read;
704 }
705
706 if (!PageUptodate(page)) {
707 ret = ext4_read_inline_page(inode, page);
708 if (ret < 0) {
709 unlock_page(page);
710 put_page(page);
711 goto out_up_read;
712 }
713 }
714
715 ret = 1;
716 handle = NULL;
717out_up_read:
718 up_read(&EXT4_I(inode)->xattr_sem);
719out:
720 if (handle && (ret != 1))
721 ext4_journal_stop(handle);
722 brelse(iloc.bh);
723 return ret;
724convert:
725 return ext4_convert_inline_data_to_extent(mapping,
726 inode, flags);
727}
728
729int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
730 unsigned copied, struct page *page)
731{
732 int ret, no_expand;
733 void *kaddr;
734 struct ext4_iloc iloc;
735
736 if (unlikely(copied < len)) {
737 if (!PageUptodate(page)) {
738 copied = 0;
739 goto out;
740 }
741 }
742
743 ret = ext4_get_inode_loc(inode, &iloc);
744 if (ret) {
745 ext4_std_error(inode->i_sb, ret);
746 copied = 0;
747 goto out;
748 }
749
750 ext4_write_lock_xattr(inode, &no_expand);
751 BUG_ON(!ext4_has_inline_data(inode));
752
753 kaddr = kmap_atomic(page);
754 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
755 kunmap_atomic(kaddr);
756 SetPageUptodate(page);
757 /* clear page dirty so that writepages wouldn't work for us. */
758 ClearPageDirty(page);
759
760 ext4_write_unlock_xattr(inode, &no_expand);
761 brelse(iloc.bh);
762 mark_inode_dirty(inode);
763out:
764 return copied;
765}
766
767struct buffer_head *
768ext4_journalled_write_inline_data(struct inode *inode,
769 unsigned len,
770 struct page *page)
771{
772 int ret, no_expand;
773 void *kaddr;
774 struct ext4_iloc iloc;
775
776 ret = ext4_get_inode_loc(inode, &iloc);
777 if (ret) {
778 ext4_std_error(inode->i_sb, ret);
779 return NULL;
780 }
781
782 ext4_write_lock_xattr(inode, &no_expand);
783 kaddr = kmap_atomic(page);
784 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
785 kunmap_atomic(kaddr);
786 ext4_write_unlock_xattr(inode, &no_expand);
787
788 return iloc.bh;
789}
790
791/*
792 * Try to make the page cache and handle ready for the inline data case.
793 * We can call this function in 2 cases:
794 * 1. The inode is created and the first write exceeds inline size. We can
795 * clear the inode state safely.
796 * 2. The inode has inline data, then we need to read the data, make it
797 * update and dirty so that ext4_da_writepages can handle it. We don't
798 * need to start the journal since the file's metatdata isn't changed now.
799 */
800static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
801 struct inode *inode,
802 unsigned flags,
803 void **fsdata)
804{
805 int ret = 0, inline_size;
806 struct page *page;
807
808 page = grab_cache_page_write_begin(mapping, 0, flags);
809 if (!page)
810 return -ENOMEM;
811
812 down_read(&EXT4_I(inode)->xattr_sem);
813 if (!ext4_has_inline_data(inode)) {
814 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
815 goto out;
816 }
817
818 inline_size = ext4_get_inline_size(inode);
819
820 if (!PageUptodate(page)) {
821 ret = ext4_read_inline_page(inode, page);
822 if (ret < 0)
823 goto out;
824 }
825
826 ret = __block_write_begin(page, 0, inline_size,
827 ext4_da_get_block_prep);
828 if (ret) {
829 up_read(&EXT4_I(inode)->xattr_sem);
830 unlock_page(page);
831 put_page(page);
832 ext4_truncate_failed_write(inode);
833 return ret;
834 }
835
836 SetPageDirty(page);
837 SetPageUptodate(page);
838 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
839 *fsdata = (void *)CONVERT_INLINE_DATA;
840
841out:
842 up_read(&EXT4_I(inode)->xattr_sem);
843 if (page) {
844 unlock_page(page);
845 put_page(page);
846 }
847 return ret;
848}
849
850/*
851 * Prepare the write for the inline data.
852 * If the the data can be written into the inode, we just read
853 * the page and make it uptodate, and start the journal.
854 * Otherwise read the page, makes it dirty so that it can be
855 * handle in writepages(the i_disksize update is left to the
856 * normal ext4_da_write_end).
857 */
858int ext4_da_write_inline_data_begin(struct address_space *mapping,
859 struct inode *inode,
860 loff_t pos, unsigned len,
861 unsigned flags,
862 struct page **pagep,
863 void **fsdata)
864{
865 int ret, inline_size;
866 handle_t *handle;
867 struct page *page;
868 struct ext4_iloc iloc;
869 int retries = 0;
870
871 ret = ext4_get_inode_loc(inode, &iloc);
872 if (ret)
873 return ret;
874
875retry_journal:
876 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
877 if (IS_ERR(handle)) {
878 ret = PTR_ERR(handle);
879 goto out;
880 }
881
882 inline_size = ext4_get_max_inline_size(inode);
883
884 ret = -ENOSPC;
885 if (inline_size >= pos + len) {
886 ret = ext4_prepare_inline_data(handle, inode, pos + len);
887 if (ret && ret != -ENOSPC)
888 goto out_journal;
889 }
890
891 /*
892 * We cannot recurse into the filesystem as the transaction
893 * is already started.
894 */
895 flags |= AOP_FLAG_NOFS;
896
897 if (ret == -ENOSPC) {
898 ext4_journal_stop(handle);
899 ret = ext4_da_convert_inline_data_to_extent(mapping,
900 inode,
901 flags,
902 fsdata);
903 if (ret == -ENOSPC &&
904 ext4_should_retry_alloc(inode->i_sb, &retries))
905 goto retry_journal;
906 goto out;
907 }
908
909 page = grab_cache_page_write_begin(mapping, 0, flags);
910 if (!page) {
911 ret = -ENOMEM;
912 goto out_journal;
913 }
914
915 down_read(&EXT4_I(inode)->xattr_sem);
916 if (!ext4_has_inline_data(inode)) {
917 ret = 0;
918 goto out_release_page;
919 }
920
921 if (!PageUptodate(page)) {
922 ret = ext4_read_inline_page(inode, page);
923 if (ret < 0)
924 goto out_release_page;
925 }
926 ret = ext4_journal_get_write_access(handle, iloc.bh);
927 if (ret)
928 goto out_release_page;
929
930 up_read(&EXT4_I(inode)->xattr_sem);
931 *pagep = page;
932 brelse(iloc.bh);
933 return 1;
934out_release_page:
935 up_read(&EXT4_I(inode)->xattr_sem);
936 unlock_page(page);
937 put_page(page);
938out_journal:
939 ext4_journal_stop(handle);
940out:
941 brelse(iloc.bh);
942 return ret;
943}
944
945int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
946 unsigned len, unsigned copied,
947 struct page *page)
948{
949 int ret;
950
951 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
952 if (ret < 0) {
953 unlock_page(page);
954 put_page(page);
955 return ret;
956 }
957 copied = ret;
958
959 /*
960 * No need to use i_size_read() here, the i_size
961 * cannot change under us because we hold i_mutex.
962 *
963 * But it's important to update i_size while still holding page lock:
964 * page writeout could otherwise come in and zero beyond i_size.
965 */
966 if (pos+copied > inode->i_size)
967 i_size_write(inode, pos+copied);
968 unlock_page(page);
969 put_page(page);
970
971 /*
972 * Don't mark the inode dirty under page lock. First, it unnecessarily
973 * makes the holding time of page lock longer. Second, it forces lock
974 * ordering of page lock and transaction start for journaling
975 * filesystems.
976 */
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(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 inode_inc_iversion(dir);
1052 return 1;
1053}
1054
1055static void *ext4_get_inline_xattr_pos(struct inode *inode,
1056 struct ext4_iloc *iloc)
1057{
1058 struct ext4_xattr_entry *entry;
1059 struct ext4_xattr_ibody_header *header;
1060
1061 BUG_ON(!EXT4_I(inode)->i_inline_off);
1062
1063 header = IHDR(inode, ext4_raw_inode(iloc));
1064 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1065 EXT4_I(inode)->i_inline_off);
1066
1067 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1068}
1069
1070/* Set the final de to cover the whole block. */
1071static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1072{
1073 struct ext4_dir_entry_2 *de, *prev_de;
1074 void *limit;
1075 int de_len;
1076
1077 de = (struct ext4_dir_entry_2 *)de_buf;
1078 if (old_size) {
1079 limit = de_buf + old_size;
1080 do {
1081 prev_de = de;
1082 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1083 de_buf += de_len;
1084 de = (struct ext4_dir_entry_2 *)de_buf;
1085 } while (de_buf < limit);
1086
1087 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1088 old_size, new_size);
1089 } else {
1090 /* this is just created, so create an empty entry. */
1091 de->inode = 0;
1092 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1093 }
1094}
1095
1096static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1097 struct ext4_iloc *iloc)
1098{
1099 int ret;
1100 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1101 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1102
1103 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1104 return -ENOSPC;
1105
1106 ret = ext4_update_inline_data(handle, dir,
1107 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1108 if (ret)
1109 return ret;
1110
1111 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1112 EXT4_I(dir)->i_inline_size -
1113 EXT4_MIN_INLINE_DATA_SIZE);
1114 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1115 return 0;
1116}
1117
1118static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1119 struct ext4_iloc *iloc,
1120 void *buf, int inline_size)
1121{
1122 ext4_create_inline_data(handle, inode, inline_size);
1123 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1124 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1125}
1126
1127static int ext4_finish_convert_inline_dir(handle_t *handle,
1128 struct inode *inode,
1129 struct buffer_head *dir_block,
1130 void *buf,
1131 int inline_size)
1132{
1133 int err, csum_size = 0, header_size = 0;
1134 struct ext4_dir_entry_2 *de;
1135 void *target = dir_block->b_data;
1136
1137 /*
1138 * First create "." and ".." and then copy the dir information
1139 * back to the block.
1140 */
1141 de = (struct ext4_dir_entry_2 *)target;
1142 de = ext4_init_dot_dotdot(inode, de,
1143 inode->i_sb->s_blocksize, csum_size,
1144 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1145 header_size = (void *)de - target;
1146
1147 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1148 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1149
1150 if (ext4_has_metadata_csum(inode->i_sb))
1151 csum_size = sizeof(struct ext4_dir_entry_tail);
1152
1153 inode->i_size = inode->i_sb->s_blocksize;
1154 i_size_write(inode, inode->i_sb->s_blocksize);
1155 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1156 ext4_update_final_de(dir_block->b_data,
1157 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1158 inode->i_sb->s_blocksize - csum_size);
1159
1160 if (csum_size)
1161 ext4_initialize_dirent_tail(dir_block,
1162 inode->i_sb->s_blocksize);
1163 set_buffer_uptodate(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, data_bh);
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 error = ext4_handle_dirty_metadata(handle,
1238 inode, data_bh);
1239 } else {
1240 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1241 buf, inline_size);
1242 }
1243
1244 unlock_buffer(data_bh);
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, 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 ext4_mark_inode_dirty(handle, dir);
1318 brelse(iloc.bh);
1319 return ret;
1320}
1321
1322/*
1323 * This function fills a red-black tree with information from an
1324 * inlined dir. It returns the number directory entries loaded
1325 * into the tree. If there is an error it is returned in err.
1326 */
1327int ext4_inlinedir_to_tree(struct file *dir_file,
1328 struct inode *dir, ext4_lblk_t block,
1329 struct dx_hash_info *hinfo,
1330 __u32 start_hash, __u32 start_minor_hash,
1331 int *has_inline_data)
1332{
1333 int err = 0, count = 0;
1334 unsigned int parent_ino;
1335 int pos;
1336 struct ext4_dir_entry_2 *de;
1337 struct inode *inode = file_inode(dir_file);
1338 int ret, inline_size = 0;
1339 struct ext4_iloc iloc;
1340 void *dir_buf = NULL;
1341 struct ext4_dir_entry_2 fake;
1342 struct fscrypt_str tmp_str;
1343
1344 ret = ext4_get_inode_loc(inode, &iloc);
1345 if (ret)
1346 return ret;
1347
1348 down_read(&EXT4_I(inode)->xattr_sem);
1349 if (!ext4_has_inline_data(inode)) {
1350 up_read(&EXT4_I(inode)->xattr_sem);
1351 *has_inline_data = 0;
1352 goto out;
1353 }
1354
1355 inline_size = ext4_get_inline_size(inode);
1356 dir_buf = kmalloc(inline_size, GFP_NOFS);
1357 if (!dir_buf) {
1358 ret = -ENOMEM;
1359 up_read(&EXT4_I(inode)->xattr_sem);
1360 goto out;
1361 }
1362
1363 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1364 up_read(&EXT4_I(inode)->xattr_sem);
1365 if (ret < 0)
1366 goto out;
1367
1368 pos = 0;
1369 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1370 while (pos < inline_size) {
1371 /*
1372 * As inlined dir doesn't store any information about '.' and
1373 * only the inode number of '..' is stored, we have to handle
1374 * them differently.
1375 */
1376 if (pos == 0) {
1377 fake.inode = cpu_to_le32(inode->i_ino);
1378 fake.name_len = 1;
1379 strcpy(fake.name, ".");
1380 fake.rec_len = ext4_rec_len_to_disk(
1381 EXT4_DIR_REC_LEN(fake.name_len),
1382 inline_size);
1383 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1384 de = &fake;
1385 pos = EXT4_INLINE_DOTDOT_OFFSET;
1386 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1387 fake.inode = cpu_to_le32(parent_ino);
1388 fake.name_len = 2;
1389 strcpy(fake.name, "..");
1390 fake.rec_len = ext4_rec_len_to_disk(
1391 EXT4_DIR_REC_LEN(fake.name_len),
1392 inline_size);
1393 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1394 de = &fake;
1395 pos = EXT4_INLINE_DOTDOT_SIZE;
1396 } else {
1397 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1398 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1399 if (ext4_check_dir_entry(inode, dir_file, de,
1400 iloc.bh, dir_buf,
1401 inline_size, pos)) {
1402 ret = count;
1403 goto out;
1404 }
1405 }
1406
1407 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1408 if ((hinfo->hash < start_hash) ||
1409 ((hinfo->hash == start_hash) &&
1410 (hinfo->minor_hash < start_minor_hash)))
1411 continue;
1412 if (de->inode == 0)
1413 continue;
1414 tmp_str.name = de->name;
1415 tmp_str.len = de->name_len;
1416 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1417 hinfo->minor_hash, de, &tmp_str);
1418 if (err) {
1419 ret = err;
1420 goto out;
1421 }
1422 count++;
1423 }
1424 ret = count;
1425out:
1426 kfree(dir_buf);
1427 brelse(iloc.bh);
1428 return ret;
1429}
1430
1431/*
1432 * So this function is called when the volume is mkfsed with
1433 * dir_index disabled. In order to keep f_pos persistent
1434 * after we convert from an inlined dir to a blocked based,
1435 * we just pretend that we are a normal dir and return the
1436 * offset as if '.' and '..' really take place.
1437 *
1438 */
1439int ext4_read_inline_dir(struct file *file,
1440 struct dir_context *ctx,
1441 int *has_inline_data)
1442{
1443 unsigned int offset, parent_ino;
1444 int i;
1445 struct ext4_dir_entry_2 *de;
1446 struct super_block *sb;
1447 struct inode *inode = file_inode(file);
1448 int ret, inline_size = 0;
1449 struct ext4_iloc iloc;
1450 void *dir_buf = NULL;
1451 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1452
1453 ret = ext4_get_inode_loc(inode, &iloc);
1454 if (ret)
1455 return ret;
1456
1457 down_read(&EXT4_I(inode)->xattr_sem);
1458 if (!ext4_has_inline_data(inode)) {
1459 up_read(&EXT4_I(inode)->xattr_sem);
1460 *has_inline_data = 0;
1461 goto out;
1462 }
1463
1464 inline_size = ext4_get_inline_size(inode);
1465 dir_buf = kmalloc(inline_size, GFP_NOFS);
1466 if (!dir_buf) {
1467 ret = -ENOMEM;
1468 up_read(&EXT4_I(inode)->xattr_sem);
1469 goto out;
1470 }
1471
1472 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1473 up_read(&EXT4_I(inode)->xattr_sem);
1474 if (ret < 0)
1475 goto out;
1476
1477 ret = 0;
1478 sb = inode->i_sb;
1479 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1480 offset = ctx->pos;
1481
1482 /*
1483 * dotdot_offset and dotdot_size is the real offset and
1484 * size for ".." and "." if the dir is block based while
1485 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1486 * So we will use extra_offset and extra_size to indicate them
1487 * during the inline dir iteration.
1488 */
1489 dotdot_offset = EXT4_DIR_REC_LEN(1);
1490 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1491 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1492 extra_size = extra_offset + inline_size;
1493
1494 /*
1495 * If the version has changed since the last call to
1496 * readdir(2), then we might be pointing to an invalid
1497 * dirent right now. Scan from the start of the inline
1498 * dir to make sure.
1499 */
1500 if (!inode_eq_iversion(inode, file->f_version)) {
1501 for (i = 0; i < extra_size && i < offset;) {
1502 /*
1503 * "." is with offset 0 and
1504 * ".." is dotdot_offset.
1505 */
1506 if (!i) {
1507 i = dotdot_offset;
1508 continue;
1509 } else if (i == dotdot_offset) {
1510 i = dotdot_size;
1511 continue;
1512 }
1513 /* for other entry, the real offset in
1514 * the buf has to be tuned accordingly.
1515 */
1516 de = (struct ext4_dir_entry_2 *)
1517 (dir_buf + i - extra_offset);
1518 /* It's too expensive to do a full
1519 * dirent test each time round this
1520 * loop, but we do have to test at
1521 * least that it is non-zero. A
1522 * failure will be detected in the
1523 * dirent test below. */
1524 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1525 < EXT4_DIR_REC_LEN(1))
1526 break;
1527 i += ext4_rec_len_from_disk(de->rec_len,
1528 extra_size);
1529 }
1530 offset = i;
1531 ctx->pos = offset;
1532 file->f_version = inode_query_iversion(inode);
1533 }
1534
1535 while (ctx->pos < extra_size) {
1536 if (ctx->pos == 0) {
1537 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1538 goto out;
1539 ctx->pos = dotdot_offset;
1540 continue;
1541 }
1542
1543 if (ctx->pos == dotdot_offset) {
1544 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1545 goto out;
1546 ctx->pos = dotdot_size;
1547 continue;
1548 }
1549
1550 de = (struct ext4_dir_entry_2 *)
1551 (dir_buf + ctx->pos - extra_offset);
1552 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1553 extra_size, ctx->pos))
1554 goto out;
1555 if (le32_to_cpu(de->inode)) {
1556 if (!dir_emit(ctx, de->name, de->name_len,
1557 le32_to_cpu(de->inode),
1558 get_dtype(sb, de->file_type)))
1559 goto out;
1560 }
1561 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1562 }
1563out:
1564 kfree(dir_buf);
1565 brelse(iloc.bh);
1566 return ret;
1567}
1568
1569struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1570 struct ext4_dir_entry_2 **parent_de,
1571 int *retval)
1572{
1573 struct ext4_iloc iloc;
1574
1575 *retval = ext4_get_inode_loc(inode, &iloc);
1576 if (*retval)
1577 return NULL;
1578
1579 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1580
1581 return iloc.bh;
1582}
1583
1584/*
1585 * Try to create the inline data for the new dir.
1586 * If it succeeds, return 0, otherwise return the error.
1587 * In case of ENOSPC, the caller should create the normal disk layout dir.
1588 */
1589int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1590 struct inode *inode)
1591{
1592 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1593 struct ext4_iloc iloc;
1594 struct ext4_dir_entry_2 *de;
1595
1596 ret = ext4_get_inode_loc(inode, &iloc);
1597 if (ret)
1598 return ret;
1599
1600 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1601 if (ret)
1602 goto out;
1603
1604 /*
1605 * For inline dir, we only save the inode information for the ".."
1606 * and create a fake dentry to cover the left space.
1607 */
1608 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1609 de->inode = cpu_to_le32(parent->i_ino);
1610 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1611 de->inode = 0;
1612 de->rec_len = ext4_rec_len_to_disk(
1613 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1614 inline_size);
1615 set_nlink(inode, 2);
1616 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1617out:
1618 brelse(iloc.bh);
1619 return ret;
1620}
1621
1622struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1623 struct ext4_filename *fname,
1624 struct ext4_dir_entry_2 **res_dir,
1625 int *has_inline_data)
1626{
1627 int ret;
1628 struct ext4_iloc iloc;
1629 void *inline_start;
1630 int inline_size;
1631
1632 if (ext4_get_inode_loc(dir, &iloc))
1633 return NULL;
1634
1635 down_read(&EXT4_I(dir)->xattr_sem);
1636 if (!ext4_has_inline_data(dir)) {
1637 *has_inline_data = 0;
1638 goto out;
1639 }
1640
1641 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1642 EXT4_INLINE_DOTDOT_SIZE;
1643 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1644 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1645 dir, fname, 0, res_dir);
1646 if (ret == 1)
1647 goto out_find;
1648 if (ret < 0)
1649 goto out;
1650
1651 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1652 goto out;
1653
1654 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1655 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1656
1657 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1658 dir, fname, 0, res_dir);
1659 if (ret == 1)
1660 goto out_find;
1661
1662out:
1663 brelse(iloc.bh);
1664 iloc.bh = NULL;
1665out_find:
1666 up_read(&EXT4_I(dir)->xattr_sem);
1667 return iloc.bh;
1668}
1669
1670int ext4_delete_inline_entry(handle_t *handle,
1671 struct inode *dir,
1672 struct ext4_dir_entry_2 *de_del,
1673 struct buffer_head *bh,
1674 int *has_inline_data)
1675{
1676 int err, inline_size, no_expand;
1677 struct ext4_iloc iloc;
1678 void *inline_start;
1679
1680 err = ext4_get_inode_loc(dir, &iloc);
1681 if (err)
1682 return err;
1683
1684 ext4_write_lock_xattr(dir, &no_expand);
1685 if (!ext4_has_inline_data(dir)) {
1686 *has_inline_data = 0;
1687 goto out;
1688 }
1689
1690 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1691 EXT4_MIN_INLINE_DATA_SIZE) {
1692 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1693 EXT4_INLINE_DOTDOT_SIZE;
1694 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1695 EXT4_INLINE_DOTDOT_SIZE;
1696 } else {
1697 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1698 inline_size = ext4_get_inline_size(dir) -
1699 EXT4_MIN_INLINE_DATA_SIZE;
1700 }
1701
1702 BUFFER_TRACE(bh, "get_write_access");
1703 err = ext4_journal_get_write_access(handle, bh);
1704 if (err)
1705 goto out;
1706
1707 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1708 inline_start, inline_size, 0);
1709 if (err)
1710 goto out;
1711
1712 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1713out:
1714 ext4_write_unlock_xattr(dir, &no_expand);
1715 if (likely(err == 0))
1716 err = ext4_mark_inode_dirty(handle, dir);
1717 brelse(iloc.bh);
1718 if (err != -ENOENT)
1719 ext4_std_error(dir->i_sb, err);
1720 return err;
1721}
1722
1723/*
1724 * Get the inline dentry at offset.
1725 */
1726static inline struct ext4_dir_entry_2 *
1727ext4_get_inline_entry(struct inode *inode,
1728 struct ext4_iloc *iloc,
1729 unsigned int offset,
1730 void **inline_start,
1731 int *inline_size)
1732{
1733 void *inline_pos;
1734
1735 BUG_ON(offset > ext4_get_inline_size(inode));
1736
1737 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1738 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1739 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1740 } else {
1741 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1742 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1743 *inline_size = ext4_get_inline_size(inode) -
1744 EXT4_MIN_INLINE_DATA_SIZE;
1745 }
1746
1747 if (inline_start)
1748 *inline_start = inline_pos;
1749 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1750}
1751
1752bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1753{
1754 int err, inline_size;
1755 struct ext4_iloc iloc;
1756 size_t inline_len;
1757 void *inline_pos;
1758 unsigned int offset;
1759 struct ext4_dir_entry_2 *de;
1760 bool ret = true;
1761
1762 err = ext4_get_inode_loc(dir, &iloc);
1763 if (err) {
1764 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1765 err, dir->i_ino);
1766 return true;
1767 }
1768
1769 down_read(&EXT4_I(dir)->xattr_sem);
1770 if (!ext4_has_inline_data(dir)) {
1771 *has_inline_data = 0;
1772 goto out;
1773 }
1774
1775 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1776 if (!le32_to_cpu(de->inode)) {
1777 ext4_warning(dir->i_sb,
1778 "bad inline directory (dir #%lu) - no `..'",
1779 dir->i_ino);
1780 ret = true;
1781 goto out;
1782 }
1783
1784 inline_len = ext4_get_inline_size(dir);
1785 offset = EXT4_INLINE_DOTDOT_SIZE;
1786 while (offset < inline_len) {
1787 de = ext4_get_inline_entry(dir, &iloc, offset,
1788 &inline_pos, &inline_size);
1789 if (ext4_check_dir_entry(dir, NULL, de,
1790 iloc.bh, inline_pos,
1791 inline_size, offset)) {
1792 ext4_warning(dir->i_sb,
1793 "bad inline directory (dir #%lu) - "
1794 "inode %u, rec_len %u, name_len %d"
1795 "inline size %d",
1796 dir->i_ino, le32_to_cpu(de->inode),
1797 le16_to_cpu(de->rec_len), de->name_len,
1798 inline_size);
1799 ret = true;
1800 goto out;
1801 }
1802 if (le32_to_cpu(de->inode)) {
1803 ret = false;
1804 goto out;
1805 }
1806 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1807 }
1808
1809out:
1810 up_read(&EXT4_I(dir)->xattr_sem);
1811 brelse(iloc.bh);
1812 return ret;
1813}
1814
1815int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1816{
1817 int ret, no_expand;
1818
1819 ext4_write_lock_xattr(inode, &no_expand);
1820 ret = ext4_destroy_inline_data_nolock(handle, inode);
1821 ext4_write_unlock_xattr(inode, &no_expand);
1822
1823 return ret;
1824}
1825
1826int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1827{
1828 __u64 addr;
1829 int error = -EAGAIN;
1830 struct ext4_iloc iloc;
1831
1832 down_read(&EXT4_I(inode)->xattr_sem);
1833 if (!ext4_has_inline_data(inode))
1834 goto out;
1835
1836 error = ext4_get_inode_loc(inode, &iloc);
1837 if (error)
1838 goto out;
1839
1840 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1841 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1842 addr += offsetof(struct ext4_inode, i_block);
1843
1844 brelse(iloc.bh);
1845
1846 iomap->addr = addr;
1847 iomap->offset = 0;
1848 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1849 i_size_read(inode));
1850 iomap->type = IOMAP_INLINE;
1851 iomap->flags = 0;
1852
1853out:
1854 up_read(&EXT4_I(inode)->xattr_sem);
1855 return error;
1856}
1857
1858int ext4_inline_data_fiemap(struct inode *inode,
1859 struct fiemap_extent_info *fieinfo,
1860 int *has_inline, __u64 start, __u64 len)
1861{
1862 __u64 physical = 0;
1863 __u64 inline_len;
1864 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1865 FIEMAP_EXTENT_LAST;
1866 int error = 0;
1867 struct ext4_iloc iloc;
1868
1869 down_read(&EXT4_I(inode)->xattr_sem);
1870 if (!ext4_has_inline_data(inode)) {
1871 *has_inline = 0;
1872 goto out;
1873 }
1874 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1875 i_size_read(inode));
1876 if (start >= inline_len)
1877 goto out;
1878 if (start + len < inline_len)
1879 inline_len = start + len;
1880 inline_len -= start;
1881
1882 error = ext4_get_inode_loc(inode, &iloc);
1883 if (error)
1884 goto out;
1885
1886 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1887 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1888 physical += offsetof(struct ext4_inode, i_block);
1889
1890 brelse(iloc.bh);
1891out:
1892 up_read(&EXT4_I(inode)->xattr_sem);
1893 if (physical)
1894 error = fiemap_fill_next_extent(fieinfo, start, physical,
1895 inline_len, flags);
1896 return (error < 0 ? error : 0);
1897}
1898
1899int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1900{
1901 handle_t *handle;
1902 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1903 size_t i_size;
1904 void *value = NULL;
1905 struct ext4_xattr_ibody_find is = {
1906 .s = { .not_found = -ENODATA, },
1907 };
1908 struct ext4_xattr_info i = {
1909 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1910 .name = EXT4_XATTR_SYSTEM_DATA,
1911 };
1912
1913
1914 needed_blocks = ext4_writepage_trans_blocks(inode);
1915 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1916 if (IS_ERR(handle))
1917 return PTR_ERR(handle);
1918
1919 ext4_write_lock_xattr(inode, &no_expand);
1920 if (!ext4_has_inline_data(inode)) {
1921 *has_inline = 0;
1922 ext4_journal_stop(handle);
1923 return 0;
1924 }
1925
1926 if ((err = ext4_orphan_add(handle, inode)) != 0)
1927 goto out;
1928
1929 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1930 goto out;
1931
1932 down_write(&EXT4_I(inode)->i_data_sem);
1933 i_size = inode->i_size;
1934 inline_size = ext4_get_inline_size(inode);
1935 EXT4_I(inode)->i_disksize = i_size;
1936
1937 if (i_size < inline_size) {
1938 /* Clear the content in the xattr space. */
1939 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1940 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1941 goto out_error;
1942
1943 BUG_ON(is.s.not_found);
1944
1945 value_len = le32_to_cpu(is.s.here->e_value_size);
1946 value = kmalloc(value_len, GFP_NOFS);
1947 if (!value) {
1948 err = -ENOMEM;
1949 goto out_error;
1950 }
1951
1952 err = ext4_xattr_ibody_get(inode, i.name_index,
1953 i.name, value, value_len);
1954 if (err <= 0)
1955 goto out_error;
1956
1957 i.value = value;
1958 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1959 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1960 err = ext4_xattr_ibody_inline_set(handle, inode,
1961 &i, &is);
1962 if (err)
1963 goto out_error;
1964 }
1965
1966 /* Clear the content within i_blocks. */
1967 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1968 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1969 memset(p + i_size, 0,
1970 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1971 }
1972
1973 EXT4_I(inode)->i_inline_size = i_size <
1974 EXT4_MIN_INLINE_DATA_SIZE ?
1975 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1976 }
1977
1978out_error:
1979 up_write(&EXT4_I(inode)->i_data_sem);
1980out:
1981 brelse(is.iloc.bh);
1982 ext4_write_unlock_xattr(inode, &no_expand);
1983 kfree(value);
1984 if (inode->i_nlink)
1985 ext4_orphan_del(handle, inode);
1986
1987 if (err == 0) {
1988 inode->i_mtime = inode->i_ctime = current_time(inode);
1989 err = ext4_mark_inode_dirty(handle, inode);
1990 if (IS_SYNC(inode))
1991 ext4_handle_sync(handle);
1992 }
1993 ext4_journal_stop(handle);
1994 return err;
1995}
1996
1997int ext4_convert_inline_data(struct inode *inode)
1998{
1999 int error, needed_blocks, no_expand;
2000 handle_t *handle;
2001 struct ext4_iloc iloc;
2002
2003 if (!ext4_has_inline_data(inode)) {
2004 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2005 return 0;
2006 }
2007
2008 needed_blocks = ext4_writepage_trans_blocks(inode);
2009
2010 iloc.bh = NULL;
2011 error = ext4_get_inode_loc(inode, &iloc);
2012 if (error)
2013 return error;
2014
2015 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2016 if (IS_ERR(handle)) {
2017 error = PTR_ERR(handle);
2018 goto out_free;
2019 }
2020
2021 ext4_write_lock_xattr(inode, &no_expand);
2022 if (ext4_has_inline_data(inode))
2023 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2024 ext4_write_unlock_xattr(inode, &no_expand);
2025 ext4_journal_stop(handle);
2026out_free:
2027 brelse(iloc.bh);
2028 return error;
2029}