Loading...
1/*
2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 */
14
15#include <linux/fiemap.h>
16
17#include "ext4_jbd2.h"
18#include "ext4.h"
19#include "xattr.h"
20#include "truncate.h"
21
22#define EXT4_XATTR_SYSTEM_DATA "data"
23#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
24#define EXT4_INLINE_DOTDOT_OFFSET 2
25#define EXT4_INLINE_DOTDOT_SIZE 4
26
27static int ext4_get_inline_size(struct inode *inode)
28{
29 if (EXT4_I(inode)->i_inline_off)
30 return EXT4_I(inode)->i_inline_size;
31
32 return 0;
33}
34
35static int get_max_inline_xattr_value_size(struct inode *inode,
36 struct ext4_iloc *iloc)
37{
38 struct ext4_xattr_ibody_header *header;
39 struct ext4_xattr_entry *entry;
40 struct ext4_inode *raw_inode;
41 int free, min_offs;
42
43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 EXT4_GOOD_OLD_INODE_SIZE -
45 EXT4_I(inode)->i_extra_isize -
46 sizeof(struct ext4_xattr_ibody_header);
47
48 /*
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
52 */
53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 return EXT4_XATTR_SIZE(min_offs -
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 EXT4_XATTR_ROUND - sizeof(__u32));
57
58 raw_inode = ext4_raw_inode(iloc);
59 header = IHDR(inode, raw_inode);
60 entry = IFIRST(header);
61
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64 if (!entry->e_value_block && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
66 if (offs < min_offs)
67 min_offs = offs;
68 }
69 }
70 free = min_offs -
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78 goto out;
79 }
80
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85 else
86 free = 0;
87
88out:
89 return free;
90}
91
92/*
93 * Get the maximum size we now can store in an inode.
94 * If we can't find the space for a xattr entry, don't use the space
95 * of the extents since we have no space to indicate the inline data.
96 */
97int ext4_get_max_inline_size(struct inode *inode)
98{
99 int error, max_inline_size;
100 struct ext4_iloc iloc;
101
102 if (EXT4_I(inode)->i_extra_isize == 0)
103 return 0;
104
105 error = ext4_get_inode_loc(inode, &iloc);
106 if (error) {
107 ext4_error_inode(inode, __func__, __LINE__, 0,
108 "can't get inode location %lu",
109 inode->i_ino);
110 return 0;
111 }
112
113 down_read(&EXT4_I(inode)->xattr_sem);
114 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
115 up_read(&EXT4_I(inode)->xattr_sem);
116
117 brelse(iloc.bh);
118
119 if (!max_inline_size)
120 return 0;
121
122 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
123}
124
125/*
126 * this function does not take xattr_sem, which is OK because it is
127 * currently only used in a code path coming form ext4_iget, before
128 * the new inode has been unlocked
129 */
130int ext4_find_inline_data_nolock(struct inode *inode)
131{
132 struct ext4_xattr_ibody_find is = {
133 .s = { .not_found = -ENODATA, },
134 };
135 struct ext4_xattr_info i = {
136 .name_index = EXT4_XATTR_INDEX_SYSTEM,
137 .name = EXT4_XATTR_SYSTEM_DATA,
138 };
139 int error;
140
141 if (EXT4_I(inode)->i_extra_isize == 0)
142 return 0;
143
144 error = ext4_get_inode_loc(inode, &is.iloc);
145 if (error)
146 return error;
147
148 error = ext4_xattr_ibody_find(inode, &i, &is);
149 if (error)
150 goto out;
151
152 if (!is.s.not_found) {
153 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154 (void *)ext4_raw_inode(&is.iloc));
155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156 le32_to_cpu(is.s.here->e_value_size);
157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158 }
159out:
160 brelse(is.iloc.bh);
161 return error;
162}
163
164static int ext4_read_inline_data(struct inode *inode, void *buffer,
165 unsigned int len,
166 struct ext4_iloc *iloc)
167{
168 struct ext4_xattr_entry *entry;
169 struct ext4_xattr_ibody_header *header;
170 int cp_len = 0;
171 struct ext4_inode *raw_inode;
172
173 if (!len)
174 return 0;
175
176 BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179 len : EXT4_MIN_INLINE_DATA_SIZE;
180
181 raw_inode = ext4_raw_inode(iloc);
182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184 len -= cp_len;
185 buffer += cp_len;
186
187 if (!len)
188 goto out;
189
190 header = IHDR(inode, raw_inode);
191 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192 EXT4_I(inode)->i_inline_off);
193 len = min_t(unsigned int, len,
194 (unsigned int)le32_to_cpu(entry->e_value_size));
195
196 memcpy(buffer,
197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198 cp_len += len;
199
200out:
201 return cp_len;
202}
203
204/*
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
209 */
210static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211 void *buffer, loff_t pos, unsigned int len)
212{
213 struct ext4_xattr_entry *entry;
214 struct ext4_xattr_ibody_header *header;
215 struct ext4_inode *raw_inode;
216 int cp_len = 0;
217
218 BUG_ON(!EXT4_I(inode)->i_inline_off);
219 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
220
221 raw_inode = ext4_raw_inode(iloc);
222 buffer += pos;
223
224 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
225 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
226 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
227 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
228
229 len -= cp_len;
230 buffer += cp_len;
231 pos += cp_len;
232 }
233
234 if (!len)
235 return;
236
237 pos -= EXT4_MIN_INLINE_DATA_SIZE;
238 header = IHDR(inode, raw_inode);
239 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
240 EXT4_I(inode)->i_inline_off);
241
242 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
243 buffer, len);
244}
245
246static int ext4_create_inline_data(handle_t *handle,
247 struct inode *inode, unsigned len)
248{
249 int error;
250 void *value = NULL;
251 struct ext4_xattr_ibody_find is = {
252 .s = { .not_found = -ENODATA, },
253 };
254 struct ext4_xattr_info i = {
255 .name_index = EXT4_XATTR_INDEX_SYSTEM,
256 .name = EXT4_XATTR_SYSTEM_DATA,
257 };
258
259 error = ext4_get_inode_loc(inode, &is.iloc);
260 if (error)
261 return error;
262
263 BUFFER_TRACE(is.iloc.bh, "get_write_access");
264 error = ext4_journal_get_write_access(handle, is.iloc.bh);
265 if (error)
266 goto out;
267
268 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
269 value = EXT4_ZERO_XATTR_VALUE;
270 len -= EXT4_MIN_INLINE_DATA_SIZE;
271 } else {
272 value = "";
273 len = 0;
274 }
275
276 /* Insert the the xttr entry. */
277 i.value = value;
278 i.value_len = len;
279
280 error = ext4_xattr_ibody_find(inode, &i, &is);
281 if (error)
282 goto out;
283
284 BUG_ON(!is.s.not_found);
285
286 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
287 if (error) {
288 if (error == -ENOSPC)
289 ext4_clear_inode_state(inode,
290 EXT4_STATE_MAY_INLINE_DATA);
291 goto out;
292 }
293
294 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
295 0, EXT4_MIN_INLINE_DATA_SIZE);
296
297 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
298 (void *)ext4_raw_inode(&is.iloc));
299 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
300 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
301 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
302 get_bh(is.iloc.bh);
303 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
304
305out:
306 brelse(is.iloc.bh);
307 return error;
308}
309
310static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
311 unsigned int len)
312{
313 int error;
314 void *value = NULL;
315 struct ext4_xattr_ibody_find is = {
316 .s = { .not_found = -ENODATA, },
317 };
318 struct ext4_xattr_info i = {
319 .name_index = EXT4_XATTR_INDEX_SYSTEM,
320 .name = EXT4_XATTR_SYSTEM_DATA,
321 };
322
323 /* If the old space is ok, write the data directly. */
324 if (len <= EXT4_I(inode)->i_inline_size)
325 return 0;
326
327 error = ext4_get_inode_loc(inode, &is.iloc);
328 if (error)
329 return error;
330
331 error = ext4_xattr_ibody_find(inode, &i, &is);
332 if (error)
333 goto out;
334
335 BUG_ON(is.s.not_found);
336
337 len -= EXT4_MIN_INLINE_DATA_SIZE;
338 value = kzalloc(len, GFP_NOFS);
339 if (!value)
340 goto out;
341
342 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
343 value, len);
344 if (error == -ENODATA)
345 goto out;
346
347 BUFFER_TRACE(is.iloc.bh, "get_write_access");
348 error = ext4_journal_get_write_access(handle, is.iloc.bh);
349 if (error)
350 goto out;
351
352 /* Update the xttr entry. */
353 i.value = value;
354 i.value_len = len;
355
356 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
357 if (error)
358 goto out;
359
360 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
361 (void *)ext4_raw_inode(&is.iloc));
362 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
363 le32_to_cpu(is.s.here->e_value_size);
364 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
365 get_bh(is.iloc.bh);
366 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
367
368out:
369 kfree(value);
370 brelse(is.iloc.bh);
371 return error;
372}
373
374static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
375 unsigned int len)
376{
377 int ret, size;
378 struct ext4_inode_info *ei = EXT4_I(inode);
379
380 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
381 return -ENOSPC;
382
383 size = ext4_get_max_inline_size(inode);
384 if (size < len)
385 return -ENOSPC;
386
387 down_write(&EXT4_I(inode)->xattr_sem);
388
389 if (ei->i_inline_off)
390 ret = ext4_update_inline_data(handle, inode, len);
391 else
392 ret = ext4_create_inline_data(handle, inode, len);
393
394 up_write(&EXT4_I(inode)->xattr_sem);
395
396 return ret;
397}
398
399static int ext4_destroy_inline_data_nolock(handle_t *handle,
400 struct inode *inode)
401{
402 struct ext4_inode_info *ei = EXT4_I(inode);
403 struct ext4_xattr_ibody_find is = {
404 .s = { .not_found = 0, },
405 };
406 struct ext4_xattr_info i = {
407 .name_index = EXT4_XATTR_INDEX_SYSTEM,
408 .name = EXT4_XATTR_SYSTEM_DATA,
409 .value = NULL,
410 .value_len = 0,
411 };
412 int error;
413
414 if (!ei->i_inline_off)
415 return 0;
416
417 error = ext4_get_inode_loc(inode, &is.iloc);
418 if (error)
419 return error;
420
421 error = ext4_xattr_ibody_find(inode, &i, &is);
422 if (error)
423 goto out;
424
425 BUFFER_TRACE(is.iloc.bh, "get_write_access");
426 error = ext4_journal_get_write_access(handle, is.iloc.bh);
427 if (error)
428 goto out;
429
430 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
431 if (error)
432 goto out;
433
434 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
435 0, EXT4_MIN_INLINE_DATA_SIZE);
436
437 if (ext4_has_feature_extents(inode->i_sb)) {
438 if (S_ISDIR(inode->i_mode) ||
439 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
440 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
441 ext4_ext_tree_init(handle, inode);
442 }
443 }
444 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
445
446 get_bh(is.iloc.bh);
447 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
448
449 EXT4_I(inode)->i_inline_off = 0;
450 EXT4_I(inode)->i_inline_size = 0;
451 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
452out:
453 brelse(is.iloc.bh);
454 if (error == -ENODATA)
455 error = 0;
456 return error;
457}
458
459static int ext4_read_inline_page(struct inode *inode, struct page *page)
460{
461 void *kaddr;
462 int ret = 0;
463 size_t len;
464 struct ext4_iloc iloc;
465
466 BUG_ON(!PageLocked(page));
467 BUG_ON(!ext4_has_inline_data(inode));
468 BUG_ON(page->index);
469
470 if (!EXT4_I(inode)->i_inline_off) {
471 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
472 inode->i_ino);
473 goto out;
474 }
475
476 ret = ext4_get_inode_loc(inode, &iloc);
477 if (ret)
478 goto out;
479
480 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
481 kaddr = kmap_atomic(page);
482 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
483 flush_dcache_page(page);
484 kunmap_atomic(kaddr);
485 zero_user_segment(page, len, PAGE_SIZE);
486 SetPageUptodate(page);
487 brelse(iloc.bh);
488
489out:
490 return ret;
491}
492
493int ext4_readpage_inline(struct inode *inode, struct page *page)
494{
495 int ret = 0;
496
497 down_read(&EXT4_I(inode)->xattr_sem);
498 if (!ext4_has_inline_data(inode)) {
499 up_read(&EXT4_I(inode)->xattr_sem);
500 return -EAGAIN;
501 }
502
503 /*
504 * Current inline data can only exist in the 1st page,
505 * So for all the other pages, just set them uptodate.
506 */
507 if (!page->index)
508 ret = ext4_read_inline_page(inode, page);
509 else if (!PageUptodate(page)) {
510 zero_user_segment(page, 0, PAGE_SIZE);
511 SetPageUptodate(page);
512 }
513
514 up_read(&EXT4_I(inode)->xattr_sem);
515
516 unlock_page(page);
517 return ret >= 0 ? 0 : ret;
518}
519
520static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
521 struct inode *inode,
522 unsigned flags)
523{
524 int ret, needed_blocks;
525 handle_t *handle = NULL;
526 int retries = 0, sem_held = 0;
527 struct page *page = NULL;
528 unsigned from, to;
529 struct ext4_iloc iloc;
530
531 if (!ext4_has_inline_data(inode)) {
532 /*
533 * clear the flag so that no new write
534 * will trap here again.
535 */
536 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
537 return 0;
538 }
539
540 needed_blocks = ext4_writepage_trans_blocks(inode);
541
542 ret = ext4_get_inode_loc(inode, &iloc);
543 if (ret)
544 return ret;
545
546retry:
547 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
548 if (IS_ERR(handle)) {
549 ret = PTR_ERR(handle);
550 handle = NULL;
551 goto out;
552 }
553
554 /* We cannot recurse into the filesystem as the transaction is already
555 * started */
556 flags |= AOP_FLAG_NOFS;
557
558 page = grab_cache_page_write_begin(mapping, 0, flags);
559 if (!page) {
560 ret = -ENOMEM;
561 goto out;
562 }
563
564 down_write(&EXT4_I(inode)->xattr_sem);
565 sem_held = 1;
566 /* If some one has already done this for us, just exit. */
567 if (!ext4_has_inline_data(inode)) {
568 ret = 0;
569 goto out;
570 }
571
572 from = 0;
573 to = ext4_get_inline_size(inode);
574 if (!PageUptodate(page)) {
575 ret = ext4_read_inline_page(inode, page);
576 if (ret < 0)
577 goto out;
578 }
579
580 ret = ext4_destroy_inline_data_nolock(handle, inode);
581 if (ret)
582 goto out;
583
584 if (ext4_should_dioread_nolock(inode)) {
585 ret = __block_write_begin(page, from, to,
586 ext4_get_block_unwritten);
587 } else
588 ret = __block_write_begin(page, from, to, ext4_get_block);
589
590 if (!ret && ext4_should_journal_data(inode)) {
591 ret = ext4_walk_page_buffers(handle, page_buffers(page),
592 from, to, NULL,
593 do_journal_get_write_access);
594 }
595
596 if (ret) {
597 unlock_page(page);
598 put_page(page);
599 page = NULL;
600 ext4_orphan_add(handle, inode);
601 up_write(&EXT4_I(inode)->xattr_sem);
602 sem_held = 0;
603 ext4_journal_stop(handle);
604 handle = NULL;
605 ext4_truncate_failed_write(inode);
606 /*
607 * If truncate failed early the inode might
608 * still be on the orphan list; we need to
609 * make sure the inode is removed from the
610 * orphan list in that case.
611 */
612 if (inode->i_nlink)
613 ext4_orphan_del(NULL, inode);
614 }
615
616 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617 goto retry;
618
619 if (page)
620 block_commit_write(page, from, to);
621out:
622 if (page) {
623 unlock_page(page);
624 put_page(page);
625 }
626 if (sem_held)
627 up_write(&EXT4_I(inode)->xattr_sem);
628 if (handle)
629 ext4_journal_stop(handle);
630 brelse(iloc.bh);
631 return ret;
632}
633
634/*
635 * Try to write data in the inode.
636 * If the inode has inline data, check whether the new write can be
637 * in the inode also. If not, create the page the handle, move the data
638 * to the page make it update and let the later codes create extent for it.
639 */
640int ext4_try_to_write_inline_data(struct address_space *mapping,
641 struct inode *inode,
642 loff_t pos, unsigned len,
643 unsigned flags,
644 struct page **pagep)
645{
646 int ret;
647 handle_t *handle;
648 struct page *page;
649 struct ext4_iloc iloc;
650
651 if (pos + len > ext4_get_max_inline_size(inode))
652 goto convert;
653
654 ret = ext4_get_inode_loc(inode, &iloc);
655 if (ret)
656 return ret;
657
658 /*
659 * The possible write could happen in the inode,
660 * so try to reserve the space in inode first.
661 */
662 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
663 if (IS_ERR(handle)) {
664 ret = PTR_ERR(handle);
665 handle = NULL;
666 goto out;
667 }
668
669 ret = ext4_prepare_inline_data(handle, inode, pos + len);
670 if (ret && ret != -ENOSPC)
671 goto out;
672
673 /* We don't have space in inline inode, so convert it to extent. */
674 if (ret == -ENOSPC) {
675 ext4_journal_stop(handle);
676 brelse(iloc.bh);
677 goto convert;
678 }
679
680 flags |= AOP_FLAG_NOFS;
681
682 page = grab_cache_page_write_begin(mapping, 0, flags);
683 if (!page) {
684 ret = -ENOMEM;
685 goto out;
686 }
687
688 *pagep = page;
689 down_read(&EXT4_I(inode)->xattr_sem);
690 if (!ext4_has_inline_data(inode)) {
691 ret = 0;
692 unlock_page(page);
693 put_page(page);
694 goto out_up_read;
695 }
696
697 if (!PageUptodate(page)) {
698 ret = ext4_read_inline_page(inode, page);
699 if (ret < 0)
700 goto out_up_read;
701 }
702
703 ret = 1;
704 handle = NULL;
705out_up_read:
706 up_read(&EXT4_I(inode)->xattr_sem);
707out:
708 if (handle)
709 ext4_journal_stop(handle);
710 brelse(iloc.bh);
711 return ret;
712convert:
713 return ext4_convert_inline_data_to_extent(mapping,
714 inode, flags);
715}
716
717int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
718 unsigned copied, struct page *page)
719{
720 int ret;
721 void *kaddr;
722 struct ext4_iloc iloc;
723
724 if (unlikely(copied < len)) {
725 if (!PageUptodate(page)) {
726 copied = 0;
727 goto out;
728 }
729 }
730
731 ret = ext4_get_inode_loc(inode, &iloc);
732 if (ret) {
733 ext4_std_error(inode->i_sb, ret);
734 copied = 0;
735 goto out;
736 }
737
738 down_write(&EXT4_I(inode)->xattr_sem);
739 BUG_ON(!ext4_has_inline_data(inode));
740
741 kaddr = kmap_atomic(page);
742 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
743 kunmap_atomic(kaddr);
744 SetPageUptodate(page);
745 /* clear page dirty so that writepages wouldn't work for us. */
746 ClearPageDirty(page);
747
748 up_write(&EXT4_I(inode)->xattr_sem);
749 brelse(iloc.bh);
750out:
751 return copied;
752}
753
754struct buffer_head *
755ext4_journalled_write_inline_data(struct inode *inode,
756 unsigned len,
757 struct page *page)
758{
759 int ret;
760 void *kaddr;
761 struct ext4_iloc iloc;
762
763 ret = ext4_get_inode_loc(inode, &iloc);
764 if (ret) {
765 ext4_std_error(inode->i_sb, ret);
766 return NULL;
767 }
768
769 down_write(&EXT4_I(inode)->xattr_sem);
770 kaddr = kmap_atomic(page);
771 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
772 kunmap_atomic(kaddr);
773 up_write(&EXT4_I(inode)->xattr_sem);
774
775 return iloc.bh;
776}
777
778/*
779 * Try to make the page cache and handle ready for the inline data case.
780 * We can call this function in 2 cases:
781 * 1. The inode is created and the first write exceeds inline size. We can
782 * clear the inode state safely.
783 * 2. The inode has inline data, then we need to read the data, make it
784 * update and dirty so that ext4_da_writepages can handle it. We don't
785 * need to start the journal since the file's metatdata isn't changed now.
786 */
787static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
788 struct inode *inode,
789 unsigned flags,
790 void **fsdata)
791{
792 int ret = 0, inline_size;
793 struct page *page;
794
795 page = grab_cache_page_write_begin(mapping, 0, flags);
796 if (!page)
797 return -ENOMEM;
798
799 down_read(&EXT4_I(inode)->xattr_sem);
800 if (!ext4_has_inline_data(inode)) {
801 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
802 goto out;
803 }
804
805 inline_size = ext4_get_inline_size(inode);
806
807 if (!PageUptodate(page)) {
808 ret = ext4_read_inline_page(inode, page);
809 if (ret < 0)
810 goto out;
811 }
812
813 ret = __block_write_begin(page, 0, inline_size,
814 ext4_da_get_block_prep);
815 if (ret) {
816 up_read(&EXT4_I(inode)->xattr_sem);
817 unlock_page(page);
818 put_page(page);
819 ext4_truncate_failed_write(inode);
820 return ret;
821 }
822
823 SetPageDirty(page);
824 SetPageUptodate(page);
825 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
826 *fsdata = (void *)CONVERT_INLINE_DATA;
827
828out:
829 up_read(&EXT4_I(inode)->xattr_sem);
830 if (page) {
831 unlock_page(page);
832 put_page(page);
833 }
834 return ret;
835}
836
837/*
838 * Prepare the write for the inline data.
839 * If the the data can be written into the inode, we just read
840 * the page and make it uptodate, and start the journal.
841 * Otherwise read the page, makes it dirty so that it can be
842 * handle in writepages(the i_disksize update is left to the
843 * normal ext4_da_write_end).
844 */
845int ext4_da_write_inline_data_begin(struct address_space *mapping,
846 struct inode *inode,
847 loff_t pos, unsigned len,
848 unsigned flags,
849 struct page **pagep,
850 void **fsdata)
851{
852 int ret, inline_size;
853 handle_t *handle;
854 struct page *page;
855 struct ext4_iloc iloc;
856 int retries;
857
858 ret = ext4_get_inode_loc(inode, &iloc);
859 if (ret)
860 return ret;
861
862retry_journal:
863 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
864 if (IS_ERR(handle)) {
865 ret = PTR_ERR(handle);
866 goto out;
867 }
868
869 inline_size = ext4_get_max_inline_size(inode);
870
871 ret = -ENOSPC;
872 if (inline_size >= pos + len) {
873 ret = ext4_prepare_inline_data(handle, inode, pos + len);
874 if (ret && ret != -ENOSPC)
875 goto out_journal;
876 }
877
878 /*
879 * We cannot recurse into the filesystem as the transaction
880 * is already started.
881 */
882 flags |= AOP_FLAG_NOFS;
883
884 if (ret == -ENOSPC) {
885 ret = ext4_da_convert_inline_data_to_extent(mapping,
886 inode,
887 flags,
888 fsdata);
889 ext4_journal_stop(handle);
890 if (ret == -ENOSPC &&
891 ext4_should_retry_alloc(inode->i_sb, &retries))
892 goto retry_journal;
893 goto out;
894 }
895
896
897 page = grab_cache_page_write_begin(mapping, 0, flags);
898 if (!page) {
899 ret = -ENOMEM;
900 goto out_journal;
901 }
902
903 down_read(&EXT4_I(inode)->xattr_sem);
904 if (!ext4_has_inline_data(inode)) {
905 ret = 0;
906 goto out_release_page;
907 }
908
909 if (!PageUptodate(page)) {
910 ret = ext4_read_inline_page(inode, page);
911 if (ret < 0)
912 goto out_release_page;
913 }
914
915 up_read(&EXT4_I(inode)->xattr_sem);
916 *pagep = page;
917 brelse(iloc.bh);
918 return 1;
919out_release_page:
920 up_read(&EXT4_I(inode)->xattr_sem);
921 unlock_page(page);
922 put_page(page);
923out_journal:
924 ext4_journal_stop(handle);
925out:
926 brelse(iloc.bh);
927 return ret;
928}
929
930int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
931 unsigned len, unsigned copied,
932 struct page *page)
933{
934 int i_size_changed = 0;
935
936 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
937
938 /*
939 * No need to use i_size_read() here, the i_size
940 * cannot change under us because we hold i_mutex.
941 *
942 * But it's important to update i_size while still holding page lock:
943 * page writeout could otherwise come in and zero beyond i_size.
944 */
945 if (pos+copied > inode->i_size) {
946 i_size_write(inode, pos+copied);
947 i_size_changed = 1;
948 }
949 unlock_page(page);
950 put_page(page);
951
952 /*
953 * Don't mark the inode dirty under page lock. First, it unnecessarily
954 * makes the holding time of page lock longer. Second, it forces lock
955 * ordering of page lock and transaction start for journaling
956 * filesystems.
957 */
958 if (i_size_changed)
959 mark_inode_dirty(inode);
960
961 return copied;
962}
963
964#ifdef INLINE_DIR_DEBUG
965void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
966 void *inline_start, int inline_size)
967{
968 int offset;
969 unsigned short de_len;
970 struct ext4_dir_entry_2 *de = inline_start;
971 void *dlimit = inline_start + inline_size;
972
973 trace_printk("inode %lu\n", dir->i_ino);
974 offset = 0;
975 while ((void *)de < dlimit) {
976 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
977 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
978 offset, de_len, de->name_len, de->name,
979 de->name_len, le32_to_cpu(de->inode));
980 if (ext4_check_dir_entry(dir, NULL, de, bh,
981 inline_start, inline_size, offset))
982 BUG();
983
984 offset += de_len;
985 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
986 }
987}
988#else
989#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
990#endif
991
992/*
993 * Add a new entry into a inline dir.
994 * It will return -ENOSPC if no space is available, and -EIO
995 * and -EEXIST if directory entry already exists.
996 */
997static int ext4_add_dirent_to_inline(handle_t *handle,
998 struct ext4_filename *fname,
999 struct inode *dir,
1000 struct inode *inode,
1001 struct ext4_iloc *iloc,
1002 void *inline_start, int inline_size)
1003{
1004 int err;
1005 struct ext4_dir_entry_2 *de;
1006
1007 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1008 inline_size, fname, &de);
1009 if (err)
1010 return err;
1011
1012 BUFFER_TRACE(iloc->bh, "get_write_access");
1013 err = ext4_journal_get_write_access(handle, iloc->bh);
1014 if (err)
1015 return err;
1016 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1017
1018 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1019
1020 /*
1021 * XXX shouldn't update any times until successful
1022 * completion of syscall, but too many callers depend
1023 * on this.
1024 *
1025 * XXX similarly, too many callers depend on
1026 * ext4_new_inode() setting the times, but error
1027 * recovery deletes the inode, so the worst that can
1028 * happen is that the times are slightly out of date
1029 * and/or different from the directory change time.
1030 */
1031 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1032 ext4_update_dx_flag(dir);
1033 dir->i_version++;
1034 ext4_mark_inode_dirty(handle, dir);
1035 return 1;
1036}
1037
1038static void *ext4_get_inline_xattr_pos(struct inode *inode,
1039 struct ext4_iloc *iloc)
1040{
1041 struct ext4_xattr_entry *entry;
1042 struct ext4_xattr_ibody_header *header;
1043
1044 BUG_ON(!EXT4_I(inode)->i_inline_off);
1045
1046 header = IHDR(inode, ext4_raw_inode(iloc));
1047 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1048 EXT4_I(inode)->i_inline_off);
1049
1050 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1051}
1052
1053/* Set the final de to cover the whole block. */
1054static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1055{
1056 struct ext4_dir_entry_2 *de, *prev_de;
1057 void *limit;
1058 int de_len;
1059
1060 de = (struct ext4_dir_entry_2 *)de_buf;
1061 if (old_size) {
1062 limit = de_buf + old_size;
1063 do {
1064 prev_de = de;
1065 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1066 de_buf += de_len;
1067 de = (struct ext4_dir_entry_2 *)de_buf;
1068 } while (de_buf < limit);
1069
1070 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1071 old_size, new_size);
1072 } else {
1073 /* this is just created, so create an empty entry. */
1074 de->inode = 0;
1075 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1076 }
1077}
1078
1079static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1080 struct ext4_iloc *iloc)
1081{
1082 int ret;
1083 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1084 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1085
1086 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1087 return -ENOSPC;
1088
1089 ret = ext4_update_inline_data(handle, dir,
1090 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1091 if (ret)
1092 return ret;
1093
1094 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1095 EXT4_I(dir)->i_inline_size -
1096 EXT4_MIN_INLINE_DATA_SIZE);
1097 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1098 return 0;
1099}
1100
1101static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1102 struct ext4_iloc *iloc,
1103 void *buf, int inline_size)
1104{
1105 ext4_create_inline_data(handle, inode, inline_size);
1106 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1107 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1108}
1109
1110static int ext4_finish_convert_inline_dir(handle_t *handle,
1111 struct inode *inode,
1112 struct buffer_head *dir_block,
1113 void *buf,
1114 int inline_size)
1115{
1116 int err, csum_size = 0, header_size = 0;
1117 struct ext4_dir_entry_2 *de;
1118 struct ext4_dir_entry_tail *t;
1119 void *target = dir_block->b_data;
1120
1121 /*
1122 * First create "." and ".." and then copy the dir information
1123 * back to the block.
1124 */
1125 de = (struct ext4_dir_entry_2 *)target;
1126 de = ext4_init_dot_dotdot(inode, de,
1127 inode->i_sb->s_blocksize, csum_size,
1128 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1129 header_size = (void *)de - target;
1130
1131 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1132 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1133
1134 if (ext4_has_metadata_csum(inode->i_sb))
1135 csum_size = sizeof(struct ext4_dir_entry_tail);
1136
1137 inode->i_size = inode->i_sb->s_blocksize;
1138 i_size_write(inode, inode->i_sb->s_blocksize);
1139 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1140 ext4_update_final_de(dir_block->b_data,
1141 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1142 inode->i_sb->s_blocksize - csum_size);
1143
1144 if (csum_size) {
1145 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1146 inode->i_sb->s_blocksize);
1147 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1148 }
1149 set_buffer_uptodate(dir_block);
1150 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1151 if (err)
1152 goto out;
1153 set_buffer_verified(dir_block);
1154out:
1155 return err;
1156}
1157
1158static int ext4_convert_inline_data_nolock(handle_t *handle,
1159 struct inode *inode,
1160 struct ext4_iloc *iloc)
1161{
1162 int error;
1163 void *buf = NULL;
1164 struct buffer_head *data_bh = NULL;
1165 struct ext4_map_blocks map;
1166 int inline_size;
1167
1168 inline_size = ext4_get_inline_size(inode);
1169 buf = kmalloc(inline_size, GFP_NOFS);
1170 if (!buf) {
1171 error = -ENOMEM;
1172 goto out;
1173 }
1174
1175 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1176 if (error < 0)
1177 goto out;
1178
1179 /*
1180 * Make sure the inline directory entries pass checks before we try to
1181 * convert them, so that we avoid touching stuff that needs fsck.
1182 */
1183 if (S_ISDIR(inode->i_mode)) {
1184 error = ext4_check_all_de(inode, iloc->bh,
1185 buf + EXT4_INLINE_DOTDOT_SIZE,
1186 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1187 if (error)
1188 goto out;
1189 }
1190
1191 error = ext4_destroy_inline_data_nolock(handle, inode);
1192 if (error)
1193 goto out;
1194
1195 map.m_lblk = 0;
1196 map.m_len = 1;
1197 map.m_flags = 0;
1198 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1199 if (error < 0)
1200 goto out_restore;
1201 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1202 error = -EIO;
1203 goto out_restore;
1204 }
1205
1206 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1207 if (!data_bh) {
1208 error = -ENOMEM;
1209 goto out_restore;
1210 }
1211
1212 lock_buffer(data_bh);
1213 error = ext4_journal_get_create_access(handle, data_bh);
1214 if (error) {
1215 unlock_buffer(data_bh);
1216 error = -EIO;
1217 goto out_restore;
1218 }
1219 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1220
1221 if (!S_ISDIR(inode->i_mode)) {
1222 memcpy(data_bh->b_data, buf, inline_size);
1223 set_buffer_uptodate(data_bh);
1224 error = ext4_handle_dirty_metadata(handle,
1225 inode, data_bh);
1226 } else {
1227 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1228 buf, inline_size);
1229 }
1230
1231 unlock_buffer(data_bh);
1232out_restore:
1233 if (error)
1234 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1235
1236out:
1237 brelse(data_bh);
1238 kfree(buf);
1239 return error;
1240}
1241
1242/*
1243 * Try to add the new entry to the inline data.
1244 * If succeeds, return 0. If not, extended the inline dir and copied data to
1245 * the new created block.
1246 */
1247int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1248 struct inode *dir, struct inode *inode)
1249{
1250 int ret, inline_size;
1251 void *inline_start;
1252 struct ext4_iloc iloc;
1253
1254 ret = ext4_get_inode_loc(dir, &iloc);
1255 if (ret)
1256 return ret;
1257
1258 down_write(&EXT4_I(dir)->xattr_sem);
1259 if (!ext4_has_inline_data(dir))
1260 goto out;
1261
1262 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1263 EXT4_INLINE_DOTDOT_SIZE;
1264 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1265
1266 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1267 inline_start, inline_size);
1268 if (ret != -ENOSPC)
1269 goto out;
1270
1271 /* check whether it can be inserted to inline xattr space. */
1272 inline_size = EXT4_I(dir)->i_inline_size -
1273 EXT4_MIN_INLINE_DATA_SIZE;
1274 if (!inline_size) {
1275 /* Try to use the xattr space.*/
1276 ret = ext4_update_inline_dir(handle, dir, &iloc);
1277 if (ret && ret != -ENOSPC)
1278 goto out;
1279
1280 inline_size = EXT4_I(dir)->i_inline_size -
1281 EXT4_MIN_INLINE_DATA_SIZE;
1282 }
1283
1284 if (inline_size) {
1285 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1286
1287 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1288 inode, &iloc, inline_start,
1289 inline_size);
1290
1291 if (ret != -ENOSPC)
1292 goto out;
1293 }
1294
1295 /*
1296 * The inline space is filled up, so create a new block for it.
1297 * As the extent tree will be created, we have to save the inline
1298 * dir first.
1299 */
1300 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1301
1302out:
1303 ext4_mark_inode_dirty(handle, dir);
1304 up_write(&EXT4_I(dir)->xattr_sem);
1305 brelse(iloc.bh);
1306 return ret;
1307}
1308
1309/*
1310 * This function fills a red-black tree with information from an
1311 * inlined dir. It returns the number directory entries loaded
1312 * into the tree. If there is an error it is returned in err.
1313 */
1314int htree_inlinedir_to_tree(struct file *dir_file,
1315 struct inode *dir, ext4_lblk_t block,
1316 struct dx_hash_info *hinfo,
1317 __u32 start_hash, __u32 start_minor_hash,
1318 int *has_inline_data)
1319{
1320 int err = 0, count = 0;
1321 unsigned int parent_ino;
1322 int pos;
1323 struct ext4_dir_entry_2 *de;
1324 struct inode *inode = file_inode(dir_file);
1325 int ret, inline_size = 0;
1326 struct ext4_iloc iloc;
1327 void *dir_buf = NULL;
1328 struct ext4_dir_entry_2 fake;
1329 struct ext4_str tmp_str;
1330
1331 ret = ext4_get_inode_loc(inode, &iloc);
1332 if (ret)
1333 return ret;
1334
1335 down_read(&EXT4_I(inode)->xattr_sem);
1336 if (!ext4_has_inline_data(inode)) {
1337 up_read(&EXT4_I(inode)->xattr_sem);
1338 *has_inline_data = 0;
1339 goto out;
1340 }
1341
1342 inline_size = ext4_get_inline_size(inode);
1343 dir_buf = kmalloc(inline_size, GFP_NOFS);
1344 if (!dir_buf) {
1345 ret = -ENOMEM;
1346 up_read(&EXT4_I(inode)->xattr_sem);
1347 goto out;
1348 }
1349
1350 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1351 up_read(&EXT4_I(inode)->xattr_sem);
1352 if (ret < 0)
1353 goto out;
1354
1355 pos = 0;
1356 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1357 while (pos < inline_size) {
1358 /*
1359 * As inlined dir doesn't store any information about '.' and
1360 * only the inode number of '..' is stored, we have to handle
1361 * them differently.
1362 */
1363 if (pos == 0) {
1364 fake.inode = cpu_to_le32(inode->i_ino);
1365 fake.name_len = 1;
1366 strcpy(fake.name, ".");
1367 fake.rec_len = ext4_rec_len_to_disk(
1368 EXT4_DIR_REC_LEN(fake.name_len),
1369 inline_size);
1370 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1371 de = &fake;
1372 pos = EXT4_INLINE_DOTDOT_OFFSET;
1373 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1374 fake.inode = cpu_to_le32(parent_ino);
1375 fake.name_len = 2;
1376 strcpy(fake.name, "..");
1377 fake.rec_len = ext4_rec_len_to_disk(
1378 EXT4_DIR_REC_LEN(fake.name_len),
1379 inline_size);
1380 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1381 de = &fake;
1382 pos = EXT4_INLINE_DOTDOT_SIZE;
1383 } else {
1384 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1385 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1386 if (ext4_check_dir_entry(inode, dir_file, de,
1387 iloc.bh, dir_buf,
1388 inline_size, pos)) {
1389 ret = count;
1390 goto out;
1391 }
1392 }
1393
1394 ext4fs_dirhash(de->name, de->name_len, hinfo);
1395 if ((hinfo->hash < start_hash) ||
1396 ((hinfo->hash == start_hash) &&
1397 (hinfo->minor_hash < start_minor_hash)))
1398 continue;
1399 if (de->inode == 0)
1400 continue;
1401 tmp_str.name = de->name;
1402 tmp_str.len = de->name_len;
1403 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1404 hinfo->minor_hash, de, &tmp_str);
1405 if (err) {
1406 count = err;
1407 goto out;
1408 }
1409 count++;
1410 }
1411 ret = count;
1412out:
1413 kfree(dir_buf);
1414 brelse(iloc.bh);
1415 return ret;
1416}
1417
1418/*
1419 * So this function is called when the volume is mkfsed with
1420 * dir_index disabled. In order to keep f_pos persistent
1421 * after we convert from an inlined dir to a blocked based,
1422 * we just pretend that we are a normal dir and return the
1423 * offset as if '.' and '..' really take place.
1424 *
1425 */
1426int ext4_read_inline_dir(struct file *file,
1427 struct dir_context *ctx,
1428 int *has_inline_data)
1429{
1430 unsigned int offset, parent_ino;
1431 int i;
1432 struct ext4_dir_entry_2 *de;
1433 struct super_block *sb;
1434 struct inode *inode = file_inode(file);
1435 int ret, inline_size = 0;
1436 struct ext4_iloc iloc;
1437 void *dir_buf = NULL;
1438 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1439
1440 ret = ext4_get_inode_loc(inode, &iloc);
1441 if (ret)
1442 return ret;
1443
1444 down_read(&EXT4_I(inode)->xattr_sem);
1445 if (!ext4_has_inline_data(inode)) {
1446 up_read(&EXT4_I(inode)->xattr_sem);
1447 *has_inline_data = 0;
1448 goto out;
1449 }
1450
1451 inline_size = ext4_get_inline_size(inode);
1452 dir_buf = kmalloc(inline_size, GFP_NOFS);
1453 if (!dir_buf) {
1454 ret = -ENOMEM;
1455 up_read(&EXT4_I(inode)->xattr_sem);
1456 goto out;
1457 }
1458
1459 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1460 up_read(&EXT4_I(inode)->xattr_sem);
1461 if (ret < 0)
1462 goto out;
1463
1464 ret = 0;
1465 sb = inode->i_sb;
1466 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1467 offset = ctx->pos;
1468
1469 /*
1470 * dotdot_offset and dotdot_size is the real offset and
1471 * size for ".." and "." if the dir is block based while
1472 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1473 * So we will use extra_offset and extra_size to indicate them
1474 * during the inline dir iteration.
1475 */
1476 dotdot_offset = EXT4_DIR_REC_LEN(1);
1477 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1478 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1479 extra_size = extra_offset + inline_size;
1480
1481 /*
1482 * If the version has changed since the last call to
1483 * readdir(2), then we might be pointing to an invalid
1484 * dirent right now. Scan from the start of the inline
1485 * dir to make sure.
1486 */
1487 if (file->f_version != inode->i_version) {
1488 for (i = 0; i < extra_size && i < offset;) {
1489 /*
1490 * "." is with offset 0 and
1491 * ".." is dotdot_offset.
1492 */
1493 if (!i) {
1494 i = dotdot_offset;
1495 continue;
1496 } else if (i == dotdot_offset) {
1497 i = dotdot_size;
1498 continue;
1499 }
1500 /* for other entry, the real offset in
1501 * the buf has to be tuned accordingly.
1502 */
1503 de = (struct ext4_dir_entry_2 *)
1504 (dir_buf + i - extra_offset);
1505 /* It's too expensive to do a full
1506 * dirent test each time round this
1507 * loop, but we do have to test at
1508 * least that it is non-zero. A
1509 * failure will be detected in the
1510 * dirent test below. */
1511 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1512 < EXT4_DIR_REC_LEN(1))
1513 break;
1514 i += ext4_rec_len_from_disk(de->rec_len,
1515 extra_size);
1516 }
1517 offset = i;
1518 ctx->pos = offset;
1519 file->f_version = inode->i_version;
1520 }
1521
1522 while (ctx->pos < extra_size) {
1523 if (ctx->pos == 0) {
1524 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1525 goto out;
1526 ctx->pos = dotdot_offset;
1527 continue;
1528 }
1529
1530 if (ctx->pos == dotdot_offset) {
1531 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1532 goto out;
1533 ctx->pos = dotdot_size;
1534 continue;
1535 }
1536
1537 de = (struct ext4_dir_entry_2 *)
1538 (dir_buf + ctx->pos - extra_offset);
1539 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1540 extra_size, ctx->pos))
1541 goto out;
1542 if (le32_to_cpu(de->inode)) {
1543 if (!dir_emit(ctx, de->name, de->name_len,
1544 le32_to_cpu(de->inode),
1545 get_dtype(sb, de->file_type)))
1546 goto out;
1547 }
1548 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1549 }
1550out:
1551 kfree(dir_buf);
1552 brelse(iloc.bh);
1553 return ret;
1554}
1555
1556struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1557 struct ext4_dir_entry_2 **parent_de,
1558 int *retval)
1559{
1560 struct ext4_iloc iloc;
1561
1562 *retval = ext4_get_inode_loc(inode, &iloc);
1563 if (*retval)
1564 return NULL;
1565
1566 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1567
1568 return iloc.bh;
1569}
1570
1571/*
1572 * Try to create the inline data for the new dir.
1573 * If it succeeds, return 0, otherwise return the error.
1574 * In case of ENOSPC, the caller should create the normal disk layout dir.
1575 */
1576int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1577 struct inode *inode)
1578{
1579 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1580 struct ext4_iloc iloc;
1581 struct ext4_dir_entry_2 *de;
1582
1583 ret = ext4_get_inode_loc(inode, &iloc);
1584 if (ret)
1585 return ret;
1586
1587 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1588 if (ret)
1589 goto out;
1590
1591 /*
1592 * For inline dir, we only save the inode information for the ".."
1593 * and create a fake dentry to cover the left space.
1594 */
1595 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1596 de->inode = cpu_to_le32(parent->i_ino);
1597 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1598 de->inode = 0;
1599 de->rec_len = ext4_rec_len_to_disk(
1600 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1601 inline_size);
1602 set_nlink(inode, 2);
1603 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1604out:
1605 brelse(iloc.bh);
1606 return ret;
1607}
1608
1609struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1610 struct ext4_filename *fname,
1611 const struct qstr *d_name,
1612 struct ext4_dir_entry_2 **res_dir,
1613 int *has_inline_data)
1614{
1615 int ret;
1616 struct ext4_iloc iloc;
1617 void *inline_start;
1618 int inline_size;
1619
1620 if (ext4_get_inode_loc(dir, &iloc))
1621 return NULL;
1622
1623 down_read(&EXT4_I(dir)->xattr_sem);
1624 if (!ext4_has_inline_data(dir)) {
1625 *has_inline_data = 0;
1626 goto out;
1627 }
1628
1629 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1630 EXT4_INLINE_DOTDOT_SIZE;
1631 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1632 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1633 dir, fname, d_name, 0, res_dir);
1634 if (ret == 1)
1635 goto out_find;
1636 if (ret < 0)
1637 goto out;
1638
1639 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1640 goto out;
1641
1642 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1643 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1644
1645 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1646 dir, fname, d_name, 0, res_dir);
1647 if (ret == 1)
1648 goto out_find;
1649
1650out:
1651 brelse(iloc.bh);
1652 iloc.bh = NULL;
1653out_find:
1654 up_read(&EXT4_I(dir)->xattr_sem);
1655 return iloc.bh;
1656}
1657
1658int ext4_delete_inline_entry(handle_t *handle,
1659 struct inode *dir,
1660 struct ext4_dir_entry_2 *de_del,
1661 struct buffer_head *bh,
1662 int *has_inline_data)
1663{
1664 int err, inline_size;
1665 struct ext4_iloc iloc;
1666 void *inline_start;
1667
1668 err = ext4_get_inode_loc(dir, &iloc);
1669 if (err)
1670 return err;
1671
1672 down_write(&EXT4_I(dir)->xattr_sem);
1673 if (!ext4_has_inline_data(dir)) {
1674 *has_inline_data = 0;
1675 goto out;
1676 }
1677
1678 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1679 EXT4_MIN_INLINE_DATA_SIZE) {
1680 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1681 EXT4_INLINE_DOTDOT_SIZE;
1682 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1683 EXT4_INLINE_DOTDOT_SIZE;
1684 } else {
1685 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1686 inline_size = ext4_get_inline_size(dir) -
1687 EXT4_MIN_INLINE_DATA_SIZE;
1688 }
1689
1690 BUFFER_TRACE(bh, "get_write_access");
1691 err = ext4_journal_get_write_access(handle, bh);
1692 if (err)
1693 goto out;
1694
1695 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1696 inline_start, inline_size, 0);
1697 if (err)
1698 goto out;
1699
1700 err = ext4_mark_inode_dirty(handle, dir);
1701 if (unlikely(err))
1702 goto out;
1703
1704 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1705out:
1706 up_write(&EXT4_I(dir)->xattr_sem);
1707 brelse(iloc.bh);
1708 if (err != -ENOENT)
1709 ext4_std_error(dir->i_sb, err);
1710 return err;
1711}
1712
1713/*
1714 * Get the inline dentry at offset.
1715 */
1716static inline struct ext4_dir_entry_2 *
1717ext4_get_inline_entry(struct inode *inode,
1718 struct ext4_iloc *iloc,
1719 unsigned int offset,
1720 void **inline_start,
1721 int *inline_size)
1722{
1723 void *inline_pos;
1724
1725 BUG_ON(offset > ext4_get_inline_size(inode));
1726
1727 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1728 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1729 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1730 } else {
1731 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1732 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1733 *inline_size = ext4_get_inline_size(inode) -
1734 EXT4_MIN_INLINE_DATA_SIZE;
1735 }
1736
1737 if (inline_start)
1738 *inline_start = inline_pos;
1739 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1740}
1741
1742int empty_inline_dir(struct inode *dir, int *has_inline_data)
1743{
1744 int err, inline_size;
1745 struct ext4_iloc iloc;
1746 void *inline_pos;
1747 unsigned int offset;
1748 struct ext4_dir_entry_2 *de;
1749 int ret = 1;
1750
1751 err = ext4_get_inode_loc(dir, &iloc);
1752 if (err) {
1753 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1754 err, dir->i_ino);
1755 return 1;
1756 }
1757
1758 down_read(&EXT4_I(dir)->xattr_sem);
1759 if (!ext4_has_inline_data(dir)) {
1760 *has_inline_data = 0;
1761 goto out;
1762 }
1763
1764 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1765 if (!le32_to_cpu(de->inode)) {
1766 ext4_warning(dir->i_sb,
1767 "bad inline directory (dir #%lu) - no `..'",
1768 dir->i_ino);
1769 ret = 1;
1770 goto out;
1771 }
1772
1773 offset = EXT4_INLINE_DOTDOT_SIZE;
1774 while (offset < dir->i_size) {
1775 de = ext4_get_inline_entry(dir, &iloc, offset,
1776 &inline_pos, &inline_size);
1777 if (ext4_check_dir_entry(dir, NULL, de,
1778 iloc.bh, inline_pos,
1779 inline_size, offset)) {
1780 ext4_warning(dir->i_sb,
1781 "bad inline directory (dir #%lu) - "
1782 "inode %u, rec_len %u, name_len %d"
1783 "inline size %d\n",
1784 dir->i_ino, le32_to_cpu(de->inode),
1785 le16_to_cpu(de->rec_len), de->name_len,
1786 inline_size);
1787 ret = 1;
1788 goto out;
1789 }
1790 if (le32_to_cpu(de->inode)) {
1791 ret = 0;
1792 goto out;
1793 }
1794 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1795 }
1796
1797out:
1798 up_read(&EXT4_I(dir)->xattr_sem);
1799 brelse(iloc.bh);
1800 return ret;
1801}
1802
1803int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1804{
1805 int ret;
1806
1807 down_write(&EXT4_I(inode)->xattr_sem);
1808 ret = ext4_destroy_inline_data_nolock(handle, inode);
1809 up_write(&EXT4_I(inode)->xattr_sem);
1810
1811 return ret;
1812}
1813
1814int ext4_inline_data_fiemap(struct inode *inode,
1815 struct fiemap_extent_info *fieinfo,
1816 int *has_inline, __u64 start, __u64 len)
1817{
1818 __u64 physical = 0;
1819 __u64 inline_len;
1820 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1821 FIEMAP_EXTENT_LAST;
1822 int error = 0;
1823 struct ext4_iloc iloc;
1824
1825 down_read(&EXT4_I(inode)->xattr_sem);
1826 if (!ext4_has_inline_data(inode)) {
1827 *has_inline = 0;
1828 goto out;
1829 }
1830 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1831 i_size_read(inode));
1832 if (start >= inline_len)
1833 goto out;
1834 if (start + len < inline_len)
1835 inline_len = start + len;
1836 inline_len -= start;
1837
1838 error = ext4_get_inode_loc(inode, &iloc);
1839 if (error)
1840 goto out;
1841
1842 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1843 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1844 physical += offsetof(struct ext4_inode, i_block);
1845
1846 if (physical)
1847 error = fiemap_fill_next_extent(fieinfo, start, physical,
1848 inline_len, flags);
1849 brelse(iloc.bh);
1850out:
1851 up_read(&EXT4_I(inode)->xattr_sem);
1852 return (error < 0 ? error : 0);
1853}
1854
1855/*
1856 * Called during xattr set, and if we can sparse space 'needed',
1857 * just create the extent tree evict the data to the outer block.
1858 *
1859 * We use jbd2 instead of page cache to move data to the 1st block
1860 * so that the whole transaction can be committed as a whole and
1861 * the data isn't lost because of the delayed page cache write.
1862 */
1863int ext4_try_to_evict_inline_data(handle_t *handle,
1864 struct inode *inode,
1865 int needed)
1866{
1867 int error;
1868 struct ext4_xattr_entry *entry;
1869 struct ext4_inode *raw_inode;
1870 struct ext4_iloc iloc;
1871
1872 error = ext4_get_inode_loc(inode, &iloc);
1873 if (error)
1874 return error;
1875
1876 raw_inode = ext4_raw_inode(&iloc);
1877 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1878 EXT4_I(inode)->i_inline_off);
1879 if (EXT4_XATTR_LEN(entry->e_name_len) +
1880 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1881 error = -ENOSPC;
1882 goto out;
1883 }
1884
1885 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1886out:
1887 brelse(iloc.bh);
1888 return error;
1889}
1890
1891void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1892{
1893 handle_t *handle;
1894 int inline_size, value_len, needed_blocks;
1895 size_t i_size;
1896 void *value = NULL;
1897 struct ext4_xattr_ibody_find is = {
1898 .s = { .not_found = -ENODATA, },
1899 };
1900 struct ext4_xattr_info i = {
1901 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1902 .name = EXT4_XATTR_SYSTEM_DATA,
1903 };
1904
1905
1906 needed_blocks = ext4_writepage_trans_blocks(inode);
1907 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1908 if (IS_ERR(handle))
1909 return;
1910
1911 down_write(&EXT4_I(inode)->xattr_sem);
1912 if (!ext4_has_inline_data(inode)) {
1913 *has_inline = 0;
1914 ext4_journal_stop(handle);
1915 return;
1916 }
1917
1918 if (ext4_orphan_add(handle, inode))
1919 goto out;
1920
1921 if (ext4_get_inode_loc(inode, &is.iloc))
1922 goto out;
1923
1924 down_write(&EXT4_I(inode)->i_data_sem);
1925 i_size = inode->i_size;
1926 inline_size = ext4_get_inline_size(inode);
1927 EXT4_I(inode)->i_disksize = i_size;
1928
1929 if (i_size < inline_size) {
1930 /* Clear the content in the xattr space. */
1931 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1932 if (ext4_xattr_ibody_find(inode, &i, &is))
1933 goto out_error;
1934
1935 BUG_ON(is.s.not_found);
1936
1937 value_len = le32_to_cpu(is.s.here->e_value_size);
1938 value = kmalloc(value_len, GFP_NOFS);
1939 if (!value)
1940 goto out_error;
1941
1942 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1943 value, value_len))
1944 goto out_error;
1945
1946 i.value = value;
1947 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1948 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1949 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1950 goto out_error;
1951 }
1952
1953 /* Clear the content within i_blocks. */
1954 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1955 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1956 memset(p + i_size, 0,
1957 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1958 }
1959
1960 EXT4_I(inode)->i_inline_size = i_size <
1961 EXT4_MIN_INLINE_DATA_SIZE ?
1962 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1963 }
1964
1965out_error:
1966 up_write(&EXT4_I(inode)->i_data_sem);
1967out:
1968 brelse(is.iloc.bh);
1969 up_write(&EXT4_I(inode)->xattr_sem);
1970 kfree(value);
1971 if (inode->i_nlink)
1972 ext4_orphan_del(handle, inode);
1973
1974 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1975 ext4_mark_inode_dirty(handle, inode);
1976 if (IS_SYNC(inode))
1977 ext4_handle_sync(handle);
1978
1979 ext4_journal_stop(handle);
1980 return;
1981}
1982
1983int ext4_convert_inline_data(struct inode *inode)
1984{
1985 int error, needed_blocks;
1986 handle_t *handle;
1987 struct ext4_iloc iloc;
1988
1989 if (!ext4_has_inline_data(inode)) {
1990 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1991 return 0;
1992 }
1993
1994 needed_blocks = ext4_writepage_trans_blocks(inode);
1995
1996 iloc.bh = NULL;
1997 error = ext4_get_inode_loc(inode, &iloc);
1998 if (error)
1999 return error;
2000
2001 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2002 if (IS_ERR(handle)) {
2003 error = PTR_ERR(handle);
2004 goto out_free;
2005 }
2006
2007 down_write(&EXT4_I(inode)->xattr_sem);
2008 if (!ext4_has_inline_data(inode)) {
2009 up_write(&EXT4_I(inode)->xattr_sem);
2010 goto out;
2011 }
2012
2013 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2014 up_write(&EXT4_I(inode)->xattr_sem);
2015out:
2016 ext4_journal_stop(handle);
2017out_free:
2018 brelse(iloc.bh);
2019 return error;
2020}
1// SPDX-License-Identifier: LGPL-2.1
2/*
3 * Copyright (c) 2012 Taobao.
4 * Written by Tao Ma <boyu.mt@taobao.com>
5 */
6
7#include <linux/iomap.h>
8#include <linux/fiemap.h>
9#include <linux/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}