Loading...
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_err(inode, __func__, __LINE__, 0, -error,
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_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 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_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 xattr entry. */
358 i.value = value;
359 i.value_len = len;
360
361 error = ext4_xattr_ibody_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_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) && !PageUptodate(page))
737 return 0;
738
739 ret = ext4_get_inode_loc(inode, &iloc);
740 if (ret) {
741 ext4_std_error(inode->i_sb, ret);
742 return ret;
743 }
744
745 ext4_write_lock_xattr(inode, &no_expand);
746 BUG_ON(!ext4_has_inline_data(inode));
747
748 /*
749 * ei->i_inline_off may have changed since ext4_write_begin()
750 * called ext4_try_to_write_inline_data()
751 */
752 (void) ext4_find_inline_data_nolock(inode);
753
754 kaddr = kmap_atomic(page);
755 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
756 kunmap_atomic(kaddr);
757 SetPageUptodate(page);
758 /* clear page dirty so that writepages wouldn't work for us. */
759 ClearPageDirty(page);
760
761 ext4_write_unlock_xattr(inode, &no_expand);
762 brelse(iloc.bh);
763 mark_inode_dirty(inode);
764
765 return copied;
766}
767
768struct buffer_head *
769ext4_journalled_write_inline_data(struct inode *inode,
770 unsigned len,
771 struct page *page)
772{
773 int ret, no_expand;
774 void *kaddr;
775 struct ext4_iloc iloc;
776
777 ret = ext4_get_inode_loc(inode, &iloc);
778 if (ret) {
779 ext4_std_error(inode->i_sb, ret);
780 return NULL;
781 }
782
783 ext4_write_lock_xattr(inode, &no_expand);
784 kaddr = kmap_atomic(page);
785 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
786 kunmap_atomic(kaddr);
787 ext4_write_unlock_xattr(inode, &no_expand);
788
789 return iloc.bh;
790}
791
792/*
793 * Try to make the page cache and handle ready for the inline data case.
794 * We can call this function in 2 cases:
795 * 1. The inode is created and the first write exceeds inline size. We can
796 * clear the inode state safely.
797 * 2. The inode has inline data, then we need to read the data, make it
798 * update and dirty so that ext4_da_writepages can handle it. We don't
799 * need to start the journal since the file's metadata isn't changed now.
800 */
801static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
802 struct inode *inode,
803 unsigned flags,
804 void **fsdata)
805{
806 int ret = 0, inline_size;
807 struct page *page;
808
809 page = grab_cache_page_write_begin(mapping, 0, flags);
810 if (!page)
811 return -ENOMEM;
812
813 down_read(&EXT4_I(inode)->xattr_sem);
814 if (!ext4_has_inline_data(inode)) {
815 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
816 goto out;
817 }
818
819 inline_size = ext4_get_inline_size(inode);
820
821 if (!PageUptodate(page)) {
822 ret = ext4_read_inline_page(inode, page);
823 if (ret < 0)
824 goto out;
825 }
826
827 ret = __block_write_begin(page, 0, inline_size,
828 ext4_da_get_block_prep);
829 if (ret) {
830 up_read(&EXT4_I(inode)->xattr_sem);
831 unlock_page(page);
832 put_page(page);
833 ext4_truncate_failed_write(inode);
834 return ret;
835 }
836
837 SetPageDirty(page);
838 SetPageUptodate(page);
839 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
840 *fsdata = (void *)CONVERT_INLINE_DATA;
841
842out:
843 up_read(&EXT4_I(inode)->xattr_sem);
844 if (page) {
845 unlock_page(page);
846 put_page(page);
847 }
848 return ret;
849}
850
851/*
852 * Prepare the write for the inline data.
853 * If the data can be written into the inode, we just read
854 * the page and make it uptodate, and start the journal.
855 * Otherwise read the page, makes it dirty so that it can be
856 * handle in writepages(the i_disksize update is left to the
857 * normal ext4_da_write_end).
858 */
859int ext4_da_write_inline_data_begin(struct address_space *mapping,
860 struct inode *inode,
861 loff_t pos, unsigned len,
862 unsigned flags,
863 struct page **pagep,
864 void **fsdata)
865{
866 int ret, inline_size;
867 handle_t *handle;
868 struct page *page;
869 struct ext4_iloc iloc;
870 int retries = 0;
871
872 ret = ext4_get_inode_loc(inode, &iloc);
873 if (ret)
874 return ret;
875
876retry_journal:
877 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
878 if (IS_ERR(handle)) {
879 ret = PTR_ERR(handle);
880 goto out;
881 }
882
883 inline_size = ext4_get_max_inline_size(inode);
884
885 ret = -ENOSPC;
886 if (inline_size >= pos + len) {
887 ret = ext4_prepare_inline_data(handle, inode, pos + len);
888 if (ret && ret != -ENOSPC)
889 goto out_journal;
890 }
891
892 /*
893 * We cannot recurse into the filesystem as the transaction
894 * is already started.
895 */
896 flags |= AOP_FLAG_NOFS;
897
898 if (ret == -ENOSPC) {
899 ext4_journal_stop(handle);
900 ret = ext4_da_convert_inline_data_to_extent(mapping,
901 inode,
902 flags,
903 fsdata);
904 if (ret == -ENOSPC &&
905 ext4_should_retry_alloc(inode->i_sb, &retries))
906 goto retry_journal;
907 goto out;
908 }
909
910 page = grab_cache_page_write_begin(mapping, 0, flags);
911 if (!page) {
912 ret = -ENOMEM;
913 goto out_journal;
914 }
915
916 down_read(&EXT4_I(inode)->xattr_sem);
917 if (!ext4_has_inline_data(inode)) {
918 ret = 0;
919 goto out_release_page;
920 }
921
922 if (!PageUptodate(page)) {
923 ret = ext4_read_inline_page(inode, page);
924 if (ret < 0)
925 goto out_release_page;
926 }
927 ret = ext4_journal_get_write_access(handle, iloc.bh);
928 if (ret)
929 goto out_release_page;
930
931 up_read(&EXT4_I(inode)->xattr_sem);
932 *pagep = page;
933 brelse(iloc.bh);
934 return 1;
935out_release_page:
936 up_read(&EXT4_I(inode)->xattr_sem);
937 unlock_page(page);
938 put_page(page);
939out_journal:
940 ext4_journal_stop(handle);
941out:
942 brelse(iloc.bh);
943 return ret;
944}
945
946int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
947 unsigned len, unsigned copied,
948 struct page *page)
949{
950 int ret;
951
952 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
953 if (ret < 0) {
954 unlock_page(page);
955 put_page(page);
956 return ret;
957 }
958 copied = ret;
959
960 /*
961 * No need to use i_size_read() here, the i_size
962 * cannot change under us because we hold i_mutex.
963 *
964 * But it's important to update i_size while still holding page lock:
965 * page writeout could otherwise come in and zero beyond i_size.
966 */
967 if (pos+copied > inode->i_size)
968 i_size_write(inode, pos+copied);
969 unlock_page(page);
970 put_page(page);
971
972 /*
973 * Don't mark the inode dirty under page lock. First, it unnecessarily
974 * makes the holding time of page lock longer. Second, it forces lock
975 * ordering of page lock and transaction start for journaling
976 * filesystems.
977 */
978 mark_inode_dirty(inode);
979
980 return copied;
981}
982
983#ifdef INLINE_DIR_DEBUG
984void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
985 void *inline_start, int inline_size)
986{
987 int offset;
988 unsigned short de_len;
989 struct ext4_dir_entry_2 *de = inline_start;
990 void *dlimit = inline_start + inline_size;
991
992 trace_printk("inode %lu\n", dir->i_ino);
993 offset = 0;
994 while ((void *)de < dlimit) {
995 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
996 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
997 offset, de_len, de->name_len, de->name,
998 de->name_len, le32_to_cpu(de->inode));
999 if (ext4_check_dir_entry(dir, NULL, de, bh,
1000 inline_start, inline_size, offset))
1001 BUG();
1002
1003 offset += de_len;
1004 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1005 }
1006}
1007#else
1008#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1009#endif
1010
1011/*
1012 * Add a new entry into a inline dir.
1013 * It will return -ENOSPC if no space is available, and -EIO
1014 * and -EEXIST if directory entry already exists.
1015 */
1016static int ext4_add_dirent_to_inline(handle_t *handle,
1017 struct ext4_filename *fname,
1018 struct inode *dir,
1019 struct inode *inode,
1020 struct ext4_iloc *iloc,
1021 void *inline_start, int inline_size)
1022{
1023 int err;
1024 struct ext4_dir_entry_2 *de;
1025
1026 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1027 inline_size, fname, &de);
1028 if (err)
1029 return err;
1030
1031 BUFFER_TRACE(iloc->bh, "get_write_access");
1032 err = ext4_journal_get_write_access(handle, iloc->bh);
1033 if (err)
1034 return err;
1035 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1036
1037 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1038
1039 /*
1040 * XXX shouldn't update any times until successful
1041 * completion of syscall, but too many callers depend
1042 * on this.
1043 *
1044 * XXX similarly, too many callers depend on
1045 * ext4_new_inode() setting the times, but error
1046 * recovery deletes the inode, so the worst that can
1047 * happen is that the times are slightly out of date
1048 * and/or different from the directory change time.
1049 */
1050 dir->i_mtime = dir->i_ctime = current_time(dir);
1051 ext4_update_dx_flag(dir);
1052 inode_inc_iversion(dir);
1053 return 1;
1054}
1055
1056static void *ext4_get_inline_xattr_pos(struct inode *inode,
1057 struct ext4_iloc *iloc)
1058{
1059 struct ext4_xattr_entry *entry;
1060 struct ext4_xattr_ibody_header *header;
1061
1062 BUG_ON(!EXT4_I(inode)->i_inline_off);
1063
1064 header = IHDR(inode, ext4_raw_inode(iloc));
1065 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1066 EXT4_I(inode)->i_inline_off);
1067
1068 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1069}
1070
1071/* Set the final de to cover the whole block. */
1072static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1073{
1074 struct ext4_dir_entry_2 *de, *prev_de;
1075 void *limit;
1076 int de_len;
1077
1078 de = (struct ext4_dir_entry_2 *)de_buf;
1079 if (old_size) {
1080 limit = de_buf + old_size;
1081 do {
1082 prev_de = de;
1083 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1084 de_buf += de_len;
1085 de = (struct ext4_dir_entry_2 *)de_buf;
1086 } while (de_buf < limit);
1087
1088 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1089 old_size, new_size);
1090 } else {
1091 /* this is just created, so create an empty entry. */
1092 de->inode = 0;
1093 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1094 }
1095}
1096
1097static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1098 struct ext4_iloc *iloc)
1099{
1100 int ret;
1101 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1102 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1103
1104 if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1105 return -ENOSPC;
1106
1107 ret = ext4_update_inline_data(handle, dir,
1108 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1109 if (ret)
1110 return ret;
1111
1112 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1113 EXT4_I(dir)->i_inline_size -
1114 EXT4_MIN_INLINE_DATA_SIZE);
1115 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1116 return 0;
1117}
1118
1119static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1120 struct ext4_iloc *iloc,
1121 void *buf, int inline_size)
1122{
1123 ext4_create_inline_data(handle, inode, inline_size);
1124 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1125 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1126}
1127
1128static int ext4_finish_convert_inline_dir(handle_t *handle,
1129 struct inode *inode,
1130 struct buffer_head *dir_block,
1131 void *buf,
1132 int inline_size)
1133{
1134 int err, csum_size = 0, header_size = 0;
1135 struct ext4_dir_entry_2 *de;
1136 void *target = dir_block->b_data;
1137
1138 /*
1139 * First create "." and ".." and then copy the dir information
1140 * back to the block.
1141 */
1142 de = (struct ext4_dir_entry_2 *)target;
1143 de = ext4_init_dot_dotdot(inode, de,
1144 inode->i_sb->s_blocksize, csum_size,
1145 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1146 header_size = (void *)de - target;
1147
1148 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1149 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1150
1151 if (ext4_has_metadata_csum(inode->i_sb))
1152 csum_size = sizeof(struct ext4_dir_entry_tail);
1153
1154 inode->i_size = inode->i_sb->s_blocksize;
1155 i_size_write(inode, inode->i_sb->s_blocksize);
1156 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1157 ext4_update_final_de(dir_block->b_data,
1158 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1159 inode->i_sb->s_blocksize - csum_size);
1160
1161 if (csum_size)
1162 ext4_initialize_dirent_tail(dir_block,
1163 inode->i_sb->s_blocksize);
1164 set_buffer_uptodate(dir_block);
1165 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1166 if (err)
1167 return err;
1168 set_buffer_verified(dir_block);
1169 return ext4_mark_inode_dirty(handle, inode);
1170}
1171
1172static int ext4_convert_inline_data_nolock(handle_t *handle,
1173 struct inode *inode,
1174 struct ext4_iloc *iloc)
1175{
1176 int error;
1177 void *buf = NULL;
1178 struct buffer_head *data_bh = NULL;
1179 struct ext4_map_blocks map;
1180 int inline_size;
1181
1182 inline_size = ext4_get_inline_size(inode);
1183 buf = kmalloc(inline_size, GFP_NOFS);
1184 if (!buf) {
1185 error = -ENOMEM;
1186 goto out;
1187 }
1188
1189 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1190 if (error < 0)
1191 goto out;
1192
1193 /*
1194 * Make sure the inline directory entries pass checks before we try to
1195 * convert them, so that we avoid touching stuff that needs fsck.
1196 */
1197 if (S_ISDIR(inode->i_mode)) {
1198 error = ext4_check_all_de(inode, iloc->bh,
1199 buf + EXT4_INLINE_DOTDOT_SIZE,
1200 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1201 if (error)
1202 goto out;
1203 }
1204
1205 error = ext4_destroy_inline_data_nolock(handle, inode);
1206 if (error)
1207 goto out;
1208
1209 map.m_lblk = 0;
1210 map.m_len = 1;
1211 map.m_flags = 0;
1212 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1213 if (error < 0)
1214 goto out_restore;
1215 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1216 error = -EIO;
1217 goto out_restore;
1218 }
1219
1220 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1221 if (!data_bh) {
1222 error = -ENOMEM;
1223 goto out_restore;
1224 }
1225
1226 lock_buffer(data_bh);
1227 error = ext4_journal_get_create_access(handle, data_bh);
1228 if (error) {
1229 unlock_buffer(data_bh);
1230 error = -EIO;
1231 goto out_restore;
1232 }
1233 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1234
1235 if (!S_ISDIR(inode->i_mode)) {
1236 memcpy(data_bh->b_data, buf, inline_size);
1237 set_buffer_uptodate(data_bh);
1238 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
1245 unlock_buffer(data_bh);
1246out_restore:
1247 if (error)
1248 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1249
1250out:
1251 brelse(data_bh);
1252 kfree(buf);
1253 return error;
1254}
1255
1256/*
1257 * Try to add the new entry to the inline data.
1258 * If succeeds, return 0. If not, extended the inline dir and copied data to
1259 * the new created block.
1260 */
1261int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1262 struct inode *dir, struct inode *inode)
1263{
1264 int ret, ret2, inline_size, no_expand;
1265 void *inline_start;
1266 struct ext4_iloc iloc;
1267
1268 ret = ext4_get_inode_loc(dir, &iloc);
1269 if (ret)
1270 return ret;
1271
1272 ext4_write_lock_xattr(dir, &no_expand);
1273 if (!ext4_has_inline_data(dir))
1274 goto out;
1275
1276 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1277 EXT4_INLINE_DOTDOT_SIZE;
1278 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1279
1280 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1281 inline_start, inline_size);
1282 if (ret != -ENOSPC)
1283 goto out;
1284
1285 /* check whether it can be inserted to inline xattr space. */
1286 inline_size = EXT4_I(dir)->i_inline_size -
1287 EXT4_MIN_INLINE_DATA_SIZE;
1288 if (!inline_size) {
1289 /* Try to use the xattr space.*/
1290 ret = ext4_update_inline_dir(handle, dir, &iloc);
1291 if (ret && ret != -ENOSPC)
1292 goto out;
1293
1294 inline_size = EXT4_I(dir)->i_inline_size -
1295 EXT4_MIN_INLINE_DATA_SIZE;
1296 }
1297
1298 if (inline_size) {
1299 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1300
1301 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1302 inode, &iloc, inline_start,
1303 inline_size);
1304
1305 if (ret != -ENOSPC)
1306 goto out;
1307 }
1308
1309 /*
1310 * The inline space is filled up, so create a new block for it.
1311 * As the extent tree will be created, we have to save the inline
1312 * dir first.
1313 */
1314 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1315
1316out:
1317 ext4_write_unlock_xattr(dir, &no_expand);
1318 ret2 = ext4_mark_inode_dirty(handle, dir);
1319 if (unlikely(ret2 && !ret))
1320 ret = ret2;
1321 brelse(iloc.bh);
1322 return ret;
1323}
1324
1325/*
1326 * This function fills a red-black tree with information from an
1327 * inlined dir. It returns the number directory entries loaded
1328 * into the tree. If there is an error it is returned in err.
1329 */
1330int ext4_inlinedir_to_tree(struct file *dir_file,
1331 struct inode *dir, ext4_lblk_t block,
1332 struct dx_hash_info *hinfo,
1333 __u32 start_hash, __u32 start_minor_hash,
1334 int *has_inline_data)
1335{
1336 int err = 0, count = 0;
1337 unsigned int parent_ino;
1338 int pos;
1339 struct ext4_dir_entry_2 *de;
1340 struct inode *inode = file_inode(dir_file);
1341 int ret, inline_size = 0;
1342 struct ext4_iloc iloc;
1343 void *dir_buf = NULL;
1344 struct ext4_dir_entry_2 fake;
1345 struct fscrypt_str tmp_str;
1346
1347 ret = ext4_get_inode_loc(inode, &iloc);
1348 if (ret)
1349 return ret;
1350
1351 down_read(&EXT4_I(inode)->xattr_sem);
1352 if (!ext4_has_inline_data(inode)) {
1353 up_read(&EXT4_I(inode)->xattr_sem);
1354 *has_inline_data = 0;
1355 goto out;
1356 }
1357
1358 inline_size = ext4_get_inline_size(inode);
1359 dir_buf = kmalloc(inline_size, GFP_NOFS);
1360 if (!dir_buf) {
1361 ret = -ENOMEM;
1362 up_read(&EXT4_I(inode)->xattr_sem);
1363 goto out;
1364 }
1365
1366 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1367 up_read(&EXT4_I(inode)->xattr_sem);
1368 if (ret < 0)
1369 goto out;
1370
1371 pos = 0;
1372 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1373 while (pos < inline_size) {
1374 /*
1375 * As inlined dir doesn't store any information about '.' and
1376 * only the inode number of '..' is stored, we have to handle
1377 * them differently.
1378 */
1379 if (pos == 0) {
1380 fake.inode = cpu_to_le32(inode->i_ino);
1381 fake.name_len = 1;
1382 strcpy(fake.name, ".");
1383 fake.rec_len = ext4_rec_len_to_disk(
1384 ext4_dir_rec_len(fake.name_len, NULL),
1385 inline_size);
1386 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1387 de = &fake;
1388 pos = EXT4_INLINE_DOTDOT_OFFSET;
1389 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1390 fake.inode = cpu_to_le32(parent_ino);
1391 fake.name_len = 2;
1392 strcpy(fake.name, "..");
1393 fake.rec_len = ext4_rec_len_to_disk(
1394 ext4_dir_rec_len(fake.name_len, NULL),
1395 inline_size);
1396 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1397 de = &fake;
1398 pos = EXT4_INLINE_DOTDOT_SIZE;
1399 } else {
1400 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1401 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1402 if (ext4_check_dir_entry(inode, dir_file, de,
1403 iloc.bh, dir_buf,
1404 inline_size, pos)) {
1405 ret = count;
1406 goto out;
1407 }
1408 }
1409
1410 if (ext4_hash_in_dirent(dir)) {
1411 hinfo->hash = EXT4_DIRENT_HASH(de);
1412 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1413 } else {
1414 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1415 }
1416 if ((hinfo->hash < start_hash) ||
1417 ((hinfo->hash == start_hash) &&
1418 (hinfo->minor_hash < start_minor_hash)))
1419 continue;
1420 if (de->inode == 0)
1421 continue;
1422 tmp_str.name = de->name;
1423 tmp_str.len = de->name_len;
1424 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1425 hinfo->minor_hash, de, &tmp_str);
1426 if (err) {
1427 ret = err;
1428 goto out;
1429 }
1430 count++;
1431 }
1432 ret = count;
1433out:
1434 kfree(dir_buf);
1435 brelse(iloc.bh);
1436 return ret;
1437}
1438
1439/*
1440 * So this function is called when the volume is mkfsed with
1441 * dir_index disabled. In order to keep f_pos persistent
1442 * after we convert from an inlined dir to a blocked based,
1443 * we just pretend that we are a normal dir and return the
1444 * offset as if '.' and '..' really take place.
1445 *
1446 */
1447int ext4_read_inline_dir(struct file *file,
1448 struct dir_context *ctx,
1449 int *has_inline_data)
1450{
1451 unsigned int offset, parent_ino;
1452 int i;
1453 struct ext4_dir_entry_2 *de;
1454 struct super_block *sb;
1455 struct inode *inode = file_inode(file);
1456 int ret, inline_size = 0;
1457 struct ext4_iloc iloc;
1458 void *dir_buf = NULL;
1459 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1460
1461 ret = ext4_get_inode_loc(inode, &iloc);
1462 if (ret)
1463 return ret;
1464
1465 down_read(&EXT4_I(inode)->xattr_sem);
1466 if (!ext4_has_inline_data(inode)) {
1467 up_read(&EXT4_I(inode)->xattr_sem);
1468 *has_inline_data = 0;
1469 goto out;
1470 }
1471
1472 inline_size = ext4_get_inline_size(inode);
1473 dir_buf = kmalloc(inline_size, GFP_NOFS);
1474 if (!dir_buf) {
1475 ret = -ENOMEM;
1476 up_read(&EXT4_I(inode)->xattr_sem);
1477 goto out;
1478 }
1479
1480 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1481 up_read(&EXT4_I(inode)->xattr_sem);
1482 if (ret < 0)
1483 goto out;
1484
1485 ret = 0;
1486 sb = inode->i_sb;
1487 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1488 offset = ctx->pos;
1489
1490 /*
1491 * dotdot_offset and dotdot_size is the real offset and
1492 * size for ".." and "." if the dir is block based while
1493 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1494 * So we will use extra_offset and extra_size to indicate them
1495 * during the inline dir iteration.
1496 */
1497 dotdot_offset = ext4_dir_rec_len(1, NULL);
1498 dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1499 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1500 extra_size = extra_offset + inline_size;
1501
1502 /*
1503 * If the version has changed since the last call to
1504 * readdir(2), then we might be pointing to an invalid
1505 * dirent right now. Scan from the start of the inline
1506 * dir to make sure.
1507 */
1508 if (!inode_eq_iversion(inode, file->f_version)) {
1509 for (i = 0; i < extra_size && i < offset;) {
1510 /*
1511 * "." is with offset 0 and
1512 * ".." is dotdot_offset.
1513 */
1514 if (!i) {
1515 i = dotdot_offset;
1516 continue;
1517 } else if (i == dotdot_offset) {
1518 i = dotdot_size;
1519 continue;
1520 }
1521 /* for other entry, the real offset in
1522 * the buf has to be tuned accordingly.
1523 */
1524 de = (struct ext4_dir_entry_2 *)
1525 (dir_buf + i - extra_offset);
1526 /* It's too expensive to do a full
1527 * dirent test each time round this
1528 * loop, but we do have to test at
1529 * least that it is non-zero. A
1530 * failure will be detected in the
1531 * dirent test below. */
1532 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1533 < ext4_dir_rec_len(1, NULL))
1534 break;
1535 i += ext4_rec_len_from_disk(de->rec_len,
1536 extra_size);
1537 }
1538 offset = i;
1539 ctx->pos = offset;
1540 file->f_version = inode_query_iversion(inode);
1541 }
1542
1543 while (ctx->pos < extra_size) {
1544 if (ctx->pos == 0) {
1545 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1546 goto out;
1547 ctx->pos = dotdot_offset;
1548 continue;
1549 }
1550
1551 if (ctx->pos == dotdot_offset) {
1552 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1553 goto out;
1554 ctx->pos = dotdot_size;
1555 continue;
1556 }
1557
1558 de = (struct ext4_dir_entry_2 *)
1559 (dir_buf + ctx->pos - extra_offset);
1560 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1561 extra_size, ctx->pos))
1562 goto out;
1563 if (le32_to_cpu(de->inode)) {
1564 if (!dir_emit(ctx, de->name, de->name_len,
1565 le32_to_cpu(de->inode),
1566 get_dtype(sb, de->file_type)))
1567 goto out;
1568 }
1569 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1570 }
1571out:
1572 kfree(dir_buf);
1573 brelse(iloc.bh);
1574 return ret;
1575}
1576
1577struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1578 struct ext4_dir_entry_2 **parent_de,
1579 int *retval)
1580{
1581 struct ext4_iloc iloc;
1582
1583 *retval = ext4_get_inode_loc(inode, &iloc);
1584 if (*retval)
1585 return NULL;
1586
1587 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1588
1589 return iloc.bh;
1590}
1591
1592/*
1593 * Try to create the inline data for the new dir.
1594 * If it succeeds, return 0, otherwise return the error.
1595 * In case of ENOSPC, the caller should create the normal disk layout dir.
1596 */
1597int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1598 struct inode *inode)
1599{
1600 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1601 struct ext4_iloc iloc;
1602 struct ext4_dir_entry_2 *de;
1603
1604 ret = ext4_get_inode_loc(inode, &iloc);
1605 if (ret)
1606 return ret;
1607
1608 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1609 if (ret)
1610 goto out;
1611
1612 /*
1613 * For inline dir, we only save the inode information for the ".."
1614 * and create a fake dentry to cover the left space.
1615 */
1616 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1617 de->inode = cpu_to_le32(parent->i_ino);
1618 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1619 de->inode = 0;
1620 de->rec_len = ext4_rec_len_to_disk(
1621 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1622 inline_size);
1623 set_nlink(inode, 2);
1624 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1625out:
1626 brelse(iloc.bh);
1627 return ret;
1628}
1629
1630struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1631 struct ext4_filename *fname,
1632 struct ext4_dir_entry_2 **res_dir,
1633 int *has_inline_data)
1634{
1635 int ret;
1636 struct ext4_iloc iloc;
1637 void *inline_start;
1638 int inline_size;
1639
1640 if (ext4_get_inode_loc(dir, &iloc))
1641 return NULL;
1642
1643 down_read(&EXT4_I(dir)->xattr_sem);
1644 if (!ext4_has_inline_data(dir)) {
1645 *has_inline_data = 0;
1646 goto out;
1647 }
1648
1649 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1650 EXT4_INLINE_DOTDOT_SIZE;
1651 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1652 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1653 dir, fname, 0, res_dir);
1654 if (ret == 1)
1655 goto out_find;
1656 if (ret < 0)
1657 goto out;
1658
1659 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1660 goto out;
1661
1662 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1663 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1664
1665 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1666 dir, fname, 0, res_dir);
1667 if (ret == 1)
1668 goto out_find;
1669
1670out:
1671 brelse(iloc.bh);
1672 iloc.bh = NULL;
1673out_find:
1674 up_read(&EXT4_I(dir)->xattr_sem);
1675 return iloc.bh;
1676}
1677
1678int ext4_delete_inline_entry(handle_t *handle,
1679 struct inode *dir,
1680 struct ext4_dir_entry_2 *de_del,
1681 struct buffer_head *bh,
1682 int *has_inline_data)
1683{
1684 int err, inline_size, no_expand;
1685 struct ext4_iloc iloc;
1686 void *inline_start;
1687
1688 err = ext4_get_inode_loc(dir, &iloc);
1689 if (err)
1690 return err;
1691
1692 ext4_write_lock_xattr(dir, &no_expand);
1693 if (!ext4_has_inline_data(dir)) {
1694 *has_inline_data = 0;
1695 goto out;
1696 }
1697
1698 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1699 EXT4_MIN_INLINE_DATA_SIZE) {
1700 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1701 EXT4_INLINE_DOTDOT_SIZE;
1702 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1703 EXT4_INLINE_DOTDOT_SIZE;
1704 } else {
1705 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1706 inline_size = ext4_get_inline_size(dir) -
1707 EXT4_MIN_INLINE_DATA_SIZE;
1708 }
1709
1710 BUFFER_TRACE(bh, "get_write_access");
1711 err = ext4_journal_get_write_access(handle, bh);
1712 if (err)
1713 goto out;
1714
1715 err = ext4_generic_delete_entry(dir, de_del, bh,
1716 inline_start, inline_size, 0);
1717 if (err)
1718 goto out;
1719
1720 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1721out:
1722 ext4_write_unlock_xattr(dir, &no_expand);
1723 if (likely(err == 0))
1724 err = ext4_mark_inode_dirty(handle, dir);
1725 brelse(iloc.bh);
1726 if (err != -ENOENT)
1727 ext4_std_error(dir->i_sb, err);
1728 return err;
1729}
1730
1731/*
1732 * Get the inline dentry at offset.
1733 */
1734static inline struct ext4_dir_entry_2 *
1735ext4_get_inline_entry(struct inode *inode,
1736 struct ext4_iloc *iloc,
1737 unsigned int offset,
1738 void **inline_start,
1739 int *inline_size)
1740{
1741 void *inline_pos;
1742
1743 BUG_ON(offset > ext4_get_inline_size(inode));
1744
1745 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1746 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1747 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1748 } else {
1749 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1750 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1751 *inline_size = ext4_get_inline_size(inode) -
1752 EXT4_MIN_INLINE_DATA_SIZE;
1753 }
1754
1755 if (inline_start)
1756 *inline_start = inline_pos;
1757 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1758}
1759
1760bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1761{
1762 int err, inline_size;
1763 struct ext4_iloc iloc;
1764 size_t inline_len;
1765 void *inline_pos;
1766 unsigned int offset;
1767 struct ext4_dir_entry_2 *de;
1768 bool ret = true;
1769
1770 err = ext4_get_inode_loc(dir, &iloc);
1771 if (err) {
1772 EXT4_ERROR_INODE_ERR(dir, -err,
1773 "error %d getting inode %lu block",
1774 err, dir->i_ino);
1775 return true;
1776 }
1777
1778 down_read(&EXT4_I(dir)->xattr_sem);
1779 if (!ext4_has_inline_data(dir)) {
1780 *has_inline_data = 0;
1781 goto out;
1782 }
1783
1784 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1785 if (!le32_to_cpu(de->inode)) {
1786 ext4_warning(dir->i_sb,
1787 "bad inline directory (dir #%lu) - no `..'",
1788 dir->i_ino);
1789 ret = true;
1790 goto out;
1791 }
1792
1793 inline_len = ext4_get_inline_size(dir);
1794 offset = EXT4_INLINE_DOTDOT_SIZE;
1795 while (offset < inline_len) {
1796 de = ext4_get_inline_entry(dir, &iloc, offset,
1797 &inline_pos, &inline_size);
1798 if (ext4_check_dir_entry(dir, NULL, de,
1799 iloc.bh, inline_pos,
1800 inline_size, offset)) {
1801 ext4_warning(dir->i_sb,
1802 "bad inline directory (dir #%lu) - "
1803 "inode %u, rec_len %u, name_len %d"
1804 "inline size %d",
1805 dir->i_ino, le32_to_cpu(de->inode),
1806 le16_to_cpu(de->rec_len), de->name_len,
1807 inline_size);
1808 ret = true;
1809 goto out;
1810 }
1811 if (le32_to_cpu(de->inode)) {
1812 ret = false;
1813 goto out;
1814 }
1815 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1816 }
1817
1818out:
1819 up_read(&EXT4_I(dir)->xattr_sem);
1820 brelse(iloc.bh);
1821 return ret;
1822}
1823
1824int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1825{
1826 int ret, no_expand;
1827
1828 ext4_write_lock_xattr(inode, &no_expand);
1829 ret = ext4_destroy_inline_data_nolock(handle, inode);
1830 ext4_write_unlock_xattr(inode, &no_expand);
1831
1832 return ret;
1833}
1834
1835int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1836{
1837 __u64 addr;
1838 int error = -EAGAIN;
1839 struct ext4_iloc iloc;
1840
1841 down_read(&EXT4_I(inode)->xattr_sem);
1842 if (!ext4_has_inline_data(inode))
1843 goto out;
1844
1845 error = ext4_get_inode_loc(inode, &iloc);
1846 if (error)
1847 goto out;
1848
1849 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1850 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1851 addr += offsetof(struct ext4_inode, i_block);
1852
1853 brelse(iloc.bh);
1854
1855 iomap->addr = addr;
1856 iomap->offset = 0;
1857 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1858 i_size_read(inode));
1859 iomap->type = IOMAP_INLINE;
1860 iomap->flags = 0;
1861
1862out:
1863 up_read(&EXT4_I(inode)->xattr_sem);
1864 return error;
1865}
1866
1867int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1868{
1869 handle_t *handle;
1870 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1871 size_t i_size;
1872 void *value = NULL;
1873 struct ext4_xattr_ibody_find is = {
1874 .s = { .not_found = -ENODATA, },
1875 };
1876 struct ext4_xattr_info i = {
1877 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1878 .name = EXT4_XATTR_SYSTEM_DATA,
1879 };
1880
1881
1882 needed_blocks = ext4_writepage_trans_blocks(inode);
1883 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1884 if (IS_ERR(handle))
1885 return PTR_ERR(handle);
1886
1887 ext4_write_lock_xattr(inode, &no_expand);
1888 if (!ext4_has_inline_data(inode)) {
1889 ext4_write_unlock_xattr(inode, &no_expand);
1890 *has_inline = 0;
1891 ext4_journal_stop(handle);
1892 return 0;
1893 }
1894
1895 if ((err = ext4_orphan_add(handle, inode)) != 0)
1896 goto out;
1897
1898 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1899 goto out;
1900
1901 down_write(&EXT4_I(inode)->i_data_sem);
1902 i_size = inode->i_size;
1903 inline_size = ext4_get_inline_size(inode);
1904 EXT4_I(inode)->i_disksize = i_size;
1905
1906 if (i_size < inline_size) {
1907 /* Clear the content in the xattr space. */
1908 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1909 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1910 goto out_error;
1911
1912 BUG_ON(is.s.not_found);
1913
1914 value_len = le32_to_cpu(is.s.here->e_value_size);
1915 value = kmalloc(value_len, GFP_NOFS);
1916 if (!value) {
1917 err = -ENOMEM;
1918 goto out_error;
1919 }
1920
1921 err = ext4_xattr_ibody_get(inode, i.name_index,
1922 i.name, value, value_len);
1923 if (err <= 0)
1924 goto out_error;
1925
1926 i.value = value;
1927 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1928 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1929 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1930 if (err)
1931 goto out_error;
1932 }
1933
1934 /* Clear the content within i_blocks. */
1935 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1936 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1937 memset(p + i_size, 0,
1938 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1939 }
1940
1941 EXT4_I(inode)->i_inline_size = i_size <
1942 EXT4_MIN_INLINE_DATA_SIZE ?
1943 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1944 }
1945
1946out_error:
1947 up_write(&EXT4_I(inode)->i_data_sem);
1948out:
1949 brelse(is.iloc.bh);
1950 ext4_write_unlock_xattr(inode, &no_expand);
1951 kfree(value);
1952 if (inode->i_nlink)
1953 ext4_orphan_del(handle, inode);
1954
1955 if (err == 0) {
1956 inode->i_mtime = inode->i_ctime = current_time(inode);
1957 err = ext4_mark_inode_dirty(handle, inode);
1958 if (IS_SYNC(inode))
1959 ext4_handle_sync(handle);
1960 }
1961 ext4_journal_stop(handle);
1962 return err;
1963}
1964
1965int ext4_convert_inline_data(struct inode *inode)
1966{
1967 int error, needed_blocks, no_expand;
1968 handle_t *handle;
1969 struct ext4_iloc iloc;
1970
1971 if (!ext4_has_inline_data(inode)) {
1972 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1973 return 0;
1974 }
1975
1976 needed_blocks = ext4_writepage_trans_blocks(inode);
1977
1978 iloc.bh = NULL;
1979 error = ext4_get_inode_loc(inode, &iloc);
1980 if (error)
1981 return error;
1982
1983 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1984 if (IS_ERR(handle)) {
1985 error = PTR_ERR(handle);
1986 goto out_free;
1987 }
1988
1989 ext4_write_lock_xattr(inode, &no_expand);
1990 if (ext4_has_inline_data(inode))
1991 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1992 ext4_write_unlock_xattr(inode, &no_expand);
1993 ext4_journal_stop(handle);
1994out_free:
1995 brelse(iloc.bh);
1996 return error;
1997}
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 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
148 (void *)ext4_raw_inode(&is.iloc));
149 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
150 le32_to_cpu(is.s.here->e_value_size);
151 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
152 }
153out:
154 brelse(is.iloc.bh);
155 return error;
156}
157
158static int ext4_read_inline_data(struct inode *inode, void *buffer,
159 unsigned int len,
160 struct ext4_iloc *iloc)
161{
162 struct ext4_xattr_entry *entry;
163 struct ext4_xattr_ibody_header *header;
164 int cp_len = 0;
165 struct ext4_inode *raw_inode;
166
167 if (!len)
168 return 0;
169
170 BUG_ON(len > EXT4_I(inode)->i_inline_size);
171
172 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
173 len : EXT4_MIN_INLINE_DATA_SIZE;
174
175 raw_inode = ext4_raw_inode(iloc);
176 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
177
178 len -= cp_len;
179 buffer += cp_len;
180
181 if (!len)
182 goto out;
183
184 header = IHDR(inode, raw_inode);
185 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
186 EXT4_I(inode)->i_inline_off);
187 len = min_t(unsigned int, len,
188 (unsigned int)le32_to_cpu(entry->e_value_size));
189
190 memcpy(buffer,
191 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
192 cp_len += len;
193
194out:
195 return cp_len;
196}
197
198/*
199 * write the buffer to the inline inode.
200 * If 'create' is set, we don't need to do the extra copy in the xattr
201 * value since it is already handled by ext4_xattr_ibody_inline_set.
202 * That saves us one memcpy.
203 */
204static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
205 void *buffer, loff_t pos, unsigned int len)
206{
207 struct ext4_xattr_entry *entry;
208 struct ext4_xattr_ibody_header *header;
209 struct ext4_inode *raw_inode;
210 int cp_len = 0;
211
212 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
213 return;
214
215 BUG_ON(!EXT4_I(inode)->i_inline_off);
216 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
217
218 raw_inode = ext4_raw_inode(iloc);
219 buffer += pos;
220
221 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
222 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
223 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
224 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
225
226 len -= cp_len;
227 buffer += cp_len;
228 pos += cp_len;
229 }
230
231 if (!len)
232 return;
233
234 pos -= EXT4_MIN_INLINE_DATA_SIZE;
235 header = IHDR(inode, raw_inode);
236 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
237 EXT4_I(inode)->i_inline_off);
238
239 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
240 buffer, len);
241}
242
243static int ext4_create_inline_data(handle_t *handle,
244 struct inode *inode, unsigned len)
245{
246 int error;
247 void *value = NULL;
248 struct ext4_xattr_ibody_find is = {
249 .s = { .not_found = -ENODATA, },
250 };
251 struct ext4_xattr_info i = {
252 .name_index = EXT4_XATTR_INDEX_SYSTEM,
253 .name = EXT4_XATTR_SYSTEM_DATA,
254 };
255
256 error = ext4_get_inode_loc(inode, &is.iloc);
257 if (error)
258 return error;
259
260 BUFFER_TRACE(is.iloc.bh, "get_write_access");
261 error = ext4_journal_get_write_access(handle, is.iloc.bh);
262 if (error)
263 goto out;
264
265 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
266 value = EXT4_ZERO_XATTR_VALUE;
267 len -= EXT4_MIN_INLINE_DATA_SIZE;
268 } else {
269 value = "";
270 len = 0;
271 }
272
273 /* Insert the the xttr entry. */
274 i.value = value;
275 i.value_len = len;
276
277 error = ext4_xattr_ibody_find(inode, &i, &is);
278 if (error)
279 goto out;
280
281 BUG_ON(!is.s.not_found);
282
283 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
284 if (error) {
285 if (error == -ENOSPC)
286 ext4_clear_inode_state(inode,
287 EXT4_STATE_MAY_INLINE_DATA);
288 goto out;
289 }
290
291 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
292 0, EXT4_MIN_INLINE_DATA_SIZE);
293
294 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
295 (void *)ext4_raw_inode(&is.iloc));
296 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
297 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
298 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
299 get_bh(is.iloc.bh);
300 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
301
302out:
303 brelse(is.iloc.bh);
304 return error;
305}
306
307static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
308 unsigned int len)
309{
310 int error;
311 void *value = NULL;
312 struct ext4_xattr_ibody_find is = {
313 .s = { .not_found = -ENODATA, },
314 };
315 struct ext4_xattr_info i = {
316 .name_index = EXT4_XATTR_INDEX_SYSTEM,
317 .name = EXT4_XATTR_SYSTEM_DATA,
318 };
319
320 /* If the old space is ok, write the data directly. */
321 if (len <= EXT4_I(inode)->i_inline_size)
322 return 0;
323
324 error = ext4_get_inode_loc(inode, &is.iloc);
325 if (error)
326 return error;
327
328 error = ext4_xattr_ibody_find(inode, &i, &is);
329 if (error)
330 goto out;
331
332 BUG_ON(is.s.not_found);
333
334 len -= EXT4_MIN_INLINE_DATA_SIZE;
335 value = kzalloc(len, GFP_NOFS);
336 if (!value) {
337 error = -ENOMEM;
338 goto out;
339 }
340
341 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
342 value, len);
343 if (error == -ENODATA)
344 goto out;
345
346 BUFFER_TRACE(is.iloc.bh, "get_write_access");
347 error = ext4_journal_get_write_access(handle, is.iloc.bh);
348 if (error)
349 goto out;
350
351 /* Update the xttr entry. */
352 i.value = value;
353 i.value_len = len;
354
355 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
356 if (error)
357 goto out;
358
359 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
360 (void *)ext4_raw_inode(&is.iloc));
361 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
362 le32_to_cpu(is.s.here->e_value_size);
363 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
364 get_bh(is.iloc.bh);
365 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
366
367out:
368 kfree(value);
369 brelse(is.iloc.bh);
370 return error;
371}
372
373static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
374 unsigned int len)
375{
376 int ret, size, no_expand;
377 struct ext4_inode_info *ei = EXT4_I(inode);
378
379 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
380 return -ENOSPC;
381
382 size = ext4_get_max_inline_size(inode);
383 if (size < len)
384 return -ENOSPC;
385
386 ext4_write_lock_xattr(inode, &no_expand);
387
388 if (ei->i_inline_off)
389 ret = ext4_update_inline_data(handle, inode, len);
390 else
391 ret = ext4_create_inline_data(handle, inode, len);
392
393 ext4_write_unlock_xattr(inode, &no_expand);
394 return ret;
395}
396
397static int ext4_destroy_inline_data_nolock(handle_t *handle,
398 struct inode *inode)
399{
400 struct ext4_inode_info *ei = EXT4_I(inode);
401 struct ext4_xattr_ibody_find is = {
402 .s = { .not_found = 0, },
403 };
404 struct ext4_xattr_info i = {
405 .name_index = EXT4_XATTR_INDEX_SYSTEM,
406 .name = EXT4_XATTR_SYSTEM_DATA,
407 .value = NULL,
408 .value_len = 0,
409 };
410 int error;
411
412 if (!ei->i_inline_off)
413 return 0;
414
415 error = ext4_get_inode_loc(inode, &is.iloc);
416 if (error)
417 return error;
418
419 error = ext4_xattr_ibody_find(inode, &i, &is);
420 if (error)
421 goto out;
422
423 BUFFER_TRACE(is.iloc.bh, "get_write_access");
424 error = ext4_journal_get_write_access(handle, is.iloc.bh);
425 if (error)
426 goto out;
427
428 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
429 if (error)
430 goto out;
431
432 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
433 0, EXT4_MIN_INLINE_DATA_SIZE);
434
435 if (ext4_has_feature_extents(inode->i_sb)) {
436 if (S_ISDIR(inode->i_mode) ||
437 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
438 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
439 ext4_ext_tree_init(handle, inode);
440 }
441 }
442 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
443
444 get_bh(is.iloc.bh);
445 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
446
447 EXT4_I(inode)->i_inline_off = 0;
448 EXT4_I(inode)->i_inline_size = 0;
449 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
450out:
451 brelse(is.iloc.bh);
452 if (error == -ENODATA)
453 error = 0;
454 return error;
455}
456
457static int ext4_read_inline_page(struct inode *inode, struct page *page)
458{
459 void *kaddr;
460 int ret = 0;
461 size_t len;
462 struct ext4_iloc iloc;
463
464 BUG_ON(!PageLocked(page));
465 BUG_ON(!ext4_has_inline_data(inode));
466 BUG_ON(page->index);
467
468 if (!EXT4_I(inode)->i_inline_off) {
469 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
470 inode->i_ino);
471 goto out;
472 }
473
474 ret = ext4_get_inode_loc(inode, &iloc);
475 if (ret)
476 goto out;
477
478 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
479 kaddr = kmap_atomic(page);
480 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
481 flush_dcache_page(page);
482 kunmap_atomic(kaddr);
483 zero_user_segment(page, len, PAGE_SIZE);
484 SetPageUptodate(page);
485 brelse(iloc.bh);
486
487out:
488 return ret;
489}
490
491int ext4_readpage_inline(struct inode *inode, struct page *page)
492{
493 int ret = 0;
494
495 down_read(&EXT4_I(inode)->xattr_sem);
496 if (!ext4_has_inline_data(inode)) {
497 up_read(&EXT4_I(inode)->xattr_sem);
498 return -EAGAIN;
499 }
500
501 /*
502 * Current inline data can only exist in the 1st page,
503 * So for all the other pages, just set them uptodate.
504 */
505 if (!page->index)
506 ret = ext4_read_inline_page(inode, page);
507 else if (!PageUptodate(page)) {
508 zero_user_segment(page, 0, PAGE_SIZE);
509 SetPageUptodate(page);
510 }
511
512 up_read(&EXT4_I(inode)->xattr_sem);
513
514 unlock_page(page);
515 return ret >= 0 ? 0 : ret;
516}
517
518static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
519 struct inode *inode,
520 unsigned flags)
521{
522 int ret, needed_blocks, no_expand;
523 handle_t *handle = NULL;
524 int retries = 0, sem_held = 0;
525 struct page *page = NULL;
526 unsigned from, to;
527 struct ext4_iloc iloc;
528
529 if (!ext4_has_inline_data(inode)) {
530 /*
531 * clear the flag so that no new write
532 * will trap here again.
533 */
534 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
535 return 0;
536 }
537
538 needed_blocks = ext4_writepage_trans_blocks(inode);
539
540 ret = ext4_get_inode_loc(inode, &iloc);
541 if (ret)
542 return ret;
543
544retry:
545 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
546 if (IS_ERR(handle)) {
547 ret = PTR_ERR(handle);
548 handle = NULL;
549 goto out;
550 }
551
552 /* We cannot recurse into the filesystem as the transaction is already
553 * started */
554 flags |= AOP_FLAG_NOFS;
555
556 page = grab_cache_page_write_begin(mapping, 0, flags);
557 if (!page) {
558 ret = -ENOMEM;
559 goto out;
560 }
561
562 ext4_write_lock_xattr(inode, &no_expand);
563 sem_held = 1;
564 /* If some one has already done this for us, just exit. */
565 if (!ext4_has_inline_data(inode)) {
566 ret = 0;
567 goto out;
568 }
569
570 from = 0;
571 to = ext4_get_inline_size(inode);
572 if (!PageUptodate(page)) {
573 ret = ext4_read_inline_page(inode, page);
574 if (ret < 0)
575 goto out;
576 }
577
578 ret = ext4_destroy_inline_data_nolock(handle, inode);
579 if (ret)
580 goto out;
581
582 if (ext4_should_dioread_nolock(inode)) {
583 ret = __block_write_begin(page, from, to,
584 ext4_get_block_unwritten);
585 } else
586 ret = __block_write_begin(page, from, to, ext4_get_block);
587
588 if (!ret && ext4_should_journal_data(inode)) {
589 ret = ext4_walk_page_buffers(handle, page_buffers(page),
590 from, to, NULL,
591 do_journal_get_write_access);
592 }
593
594 if (ret) {
595 unlock_page(page);
596 put_page(page);
597 page = NULL;
598 ext4_orphan_add(handle, inode);
599 ext4_write_unlock_xattr(inode, &no_expand);
600 sem_held = 0;
601 ext4_journal_stop(handle);
602 handle = NULL;
603 ext4_truncate_failed_write(inode);
604 /*
605 * If truncate failed early the inode might
606 * still be on the orphan list; we need to
607 * make sure the inode is removed from the
608 * orphan list in that case.
609 */
610 if (inode->i_nlink)
611 ext4_orphan_del(NULL, inode);
612 }
613
614 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
615 goto retry;
616
617 if (page)
618 block_commit_write(page, from, to);
619out:
620 if (page) {
621 unlock_page(page);
622 put_page(page);
623 }
624 if (sem_held)
625 ext4_write_unlock_xattr(inode, &no_expand);
626 if (handle)
627 ext4_journal_stop(handle);
628 brelse(iloc.bh);
629 return ret;
630}
631
632/*
633 * Try to write data in the inode.
634 * If the inode has inline data, check whether the new write can be
635 * in the inode also. If not, create the page the handle, move the data
636 * to the page make it update and let the later codes create extent for it.
637 */
638int ext4_try_to_write_inline_data(struct address_space *mapping,
639 struct inode *inode,
640 loff_t pos, unsigned len,
641 unsigned flags,
642 struct page **pagep)
643{
644 int ret;
645 handle_t *handle;
646 struct page *page;
647 struct ext4_iloc iloc;
648
649 if (pos + len > ext4_get_max_inline_size(inode))
650 goto convert;
651
652 ret = ext4_get_inode_loc(inode, &iloc);
653 if (ret)
654 return ret;
655
656 /*
657 * The possible write could happen in the inode,
658 * so try to reserve the space in inode first.
659 */
660 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
661 if (IS_ERR(handle)) {
662 ret = PTR_ERR(handle);
663 handle = NULL;
664 goto out;
665 }
666
667 ret = ext4_prepare_inline_data(handle, inode, pos + len);
668 if (ret && ret != -ENOSPC)
669 goto out;
670
671 /* We don't have space in inline inode, so convert it to extent. */
672 if (ret == -ENOSPC) {
673 ext4_journal_stop(handle);
674 brelse(iloc.bh);
675 goto convert;
676 }
677
678 flags |= AOP_FLAG_NOFS;
679
680 page = grab_cache_page_write_begin(mapping, 0, flags);
681 if (!page) {
682 ret = -ENOMEM;
683 goto out;
684 }
685
686 *pagep = page;
687 down_read(&EXT4_I(inode)->xattr_sem);
688 if (!ext4_has_inline_data(inode)) {
689 ret = 0;
690 unlock_page(page);
691 put_page(page);
692 goto out_up_read;
693 }
694
695 if (!PageUptodate(page)) {
696 ret = ext4_read_inline_page(inode, page);
697 if (ret < 0)
698 goto out_up_read;
699 }
700
701 ret = 1;
702 handle = NULL;
703out_up_read:
704 up_read(&EXT4_I(inode)->xattr_sem);
705out:
706 if (handle)
707 ext4_journal_stop(handle);
708 brelse(iloc.bh);
709 return ret;
710convert:
711 return ext4_convert_inline_data_to_extent(mapping,
712 inode, flags);
713}
714
715int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
716 unsigned copied, struct page *page)
717{
718 int ret, no_expand;
719 void *kaddr;
720 struct ext4_iloc iloc;
721
722 if (unlikely(copied < len)) {
723 if (!PageUptodate(page)) {
724 copied = 0;
725 goto out;
726 }
727 }
728
729 ret = ext4_get_inode_loc(inode, &iloc);
730 if (ret) {
731 ext4_std_error(inode->i_sb, ret);
732 copied = 0;
733 goto out;
734 }
735
736 ext4_write_lock_xattr(inode, &no_expand);
737 BUG_ON(!ext4_has_inline_data(inode));
738
739 kaddr = kmap_atomic(page);
740 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
741 kunmap_atomic(kaddr);
742 SetPageUptodate(page);
743 /* clear page dirty so that writepages wouldn't work for us. */
744 ClearPageDirty(page);
745
746 ext4_write_unlock_xattr(inode, &no_expand);
747 brelse(iloc.bh);
748out:
749 return copied;
750}
751
752struct buffer_head *
753ext4_journalled_write_inline_data(struct inode *inode,
754 unsigned len,
755 struct page *page)
756{
757 int ret, no_expand;
758 void *kaddr;
759 struct ext4_iloc iloc;
760
761 ret = ext4_get_inode_loc(inode, &iloc);
762 if (ret) {
763 ext4_std_error(inode->i_sb, ret);
764 return NULL;
765 }
766
767 ext4_write_lock_xattr(inode, &no_expand);
768 kaddr = kmap_atomic(page);
769 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
770 kunmap_atomic(kaddr);
771 ext4_write_unlock_xattr(inode, &no_expand);
772
773 return iloc.bh;
774}
775
776/*
777 * Try to make the page cache and handle ready for the inline data case.
778 * We can call this function in 2 cases:
779 * 1. The inode is created and the first write exceeds inline size. We can
780 * clear the inode state safely.
781 * 2. The inode has inline data, then we need to read the data, make it
782 * update and dirty so that ext4_da_writepages can handle it. We don't
783 * need to start the journal since the file's metatdata isn't changed now.
784 */
785static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
786 struct inode *inode,
787 unsigned flags,
788 void **fsdata)
789{
790 int ret = 0, inline_size;
791 struct page *page;
792
793 page = grab_cache_page_write_begin(mapping, 0, flags);
794 if (!page)
795 return -ENOMEM;
796
797 down_read(&EXT4_I(inode)->xattr_sem);
798 if (!ext4_has_inline_data(inode)) {
799 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
800 goto out;
801 }
802
803 inline_size = ext4_get_inline_size(inode);
804
805 if (!PageUptodate(page)) {
806 ret = ext4_read_inline_page(inode, page);
807 if (ret < 0)
808 goto out;
809 }
810
811 ret = __block_write_begin(page, 0, inline_size,
812 ext4_da_get_block_prep);
813 if (ret) {
814 up_read(&EXT4_I(inode)->xattr_sem);
815 unlock_page(page);
816 put_page(page);
817 ext4_truncate_failed_write(inode);
818 return ret;
819 }
820
821 SetPageDirty(page);
822 SetPageUptodate(page);
823 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
824 *fsdata = (void *)CONVERT_INLINE_DATA;
825
826out:
827 up_read(&EXT4_I(inode)->xattr_sem);
828 if (page) {
829 unlock_page(page);
830 put_page(page);
831 }
832 return ret;
833}
834
835/*
836 * Prepare the write for the inline data.
837 * If the the data can be written into the inode, we just read
838 * the page and make it uptodate, and start the journal.
839 * Otherwise read the page, makes it dirty so that it can be
840 * handle in writepages(the i_disksize update is left to the
841 * normal ext4_da_write_end).
842 */
843int ext4_da_write_inline_data_begin(struct address_space *mapping,
844 struct inode *inode,
845 loff_t pos, unsigned len,
846 unsigned flags,
847 struct page **pagep,
848 void **fsdata)
849{
850 int ret, inline_size;
851 handle_t *handle;
852 struct page *page;
853 struct ext4_iloc iloc;
854 int retries;
855
856 ret = ext4_get_inode_loc(inode, &iloc);
857 if (ret)
858 return ret;
859
860retry_journal:
861 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
862 if (IS_ERR(handle)) {
863 ret = PTR_ERR(handle);
864 goto out;
865 }
866
867 inline_size = ext4_get_max_inline_size(inode);
868
869 ret = -ENOSPC;
870 if (inline_size >= pos + len) {
871 ret = ext4_prepare_inline_data(handle, inode, pos + len);
872 if (ret && ret != -ENOSPC)
873 goto out_journal;
874 }
875
876 /*
877 * We cannot recurse into the filesystem as the transaction
878 * is already started.
879 */
880 flags |= AOP_FLAG_NOFS;
881
882 if (ret == -ENOSPC) {
883 ret = ext4_da_convert_inline_data_to_extent(mapping,
884 inode,
885 flags,
886 fsdata);
887 ext4_journal_stop(handle);
888 if (ret == -ENOSPC &&
889 ext4_should_retry_alloc(inode->i_sb, &retries))
890 goto retry_journal;
891 goto out;
892 }
893
894
895 page = grab_cache_page_write_begin(mapping, 0, flags);
896 if (!page) {
897 ret = -ENOMEM;
898 goto out_journal;
899 }
900
901 down_read(&EXT4_I(inode)->xattr_sem);
902 if (!ext4_has_inline_data(inode)) {
903 ret = 0;
904 goto out_release_page;
905 }
906
907 if (!PageUptodate(page)) {
908 ret = ext4_read_inline_page(inode, page);
909 if (ret < 0)
910 goto out_release_page;
911 }
912
913 up_read(&EXT4_I(inode)->xattr_sem);
914 *pagep = page;
915 brelse(iloc.bh);
916 return 1;
917out_release_page:
918 up_read(&EXT4_I(inode)->xattr_sem);
919 unlock_page(page);
920 put_page(page);
921out_journal:
922 ext4_journal_stop(handle);
923out:
924 brelse(iloc.bh);
925 return ret;
926}
927
928int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
929 unsigned len, unsigned copied,
930 struct page *page)
931{
932 int i_size_changed = 0;
933 int ret;
934
935 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
936 if (ret < 0) {
937 unlock_page(page);
938 put_page(page);
939 return ret;
940 }
941 copied = ret;
942
943 /*
944 * No need to use i_size_read() here, the i_size
945 * cannot change under us because we hold i_mutex.
946 *
947 * But it's important to update i_size while still holding page lock:
948 * page writeout could otherwise come in and zero beyond i_size.
949 */
950 if (pos+copied > inode->i_size) {
951 i_size_write(inode, pos+copied);
952 i_size_changed = 1;
953 }
954 unlock_page(page);
955 put_page(page);
956
957 /*
958 * Don't mark the inode dirty under page lock. First, it unnecessarily
959 * makes the holding time of page lock longer. Second, it forces lock
960 * ordering of page lock and transaction start for journaling
961 * filesystems.
962 */
963 if (i_size_changed)
964 mark_inode_dirty(inode);
965
966 return copied;
967}
968
969#ifdef INLINE_DIR_DEBUG
970void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
971 void *inline_start, int inline_size)
972{
973 int offset;
974 unsigned short de_len;
975 struct ext4_dir_entry_2 *de = inline_start;
976 void *dlimit = inline_start + inline_size;
977
978 trace_printk("inode %lu\n", dir->i_ino);
979 offset = 0;
980 while ((void *)de < dlimit) {
981 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
982 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
983 offset, de_len, de->name_len, de->name,
984 de->name_len, le32_to_cpu(de->inode));
985 if (ext4_check_dir_entry(dir, NULL, de, bh,
986 inline_start, inline_size, offset))
987 BUG();
988
989 offset += de_len;
990 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
991 }
992}
993#else
994#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
995#endif
996
997/*
998 * Add a new entry into a inline dir.
999 * It will return -ENOSPC if no space is available, and -EIO
1000 * and -EEXIST if directory entry already exists.
1001 */
1002static int ext4_add_dirent_to_inline(handle_t *handle,
1003 struct ext4_filename *fname,
1004 struct inode *dir,
1005 struct inode *inode,
1006 struct ext4_iloc *iloc,
1007 void *inline_start, int inline_size)
1008{
1009 int err;
1010 struct ext4_dir_entry_2 *de;
1011
1012 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1013 inline_size, fname, &de);
1014 if (err)
1015 return err;
1016
1017 BUFFER_TRACE(iloc->bh, "get_write_access");
1018 err = ext4_journal_get_write_access(handle, iloc->bh);
1019 if (err)
1020 return err;
1021 ext4_insert_dentry(inode, de, inline_size, fname);
1022
1023 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1024
1025 /*
1026 * XXX shouldn't update any times until successful
1027 * completion of syscall, but too many callers depend
1028 * on this.
1029 *
1030 * XXX similarly, too many callers depend on
1031 * ext4_new_inode() setting the times, but error
1032 * recovery deletes the inode, so the worst that can
1033 * happen is that the times are slightly out of date
1034 * and/or different from the directory change time.
1035 */
1036 dir->i_mtime = dir->i_ctime = current_time(dir);
1037 ext4_update_dx_flag(dir);
1038 inode_inc_iversion(dir);
1039 return 1;
1040}
1041
1042static void *ext4_get_inline_xattr_pos(struct inode *inode,
1043 struct ext4_iloc *iloc)
1044{
1045 struct ext4_xattr_entry *entry;
1046 struct ext4_xattr_ibody_header *header;
1047
1048 BUG_ON(!EXT4_I(inode)->i_inline_off);
1049
1050 header = IHDR(inode, ext4_raw_inode(iloc));
1051 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1052 EXT4_I(inode)->i_inline_off);
1053
1054 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1055}
1056
1057/* Set the final de to cover the whole block. */
1058static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1059{
1060 struct ext4_dir_entry_2 *de, *prev_de;
1061 void *limit;
1062 int de_len;
1063
1064 de = (struct ext4_dir_entry_2 *)de_buf;
1065 if (old_size) {
1066 limit = de_buf + old_size;
1067 do {
1068 prev_de = de;
1069 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1070 de_buf += de_len;
1071 de = (struct ext4_dir_entry_2 *)de_buf;
1072 } while (de_buf < limit);
1073
1074 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1075 old_size, new_size);
1076 } else {
1077 /* this is just created, so create an empty entry. */
1078 de->inode = 0;
1079 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1080 }
1081}
1082
1083static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1084 struct ext4_iloc *iloc)
1085{
1086 int ret;
1087 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1088 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1089
1090 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1091 return -ENOSPC;
1092
1093 ret = ext4_update_inline_data(handle, dir,
1094 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1095 if (ret)
1096 return ret;
1097
1098 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1099 EXT4_I(dir)->i_inline_size -
1100 EXT4_MIN_INLINE_DATA_SIZE);
1101 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1102 return 0;
1103}
1104
1105static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1106 struct ext4_iloc *iloc,
1107 void *buf, int inline_size)
1108{
1109 ext4_create_inline_data(handle, inode, inline_size);
1110 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1111 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1112}
1113
1114static int ext4_finish_convert_inline_dir(handle_t *handle,
1115 struct inode *inode,
1116 struct buffer_head *dir_block,
1117 void *buf,
1118 int inline_size)
1119{
1120 int err, csum_size = 0, header_size = 0;
1121 struct ext4_dir_entry_2 *de;
1122 struct ext4_dir_entry_tail *t;
1123 void *target = dir_block->b_data;
1124
1125 /*
1126 * First create "." and ".." and then copy the dir information
1127 * back to the block.
1128 */
1129 de = (struct ext4_dir_entry_2 *)target;
1130 de = ext4_init_dot_dotdot(inode, de,
1131 inode->i_sb->s_blocksize, csum_size,
1132 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1133 header_size = (void *)de - target;
1134
1135 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1136 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1137
1138 if (ext4_has_metadata_csum(inode->i_sb))
1139 csum_size = sizeof(struct ext4_dir_entry_tail);
1140
1141 inode->i_size = inode->i_sb->s_blocksize;
1142 i_size_write(inode, inode->i_sb->s_blocksize);
1143 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1144 ext4_update_final_de(dir_block->b_data,
1145 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1146 inode->i_sb->s_blocksize - csum_size);
1147
1148 if (csum_size) {
1149 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1150 inode->i_sb->s_blocksize);
1151 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1152 }
1153 set_buffer_uptodate(dir_block);
1154 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1155 if (err)
1156 return err;
1157 set_buffer_verified(dir_block);
1158 return ext4_mark_inode_dirty(handle, inode);
1159}
1160
1161static int ext4_convert_inline_data_nolock(handle_t *handle,
1162 struct inode *inode,
1163 struct ext4_iloc *iloc)
1164{
1165 int error;
1166 void *buf = NULL;
1167 struct buffer_head *data_bh = NULL;
1168 struct ext4_map_blocks map;
1169 int inline_size;
1170
1171 inline_size = ext4_get_inline_size(inode);
1172 buf = kmalloc(inline_size, GFP_NOFS);
1173 if (!buf) {
1174 error = -ENOMEM;
1175 goto out;
1176 }
1177
1178 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1179 if (error < 0)
1180 goto out;
1181
1182 /*
1183 * Make sure the inline directory entries pass checks before we try to
1184 * convert them, so that we avoid touching stuff that needs fsck.
1185 */
1186 if (S_ISDIR(inode->i_mode)) {
1187 error = ext4_check_all_de(inode, iloc->bh,
1188 buf + EXT4_INLINE_DOTDOT_SIZE,
1189 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1190 if (error)
1191 goto out;
1192 }
1193
1194 error = ext4_destroy_inline_data_nolock(handle, inode);
1195 if (error)
1196 goto out;
1197
1198 map.m_lblk = 0;
1199 map.m_len = 1;
1200 map.m_flags = 0;
1201 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1202 if (error < 0)
1203 goto out_restore;
1204 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1205 error = -EIO;
1206 goto out_restore;
1207 }
1208
1209 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1210 if (!data_bh) {
1211 error = -ENOMEM;
1212 goto out_restore;
1213 }
1214
1215 lock_buffer(data_bh);
1216 error = ext4_journal_get_create_access(handle, data_bh);
1217 if (error) {
1218 unlock_buffer(data_bh);
1219 error = -EIO;
1220 goto out_restore;
1221 }
1222 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1223
1224 if (!S_ISDIR(inode->i_mode)) {
1225 memcpy(data_bh->b_data, buf, inline_size);
1226 set_buffer_uptodate(data_bh);
1227 error = ext4_handle_dirty_metadata(handle,
1228 inode, data_bh);
1229 } else {
1230 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1231 buf, inline_size);
1232 }
1233
1234 unlock_buffer(data_bh);
1235out_restore:
1236 if (error)
1237 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1238
1239out:
1240 brelse(data_bh);
1241 kfree(buf);
1242 return error;
1243}
1244
1245/*
1246 * Try to add the new entry to the inline data.
1247 * If succeeds, return 0. If not, extended the inline dir and copied data to
1248 * the new created block.
1249 */
1250int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1251 struct inode *dir, struct inode *inode)
1252{
1253 int ret, inline_size, no_expand;
1254 void *inline_start;
1255 struct ext4_iloc iloc;
1256
1257 ret = ext4_get_inode_loc(dir, &iloc);
1258 if (ret)
1259 return ret;
1260
1261 ext4_write_lock_xattr(dir, &no_expand);
1262 if (!ext4_has_inline_data(dir))
1263 goto out;
1264
1265 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1266 EXT4_INLINE_DOTDOT_SIZE;
1267 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1268
1269 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1270 inline_start, inline_size);
1271 if (ret != -ENOSPC)
1272 goto out;
1273
1274 /* check whether it can be inserted to inline xattr space. */
1275 inline_size = EXT4_I(dir)->i_inline_size -
1276 EXT4_MIN_INLINE_DATA_SIZE;
1277 if (!inline_size) {
1278 /* Try to use the xattr space.*/
1279 ret = ext4_update_inline_dir(handle, dir, &iloc);
1280 if (ret && ret != -ENOSPC)
1281 goto out;
1282
1283 inline_size = EXT4_I(dir)->i_inline_size -
1284 EXT4_MIN_INLINE_DATA_SIZE;
1285 }
1286
1287 if (inline_size) {
1288 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1289
1290 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1291 inode, &iloc, inline_start,
1292 inline_size);
1293
1294 if (ret != -ENOSPC)
1295 goto out;
1296 }
1297
1298 /*
1299 * The inline space is filled up, so create a new block for it.
1300 * As the extent tree will be created, we have to save the inline
1301 * dir first.
1302 */
1303 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1304
1305out:
1306 ext4_write_unlock_xattr(dir, &no_expand);
1307 ext4_mark_inode_dirty(handle, dir);
1308 brelse(iloc.bh);
1309 return ret;
1310}
1311
1312/*
1313 * This function fills a red-black tree with information from an
1314 * inlined dir. It returns the number directory entries loaded
1315 * into the tree. If there is an error it is returned in err.
1316 */
1317int htree_inlinedir_to_tree(struct file *dir_file,
1318 struct inode *dir, ext4_lblk_t block,
1319 struct dx_hash_info *hinfo,
1320 __u32 start_hash, __u32 start_minor_hash,
1321 int *has_inline_data)
1322{
1323 int err = 0, count = 0;
1324 unsigned int parent_ino;
1325 int pos;
1326 struct ext4_dir_entry_2 *de;
1327 struct inode *inode = file_inode(dir_file);
1328 int ret, inline_size = 0;
1329 struct ext4_iloc iloc;
1330 void *dir_buf = NULL;
1331 struct ext4_dir_entry_2 fake;
1332 struct fscrypt_str tmp_str;
1333
1334 ret = ext4_get_inode_loc(inode, &iloc);
1335 if (ret)
1336 return ret;
1337
1338 down_read(&EXT4_I(inode)->xattr_sem);
1339 if (!ext4_has_inline_data(inode)) {
1340 up_read(&EXT4_I(inode)->xattr_sem);
1341 *has_inline_data = 0;
1342 goto out;
1343 }
1344
1345 inline_size = ext4_get_inline_size(inode);
1346 dir_buf = kmalloc(inline_size, GFP_NOFS);
1347 if (!dir_buf) {
1348 ret = -ENOMEM;
1349 up_read(&EXT4_I(inode)->xattr_sem);
1350 goto out;
1351 }
1352
1353 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1354 up_read(&EXT4_I(inode)->xattr_sem);
1355 if (ret < 0)
1356 goto out;
1357
1358 pos = 0;
1359 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1360 while (pos < inline_size) {
1361 /*
1362 * As inlined dir doesn't store any information about '.' and
1363 * only the inode number of '..' is stored, we have to handle
1364 * them differently.
1365 */
1366 if (pos == 0) {
1367 fake.inode = cpu_to_le32(inode->i_ino);
1368 fake.name_len = 1;
1369 strcpy(fake.name, ".");
1370 fake.rec_len = ext4_rec_len_to_disk(
1371 EXT4_DIR_REC_LEN(fake.name_len),
1372 inline_size);
1373 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1374 de = &fake;
1375 pos = EXT4_INLINE_DOTDOT_OFFSET;
1376 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1377 fake.inode = cpu_to_le32(parent_ino);
1378 fake.name_len = 2;
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_SIZE;
1386 } else {
1387 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1388 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1389 if (ext4_check_dir_entry(inode, dir_file, de,
1390 iloc.bh, dir_buf,
1391 inline_size, pos)) {
1392 ret = count;
1393 goto out;
1394 }
1395 }
1396
1397 ext4fs_dirhash(de->name, de->name_len, hinfo);
1398 if ((hinfo->hash < start_hash) ||
1399 ((hinfo->hash == start_hash) &&
1400 (hinfo->minor_hash < start_minor_hash)))
1401 continue;
1402 if (de->inode == 0)
1403 continue;
1404 tmp_str.name = de->name;
1405 tmp_str.len = de->name_len;
1406 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1407 hinfo->minor_hash, de, &tmp_str);
1408 if (err) {
1409 count = err;
1410 goto out;
1411 }
1412 count++;
1413 }
1414 ret = count;
1415out:
1416 kfree(dir_buf);
1417 brelse(iloc.bh);
1418 return ret;
1419}
1420
1421/*
1422 * So this function is called when the volume is mkfsed with
1423 * dir_index disabled. In order to keep f_pos persistent
1424 * after we convert from an inlined dir to a blocked based,
1425 * we just pretend that we are a normal dir and return the
1426 * offset as if '.' and '..' really take place.
1427 *
1428 */
1429int ext4_read_inline_dir(struct file *file,
1430 struct dir_context *ctx,
1431 int *has_inline_data)
1432{
1433 unsigned int offset, parent_ino;
1434 int i;
1435 struct ext4_dir_entry_2 *de;
1436 struct super_block *sb;
1437 struct inode *inode = file_inode(file);
1438 int ret, inline_size = 0;
1439 struct ext4_iloc iloc;
1440 void *dir_buf = NULL;
1441 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1442
1443 ret = ext4_get_inode_loc(inode, &iloc);
1444 if (ret)
1445 return ret;
1446
1447 down_read(&EXT4_I(inode)->xattr_sem);
1448 if (!ext4_has_inline_data(inode)) {
1449 up_read(&EXT4_I(inode)->xattr_sem);
1450 *has_inline_data = 0;
1451 goto out;
1452 }
1453
1454 inline_size = ext4_get_inline_size(inode);
1455 dir_buf = kmalloc(inline_size, GFP_NOFS);
1456 if (!dir_buf) {
1457 ret = -ENOMEM;
1458 up_read(&EXT4_I(inode)->xattr_sem);
1459 goto out;
1460 }
1461
1462 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1463 up_read(&EXT4_I(inode)->xattr_sem);
1464 if (ret < 0)
1465 goto out;
1466
1467 ret = 0;
1468 sb = inode->i_sb;
1469 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1470 offset = ctx->pos;
1471
1472 /*
1473 * dotdot_offset and dotdot_size is the real offset and
1474 * size for ".." and "." if the dir is block based while
1475 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1476 * So we will use extra_offset and extra_size to indicate them
1477 * during the inline dir iteration.
1478 */
1479 dotdot_offset = EXT4_DIR_REC_LEN(1);
1480 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1481 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1482 extra_size = extra_offset + inline_size;
1483
1484 /*
1485 * If the version has changed since the last call to
1486 * readdir(2), then we might be pointing to an invalid
1487 * dirent right now. Scan from the start of the inline
1488 * dir to make sure.
1489 */
1490 if (!inode_eq_iversion(inode, file->f_version)) {
1491 for (i = 0; i < extra_size && i < offset;) {
1492 /*
1493 * "." is with offset 0 and
1494 * ".." is dotdot_offset.
1495 */
1496 if (!i) {
1497 i = dotdot_offset;
1498 continue;
1499 } else if (i == dotdot_offset) {
1500 i = dotdot_size;
1501 continue;
1502 }
1503 /* for other entry, the real offset in
1504 * the buf has to be tuned accordingly.
1505 */
1506 de = (struct ext4_dir_entry_2 *)
1507 (dir_buf + i - extra_offset);
1508 /* It's too expensive to do a full
1509 * dirent test each time round this
1510 * loop, but we do have to test at
1511 * least that it is non-zero. A
1512 * failure will be detected in the
1513 * dirent test below. */
1514 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1515 < EXT4_DIR_REC_LEN(1))
1516 break;
1517 i += ext4_rec_len_from_disk(de->rec_len,
1518 extra_size);
1519 }
1520 offset = i;
1521 ctx->pos = offset;
1522 file->f_version = inode_query_iversion(inode);
1523 }
1524
1525 while (ctx->pos < extra_size) {
1526 if (ctx->pos == 0) {
1527 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1528 goto out;
1529 ctx->pos = dotdot_offset;
1530 continue;
1531 }
1532
1533 if (ctx->pos == dotdot_offset) {
1534 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1535 goto out;
1536 ctx->pos = dotdot_size;
1537 continue;
1538 }
1539
1540 de = (struct ext4_dir_entry_2 *)
1541 (dir_buf + ctx->pos - extra_offset);
1542 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1543 extra_size, ctx->pos))
1544 goto out;
1545 if (le32_to_cpu(de->inode)) {
1546 if (!dir_emit(ctx, de->name, de->name_len,
1547 le32_to_cpu(de->inode),
1548 get_dtype(sb, de->file_type)))
1549 goto out;
1550 }
1551 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1552 }
1553out:
1554 kfree(dir_buf);
1555 brelse(iloc.bh);
1556 return ret;
1557}
1558
1559struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1560 struct ext4_dir_entry_2 **parent_de,
1561 int *retval)
1562{
1563 struct ext4_iloc iloc;
1564
1565 *retval = ext4_get_inode_loc(inode, &iloc);
1566 if (*retval)
1567 return NULL;
1568
1569 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1570
1571 return iloc.bh;
1572}
1573
1574/*
1575 * Try to create the inline data for the new dir.
1576 * If it succeeds, return 0, otherwise return the error.
1577 * In case of ENOSPC, the caller should create the normal disk layout dir.
1578 */
1579int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1580 struct inode *inode)
1581{
1582 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1583 struct ext4_iloc iloc;
1584 struct ext4_dir_entry_2 *de;
1585
1586 ret = ext4_get_inode_loc(inode, &iloc);
1587 if (ret)
1588 return ret;
1589
1590 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1591 if (ret)
1592 goto out;
1593
1594 /*
1595 * For inline dir, we only save the inode information for the ".."
1596 * and create a fake dentry to cover the left space.
1597 */
1598 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1599 de->inode = cpu_to_le32(parent->i_ino);
1600 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1601 de->inode = 0;
1602 de->rec_len = ext4_rec_len_to_disk(
1603 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1604 inline_size);
1605 set_nlink(inode, 2);
1606 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1607out:
1608 brelse(iloc.bh);
1609 return ret;
1610}
1611
1612struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1613 struct ext4_filename *fname,
1614 struct ext4_dir_entry_2 **res_dir,
1615 int *has_inline_data)
1616{
1617 int ret;
1618 struct ext4_iloc iloc;
1619 void *inline_start;
1620 int inline_size;
1621
1622 if (ext4_get_inode_loc(dir, &iloc))
1623 return NULL;
1624
1625 down_read(&EXT4_I(dir)->xattr_sem);
1626 if (!ext4_has_inline_data(dir)) {
1627 *has_inline_data = 0;
1628 goto out;
1629 }
1630
1631 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1632 EXT4_INLINE_DOTDOT_SIZE;
1633 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1634 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1635 dir, fname, 0, res_dir);
1636 if (ret == 1)
1637 goto out_find;
1638 if (ret < 0)
1639 goto out;
1640
1641 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1642 goto out;
1643
1644 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1645 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1646
1647 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1648 dir, fname, 0, res_dir);
1649 if (ret == 1)
1650 goto out_find;
1651
1652out:
1653 brelse(iloc.bh);
1654 iloc.bh = NULL;
1655out_find:
1656 up_read(&EXT4_I(dir)->xattr_sem);
1657 return iloc.bh;
1658}
1659
1660int ext4_delete_inline_entry(handle_t *handle,
1661 struct inode *dir,
1662 struct ext4_dir_entry_2 *de_del,
1663 struct buffer_head *bh,
1664 int *has_inline_data)
1665{
1666 int err, inline_size, no_expand;
1667 struct ext4_iloc iloc;
1668 void *inline_start;
1669
1670 err = ext4_get_inode_loc(dir, &iloc);
1671 if (err)
1672 return err;
1673
1674 ext4_write_lock_xattr(dir, &no_expand);
1675 if (!ext4_has_inline_data(dir)) {
1676 *has_inline_data = 0;
1677 goto out;
1678 }
1679
1680 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1681 EXT4_MIN_INLINE_DATA_SIZE) {
1682 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1683 EXT4_INLINE_DOTDOT_SIZE;
1684 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1685 EXT4_INLINE_DOTDOT_SIZE;
1686 } else {
1687 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1688 inline_size = ext4_get_inline_size(dir) -
1689 EXT4_MIN_INLINE_DATA_SIZE;
1690 }
1691
1692 BUFFER_TRACE(bh, "get_write_access");
1693 err = ext4_journal_get_write_access(handle, bh);
1694 if (err)
1695 goto out;
1696
1697 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1698 inline_start, inline_size, 0);
1699 if (err)
1700 goto out;
1701
1702 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1703out:
1704 ext4_write_unlock_xattr(dir, &no_expand);
1705 if (likely(err == 0))
1706 err = ext4_mark_inode_dirty(handle, dir);
1707 brelse(iloc.bh);
1708 if (err != -ENOENT)
1709 ext4_std_error(dir->i_sb, err);
1710 return err;
1711}
1712
1713/*
1714 * Get the inline dentry at offset.
1715 */
1716static inline struct ext4_dir_entry_2 *
1717ext4_get_inline_entry(struct inode *inode,
1718 struct ext4_iloc *iloc,
1719 unsigned int offset,
1720 void **inline_start,
1721 int *inline_size)
1722{
1723 void *inline_pos;
1724
1725 BUG_ON(offset > ext4_get_inline_size(inode));
1726
1727 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1728 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1729 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1730 } else {
1731 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1732 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1733 *inline_size = ext4_get_inline_size(inode) -
1734 EXT4_MIN_INLINE_DATA_SIZE;
1735 }
1736
1737 if (inline_start)
1738 *inline_start = inline_pos;
1739 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1740}
1741
1742bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1743{
1744 int err, inline_size;
1745 struct ext4_iloc iloc;
1746 void *inline_pos;
1747 unsigned int offset;
1748 struct ext4_dir_entry_2 *de;
1749 bool ret = true;
1750
1751 err = ext4_get_inode_loc(dir, &iloc);
1752 if (err) {
1753 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1754 err, dir->i_ino);
1755 return true;
1756 }
1757
1758 down_read(&EXT4_I(dir)->xattr_sem);
1759 if (!ext4_has_inline_data(dir)) {
1760 *has_inline_data = 0;
1761 goto out;
1762 }
1763
1764 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1765 if (!le32_to_cpu(de->inode)) {
1766 ext4_warning(dir->i_sb,
1767 "bad inline directory (dir #%lu) - no `..'",
1768 dir->i_ino);
1769 ret = true;
1770 goto out;
1771 }
1772
1773 offset = EXT4_INLINE_DOTDOT_SIZE;
1774 while (offset < dir->i_size) {
1775 de = ext4_get_inline_entry(dir, &iloc, offset,
1776 &inline_pos, &inline_size);
1777 if (ext4_check_dir_entry(dir, NULL, de,
1778 iloc.bh, inline_pos,
1779 inline_size, offset)) {
1780 ext4_warning(dir->i_sb,
1781 "bad inline directory (dir #%lu) - "
1782 "inode %u, rec_len %u, name_len %d"
1783 "inline size %d",
1784 dir->i_ino, le32_to_cpu(de->inode),
1785 le16_to_cpu(de->rec_len), de->name_len,
1786 inline_size);
1787 ret = true;
1788 goto out;
1789 }
1790 if (le32_to_cpu(de->inode)) {
1791 ret = false;
1792 goto out;
1793 }
1794 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1795 }
1796
1797out:
1798 up_read(&EXT4_I(dir)->xattr_sem);
1799 brelse(iloc.bh);
1800 return ret;
1801}
1802
1803int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1804{
1805 int ret, no_expand;
1806
1807 ext4_write_lock_xattr(inode, &no_expand);
1808 ret = ext4_destroy_inline_data_nolock(handle, inode);
1809 ext4_write_unlock_xattr(inode, &no_expand);
1810
1811 return ret;
1812}
1813
1814int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1815{
1816 __u64 addr;
1817 int error = -EAGAIN;
1818 struct ext4_iloc iloc;
1819
1820 down_read(&EXT4_I(inode)->xattr_sem);
1821 if (!ext4_has_inline_data(inode))
1822 goto out;
1823
1824 error = ext4_get_inode_loc(inode, &iloc);
1825 if (error)
1826 goto out;
1827
1828 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1829 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1830 addr += offsetof(struct ext4_inode, i_block);
1831
1832 brelse(iloc.bh);
1833
1834 iomap->addr = addr;
1835 iomap->offset = 0;
1836 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1837 i_size_read(inode));
1838 iomap->type = 0;
1839 iomap->flags = IOMAP_F_DATA_INLINE;
1840
1841out:
1842 up_read(&EXT4_I(inode)->xattr_sem);
1843 return error;
1844}
1845
1846int ext4_inline_data_fiemap(struct inode *inode,
1847 struct fiemap_extent_info *fieinfo,
1848 int *has_inline, __u64 start, __u64 len)
1849{
1850 __u64 physical = 0;
1851 __u64 inline_len;
1852 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1853 FIEMAP_EXTENT_LAST;
1854 int error = 0;
1855 struct ext4_iloc iloc;
1856
1857 down_read(&EXT4_I(inode)->xattr_sem);
1858 if (!ext4_has_inline_data(inode)) {
1859 *has_inline = 0;
1860 goto out;
1861 }
1862 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1863 i_size_read(inode));
1864 if (start >= inline_len)
1865 goto out;
1866 if (start + len < inline_len)
1867 inline_len = start + len;
1868 inline_len -= start;
1869
1870 error = ext4_get_inode_loc(inode, &iloc);
1871 if (error)
1872 goto out;
1873
1874 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1875 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1876 physical += offsetof(struct ext4_inode, i_block);
1877
1878 if (physical)
1879 error = fiemap_fill_next_extent(fieinfo, start, physical,
1880 inline_len, flags);
1881 brelse(iloc.bh);
1882out:
1883 up_read(&EXT4_I(inode)->xattr_sem);
1884 return (error < 0 ? error : 0);
1885}
1886
1887/*
1888 * Called during xattr set, and if we can sparse space 'needed',
1889 * just create the extent tree evict the data to the outer block.
1890 *
1891 * We use jbd2 instead of page cache to move data to the 1st block
1892 * so that the whole transaction can be committed as a whole and
1893 * the data isn't lost because of the delayed page cache write.
1894 */
1895int ext4_try_to_evict_inline_data(handle_t *handle,
1896 struct inode *inode,
1897 int needed)
1898{
1899 int error;
1900 struct ext4_xattr_entry *entry;
1901 struct ext4_inode *raw_inode;
1902 struct ext4_iloc iloc;
1903
1904 error = ext4_get_inode_loc(inode, &iloc);
1905 if (error)
1906 return error;
1907
1908 raw_inode = ext4_raw_inode(&iloc);
1909 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1910 EXT4_I(inode)->i_inline_off);
1911 if (EXT4_XATTR_LEN(entry->e_name_len) +
1912 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1913 error = -ENOSPC;
1914 goto out;
1915 }
1916
1917 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1918out:
1919 brelse(iloc.bh);
1920 return error;
1921}
1922
1923int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1924{
1925 handle_t *handle;
1926 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1927 size_t i_size;
1928 void *value = NULL;
1929 struct ext4_xattr_ibody_find is = {
1930 .s = { .not_found = -ENODATA, },
1931 };
1932 struct ext4_xattr_info i = {
1933 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1934 .name = EXT4_XATTR_SYSTEM_DATA,
1935 };
1936
1937
1938 needed_blocks = ext4_writepage_trans_blocks(inode);
1939 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1940 if (IS_ERR(handle))
1941 return PTR_ERR(handle);
1942
1943 ext4_write_lock_xattr(inode, &no_expand);
1944 if (!ext4_has_inline_data(inode)) {
1945 *has_inline = 0;
1946 ext4_journal_stop(handle);
1947 return 0;
1948 }
1949
1950 if ((err = ext4_orphan_add(handle, inode)) != 0)
1951 goto out;
1952
1953 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1954 goto out;
1955
1956 down_write(&EXT4_I(inode)->i_data_sem);
1957 i_size = inode->i_size;
1958 inline_size = ext4_get_inline_size(inode);
1959 EXT4_I(inode)->i_disksize = i_size;
1960
1961 if (i_size < inline_size) {
1962 /* Clear the content in the xattr space. */
1963 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1964 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1965 goto out_error;
1966
1967 BUG_ON(is.s.not_found);
1968
1969 value_len = le32_to_cpu(is.s.here->e_value_size);
1970 value = kmalloc(value_len, GFP_NOFS);
1971 if (!value) {
1972 err = -ENOMEM;
1973 goto out_error;
1974 }
1975
1976 err = ext4_xattr_ibody_get(inode, i.name_index,
1977 i.name, value, value_len);
1978 if (err <= 0)
1979 goto out_error;
1980
1981 i.value = value;
1982 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1983 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1984 err = ext4_xattr_ibody_inline_set(handle, inode,
1985 &i, &is);
1986 if (err)
1987 goto out_error;
1988 }
1989
1990 /* Clear the content within i_blocks. */
1991 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1992 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1993 memset(p + i_size, 0,
1994 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1995 }
1996
1997 EXT4_I(inode)->i_inline_size = i_size <
1998 EXT4_MIN_INLINE_DATA_SIZE ?
1999 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2000 }
2001
2002out_error:
2003 up_write(&EXT4_I(inode)->i_data_sem);
2004out:
2005 brelse(is.iloc.bh);
2006 ext4_write_unlock_xattr(inode, &no_expand);
2007 kfree(value);
2008 if (inode->i_nlink)
2009 ext4_orphan_del(handle, inode);
2010
2011 if (err == 0) {
2012 inode->i_mtime = inode->i_ctime = current_time(inode);
2013 err = ext4_mark_inode_dirty(handle, inode);
2014 if (IS_SYNC(inode))
2015 ext4_handle_sync(handle);
2016 }
2017 ext4_journal_stop(handle);
2018 return err;
2019}
2020
2021int ext4_convert_inline_data(struct inode *inode)
2022{
2023 int error, needed_blocks, no_expand;
2024 handle_t *handle;
2025 struct ext4_iloc iloc;
2026
2027 if (!ext4_has_inline_data(inode)) {
2028 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2029 return 0;
2030 }
2031
2032 needed_blocks = ext4_writepage_trans_blocks(inode);
2033
2034 iloc.bh = NULL;
2035 error = ext4_get_inode_loc(inode, &iloc);
2036 if (error)
2037 return error;
2038
2039 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2040 if (IS_ERR(handle)) {
2041 error = PTR_ERR(handle);
2042 goto out_free;
2043 }
2044
2045 ext4_write_lock_xattr(inode, &no_expand);
2046 if (ext4_has_inline_data(inode))
2047 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2048 ext4_write_unlock_xattr(inode, &no_expand);
2049 ext4_journal_stop(handle);
2050out_free:
2051 brelse(iloc.bh);
2052 return error;
2053}