Loading...
1// SPDX-License-Identifier: LGPL-2.1
2/*
3 * Copyright (c) 2012 Taobao.
4 * Written by Tao Ma <boyu.mt@taobao.com>
5 */
6
7#include <linux/iomap.h>
8#include <linux/fiemap.h>
9#include <linux/iversion.h>
10
11#include "ext4_jbd2.h"
12#include "ext4.h"
13#include "xattr.h"
14#include "truncate.h"
15
16#define EXT4_XATTR_SYSTEM_DATA "data"
17#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
18#define EXT4_INLINE_DOTDOT_OFFSET 2
19#define EXT4_INLINE_DOTDOT_SIZE 4
20
21static int ext4_get_inline_size(struct inode *inode)
22{
23 if (EXT4_I(inode)->i_inline_off)
24 return EXT4_I(inode)->i_inline_size;
25
26 return 0;
27}
28
29static int get_max_inline_xattr_value_size(struct inode *inode,
30 struct ext4_iloc *iloc)
31{
32 struct ext4_xattr_ibody_header *header;
33 struct ext4_xattr_entry *entry;
34 struct ext4_inode *raw_inode;
35 int free, min_offs;
36
37 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
38 EXT4_GOOD_OLD_INODE_SIZE -
39 EXT4_I(inode)->i_extra_isize -
40 sizeof(struct ext4_xattr_ibody_header);
41
42 /*
43 * We need to subtract another sizeof(__u32) since an in-inode xattr
44 * needs an empty 4 bytes to indicate the gap between the xattr entry
45 * and the name/value pair.
46 */
47 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
48 return EXT4_XATTR_SIZE(min_offs -
49 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
50 EXT4_XATTR_ROUND - sizeof(__u32));
51
52 raw_inode = ext4_raw_inode(iloc);
53 header = IHDR(inode, raw_inode);
54 entry = IFIRST(header);
55
56 /* Compute min_offs. */
57 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
58 if (!entry->e_value_inum && entry->e_value_size) {
59 size_t offs = le16_to_cpu(entry->e_value_offs);
60 if (offs < min_offs)
61 min_offs = offs;
62 }
63 }
64 free = min_offs -
65 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
66
67 if (EXT4_I(inode)->i_inline_off) {
68 entry = (struct ext4_xattr_entry *)
69 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
70
71 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
72 goto out;
73 }
74
75 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
76
77 if (free > EXT4_XATTR_ROUND)
78 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
79 else
80 free = 0;
81
82out:
83 return free;
84}
85
86/*
87 * Get the maximum size we now can store in an inode.
88 * If we can't find the space for a xattr entry, don't use the space
89 * of the extents since we have no space to indicate the inline data.
90 */
91int ext4_get_max_inline_size(struct inode *inode)
92{
93 int error, max_inline_size;
94 struct ext4_iloc iloc;
95
96 if (EXT4_I(inode)->i_extra_isize == 0)
97 return 0;
98
99 error = ext4_get_inode_loc(inode, &iloc);
100 if (error) {
101 ext4_error_inode(inode, __func__, __LINE__, 0,
102 "can't get inode location %lu",
103 inode->i_ino);
104 return 0;
105 }
106
107 down_read(&EXT4_I(inode)->xattr_sem);
108 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
109 up_read(&EXT4_I(inode)->xattr_sem);
110
111 brelse(iloc.bh);
112
113 if (!max_inline_size)
114 return 0;
115
116 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
117}
118
119/*
120 * this function does not take xattr_sem, which is OK because it is
121 * currently only used in a code path coming form ext4_iget, before
122 * the new inode has been unlocked
123 */
124int ext4_find_inline_data_nolock(struct inode *inode)
125{
126 struct ext4_xattr_ibody_find is = {
127 .s = { .not_found = -ENODATA, },
128 };
129 struct ext4_xattr_info i = {
130 .name_index = EXT4_XATTR_INDEX_SYSTEM,
131 .name = EXT4_XATTR_SYSTEM_DATA,
132 };
133 int error;
134
135 if (EXT4_I(inode)->i_extra_isize == 0)
136 return 0;
137
138 error = ext4_get_inode_loc(inode, &is.iloc);
139 if (error)
140 return error;
141
142 error = ext4_xattr_ibody_find(inode, &i, &is);
143 if (error)
144 goto out;
145
146 if (!is.s.not_found) {
147 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
148 (void *)ext4_raw_inode(&is.iloc));
149 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
150 le32_to_cpu(is.s.here->e_value_size);
151 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
152 }
153out:
154 brelse(is.iloc.bh);
155 return error;
156}
157
158static int ext4_read_inline_data(struct inode *inode, void *buffer,
159 unsigned int len,
160 struct ext4_iloc *iloc)
161{
162 struct ext4_xattr_entry *entry;
163 struct ext4_xattr_ibody_header *header;
164 int cp_len = 0;
165 struct ext4_inode *raw_inode;
166
167 if (!len)
168 return 0;
169
170 BUG_ON(len > EXT4_I(inode)->i_inline_size);
171
172 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
173 len : EXT4_MIN_INLINE_DATA_SIZE;
174
175 raw_inode = ext4_raw_inode(iloc);
176 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
177
178 len -= cp_len;
179 buffer += cp_len;
180
181 if (!len)
182 goto out;
183
184 header = IHDR(inode, raw_inode);
185 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
186 EXT4_I(inode)->i_inline_off);
187 len = min_t(unsigned int, len,
188 (unsigned int)le32_to_cpu(entry->e_value_size));
189
190 memcpy(buffer,
191 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
192 cp_len += len;
193
194out:
195 return cp_len;
196}
197
198/*
199 * write the buffer to the inline inode.
200 * If 'create' is set, we don't need to do the extra copy in the xattr
201 * value since it is already handled by ext4_xattr_ibody_inline_set.
202 * That saves us one memcpy.
203 */
204static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
205 void *buffer, loff_t pos, unsigned int len)
206{
207 struct ext4_xattr_entry *entry;
208 struct ext4_xattr_ibody_header *header;
209 struct ext4_inode *raw_inode;
210 int cp_len = 0;
211
212 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
213 return;
214
215 BUG_ON(!EXT4_I(inode)->i_inline_off);
216 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
217
218 raw_inode = ext4_raw_inode(iloc);
219 buffer += pos;
220
221 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
222 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
223 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
224 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
225
226 len -= cp_len;
227 buffer += cp_len;
228 pos += cp_len;
229 }
230
231 if (!len)
232 return;
233
234 pos -= EXT4_MIN_INLINE_DATA_SIZE;
235 header = IHDR(inode, raw_inode);
236 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
237 EXT4_I(inode)->i_inline_off);
238
239 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
240 buffer, len);
241}
242
243static int ext4_create_inline_data(handle_t *handle,
244 struct inode *inode, unsigned len)
245{
246 int error;
247 void *value = NULL;
248 struct ext4_xattr_ibody_find is = {
249 .s = { .not_found = -ENODATA, },
250 };
251 struct ext4_xattr_info i = {
252 .name_index = EXT4_XATTR_INDEX_SYSTEM,
253 .name = EXT4_XATTR_SYSTEM_DATA,
254 };
255
256 error = ext4_get_inode_loc(inode, &is.iloc);
257 if (error)
258 return error;
259
260 BUFFER_TRACE(is.iloc.bh, "get_write_access");
261 error = ext4_journal_get_write_access(handle, is.iloc.bh);
262 if (error)
263 goto out;
264
265 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
266 value = EXT4_ZERO_XATTR_VALUE;
267 len -= EXT4_MIN_INLINE_DATA_SIZE;
268 } else {
269 value = "";
270 len = 0;
271 }
272
273 /* Insert the the xttr entry. */
274 i.value = value;
275 i.value_len = len;
276
277 error = ext4_xattr_ibody_find(inode, &i, &is);
278 if (error)
279 goto out;
280
281 BUG_ON(!is.s.not_found);
282
283 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
284 if (error) {
285 if (error == -ENOSPC)
286 ext4_clear_inode_state(inode,
287 EXT4_STATE_MAY_INLINE_DATA);
288 goto out;
289 }
290
291 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
292 0, EXT4_MIN_INLINE_DATA_SIZE);
293
294 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
295 (void *)ext4_raw_inode(&is.iloc));
296 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
297 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
298 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
299 get_bh(is.iloc.bh);
300 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
301
302out:
303 brelse(is.iloc.bh);
304 return error;
305}
306
307static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
308 unsigned int len)
309{
310 int error;
311 void *value = NULL;
312 struct ext4_xattr_ibody_find is = {
313 .s = { .not_found = -ENODATA, },
314 };
315 struct ext4_xattr_info i = {
316 .name_index = EXT4_XATTR_INDEX_SYSTEM,
317 .name = EXT4_XATTR_SYSTEM_DATA,
318 };
319
320 /* If the old space is ok, write the data directly. */
321 if (len <= EXT4_I(inode)->i_inline_size)
322 return 0;
323
324 error = ext4_get_inode_loc(inode, &is.iloc);
325 if (error)
326 return error;
327
328 error = ext4_xattr_ibody_find(inode, &i, &is);
329 if (error)
330 goto out;
331
332 BUG_ON(is.s.not_found);
333
334 len -= EXT4_MIN_INLINE_DATA_SIZE;
335 value = kzalloc(len, GFP_NOFS);
336 if (!value) {
337 error = -ENOMEM;
338 goto out;
339 }
340
341 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
342 value, len);
343 if (error == -ENODATA)
344 goto out;
345
346 BUFFER_TRACE(is.iloc.bh, "get_write_access");
347 error = ext4_journal_get_write_access(handle, is.iloc.bh);
348 if (error)
349 goto out;
350
351 /* Update the xttr entry. */
352 i.value = value;
353 i.value_len = len;
354
355 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
356 if (error)
357 goto out;
358
359 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
360 (void *)ext4_raw_inode(&is.iloc));
361 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
362 le32_to_cpu(is.s.here->e_value_size);
363 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
364 get_bh(is.iloc.bh);
365 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
366
367out:
368 kfree(value);
369 brelse(is.iloc.bh);
370 return error;
371}
372
373static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
374 unsigned int len)
375{
376 int ret, size, no_expand;
377 struct ext4_inode_info *ei = EXT4_I(inode);
378
379 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
380 return -ENOSPC;
381
382 size = ext4_get_max_inline_size(inode);
383 if (size < len)
384 return -ENOSPC;
385
386 ext4_write_lock_xattr(inode, &no_expand);
387
388 if (ei->i_inline_off)
389 ret = ext4_update_inline_data(handle, inode, len);
390 else
391 ret = ext4_create_inline_data(handle, inode, len);
392
393 ext4_write_unlock_xattr(inode, &no_expand);
394 return ret;
395}
396
397static int ext4_destroy_inline_data_nolock(handle_t *handle,
398 struct inode *inode)
399{
400 struct ext4_inode_info *ei = EXT4_I(inode);
401 struct ext4_xattr_ibody_find is = {
402 .s = { .not_found = 0, },
403 };
404 struct ext4_xattr_info i = {
405 .name_index = EXT4_XATTR_INDEX_SYSTEM,
406 .name = EXT4_XATTR_SYSTEM_DATA,
407 .value = NULL,
408 .value_len = 0,
409 };
410 int error;
411
412 if (!ei->i_inline_off)
413 return 0;
414
415 error = ext4_get_inode_loc(inode, &is.iloc);
416 if (error)
417 return error;
418
419 error = ext4_xattr_ibody_find(inode, &i, &is);
420 if (error)
421 goto out;
422
423 BUFFER_TRACE(is.iloc.bh, "get_write_access");
424 error = ext4_journal_get_write_access(handle, is.iloc.bh);
425 if (error)
426 goto out;
427
428 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
429 if (error)
430 goto out;
431
432 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
433 0, EXT4_MIN_INLINE_DATA_SIZE);
434
435 if (ext4_has_feature_extents(inode->i_sb)) {
436 if (S_ISDIR(inode->i_mode) ||
437 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
438 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
439 ext4_ext_tree_init(handle, inode);
440 }
441 }
442 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
443
444 get_bh(is.iloc.bh);
445 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
446
447 EXT4_I(inode)->i_inline_off = 0;
448 EXT4_I(inode)->i_inline_size = 0;
449 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
450out:
451 brelse(is.iloc.bh);
452 if (error == -ENODATA)
453 error = 0;
454 return error;
455}
456
457static int ext4_read_inline_page(struct inode *inode, struct page *page)
458{
459 void *kaddr;
460 int ret = 0;
461 size_t len;
462 struct ext4_iloc iloc;
463
464 BUG_ON(!PageLocked(page));
465 BUG_ON(!ext4_has_inline_data(inode));
466 BUG_ON(page->index);
467
468 if (!EXT4_I(inode)->i_inline_off) {
469 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
470 inode->i_ino);
471 goto out;
472 }
473
474 ret = ext4_get_inode_loc(inode, &iloc);
475 if (ret)
476 goto out;
477
478 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
479 kaddr = kmap_atomic(page);
480 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
481 flush_dcache_page(page);
482 kunmap_atomic(kaddr);
483 zero_user_segment(page, len, PAGE_SIZE);
484 SetPageUptodate(page);
485 brelse(iloc.bh);
486
487out:
488 return ret;
489}
490
491int ext4_readpage_inline(struct inode *inode, struct page *page)
492{
493 int ret = 0;
494
495 down_read(&EXT4_I(inode)->xattr_sem);
496 if (!ext4_has_inline_data(inode)) {
497 up_read(&EXT4_I(inode)->xattr_sem);
498 return -EAGAIN;
499 }
500
501 /*
502 * Current inline data can only exist in the 1st page,
503 * So for all the other pages, just set them uptodate.
504 */
505 if (!page->index)
506 ret = ext4_read_inline_page(inode, page);
507 else if (!PageUptodate(page)) {
508 zero_user_segment(page, 0, PAGE_SIZE);
509 SetPageUptodate(page);
510 }
511
512 up_read(&EXT4_I(inode)->xattr_sem);
513
514 unlock_page(page);
515 return ret >= 0 ? 0 : ret;
516}
517
518static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
519 struct inode *inode,
520 unsigned flags)
521{
522 int ret, needed_blocks, no_expand;
523 handle_t *handle = NULL;
524 int retries = 0, sem_held = 0;
525 struct page *page = NULL;
526 unsigned from, to;
527 struct ext4_iloc iloc;
528
529 if (!ext4_has_inline_data(inode)) {
530 /*
531 * clear the flag so that no new write
532 * will trap here again.
533 */
534 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
535 return 0;
536 }
537
538 needed_blocks = ext4_writepage_trans_blocks(inode);
539
540 ret = ext4_get_inode_loc(inode, &iloc);
541 if (ret)
542 return ret;
543
544retry:
545 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
546 if (IS_ERR(handle)) {
547 ret = PTR_ERR(handle);
548 handle = NULL;
549 goto out;
550 }
551
552 /* We cannot recurse into the filesystem as the transaction is already
553 * started */
554 flags |= AOP_FLAG_NOFS;
555
556 page = grab_cache_page_write_begin(mapping, 0, flags);
557 if (!page) {
558 ret = -ENOMEM;
559 goto out;
560 }
561
562 ext4_write_lock_xattr(inode, &no_expand);
563 sem_held = 1;
564 /* If some one has already done this for us, just exit. */
565 if (!ext4_has_inline_data(inode)) {
566 ret = 0;
567 goto out;
568 }
569
570 from = 0;
571 to = ext4_get_inline_size(inode);
572 if (!PageUptodate(page)) {
573 ret = ext4_read_inline_page(inode, page);
574 if (ret < 0)
575 goto out;
576 }
577
578 ret = ext4_destroy_inline_data_nolock(handle, inode);
579 if (ret)
580 goto out;
581
582 if (ext4_should_dioread_nolock(inode)) {
583 ret = __block_write_begin(page, from, to,
584 ext4_get_block_unwritten);
585 } else
586 ret = __block_write_begin(page, from, to, ext4_get_block);
587
588 if (!ret && ext4_should_journal_data(inode)) {
589 ret = ext4_walk_page_buffers(handle, page_buffers(page),
590 from, to, NULL,
591 do_journal_get_write_access);
592 }
593
594 if (ret) {
595 unlock_page(page);
596 put_page(page);
597 page = NULL;
598 ext4_orphan_add(handle, inode);
599 ext4_write_unlock_xattr(inode, &no_expand);
600 sem_held = 0;
601 ext4_journal_stop(handle);
602 handle = NULL;
603 ext4_truncate_failed_write(inode);
604 /*
605 * If truncate failed early the inode might
606 * still be on the orphan list; we need to
607 * make sure the inode is removed from the
608 * orphan list in that case.
609 */
610 if (inode->i_nlink)
611 ext4_orphan_del(NULL, inode);
612 }
613
614 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
615 goto retry;
616
617 if (page)
618 block_commit_write(page, from, to);
619out:
620 if (page) {
621 unlock_page(page);
622 put_page(page);
623 }
624 if (sem_held)
625 ext4_write_unlock_xattr(inode, &no_expand);
626 if (handle)
627 ext4_journal_stop(handle);
628 brelse(iloc.bh);
629 return ret;
630}
631
632/*
633 * Try to write data in the inode.
634 * If the inode has inline data, check whether the new write can be
635 * in the inode also. If not, create the page the handle, move the data
636 * to the page make it update and let the later codes create extent for it.
637 */
638int ext4_try_to_write_inline_data(struct address_space *mapping,
639 struct inode *inode,
640 loff_t pos, unsigned len,
641 unsigned flags,
642 struct page **pagep)
643{
644 int ret;
645 handle_t *handle;
646 struct page *page;
647 struct ext4_iloc iloc;
648
649 if (pos + len > ext4_get_max_inline_size(inode))
650 goto convert;
651
652 ret = ext4_get_inode_loc(inode, &iloc);
653 if (ret)
654 return ret;
655
656 /*
657 * The possible write could happen in the inode,
658 * so try to reserve the space in inode first.
659 */
660 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
661 if (IS_ERR(handle)) {
662 ret = PTR_ERR(handle);
663 handle = NULL;
664 goto out;
665 }
666
667 ret = ext4_prepare_inline_data(handle, inode, pos + len);
668 if (ret && ret != -ENOSPC)
669 goto out;
670
671 /* We don't have space in inline inode, so convert it to extent. */
672 if (ret == -ENOSPC) {
673 ext4_journal_stop(handle);
674 brelse(iloc.bh);
675 goto convert;
676 }
677
678 flags |= AOP_FLAG_NOFS;
679
680 page = grab_cache_page_write_begin(mapping, 0, flags);
681 if (!page) {
682 ret = -ENOMEM;
683 goto out;
684 }
685
686 *pagep = page;
687 down_read(&EXT4_I(inode)->xattr_sem);
688 if (!ext4_has_inline_data(inode)) {
689 ret = 0;
690 unlock_page(page);
691 put_page(page);
692 goto out_up_read;
693 }
694
695 if (!PageUptodate(page)) {
696 ret = ext4_read_inline_page(inode, page);
697 if (ret < 0)
698 goto out_up_read;
699 }
700
701 ret = 1;
702 handle = NULL;
703out_up_read:
704 up_read(&EXT4_I(inode)->xattr_sem);
705out:
706 if (handle)
707 ext4_journal_stop(handle);
708 brelse(iloc.bh);
709 return ret;
710convert:
711 return ext4_convert_inline_data_to_extent(mapping,
712 inode, flags);
713}
714
715int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
716 unsigned copied, struct page *page)
717{
718 int ret, no_expand;
719 void *kaddr;
720 struct ext4_iloc iloc;
721
722 if (unlikely(copied < len)) {
723 if (!PageUptodate(page)) {
724 copied = 0;
725 goto out;
726 }
727 }
728
729 ret = ext4_get_inode_loc(inode, &iloc);
730 if (ret) {
731 ext4_std_error(inode->i_sb, ret);
732 copied = 0;
733 goto out;
734 }
735
736 ext4_write_lock_xattr(inode, &no_expand);
737 BUG_ON(!ext4_has_inline_data(inode));
738
739 kaddr = kmap_atomic(page);
740 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
741 kunmap_atomic(kaddr);
742 SetPageUptodate(page);
743 /* clear page dirty so that writepages wouldn't work for us. */
744 ClearPageDirty(page);
745
746 ext4_write_unlock_xattr(inode, &no_expand);
747 brelse(iloc.bh);
748out:
749 return copied;
750}
751
752struct buffer_head *
753ext4_journalled_write_inline_data(struct inode *inode,
754 unsigned len,
755 struct page *page)
756{
757 int ret, no_expand;
758 void *kaddr;
759 struct ext4_iloc iloc;
760
761 ret = ext4_get_inode_loc(inode, &iloc);
762 if (ret) {
763 ext4_std_error(inode->i_sb, ret);
764 return NULL;
765 }
766
767 ext4_write_lock_xattr(inode, &no_expand);
768 kaddr = kmap_atomic(page);
769 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
770 kunmap_atomic(kaddr);
771 ext4_write_unlock_xattr(inode, &no_expand);
772
773 return iloc.bh;
774}
775
776/*
777 * Try to make the page cache and handle ready for the inline data case.
778 * We can call this function in 2 cases:
779 * 1. The inode is created and the first write exceeds inline size. We can
780 * clear the inode state safely.
781 * 2. The inode has inline data, then we need to read the data, make it
782 * update and dirty so that ext4_da_writepages can handle it. We don't
783 * need to start the journal since the file's metatdata isn't changed now.
784 */
785static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
786 struct inode *inode,
787 unsigned flags,
788 void **fsdata)
789{
790 int ret = 0, inline_size;
791 struct page *page;
792
793 page = grab_cache_page_write_begin(mapping, 0, flags);
794 if (!page)
795 return -ENOMEM;
796
797 down_read(&EXT4_I(inode)->xattr_sem);
798 if (!ext4_has_inline_data(inode)) {
799 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
800 goto out;
801 }
802
803 inline_size = ext4_get_inline_size(inode);
804
805 if (!PageUptodate(page)) {
806 ret = ext4_read_inline_page(inode, page);
807 if (ret < 0)
808 goto out;
809 }
810
811 ret = __block_write_begin(page, 0, inline_size,
812 ext4_da_get_block_prep);
813 if (ret) {
814 up_read(&EXT4_I(inode)->xattr_sem);
815 unlock_page(page);
816 put_page(page);
817 ext4_truncate_failed_write(inode);
818 return ret;
819 }
820
821 SetPageDirty(page);
822 SetPageUptodate(page);
823 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
824 *fsdata = (void *)CONVERT_INLINE_DATA;
825
826out:
827 up_read(&EXT4_I(inode)->xattr_sem);
828 if (page) {
829 unlock_page(page);
830 put_page(page);
831 }
832 return ret;
833}
834
835/*
836 * Prepare the write for the inline data.
837 * If the the data can be written into the inode, we just read
838 * the page and make it uptodate, and start the journal.
839 * Otherwise read the page, makes it dirty so that it can be
840 * handle in writepages(the i_disksize update is left to the
841 * normal ext4_da_write_end).
842 */
843int ext4_da_write_inline_data_begin(struct address_space *mapping,
844 struct inode *inode,
845 loff_t pos, unsigned len,
846 unsigned flags,
847 struct page **pagep,
848 void **fsdata)
849{
850 int ret, inline_size;
851 handle_t *handle;
852 struct page *page;
853 struct ext4_iloc iloc;
854 int retries;
855
856 ret = ext4_get_inode_loc(inode, &iloc);
857 if (ret)
858 return ret;
859
860retry_journal:
861 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
862 if (IS_ERR(handle)) {
863 ret = PTR_ERR(handle);
864 goto out;
865 }
866
867 inline_size = ext4_get_max_inline_size(inode);
868
869 ret = -ENOSPC;
870 if (inline_size >= pos + len) {
871 ret = ext4_prepare_inline_data(handle, inode, pos + len);
872 if (ret && ret != -ENOSPC)
873 goto out_journal;
874 }
875
876 /*
877 * We cannot recurse into the filesystem as the transaction
878 * is already started.
879 */
880 flags |= AOP_FLAG_NOFS;
881
882 if (ret == -ENOSPC) {
883 ret = ext4_da_convert_inline_data_to_extent(mapping,
884 inode,
885 flags,
886 fsdata);
887 ext4_journal_stop(handle);
888 if (ret == -ENOSPC &&
889 ext4_should_retry_alloc(inode->i_sb, &retries))
890 goto retry_journal;
891 goto out;
892 }
893
894
895 page = grab_cache_page_write_begin(mapping, 0, flags);
896 if (!page) {
897 ret = -ENOMEM;
898 goto out_journal;
899 }
900
901 down_read(&EXT4_I(inode)->xattr_sem);
902 if (!ext4_has_inline_data(inode)) {
903 ret = 0;
904 goto out_release_page;
905 }
906
907 if (!PageUptodate(page)) {
908 ret = ext4_read_inline_page(inode, page);
909 if (ret < 0)
910 goto out_release_page;
911 }
912
913 up_read(&EXT4_I(inode)->xattr_sem);
914 *pagep = page;
915 brelse(iloc.bh);
916 return 1;
917out_release_page:
918 up_read(&EXT4_I(inode)->xattr_sem);
919 unlock_page(page);
920 put_page(page);
921out_journal:
922 ext4_journal_stop(handle);
923out:
924 brelse(iloc.bh);
925 return ret;
926}
927
928int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
929 unsigned len, unsigned copied,
930 struct page *page)
931{
932 int i_size_changed = 0;
933 int ret;
934
935 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
936 if (ret < 0) {
937 unlock_page(page);
938 put_page(page);
939 return ret;
940 }
941 copied = ret;
942
943 /*
944 * No need to use i_size_read() here, the i_size
945 * cannot change under us because we hold i_mutex.
946 *
947 * But it's important to update i_size while still holding page lock:
948 * page writeout could otherwise come in and zero beyond i_size.
949 */
950 if (pos+copied > inode->i_size) {
951 i_size_write(inode, pos+copied);
952 i_size_changed = 1;
953 }
954 unlock_page(page);
955 put_page(page);
956
957 /*
958 * Don't mark the inode dirty under page lock. First, it unnecessarily
959 * makes the holding time of page lock longer. Second, it forces lock
960 * ordering of page lock and transaction start for journaling
961 * filesystems.
962 */
963 if (i_size_changed)
964 mark_inode_dirty(inode);
965
966 return copied;
967}
968
969#ifdef INLINE_DIR_DEBUG
970void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
971 void *inline_start, int inline_size)
972{
973 int offset;
974 unsigned short de_len;
975 struct ext4_dir_entry_2 *de = inline_start;
976 void *dlimit = inline_start + inline_size;
977
978 trace_printk("inode %lu\n", dir->i_ino);
979 offset = 0;
980 while ((void *)de < dlimit) {
981 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
982 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
983 offset, de_len, de->name_len, de->name,
984 de->name_len, le32_to_cpu(de->inode));
985 if (ext4_check_dir_entry(dir, NULL, de, bh,
986 inline_start, inline_size, offset))
987 BUG();
988
989 offset += de_len;
990 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
991 }
992}
993#else
994#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
995#endif
996
997/*
998 * Add a new entry into a inline dir.
999 * It will return -ENOSPC if no space is available, and -EIO
1000 * and -EEXIST if directory entry already exists.
1001 */
1002static int ext4_add_dirent_to_inline(handle_t *handle,
1003 struct ext4_filename *fname,
1004 struct inode *dir,
1005 struct inode *inode,
1006 struct ext4_iloc *iloc,
1007 void *inline_start, int inline_size)
1008{
1009 int err;
1010 struct ext4_dir_entry_2 *de;
1011
1012 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1013 inline_size, fname, &de);
1014 if (err)
1015 return err;
1016
1017 BUFFER_TRACE(iloc->bh, "get_write_access");
1018 err = ext4_journal_get_write_access(handle, iloc->bh);
1019 if (err)
1020 return err;
1021 ext4_insert_dentry(inode, de, inline_size, fname);
1022
1023 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1024
1025 /*
1026 * XXX shouldn't update any times until successful
1027 * completion of syscall, but too many callers depend
1028 * on this.
1029 *
1030 * XXX similarly, too many callers depend on
1031 * ext4_new_inode() setting the times, but error
1032 * recovery deletes the inode, so the worst that can
1033 * happen is that the times are slightly out of date
1034 * and/or different from the directory change time.
1035 */
1036 dir->i_mtime = dir->i_ctime = current_time(dir);
1037 ext4_update_dx_flag(dir);
1038 inode_inc_iversion(dir);
1039 return 1;
1040}
1041
1042static void *ext4_get_inline_xattr_pos(struct inode *inode,
1043 struct ext4_iloc *iloc)
1044{
1045 struct ext4_xattr_entry *entry;
1046 struct ext4_xattr_ibody_header *header;
1047
1048 BUG_ON(!EXT4_I(inode)->i_inline_off);
1049
1050 header = IHDR(inode, ext4_raw_inode(iloc));
1051 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1052 EXT4_I(inode)->i_inline_off);
1053
1054 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1055}
1056
1057/* Set the final de to cover the whole block. */
1058static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1059{
1060 struct ext4_dir_entry_2 *de, *prev_de;
1061 void *limit;
1062 int de_len;
1063
1064 de = (struct ext4_dir_entry_2 *)de_buf;
1065 if (old_size) {
1066 limit = de_buf + old_size;
1067 do {
1068 prev_de = de;
1069 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1070 de_buf += de_len;
1071 de = (struct ext4_dir_entry_2 *)de_buf;
1072 } while (de_buf < limit);
1073
1074 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1075 old_size, new_size);
1076 } else {
1077 /* this is just created, so create an empty entry. */
1078 de->inode = 0;
1079 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1080 }
1081}
1082
1083static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1084 struct ext4_iloc *iloc)
1085{
1086 int ret;
1087 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1088 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1089
1090 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1091 return -ENOSPC;
1092
1093 ret = ext4_update_inline_data(handle, dir,
1094 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1095 if (ret)
1096 return ret;
1097
1098 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1099 EXT4_I(dir)->i_inline_size -
1100 EXT4_MIN_INLINE_DATA_SIZE);
1101 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1102 return 0;
1103}
1104
1105static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1106 struct ext4_iloc *iloc,
1107 void *buf, int inline_size)
1108{
1109 ext4_create_inline_data(handle, inode, inline_size);
1110 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1111 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1112}
1113
1114static int ext4_finish_convert_inline_dir(handle_t *handle,
1115 struct inode *inode,
1116 struct buffer_head *dir_block,
1117 void *buf,
1118 int inline_size)
1119{
1120 int err, csum_size = 0, header_size = 0;
1121 struct ext4_dir_entry_2 *de;
1122 struct ext4_dir_entry_tail *t;
1123 void *target = dir_block->b_data;
1124
1125 /*
1126 * First create "." and ".." and then copy the dir information
1127 * back to the block.
1128 */
1129 de = (struct ext4_dir_entry_2 *)target;
1130 de = ext4_init_dot_dotdot(inode, de,
1131 inode->i_sb->s_blocksize, csum_size,
1132 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1133 header_size = (void *)de - target;
1134
1135 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1136 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1137
1138 if (ext4_has_metadata_csum(inode->i_sb))
1139 csum_size = sizeof(struct ext4_dir_entry_tail);
1140
1141 inode->i_size = inode->i_sb->s_blocksize;
1142 i_size_write(inode, inode->i_sb->s_blocksize);
1143 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1144 ext4_update_final_de(dir_block->b_data,
1145 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1146 inode->i_sb->s_blocksize - csum_size);
1147
1148 if (csum_size) {
1149 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1150 inode->i_sb->s_blocksize);
1151 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1152 }
1153 set_buffer_uptodate(dir_block);
1154 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1155 if (err)
1156 return err;
1157 set_buffer_verified(dir_block);
1158 return ext4_mark_inode_dirty(handle, inode);
1159}
1160
1161static int ext4_convert_inline_data_nolock(handle_t *handle,
1162 struct inode *inode,
1163 struct ext4_iloc *iloc)
1164{
1165 int error;
1166 void *buf = NULL;
1167 struct buffer_head *data_bh = NULL;
1168 struct ext4_map_blocks map;
1169 int inline_size;
1170
1171 inline_size = ext4_get_inline_size(inode);
1172 buf = kmalloc(inline_size, GFP_NOFS);
1173 if (!buf) {
1174 error = -ENOMEM;
1175 goto out;
1176 }
1177
1178 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1179 if (error < 0)
1180 goto out;
1181
1182 /*
1183 * Make sure the inline directory entries pass checks before we try to
1184 * convert them, so that we avoid touching stuff that needs fsck.
1185 */
1186 if (S_ISDIR(inode->i_mode)) {
1187 error = ext4_check_all_de(inode, iloc->bh,
1188 buf + EXT4_INLINE_DOTDOT_SIZE,
1189 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1190 if (error)
1191 goto out;
1192 }
1193
1194 error = ext4_destroy_inline_data_nolock(handle, inode);
1195 if (error)
1196 goto out;
1197
1198 map.m_lblk = 0;
1199 map.m_len = 1;
1200 map.m_flags = 0;
1201 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1202 if (error < 0)
1203 goto out_restore;
1204 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1205 error = -EIO;
1206 goto out_restore;
1207 }
1208
1209 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1210 if (!data_bh) {
1211 error = -ENOMEM;
1212 goto out_restore;
1213 }
1214
1215 lock_buffer(data_bh);
1216 error = ext4_journal_get_create_access(handle, data_bh);
1217 if (error) {
1218 unlock_buffer(data_bh);
1219 error = -EIO;
1220 goto out_restore;
1221 }
1222 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1223
1224 if (!S_ISDIR(inode->i_mode)) {
1225 memcpy(data_bh->b_data, buf, inline_size);
1226 set_buffer_uptodate(data_bh);
1227 error = ext4_handle_dirty_metadata(handle,
1228 inode, data_bh);
1229 } else {
1230 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1231 buf, inline_size);
1232 }
1233
1234 unlock_buffer(data_bh);
1235out_restore:
1236 if (error)
1237 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1238
1239out:
1240 brelse(data_bh);
1241 kfree(buf);
1242 return error;
1243}
1244
1245/*
1246 * Try to add the new entry to the inline data.
1247 * If succeeds, return 0. If not, extended the inline dir and copied data to
1248 * the new created block.
1249 */
1250int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1251 struct inode *dir, struct inode *inode)
1252{
1253 int ret, inline_size, no_expand;
1254 void *inline_start;
1255 struct ext4_iloc iloc;
1256
1257 ret = ext4_get_inode_loc(dir, &iloc);
1258 if (ret)
1259 return ret;
1260
1261 ext4_write_lock_xattr(dir, &no_expand);
1262 if (!ext4_has_inline_data(dir))
1263 goto out;
1264
1265 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1266 EXT4_INLINE_DOTDOT_SIZE;
1267 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1268
1269 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1270 inline_start, inline_size);
1271 if (ret != -ENOSPC)
1272 goto out;
1273
1274 /* check whether it can be inserted to inline xattr space. */
1275 inline_size = EXT4_I(dir)->i_inline_size -
1276 EXT4_MIN_INLINE_DATA_SIZE;
1277 if (!inline_size) {
1278 /* Try to use the xattr space.*/
1279 ret = ext4_update_inline_dir(handle, dir, &iloc);
1280 if (ret && ret != -ENOSPC)
1281 goto out;
1282
1283 inline_size = EXT4_I(dir)->i_inline_size -
1284 EXT4_MIN_INLINE_DATA_SIZE;
1285 }
1286
1287 if (inline_size) {
1288 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1289
1290 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1291 inode, &iloc, inline_start,
1292 inline_size);
1293
1294 if (ret != -ENOSPC)
1295 goto out;
1296 }
1297
1298 /*
1299 * The inline space is filled up, so create a new block for it.
1300 * As the extent tree will be created, we have to save the inline
1301 * dir first.
1302 */
1303 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1304
1305out:
1306 ext4_write_unlock_xattr(dir, &no_expand);
1307 ext4_mark_inode_dirty(handle, dir);
1308 brelse(iloc.bh);
1309 return ret;
1310}
1311
1312/*
1313 * This function fills a red-black tree with information from an
1314 * inlined dir. It returns the number directory entries loaded
1315 * into the tree. If there is an error it is returned in err.
1316 */
1317int htree_inlinedir_to_tree(struct file *dir_file,
1318 struct inode *dir, ext4_lblk_t block,
1319 struct dx_hash_info *hinfo,
1320 __u32 start_hash, __u32 start_minor_hash,
1321 int *has_inline_data)
1322{
1323 int err = 0, count = 0;
1324 unsigned int parent_ino;
1325 int pos;
1326 struct ext4_dir_entry_2 *de;
1327 struct inode *inode = file_inode(dir_file);
1328 int ret, inline_size = 0;
1329 struct ext4_iloc iloc;
1330 void *dir_buf = NULL;
1331 struct ext4_dir_entry_2 fake;
1332 struct fscrypt_str tmp_str;
1333
1334 ret = ext4_get_inode_loc(inode, &iloc);
1335 if (ret)
1336 return ret;
1337
1338 down_read(&EXT4_I(inode)->xattr_sem);
1339 if (!ext4_has_inline_data(inode)) {
1340 up_read(&EXT4_I(inode)->xattr_sem);
1341 *has_inline_data = 0;
1342 goto out;
1343 }
1344
1345 inline_size = ext4_get_inline_size(inode);
1346 dir_buf = kmalloc(inline_size, GFP_NOFS);
1347 if (!dir_buf) {
1348 ret = -ENOMEM;
1349 up_read(&EXT4_I(inode)->xattr_sem);
1350 goto out;
1351 }
1352
1353 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1354 up_read(&EXT4_I(inode)->xattr_sem);
1355 if (ret < 0)
1356 goto out;
1357
1358 pos = 0;
1359 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1360 while (pos < inline_size) {
1361 /*
1362 * As inlined dir doesn't store any information about '.' and
1363 * only the inode number of '..' is stored, we have to handle
1364 * them differently.
1365 */
1366 if (pos == 0) {
1367 fake.inode = cpu_to_le32(inode->i_ino);
1368 fake.name_len = 1;
1369 strcpy(fake.name, ".");
1370 fake.rec_len = ext4_rec_len_to_disk(
1371 EXT4_DIR_REC_LEN(fake.name_len),
1372 inline_size);
1373 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1374 de = &fake;
1375 pos = EXT4_INLINE_DOTDOT_OFFSET;
1376 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1377 fake.inode = cpu_to_le32(parent_ino);
1378 fake.name_len = 2;
1379 strcpy(fake.name, "..");
1380 fake.rec_len = ext4_rec_len_to_disk(
1381 EXT4_DIR_REC_LEN(fake.name_len),
1382 inline_size);
1383 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1384 de = &fake;
1385 pos = EXT4_INLINE_DOTDOT_SIZE;
1386 } else {
1387 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1388 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1389 if (ext4_check_dir_entry(inode, dir_file, de,
1390 iloc.bh, dir_buf,
1391 inline_size, pos)) {
1392 ret = count;
1393 goto out;
1394 }
1395 }
1396
1397 ext4fs_dirhash(de->name, de->name_len, hinfo);
1398 if ((hinfo->hash < start_hash) ||
1399 ((hinfo->hash == start_hash) &&
1400 (hinfo->minor_hash < start_minor_hash)))
1401 continue;
1402 if (de->inode == 0)
1403 continue;
1404 tmp_str.name = de->name;
1405 tmp_str.len = de->name_len;
1406 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1407 hinfo->minor_hash, de, &tmp_str);
1408 if (err) {
1409 count = err;
1410 goto out;
1411 }
1412 count++;
1413 }
1414 ret = count;
1415out:
1416 kfree(dir_buf);
1417 brelse(iloc.bh);
1418 return ret;
1419}
1420
1421/*
1422 * So this function is called when the volume is mkfsed with
1423 * dir_index disabled. In order to keep f_pos persistent
1424 * after we convert from an inlined dir to a blocked based,
1425 * we just pretend that we are a normal dir and return the
1426 * offset as if '.' and '..' really take place.
1427 *
1428 */
1429int ext4_read_inline_dir(struct file *file,
1430 struct dir_context *ctx,
1431 int *has_inline_data)
1432{
1433 unsigned int offset, parent_ino;
1434 int i;
1435 struct ext4_dir_entry_2 *de;
1436 struct super_block *sb;
1437 struct inode *inode = file_inode(file);
1438 int ret, inline_size = 0;
1439 struct ext4_iloc iloc;
1440 void *dir_buf = NULL;
1441 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1442
1443 ret = ext4_get_inode_loc(inode, &iloc);
1444 if (ret)
1445 return ret;
1446
1447 down_read(&EXT4_I(inode)->xattr_sem);
1448 if (!ext4_has_inline_data(inode)) {
1449 up_read(&EXT4_I(inode)->xattr_sem);
1450 *has_inline_data = 0;
1451 goto out;
1452 }
1453
1454 inline_size = ext4_get_inline_size(inode);
1455 dir_buf = kmalloc(inline_size, GFP_NOFS);
1456 if (!dir_buf) {
1457 ret = -ENOMEM;
1458 up_read(&EXT4_I(inode)->xattr_sem);
1459 goto out;
1460 }
1461
1462 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1463 up_read(&EXT4_I(inode)->xattr_sem);
1464 if (ret < 0)
1465 goto out;
1466
1467 ret = 0;
1468 sb = inode->i_sb;
1469 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1470 offset = ctx->pos;
1471
1472 /*
1473 * dotdot_offset and dotdot_size is the real offset and
1474 * size for ".." and "." if the dir is block based while
1475 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1476 * So we will use extra_offset and extra_size to indicate them
1477 * during the inline dir iteration.
1478 */
1479 dotdot_offset = EXT4_DIR_REC_LEN(1);
1480 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1481 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1482 extra_size = extra_offset + inline_size;
1483
1484 /*
1485 * If the version has changed since the last call to
1486 * readdir(2), then we might be pointing to an invalid
1487 * dirent right now. Scan from the start of the inline
1488 * dir to make sure.
1489 */
1490 if (!inode_eq_iversion(inode, file->f_version)) {
1491 for (i = 0; i < extra_size && i < offset;) {
1492 /*
1493 * "." is with offset 0 and
1494 * ".." is dotdot_offset.
1495 */
1496 if (!i) {
1497 i = dotdot_offset;
1498 continue;
1499 } else if (i == dotdot_offset) {
1500 i = dotdot_size;
1501 continue;
1502 }
1503 /* for other entry, the real offset in
1504 * the buf has to be tuned accordingly.
1505 */
1506 de = (struct ext4_dir_entry_2 *)
1507 (dir_buf + i - extra_offset);
1508 /* It's too expensive to do a full
1509 * dirent test each time round this
1510 * loop, but we do have to test at
1511 * least that it is non-zero. A
1512 * failure will be detected in the
1513 * dirent test below. */
1514 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1515 < EXT4_DIR_REC_LEN(1))
1516 break;
1517 i += ext4_rec_len_from_disk(de->rec_len,
1518 extra_size);
1519 }
1520 offset = i;
1521 ctx->pos = offset;
1522 file->f_version = inode_query_iversion(inode);
1523 }
1524
1525 while (ctx->pos < extra_size) {
1526 if (ctx->pos == 0) {
1527 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1528 goto out;
1529 ctx->pos = dotdot_offset;
1530 continue;
1531 }
1532
1533 if (ctx->pos == dotdot_offset) {
1534 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1535 goto out;
1536 ctx->pos = dotdot_size;
1537 continue;
1538 }
1539
1540 de = (struct ext4_dir_entry_2 *)
1541 (dir_buf + ctx->pos - extra_offset);
1542 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1543 extra_size, ctx->pos))
1544 goto out;
1545 if (le32_to_cpu(de->inode)) {
1546 if (!dir_emit(ctx, de->name, de->name_len,
1547 le32_to_cpu(de->inode),
1548 get_dtype(sb, de->file_type)))
1549 goto out;
1550 }
1551 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1552 }
1553out:
1554 kfree(dir_buf);
1555 brelse(iloc.bh);
1556 return ret;
1557}
1558
1559struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1560 struct ext4_dir_entry_2 **parent_de,
1561 int *retval)
1562{
1563 struct ext4_iloc iloc;
1564
1565 *retval = ext4_get_inode_loc(inode, &iloc);
1566 if (*retval)
1567 return NULL;
1568
1569 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1570
1571 return iloc.bh;
1572}
1573
1574/*
1575 * Try to create the inline data for the new dir.
1576 * If it succeeds, return 0, otherwise return the error.
1577 * In case of ENOSPC, the caller should create the normal disk layout dir.
1578 */
1579int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1580 struct inode *inode)
1581{
1582 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1583 struct ext4_iloc iloc;
1584 struct ext4_dir_entry_2 *de;
1585
1586 ret = ext4_get_inode_loc(inode, &iloc);
1587 if (ret)
1588 return ret;
1589
1590 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1591 if (ret)
1592 goto out;
1593
1594 /*
1595 * For inline dir, we only save the inode information for the ".."
1596 * and create a fake dentry to cover the left space.
1597 */
1598 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1599 de->inode = cpu_to_le32(parent->i_ino);
1600 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1601 de->inode = 0;
1602 de->rec_len = ext4_rec_len_to_disk(
1603 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1604 inline_size);
1605 set_nlink(inode, 2);
1606 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1607out:
1608 brelse(iloc.bh);
1609 return ret;
1610}
1611
1612struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1613 struct ext4_filename *fname,
1614 struct ext4_dir_entry_2 **res_dir,
1615 int *has_inline_data)
1616{
1617 int ret;
1618 struct ext4_iloc iloc;
1619 void *inline_start;
1620 int inline_size;
1621
1622 if (ext4_get_inode_loc(dir, &iloc))
1623 return NULL;
1624
1625 down_read(&EXT4_I(dir)->xattr_sem);
1626 if (!ext4_has_inline_data(dir)) {
1627 *has_inline_data = 0;
1628 goto out;
1629 }
1630
1631 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1632 EXT4_INLINE_DOTDOT_SIZE;
1633 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1634 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1635 dir, fname, 0, res_dir);
1636 if (ret == 1)
1637 goto out_find;
1638 if (ret < 0)
1639 goto out;
1640
1641 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1642 goto out;
1643
1644 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1645 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1646
1647 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1648 dir, fname, 0, res_dir);
1649 if (ret == 1)
1650 goto out_find;
1651
1652out:
1653 brelse(iloc.bh);
1654 iloc.bh = NULL;
1655out_find:
1656 up_read(&EXT4_I(dir)->xattr_sem);
1657 return iloc.bh;
1658}
1659
1660int ext4_delete_inline_entry(handle_t *handle,
1661 struct inode *dir,
1662 struct ext4_dir_entry_2 *de_del,
1663 struct buffer_head *bh,
1664 int *has_inline_data)
1665{
1666 int err, inline_size, no_expand;
1667 struct ext4_iloc iloc;
1668 void *inline_start;
1669
1670 err = ext4_get_inode_loc(dir, &iloc);
1671 if (err)
1672 return err;
1673
1674 ext4_write_lock_xattr(dir, &no_expand);
1675 if (!ext4_has_inline_data(dir)) {
1676 *has_inline_data = 0;
1677 goto out;
1678 }
1679
1680 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1681 EXT4_MIN_INLINE_DATA_SIZE) {
1682 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1683 EXT4_INLINE_DOTDOT_SIZE;
1684 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1685 EXT4_INLINE_DOTDOT_SIZE;
1686 } else {
1687 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1688 inline_size = ext4_get_inline_size(dir) -
1689 EXT4_MIN_INLINE_DATA_SIZE;
1690 }
1691
1692 BUFFER_TRACE(bh, "get_write_access");
1693 err = ext4_journal_get_write_access(handle, bh);
1694 if (err)
1695 goto out;
1696
1697 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1698 inline_start, inline_size, 0);
1699 if (err)
1700 goto out;
1701
1702 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1703out:
1704 ext4_write_unlock_xattr(dir, &no_expand);
1705 if (likely(err == 0))
1706 err = ext4_mark_inode_dirty(handle, dir);
1707 brelse(iloc.bh);
1708 if (err != -ENOENT)
1709 ext4_std_error(dir->i_sb, err);
1710 return err;
1711}
1712
1713/*
1714 * Get the inline dentry at offset.
1715 */
1716static inline struct ext4_dir_entry_2 *
1717ext4_get_inline_entry(struct inode *inode,
1718 struct ext4_iloc *iloc,
1719 unsigned int offset,
1720 void **inline_start,
1721 int *inline_size)
1722{
1723 void *inline_pos;
1724
1725 BUG_ON(offset > ext4_get_inline_size(inode));
1726
1727 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1728 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1729 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1730 } else {
1731 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1732 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1733 *inline_size = ext4_get_inline_size(inode) -
1734 EXT4_MIN_INLINE_DATA_SIZE;
1735 }
1736
1737 if (inline_start)
1738 *inline_start = inline_pos;
1739 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1740}
1741
1742bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1743{
1744 int err, inline_size;
1745 struct ext4_iloc iloc;
1746 void *inline_pos;
1747 unsigned int offset;
1748 struct ext4_dir_entry_2 *de;
1749 bool ret = true;
1750
1751 err = ext4_get_inode_loc(dir, &iloc);
1752 if (err) {
1753 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1754 err, dir->i_ino);
1755 return true;
1756 }
1757
1758 down_read(&EXT4_I(dir)->xattr_sem);
1759 if (!ext4_has_inline_data(dir)) {
1760 *has_inline_data = 0;
1761 goto out;
1762 }
1763
1764 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1765 if (!le32_to_cpu(de->inode)) {
1766 ext4_warning(dir->i_sb,
1767 "bad inline directory (dir #%lu) - no `..'",
1768 dir->i_ino);
1769 ret = true;
1770 goto out;
1771 }
1772
1773 offset = EXT4_INLINE_DOTDOT_SIZE;
1774 while (offset < dir->i_size) {
1775 de = ext4_get_inline_entry(dir, &iloc, offset,
1776 &inline_pos, &inline_size);
1777 if (ext4_check_dir_entry(dir, NULL, de,
1778 iloc.bh, inline_pos,
1779 inline_size, offset)) {
1780 ext4_warning(dir->i_sb,
1781 "bad inline directory (dir #%lu) - "
1782 "inode %u, rec_len %u, name_len %d"
1783 "inline size %d",
1784 dir->i_ino, le32_to_cpu(de->inode),
1785 le16_to_cpu(de->rec_len), de->name_len,
1786 inline_size);
1787 ret = true;
1788 goto out;
1789 }
1790 if (le32_to_cpu(de->inode)) {
1791 ret = false;
1792 goto out;
1793 }
1794 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1795 }
1796
1797out:
1798 up_read(&EXT4_I(dir)->xattr_sem);
1799 brelse(iloc.bh);
1800 return ret;
1801}
1802
1803int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1804{
1805 int ret, no_expand;
1806
1807 ext4_write_lock_xattr(inode, &no_expand);
1808 ret = ext4_destroy_inline_data_nolock(handle, inode);
1809 ext4_write_unlock_xattr(inode, &no_expand);
1810
1811 return ret;
1812}
1813
1814int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1815{
1816 __u64 addr;
1817 int error = -EAGAIN;
1818 struct ext4_iloc iloc;
1819
1820 down_read(&EXT4_I(inode)->xattr_sem);
1821 if (!ext4_has_inline_data(inode))
1822 goto out;
1823
1824 error = ext4_get_inode_loc(inode, &iloc);
1825 if (error)
1826 goto out;
1827
1828 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1829 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1830 addr += offsetof(struct ext4_inode, i_block);
1831
1832 brelse(iloc.bh);
1833
1834 iomap->addr = addr;
1835 iomap->offset = 0;
1836 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1837 i_size_read(inode));
1838 iomap->type = 0;
1839 iomap->flags = IOMAP_F_DATA_INLINE;
1840
1841out:
1842 up_read(&EXT4_I(inode)->xattr_sem);
1843 return error;
1844}
1845
1846int ext4_inline_data_fiemap(struct inode *inode,
1847 struct fiemap_extent_info *fieinfo,
1848 int *has_inline, __u64 start, __u64 len)
1849{
1850 __u64 physical = 0;
1851 __u64 inline_len;
1852 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1853 FIEMAP_EXTENT_LAST;
1854 int error = 0;
1855 struct ext4_iloc iloc;
1856
1857 down_read(&EXT4_I(inode)->xattr_sem);
1858 if (!ext4_has_inline_data(inode)) {
1859 *has_inline = 0;
1860 goto out;
1861 }
1862 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1863 i_size_read(inode));
1864 if (start >= inline_len)
1865 goto out;
1866 if (start + len < inline_len)
1867 inline_len = start + len;
1868 inline_len -= start;
1869
1870 error = ext4_get_inode_loc(inode, &iloc);
1871 if (error)
1872 goto out;
1873
1874 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1875 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1876 physical += offsetof(struct ext4_inode, i_block);
1877
1878 if (physical)
1879 error = fiemap_fill_next_extent(fieinfo, start, physical,
1880 inline_len, flags);
1881 brelse(iloc.bh);
1882out:
1883 up_read(&EXT4_I(inode)->xattr_sem);
1884 return (error < 0 ? error : 0);
1885}
1886
1887/*
1888 * Called during xattr set, and if we can sparse space 'needed',
1889 * just create the extent tree evict the data to the outer block.
1890 *
1891 * We use jbd2 instead of page cache to move data to the 1st block
1892 * so that the whole transaction can be committed as a whole and
1893 * the data isn't lost because of the delayed page cache write.
1894 */
1895int ext4_try_to_evict_inline_data(handle_t *handle,
1896 struct inode *inode,
1897 int needed)
1898{
1899 int error;
1900 struct ext4_xattr_entry *entry;
1901 struct ext4_inode *raw_inode;
1902 struct ext4_iloc iloc;
1903
1904 error = ext4_get_inode_loc(inode, &iloc);
1905 if (error)
1906 return error;
1907
1908 raw_inode = ext4_raw_inode(&iloc);
1909 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1910 EXT4_I(inode)->i_inline_off);
1911 if (EXT4_XATTR_LEN(entry->e_name_len) +
1912 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1913 error = -ENOSPC;
1914 goto out;
1915 }
1916
1917 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1918out:
1919 brelse(iloc.bh);
1920 return error;
1921}
1922
1923int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1924{
1925 handle_t *handle;
1926 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1927 size_t i_size;
1928 void *value = NULL;
1929 struct ext4_xattr_ibody_find is = {
1930 .s = { .not_found = -ENODATA, },
1931 };
1932 struct ext4_xattr_info i = {
1933 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1934 .name = EXT4_XATTR_SYSTEM_DATA,
1935 };
1936
1937
1938 needed_blocks = ext4_writepage_trans_blocks(inode);
1939 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1940 if (IS_ERR(handle))
1941 return PTR_ERR(handle);
1942
1943 ext4_write_lock_xattr(inode, &no_expand);
1944 if (!ext4_has_inline_data(inode)) {
1945 *has_inline = 0;
1946 ext4_journal_stop(handle);
1947 return 0;
1948 }
1949
1950 if ((err = ext4_orphan_add(handle, inode)) != 0)
1951 goto out;
1952
1953 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1954 goto out;
1955
1956 down_write(&EXT4_I(inode)->i_data_sem);
1957 i_size = inode->i_size;
1958 inline_size = ext4_get_inline_size(inode);
1959 EXT4_I(inode)->i_disksize = i_size;
1960
1961 if (i_size < inline_size) {
1962 /* Clear the content in the xattr space. */
1963 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1964 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1965 goto out_error;
1966
1967 BUG_ON(is.s.not_found);
1968
1969 value_len = le32_to_cpu(is.s.here->e_value_size);
1970 value = kmalloc(value_len, GFP_NOFS);
1971 if (!value) {
1972 err = -ENOMEM;
1973 goto out_error;
1974 }
1975
1976 err = ext4_xattr_ibody_get(inode, i.name_index,
1977 i.name, value, value_len);
1978 if (err <= 0)
1979 goto out_error;
1980
1981 i.value = value;
1982 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1983 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1984 err = ext4_xattr_ibody_inline_set(handle, inode,
1985 &i, &is);
1986 if (err)
1987 goto out_error;
1988 }
1989
1990 /* Clear the content within i_blocks. */
1991 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1992 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1993 memset(p + i_size, 0,
1994 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1995 }
1996
1997 EXT4_I(inode)->i_inline_size = i_size <
1998 EXT4_MIN_INLINE_DATA_SIZE ?
1999 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2000 }
2001
2002out_error:
2003 up_write(&EXT4_I(inode)->i_data_sem);
2004out:
2005 brelse(is.iloc.bh);
2006 ext4_write_unlock_xattr(inode, &no_expand);
2007 kfree(value);
2008 if (inode->i_nlink)
2009 ext4_orphan_del(handle, inode);
2010
2011 if (err == 0) {
2012 inode->i_mtime = inode->i_ctime = current_time(inode);
2013 err = ext4_mark_inode_dirty(handle, inode);
2014 if (IS_SYNC(inode))
2015 ext4_handle_sync(handle);
2016 }
2017 ext4_journal_stop(handle);
2018 return err;
2019}
2020
2021int ext4_convert_inline_data(struct inode *inode)
2022{
2023 int error, needed_blocks, no_expand;
2024 handle_t *handle;
2025 struct ext4_iloc iloc;
2026
2027 if (!ext4_has_inline_data(inode)) {
2028 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2029 return 0;
2030 }
2031
2032 needed_blocks = ext4_writepage_trans_blocks(inode);
2033
2034 iloc.bh = NULL;
2035 error = ext4_get_inode_loc(inode, &iloc);
2036 if (error)
2037 return error;
2038
2039 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2040 if (IS_ERR(handle)) {
2041 error = PTR_ERR(handle);
2042 goto out_free;
2043 }
2044
2045 ext4_write_lock_xattr(inode, &no_expand);
2046 if (ext4_has_inline_data(inode))
2047 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2048 ext4_write_unlock_xattr(inode, &no_expand);
2049 ext4_journal_stop(handle);
2050out_free:
2051 brelse(iloc.bh);
2052 return error;
2053}
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#include "ext4_jbd2.h"
15#include "ext4.h"
16#include "xattr.h"
17#include "truncate.h"
18#include <linux/fiemap.h>
19
20#define EXT4_XATTR_SYSTEM_DATA "data"
21#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
22#define EXT4_INLINE_DOTDOT_OFFSET 2
23#define EXT4_INLINE_DOTDOT_SIZE 4
24
25int ext4_get_inline_size(struct inode *inode)
26{
27 if (EXT4_I(inode)->i_inline_off)
28 return EXT4_I(inode)->i_inline_size;
29
30 return 0;
31}
32
33static int get_max_inline_xattr_value_size(struct inode *inode,
34 struct ext4_iloc *iloc)
35{
36 struct ext4_xattr_ibody_header *header;
37 struct ext4_xattr_entry *entry;
38 struct ext4_inode *raw_inode;
39 int free, min_offs;
40
41 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42 EXT4_GOOD_OLD_INODE_SIZE -
43 EXT4_I(inode)->i_extra_isize -
44 sizeof(struct ext4_xattr_ibody_header);
45
46 /*
47 * We need to subtract another sizeof(__u32) since an in-inode xattr
48 * needs an empty 4 bytes to indicate the gap between the xattr entry
49 * and the name/value pair.
50 */
51 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52 return EXT4_XATTR_SIZE(min_offs -
53 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54 EXT4_XATTR_ROUND - sizeof(__u32));
55
56 raw_inode = ext4_raw_inode(iloc);
57 header = IHDR(inode, raw_inode);
58 entry = IFIRST(header);
59
60 /* Compute min_offs. */
61 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62 if (!entry->e_value_block && entry->e_value_size) {
63 size_t offs = le16_to_cpu(entry->e_value_offs);
64 if (offs < min_offs)
65 min_offs = offs;
66 }
67 }
68 free = min_offs -
69 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71 if (EXT4_I(inode)->i_inline_off) {
72 entry = (struct ext4_xattr_entry *)
73 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76 goto out;
77 }
78
79 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81 if (free > EXT4_XATTR_ROUND)
82 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83 else
84 free = 0;
85
86out:
87 return free;
88}
89
90/*
91 * Get the maximum size we now can store in an inode.
92 * If we can't find the space for a xattr entry, don't use the space
93 * of the extents since we have no space to indicate the inline data.
94 */
95int ext4_get_max_inline_size(struct inode *inode)
96{
97 int error, max_inline_size;
98 struct ext4_iloc iloc;
99
100 if (EXT4_I(inode)->i_extra_isize == 0)
101 return 0;
102
103 error = ext4_get_inode_loc(inode, &iloc);
104 if (error) {
105 ext4_error_inode(inode, __func__, __LINE__, 0,
106 "can't get inode location %lu",
107 inode->i_ino);
108 return 0;
109 }
110
111 down_read(&EXT4_I(inode)->xattr_sem);
112 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113 up_read(&EXT4_I(inode)->xattr_sem);
114
115 brelse(iloc.bh);
116
117 if (!max_inline_size)
118 return 0;
119
120 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121}
122
123int ext4_has_inline_data(struct inode *inode)
124{
125 return ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) &&
126 EXT4_I(inode)->i_inline_off;
127}
128
129/*
130 * this function does not take xattr_sem, which is OK because it is
131 * currently only used in a code path coming form ext4_iget, before
132 * the new inode has been unlocked
133 */
134int ext4_find_inline_data_nolock(struct inode *inode)
135{
136 struct ext4_xattr_ibody_find is = {
137 .s = { .not_found = -ENODATA, },
138 };
139 struct ext4_xattr_info i = {
140 .name_index = EXT4_XATTR_INDEX_SYSTEM,
141 .name = EXT4_XATTR_SYSTEM_DATA,
142 };
143 int error;
144
145 if (EXT4_I(inode)->i_extra_isize == 0)
146 return 0;
147
148 error = ext4_get_inode_loc(inode, &is.iloc);
149 if (error)
150 return error;
151
152 error = ext4_xattr_ibody_find(inode, &i, &is);
153 if (error)
154 goto out;
155
156 if (!is.s.not_found) {
157 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
158 (void *)ext4_raw_inode(&is.iloc));
159 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
160 le32_to_cpu(is.s.here->e_value_size);
161 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
162 }
163out:
164 brelse(is.iloc.bh);
165 return error;
166}
167
168static int ext4_read_inline_data(struct inode *inode, void *buffer,
169 unsigned int len,
170 struct ext4_iloc *iloc)
171{
172 struct ext4_xattr_entry *entry;
173 struct ext4_xattr_ibody_header *header;
174 int cp_len = 0;
175 struct ext4_inode *raw_inode;
176
177 if (!len)
178 return 0;
179
180 BUG_ON(len > EXT4_I(inode)->i_inline_size);
181
182 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
183 len : EXT4_MIN_INLINE_DATA_SIZE;
184
185 raw_inode = ext4_raw_inode(iloc);
186 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
187
188 len -= cp_len;
189 buffer += cp_len;
190
191 if (!len)
192 goto out;
193
194 header = IHDR(inode, raw_inode);
195 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
196 EXT4_I(inode)->i_inline_off);
197 len = min_t(unsigned int, len,
198 (unsigned int)le32_to_cpu(entry->e_value_size));
199
200 memcpy(buffer,
201 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
202 cp_len += len;
203
204out:
205 return cp_len;
206}
207
208/*
209 * write the buffer to the inline inode.
210 * If 'create' is set, we don't need to do the extra copy in the xattr
211 * value since it is already handled by ext4_xattr_ibody_inline_set.
212 * That saves us one memcpy.
213 */
214void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
215 void *buffer, loff_t pos, unsigned int len)
216{
217 struct ext4_xattr_entry *entry;
218 struct ext4_xattr_ibody_header *header;
219 struct ext4_inode *raw_inode;
220 int cp_len = 0;
221
222 BUG_ON(!EXT4_I(inode)->i_inline_off);
223 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
224
225 raw_inode = ext4_raw_inode(iloc);
226 buffer += pos;
227
228 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
229 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
230 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
231 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
232
233 len -= cp_len;
234 buffer += cp_len;
235 pos += cp_len;
236 }
237
238 if (!len)
239 return;
240
241 pos -= EXT4_MIN_INLINE_DATA_SIZE;
242 header = IHDR(inode, raw_inode);
243 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
244 EXT4_I(inode)->i_inline_off);
245
246 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
247 buffer, len);
248}
249
250static int ext4_create_inline_data(handle_t *handle,
251 struct inode *inode, unsigned len)
252{
253 int error;
254 void *value = NULL;
255 struct ext4_xattr_ibody_find is = {
256 .s = { .not_found = -ENODATA, },
257 };
258 struct ext4_xattr_info i = {
259 .name_index = EXT4_XATTR_INDEX_SYSTEM,
260 .name = EXT4_XATTR_SYSTEM_DATA,
261 };
262
263 error = ext4_get_inode_loc(inode, &is.iloc);
264 if (error)
265 return error;
266
267 error = ext4_journal_get_write_access(handle, is.iloc.bh);
268 if (error)
269 goto out;
270
271 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272 value = EXT4_ZERO_XATTR_VALUE;
273 len -= EXT4_MIN_INLINE_DATA_SIZE;
274 } else {
275 value = "";
276 len = 0;
277 }
278
279 /* Insert the the xttr entry. */
280 i.value = value;
281 i.value_len = len;
282
283 error = ext4_xattr_ibody_find(inode, &i, &is);
284 if (error)
285 goto out;
286
287 BUG_ON(!is.s.not_found);
288
289 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290 if (error) {
291 if (error == -ENOSPC)
292 ext4_clear_inode_state(inode,
293 EXT4_STATE_MAY_INLINE_DATA);
294 goto out;
295 }
296
297 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301 (void *)ext4_raw_inode(&is.iloc));
302 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305 get_bh(is.iloc.bh);
306 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
307
308out:
309 brelse(is.iloc.bh);
310 return error;
311}
312
313static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
314 unsigned int len)
315{
316 int error;
317 void *value = NULL;
318 struct ext4_xattr_ibody_find is = {
319 .s = { .not_found = -ENODATA, },
320 };
321 struct ext4_xattr_info i = {
322 .name_index = EXT4_XATTR_INDEX_SYSTEM,
323 .name = EXT4_XATTR_SYSTEM_DATA,
324 };
325
326 /* If the old space is ok, write the data directly. */
327 if (len <= EXT4_I(inode)->i_inline_size)
328 return 0;
329
330 error = ext4_get_inode_loc(inode, &is.iloc);
331 if (error)
332 return error;
333
334 error = ext4_xattr_ibody_find(inode, &i, &is);
335 if (error)
336 goto out;
337
338 BUG_ON(is.s.not_found);
339
340 len -= EXT4_MIN_INLINE_DATA_SIZE;
341 value = kzalloc(len, GFP_NOFS);
342 if (!value)
343 goto out;
344
345 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
346 value, len);
347 if (error == -ENODATA)
348 goto out;
349
350 error = ext4_journal_get_write_access(handle, is.iloc.bh);
351 if (error)
352 goto out;
353
354 /* Update the xttr entry. */
355 i.value = value;
356 i.value_len = len;
357
358 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
359 if (error)
360 goto out;
361
362 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
363 (void *)ext4_raw_inode(&is.iloc));
364 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
365 le32_to_cpu(is.s.here->e_value_size);
366 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
367 get_bh(is.iloc.bh);
368 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
369
370out:
371 kfree(value);
372 brelse(is.iloc.bh);
373 return error;
374}
375
376int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
377 unsigned int len)
378{
379 int ret, size;
380 struct ext4_inode_info *ei = EXT4_I(inode);
381
382 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
383 return -ENOSPC;
384
385 size = ext4_get_max_inline_size(inode);
386 if (size < len)
387 return -ENOSPC;
388
389 down_write(&EXT4_I(inode)->xattr_sem);
390
391 if (ei->i_inline_off)
392 ret = ext4_update_inline_data(handle, inode, len);
393 else
394 ret = ext4_create_inline_data(handle, inode, len);
395
396 up_write(&EXT4_I(inode)->xattr_sem);
397
398 return ret;
399}
400
401static int ext4_destroy_inline_data_nolock(handle_t *handle,
402 struct inode *inode)
403{
404 struct ext4_inode_info *ei = EXT4_I(inode);
405 struct ext4_xattr_ibody_find is = {
406 .s = { .not_found = 0, },
407 };
408 struct ext4_xattr_info i = {
409 .name_index = EXT4_XATTR_INDEX_SYSTEM,
410 .name = EXT4_XATTR_SYSTEM_DATA,
411 .value = NULL,
412 .value_len = 0,
413 };
414 int error;
415
416 if (!ei->i_inline_off)
417 return 0;
418
419 error = ext4_get_inode_loc(inode, &is.iloc);
420 if (error)
421 return error;
422
423 error = ext4_xattr_ibody_find(inode, &i, &is);
424 if (error)
425 goto out;
426
427 error = ext4_journal_get_write_access(handle, is.iloc.bh);
428 if (error)
429 goto out;
430
431 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
432 if (error)
433 goto out;
434
435 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
436 0, EXT4_MIN_INLINE_DATA_SIZE);
437
438 if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
439 EXT4_FEATURE_INCOMPAT_EXTENTS)) {
440 if (S_ISDIR(inode->i_mode) ||
441 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
442 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
443 ext4_ext_tree_init(handle, inode);
444 }
445 }
446 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
447
448 get_bh(is.iloc.bh);
449 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
450
451 EXT4_I(inode)->i_inline_off = 0;
452 EXT4_I(inode)->i_inline_size = 0;
453 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
454out:
455 brelse(is.iloc.bh);
456 if (error == -ENODATA)
457 error = 0;
458 return error;
459}
460
461static int ext4_read_inline_page(struct inode *inode, struct page *page)
462{
463 void *kaddr;
464 int ret = 0;
465 size_t len;
466 struct ext4_iloc iloc;
467
468 BUG_ON(!PageLocked(page));
469 BUG_ON(!ext4_has_inline_data(inode));
470 BUG_ON(page->index);
471
472 if (!EXT4_I(inode)->i_inline_off) {
473 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
474 inode->i_ino);
475 goto out;
476 }
477
478 ret = ext4_get_inode_loc(inode, &iloc);
479 if (ret)
480 goto out;
481
482 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
483 kaddr = kmap_atomic(page);
484 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
485 flush_dcache_page(page);
486 kunmap_atomic(kaddr);
487 zero_user_segment(page, len, PAGE_CACHE_SIZE);
488 SetPageUptodate(page);
489 brelse(iloc.bh);
490
491out:
492 return ret;
493}
494
495int ext4_readpage_inline(struct inode *inode, struct page *page)
496{
497 int ret = 0;
498
499 down_read(&EXT4_I(inode)->xattr_sem);
500 if (!ext4_has_inline_data(inode)) {
501 up_read(&EXT4_I(inode)->xattr_sem);
502 return -EAGAIN;
503 }
504
505 /*
506 * Current inline data can only exist in the 1st page,
507 * So for all the other pages, just set them uptodate.
508 */
509 if (!page->index)
510 ret = ext4_read_inline_page(inode, page);
511 else if (!PageUptodate(page)) {
512 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
513 SetPageUptodate(page);
514 }
515
516 up_read(&EXT4_I(inode)->xattr_sem);
517
518 unlock_page(page);
519 return ret >= 0 ? 0 : ret;
520}
521
522static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
523 struct inode *inode,
524 unsigned flags)
525{
526 int ret, needed_blocks;
527 handle_t *handle = NULL;
528 int retries = 0, sem_held = 0;
529 struct page *page = NULL;
530 unsigned from, to;
531 struct ext4_iloc iloc;
532
533 if (!ext4_has_inline_data(inode)) {
534 /*
535 * clear the flag so that no new write
536 * will trap here again.
537 */
538 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
539 return 0;
540 }
541
542 needed_blocks = ext4_writepage_trans_blocks(inode);
543
544 ret = ext4_get_inode_loc(inode, &iloc);
545 if (ret)
546 return ret;
547
548retry:
549 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
550 if (IS_ERR(handle)) {
551 ret = PTR_ERR(handle);
552 handle = NULL;
553 goto out;
554 }
555
556 /* We cannot recurse into the filesystem as the transaction is already
557 * started */
558 flags |= AOP_FLAG_NOFS;
559
560 page = grab_cache_page_write_begin(mapping, 0, flags);
561 if (!page) {
562 ret = -ENOMEM;
563 goto out;
564 }
565
566 down_write(&EXT4_I(inode)->xattr_sem);
567 sem_held = 1;
568 /* If some one has already done this for us, just exit. */
569 if (!ext4_has_inline_data(inode)) {
570 ret = 0;
571 goto out;
572 }
573
574 from = 0;
575 to = ext4_get_inline_size(inode);
576 if (!PageUptodate(page)) {
577 ret = ext4_read_inline_page(inode, page);
578 if (ret < 0)
579 goto out;
580 }
581
582 ret = ext4_destroy_inline_data_nolock(handle, inode);
583 if (ret)
584 goto out;
585
586 if (ext4_should_dioread_nolock(inode))
587 ret = __block_write_begin(page, from, to, ext4_get_block_write);
588 else
589 ret = __block_write_begin(page, from, to, ext4_get_block);
590
591 if (!ret && ext4_should_journal_data(inode)) {
592 ret = ext4_walk_page_buffers(handle, page_buffers(page),
593 from, to, NULL,
594 do_journal_get_write_access);
595 }
596
597 if (ret) {
598 unlock_page(page);
599 page_cache_release(page);
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 block_commit_write(page, from, to);
620out:
621 if (page) {
622 unlock_page(page);
623 page_cache_release(page);
624 }
625 if (sem_held)
626 up_write(&EXT4_I(inode)->xattr_sem);
627 if (handle)
628 ext4_journal_stop(handle);
629 brelse(iloc.bh);
630 return ret;
631}
632
633/*
634 * Try to write data in the inode.
635 * If the inode has inline data, check whether the new write can be
636 * in the inode also. If not, create the page the handle, move the data
637 * to the page make it update and let the later codes create extent for it.
638 */
639int ext4_try_to_write_inline_data(struct address_space *mapping,
640 struct inode *inode,
641 loff_t pos, unsigned len,
642 unsigned flags,
643 struct page **pagep)
644{
645 int ret;
646 handle_t *handle;
647 struct page *page;
648 struct ext4_iloc iloc;
649
650 if (pos + len > ext4_get_max_inline_size(inode))
651 goto convert;
652
653 ret = ext4_get_inode_loc(inode, &iloc);
654 if (ret)
655 return ret;
656
657 /*
658 * The possible write could happen in the inode,
659 * so try to reserve the space in inode first.
660 */
661 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
662 if (IS_ERR(handle)) {
663 ret = PTR_ERR(handle);
664 handle = NULL;
665 goto out;
666 }
667
668 ret = ext4_prepare_inline_data(handle, inode, pos + len);
669 if (ret && ret != -ENOSPC)
670 goto out;
671
672 /* We don't have space in inline inode, so convert it to extent. */
673 if (ret == -ENOSPC) {
674 ext4_journal_stop(handle);
675 brelse(iloc.bh);
676 goto convert;
677 }
678
679 flags |= AOP_FLAG_NOFS;
680
681 page = grab_cache_page_write_begin(mapping, 0, flags);
682 if (!page) {
683 ret = -ENOMEM;
684 goto out;
685 }
686
687 *pagep = page;
688 down_read(&EXT4_I(inode)->xattr_sem);
689 if (!ext4_has_inline_data(inode)) {
690 ret = 0;
691 unlock_page(page);
692 page_cache_release(page);
693 goto out_up_read;
694 }
695
696 if (!PageUptodate(page)) {
697 ret = ext4_read_inline_page(inode, page);
698 if (ret < 0)
699 goto out_up_read;
700 }
701
702 ret = 1;
703 handle = NULL;
704out_up_read:
705 up_read(&EXT4_I(inode)->xattr_sem);
706out:
707 if (handle)
708 ext4_journal_stop(handle);
709 brelse(iloc.bh);
710 return ret;
711convert:
712 return ext4_convert_inline_data_to_extent(mapping,
713 inode, flags);
714}
715
716int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
717 unsigned copied, struct page *page)
718{
719 int ret;
720 void *kaddr;
721 struct ext4_iloc iloc;
722
723 if (unlikely(copied < len)) {
724 if (!PageUptodate(page)) {
725 copied = 0;
726 goto out;
727 }
728 }
729
730 ret = ext4_get_inode_loc(inode, &iloc);
731 if (ret) {
732 ext4_std_error(inode->i_sb, ret);
733 copied = 0;
734 goto out;
735 }
736
737 down_write(&EXT4_I(inode)->xattr_sem);
738 BUG_ON(!ext4_has_inline_data(inode));
739
740 kaddr = kmap_atomic(page);
741 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
742 kunmap_atomic(kaddr);
743 SetPageUptodate(page);
744 /* clear page dirty so that writepages wouldn't work for us. */
745 ClearPageDirty(page);
746
747 up_write(&EXT4_I(inode)->xattr_sem);
748 brelse(iloc.bh);
749out:
750 return copied;
751}
752
753struct buffer_head *
754ext4_journalled_write_inline_data(struct inode *inode,
755 unsigned len,
756 struct page *page)
757{
758 int ret;
759 void *kaddr;
760 struct ext4_iloc iloc;
761
762 ret = ext4_get_inode_loc(inode, &iloc);
763 if (ret) {
764 ext4_std_error(inode->i_sb, ret);
765 return NULL;
766 }
767
768 down_write(&EXT4_I(inode)->xattr_sem);
769 kaddr = kmap_atomic(page);
770 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
771 kunmap_atomic(kaddr);
772 up_write(&EXT4_I(inode)->xattr_sem);
773
774 return iloc.bh;
775}
776
777/*
778 * Try to make the page cache and handle ready for the inline data case.
779 * We can call this function in 2 cases:
780 * 1. The inode is created and the first write exceeds inline size. We can
781 * clear the inode state safely.
782 * 2. The inode has inline data, then we need to read the data, make it
783 * update and dirty so that ext4_da_writepages can handle it. We don't
784 * need to start the journal since the file's metatdata isn't changed now.
785 */
786static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
787 struct inode *inode,
788 unsigned flags,
789 void **fsdata)
790{
791 int ret = 0, inline_size;
792 struct page *page;
793
794 page = grab_cache_page_write_begin(mapping, 0, flags);
795 if (!page)
796 return -ENOMEM;
797
798 down_read(&EXT4_I(inode)->xattr_sem);
799 if (!ext4_has_inline_data(inode)) {
800 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
801 goto out;
802 }
803
804 inline_size = ext4_get_inline_size(inode);
805
806 if (!PageUptodate(page)) {
807 ret = ext4_read_inline_page(inode, page);
808 if (ret < 0)
809 goto out;
810 }
811
812 ret = __block_write_begin(page, 0, inline_size,
813 ext4_da_get_block_prep);
814 if (ret) {
815 ext4_truncate_failed_write(inode);
816 goto out;
817 }
818
819 SetPageDirty(page);
820 SetPageUptodate(page);
821 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
822 *fsdata = (void *)CONVERT_INLINE_DATA;
823
824out:
825 up_read(&EXT4_I(inode)->xattr_sem);
826 if (page) {
827 unlock_page(page);
828 page_cache_release(page);
829 }
830 return ret;
831}
832
833/*
834 * Prepare the write for the inline data.
835 * If the the data can be written into the inode, we just read
836 * the page and make it uptodate, and start the journal.
837 * Otherwise read the page, makes it dirty so that it can be
838 * handle in writepages(the i_disksize update is left to the
839 * normal ext4_da_write_end).
840 */
841int ext4_da_write_inline_data_begin(struct address_space *mapping,
842 struct inode *inode,
843 loff_t pos, unsigned len,
844 unsigned flags,
845 struct page **pagep,
846 void **fsdata)
847{
848 int ret, inline_size;
849 handle_t *handle;
850 struct page *page;
851 struct ext4_iloc iloc;
852 int retries;
853
854 ret = ext4_get_inode_loc(inode, &iloc);
855 if (ret)
856 return ret;
857
858retry_journal:
859 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
860 if (IS_ERR(handle)) {
861 ret = PTR_ERR(handle);
862 goto out;
863 }
864
865 inline_size = ext4_get_max_inline_size(inode);
866
867 ret = -ENOSPC;
868 if (inline_size >= pos + len) {
869 ret = ext4_prepare_inline_data(handle, inode, pos + len);
870 if (ret && ret != -ENOSPC)
871 goto out_journal;
872 }
873
874 if (ret == -ENOSPC) {
875 ret = ext4_da_convert_inline_data_to_extent(mapping,
876 inode,
877 flags,
878 fsdata);
879 ext4_journal_stop(handle);
880 if (ret == -ENOSPC &&
881 ext4_should_retry_alloc(inode->i_sb, &retries))
882 goto retry_journal;
883 goto out;
884 }
885
886 /*
887 * We cannot recurse into the filesystem as the transaction
888 * is already started.
889 */
890 flags |= AOP_FLAG_NOFS;
891
892 page = grab_cache_page_write_begin(mapping, 0, flags);
893 if (!page) {
894 ret = -ENOMEM;
895 goto out_journal;
896 }
897
898 down_read(&EXT4_I(inode)->xattr_sem);
899 if (!ext4_has_inline_data(inode)) {
900 ret = 0;
901 goto out_release_page;
902 }
903
904 if (!PageUptodate(page)) {
905 ret = ext4_read_inline_page(inode, page);
906 if (ret < 0)
907 goto out_release_page;
908 }
909
910 up_read(&EXT4_I(inode)->xattr_sem);
911 *pagep = page;
912 brelse(iloc.bh);
913 return 1;
914out_release_page:
915 up_read(&EXT4_I(inode)->xattr_sem);
916 unlock_page(page);
917 page_cache_release(page);
918out_journal:
919 ext4_journal_stop(handle);
920out:
921 brelse(iloc.bh);
922 return ret;
923}
924
925int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
926 unsigned len, unsigned copied,
927 struct page *page)
928{
929 int i_size_changed = 0;
930
931 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
932
933 /*
934 * No need to use i_size_read() here, the i_size
935 * cannot change under us because we hold i_mutex.
936 *
937 * But it's important to update i_size while still holding page lock:
938 * page writeout could otherwise come in and zero beyond i_size.
939 */
940 if (pos+copied > inode->i_size) {
941 i_size_write(inode, pos+copied);
942 i_size_changed = 1;
943 }
944 unlock_page(page);
945 page_cache_release(page);
946
947 /*
948 * Don't mark the inode dirty under page lock. First, it unnecessarily
949 * makes the holding time of page lock longer. Second, it forces lock
950 * ordering of page lock and transaction start for journaling
951 * filesystems.
952 */
953 if (i_size_changed)
954 mark_inode_dirty(inode);
955
956 return copied;
957}
958
959#ifdef INLINE_DIR_DEBUG
960void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
961 void *inline_start, int inline_size)
962{
963 int offset;
964 unsigned short de_len;
965 struct ext4_dir_entry_2 *de = inline_start;
966 void *dlimit = inline_start + inline_size;
967
968 trace_printk("inode %lu\n", dir->i_ino);
969 offset = 0;
970 while ((void *)de < dlimit) {
971 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
972 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
973 offset, de_len, de->name_len, de->name,
974 de->name_len, le32_to_cpu(de->inode));
975 if (ext4_check_dir_entry(dir, NULL, de, bh,
976 inline_start, inline_size, offset))
977 BUG();
978
979 offset += de_len;
980 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
981 }
982}
983#else
984#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
985#endif
986
987/*
988 * Add a new entry into a inline dir.
989 * It will return -ENOSPC if no space is available, and -EIO
990 * and -EEXIST if directory entry already exists.
991 */
992static int ext4_add_dirent_to_inline(handle_t *handle,
993 struct dentry *dentry,
994 struct inode *inode,
995 struct ext4_iloc *iloc,
996 void *inline_start, int inline_size)
997{
998 struct inode *dir = dentry->d_parent->d_inode;
999 const char *name = dentry->d_name.name;
1000 int namelen = dentry->d_name.len;
1001 int err;
1002 struct ext4_dir_entry_2 *de;
1003
1004 err = ext4_find_dest_de(dir, inode, iloc->bh,
1005 inline_start, inline_size,
1006 name, namelen, &de);
1007 if (err)
1008 return err;
1009
1010 err = ext4_journal_get_write_access(handle, iloc->bh);
1011 if (err)
1012 return err;
1013 ext4_insert_dentry(inode, de, inline_size, name, namelen);
1014
1015 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1016
1017 /*
1018 * XXX shouldn't update any times until successful
1019 * completion of syscall, but too many callers depend
1020 * on this.
1021 *
1022 * XXX similarly, too many callers depend on
1023 * ext4_new_inode() setting the times, but error
1024 * recovery deletes the inode, so the worst that can
1025 * happen is that the times are slightly out of date
1026 * and/or different from the directory change time.
1027 */
1028 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1029 ext4_update_dx_flag(dir);
1030 dir->i_version++;
1031 ext4_mark_inode_dirty(handle, dir);
1032 return 1;
1033}
1034
1035static void *ext4_get_inline_xattr_pos(struct inode *inode,
1036 struct ext4_iloc *iloc)
1037{
1038 struct ext4_xattr_entry *entry;
1039 struct ext4_xattr_ibody_header *header;
1040
1041 BUG_ON(!EXT4_I(inode)->i_inline_off);
1042
1043 header = IHDR(inode, ext4_raw_inode(iloc));
1044 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1045 EXT4_I(inode)->i_inline_off);
1046
1047 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1048}
1049
1050/* Set the final de to cover the whole block. */
1051static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1052{
1053 struct ext4_dir_entry_2 *de, *prev_de;
1054 void *limit;
1055 int de_len;
1056
1057 de = (struct ext4_dir_entry_2 *)de_buf;
1058 if (old_size) {
1059 limit = de_buf + old_size;
1060 do {
1061 prev_de = de;
1062 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1063 de_buf += de_len;
1064 de = (struct ext4_dir_entry_2 *)de_buf;
1065 } while (de_buf < limit);
1066
1067 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1068 old_size, new_size);
1069 } else {
1070 /* this is just created, so create an empty entry. */
1071 de->inode = 0;
1072 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1073 }
1074}
1075
1076static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1077 struct ext4_iloc *iloc)
1078{
1079 int ret;
1080 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1081 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1082
1083 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1084 return -ENOSPC;
1085
1086 ret = ext4_update_inline_data(handle, dir,
1087 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1088 if (ret)
1089 return ret;
1090
1091 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1092 EXT4_I(dir)->i_inline_size -
1093 EXT4_MIN_INLINE_DATA_SIZE);
1094 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1095 return 0;
1096}
1097
1098static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1099 struct ext4_iloc *iloc,
1100 void *buf, int inline_size)
1101{
1102 ext4_create_inline_data(handle, inode, inline_size);
1103 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1104 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1105}
1106
1107static int ext4_finish_convert_inline_dir(handle_t *handle,
1108 struct inode *inode,
1109 struct buffer_head *dir_block,
1110 void *buf,
1111 int inline_size)
1112{
1113 int err, csum_size = 0, header_size = 0;
1114 struct ext4_dir_entry_2 *de;
1115 struct ext4_dir_entry_tail *t;
1116 void *target = dir_block->b_data;
1117
1118 /*
1119 * First create "." and ".." and then copy the dir information
1120 * back to the block.
1121 */
1122 de = (struct ext4_dir_entry_2 *)target;
1123 de = ext4_init_dot_dotdot(inode, de,
1124 inode->i_sb->s_blocksize, csum_size,
1125 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1126 header_size = (void *)de - target;
1127
1128 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1129 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1130
1131 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1132 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1133 csum_size = sizeof(struct ext4_dir_entry_tail);
1134
1135 inode->i_size = inode->i_sb->s_blocksize;
1136 i_size_write(inode, inode->i_sb->s_blocksize);
1137 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1138 ext4_update_final_de(dir_block->b_data,
1139 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1140 inode->i_sb->s_blocksize - csum_size);
1141
1142 if (csum_size) {
1143 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1144 inode->i_sb->s_blocksize);
1145 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1146 }
1147 set_buffer_uptodate(dir_block);
1148 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1149 if (err)
1150 goto out;
1151 set_buffer_verified(dir_block);
1152out:
1153 return err;
1154}
1155
1156static int ext4_convert_inline_data_nolock(handle_t *handle,
1157 struct inode *inode,
1158 struct ext4_iloc *iloc)
1159{
1160 int error;
1161 void *buf = NULL;
1162 struct buffer_head *data_bh = NULL;
1163 struct ext4_map_blocks map;
1164 int inline_size;
1165
1166 inline_size = ext4_get_inline_size(inode);
1167 buf = kmalloc(inline_size, GFP_NOFS);
1168 if (!buf) {
1169 error = -ENOMEM;
1170 goto out;
1171 }
1172
1173 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1174 if (error < 0)
1175 goto out;
1176
1177 error = ext4_destroy_inline_data_nolock(handle, inode);
1178 if (error)
1179 goto out;
1180
1181 map.m_lblk = 0;
1182 map.m_len = 1;
1183 map.m_flags = 0;
1184 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1185 if (error < 0)
1186 goto out_restore;
1187 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1188 error = -EIO;
1189 goto out_restore;
1190 }
1191
1192 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1193 if (!data_bh) {
1194 error = -ENOMEM;
1195 goto out_restore;
1196 }
1197
1198 lock_buffer(data_bh);
1199 error = ext4_journal_get_create_access(handle, data_bh);
1200 if (error) {
1201 unlock_buffer(data_bh);
1202 error = -EIO;
1203 goto out_restore;
1204 }
1205 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1206
1207 if (!S_ISDIR(inode->i_mode)) {
1208 memcpy(data_bh->b_data, buf, inline_size);
1209 set_buffer_uptodate(data_bh);
1210 error = ext4_handle_dirty_metadata(handle,
1211 inode, data_bh);
1212 } else {
1213 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1214 buf, inline_size);
1215 }
1216
1217 unlock_buffer(data_bh);
1218out_restore:
1219 if (error)
1220 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1221
1222out:
1223 brelse(data_bh);
1224 kfree(buf);
1225 return error;
1226}
1227
1228/*
1229 * Try to add the new entry to the inline data.
1230 * If succeeds, return 0. If not, extended the inline dir and copied data to
1231 * the new created block.
1232 */
1233int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1234 struct inode *inode)
1235{
1236 int ret, inline_size;
1237 void *inline_start;
1238 struct ext4_iloc iloc;
1239 struct inode *dir = dentry->d_parent->d_inode;
1240
1241 ret = ext4_get_inode_loc(dir, &iloc);
1242 if (ret)
1243 return ret;
1244
1245 down_write(&EXT4_I(dir)->xattr_sem);
1246 if (!ext4_has_inline_data(dir))
1247 goto out;
1248
1249 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1250 EXT4_INLINE_DOTDOT_SIZE;
1251 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1252
1253 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1254 inline_start, inline_size);
1255 if (ret != -ENOSPC)
1256 goto out;
1257
1258 /* check whether it can be inserted to inline xattr space. */
1259 inline_size = EXT4_I(dir)->i_inline_size -
1260 EXT4_MIN_INLINE_DATA_SIZE;
1261 if (!inline_size) {
1262 /* Try to use the xattr space.*/
1263 ret = ext4_update_inline_dir(handle, dir, &iloc);
1264 if (ret && ret != -ENOSPC)
1265 goto out;
1266
1267 inline_size = EXT4_I(dir)->i_inline_size -
1268 EXT4_MIN_INLINE_DATA_SIZE;
1269 }
1270
1271 if (inline_size) {
1272 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1273
1274 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1275 inline_start, inline_size);
1276
1277 if (ret != -ENOSPC)
1278 goto out;
1279 }
1280
1281 /*
1282 * The inline space is filled up, so create a new block for it.
1283 * As the extent tree will be created, we have to save the inline
1284 * dir first.
1285 */
1286 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1287
1288out:
1289 ext4_mark_inode_dirty(handle, dir);
1290 up_write(&EXT4_I(dir)->xattr_sem);
1291 brelse(iloc.bh);
1292 return ret;
1293}
1294
1295/*
1296 * This function fills a red-black tree with information from an
1297 * inlined dir. It returns the number directory entries loaded
1298 * into the tree. If there is an error it is returned in err.
1299 */
1300int htree_inlinedir_to_tree(struct file *dir_file,
1301 struct inode *dir, ext4_lblk_t block,
1302 struct dx_hash_info *hinfo,
1303 __u32 start_hash, __u32 start_minor_hash,
1304 int *has_inline_data)
1305{
1306 int err = 0, count = 0;
1307 unsigned int parent_ino;
1308 int pos;
1309 struct ext4_dir_entry_2 *de;
1310 struct inode *inode = file_inode(dir_file);
1311 int ret, inline_size = 0;
1312 struct ext4_iloc iloc;
1313 void *dir_buf = NULL;
1314 struct ext4_dir_entry_2 fake;
1315
1316 ret = ext4_get_inode_loc(inode, &iloc);
1317 if (ret)
1318 return ret;
1319
1320 down_read(&EXT4_I(inode)->xattr_sem);
1321 if (!ext4_has_inline_data(inode)) {
1322 up_read(&EXT4_I(inode)->xattr_sem);
1323 *has_inline_data = 0;
1324 goto out;
1325 }
1326
1327 inline_size = ext4_get_inline_size(inode);
1328 dir_buf = kmalloc(inline_size, GFP_NOFS);
1329 if (!dir_buf) {
1330 ret = -ENOMEM;
1331 up_read(&EXT4_I(inode)->xattr_sem);
1332 goto out;
1333 }
1334
1335 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1336 up_read(&EXT4_I(inode)->xattr_sem);
1337 if (ret < 0)
1338 goto out;
1339
1340 pos = 0;
1341 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1342 while (pos < inline_size) {
1343 /*
1344 * As inlined dir doesn't store any information about '.' and
1345 * only the inode number of '..' is stored, we have to handle
1346 * them differently.
1347 */
1348 if (pos == 0) {
1349 fake.inode = cpu_to_le32(inode->i_ino);
1350 fake.name_len = 1;
1351 strcpy(fake.name, ".");
1352 fake.rec_len = ext4_rec_len_to_disk(
1353 EXT4_DIR_REC_LEN(fake.name_len),
1354 inline_size);
1355 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1356 de = &fake;
1357 pos = EXT4_INLINE_DOTDOT_OFFSET;
1358 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1359 fake.inode = cpu_to_le32(parent_ino);
1360 fake.name_len = 2;
1361 strcpy(fake.name, "..");
1362 fake.rec_len = ext4_rec_len_to_disk(
1363 EXT4_DIR_REC_LEN(fake.name_len),
1364 inline_size);
1365 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1366 de = &fake;
1367 pos = EXT4_INLINE_DOTDOT_SIZE;
1368 } else {
1369 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1370 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1371 if (ext4_check_dir_entry(inode, dir_file, de,
1372 iloc.bh, dir_buf,
1373 inline_size, pos)) {
1374 ret = count;
1375 goto out;
1376 }
1377 }
1378
1379 ext4fs_dirhash(de->name, de->name_len, hinfo);
1380 if ((hinfo->hash < start_hash) ||
1381 ((hinfo->hash == start_hash) &&
1382 (hinfo->minor_hash < start_minor_hash)))
1383 continue;
1384 if (de->inode == 0)
1385 continue;
1386 err = ext4_htree_store_dirent(dir_file,
1387 hinfo->hash, hinfo->minor_hash, de);
1388 if (err) {
1389 count = err;
1390 goto out;
1391 }
1392 count++;
1393 }
1394 ret = count;
1395out:
1396 kfree(dir_buf);
1397 brelse(iloc.bh);
1398 return ret;
1399}
1400
1401/*
1402 * So this function is called when the volume is mkfsed with
1403 * dir_index disabled. In order to keep f_pos persistent
1404 * after we convert from an inlined dir to a blocked based,
1405 * we just pretend that we are a normal dir and return the
1406 * offset as if '.' and '..' really take place.
1407 *
1408 */
1409int ext4_read_inline_dir(struct file *file,
1410 struct dir_context *ctx,
1411 int *has_inline_data)
1412{
1413 unsigned int offset, parent_ino;
1414 int i;
1415 struct ext4_dir_entry_2 *de;
1416 struct super_block *sb;
1417 struct inode *inode = file_inode(file);
1418 int ret, inline_size = 0;
1419 struct ext4_iloc iloc;
1420 void *dir_buf = NULL;
1421 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1422
1423 ret = ext4_get_inode_loc(inode, &iloc);
1424 if (ret)
1425 return ret;
1426
1427 down_read(&EXT4_I(inode)->xattr_sem);
1428 if (!ext4_has_inline_data(inode)) {
1429 up_read(&EXT4_I(inode)->xattr_sem);
1430 *has_inline_data = 0;
1431 goto out;
1432 }
1433
1434 inline_size = ext4_get_inline_size(inode);
1435 dir_buf = kmalloc(inline_size, GFP_NOFS);
1436 if (!dir_buf) {
1437 ret = -ENOMEM;
1438 up_read(&EXT4_I(inode)->xattr_sem);
1439 goto out;
1440 }
1441
1442 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1443 up_read(&EXT4_I(inode)->xattr_sem);
1444 if (ret < 0)
1445 goto out;
1446
1447 ret = 0;
1448 sb = inode->i_sb;
1449 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1450 offset = ctx->pos;
1451
1452 /*
1453 * dotdot_offset and dotdot_size is the real offset and
1454 * size for ".." and "." if the dir is block based while
1455 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1456 * So we will use extra_offset and extra_size to indicate them
1457 * during the inline dir iteration.
1458 */
1459 dotdot_offset = EXT4_DIR_REC_LEN(1);
1460 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1461 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1462 extra_size = extra_offset + inline_size;
1463
1464 /*
1465 * If the version has changed since the last call to
1466 * readdir(2), then we might be pointing to an invalid
1467 * dirent right now. Scan from the start of the inline
1468 * dir to make sure.
1469 */
1470 if (file->f_version != inode->i_version) {
1471 for (i = 0; i < extra_size && i < offset;) {
1472 /*
1473 * "." is with offset 0 and
1474 * ".." is dotdot_offset.
1475 */
1476 if (!i) {
1477 i = dotdot_offset;
1478 continue;
1479 } else if (i == dotdot_offset) {
1480 i = dotdot_size;
1481 continue;
1482 }
1483 /* for other entry, the real offset in
1484 * the buf has to be tuned accordingly.
1485 */
1486 de = (struct ext4_dir_entry_2 *)
1487 (dir_buf + i - extra_offset);
1488 /* It's too expensive to do a full
1489 * dirent test each time round this
1490 * loop, but we do have to test at
1491 * least that it is non-zero. A
1492 * failure will be detected in the
1493 * dirent test below. */
1494 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1495 < EXT4_DIR_REC_LEN(1))
1496 break;
1497 i += ext4_rec_len_from_disk(de->rec_len,
1498 extra_size);
1499 }
1500 offset = i;
1501 ctx->pos = offset;
1502 file->f_version = inode->i_version;
1503 }
1504
1505 while (ctx->pos < extra_size) {
1506 if (ctx->pos == 0) {
1507 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1508 goto out;
1509 ctx->pos = dotdot_offset;
1510 continue;
1511 }
1512
1513 if (ctx->pos == dotdot_offset) {
1514 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1515 goto out;
1516 ctx->pos = dotdot_size;
1517 continue;
1518 }
1519
1520 de = (struct ext4_dir_entry_2 *)
1521 (dir_buf + ctx->pos - extra_offset);
1522 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1523 extra_size, ctx->pos))
1524 goto out;
1525 if (le32_to_cpu(de->inode)) {
1526 if (!dir_emit(ctx, de->name, de->name_len,
1527 le32_to_cpu(de->inode),
1528 get_dtype(sb, de->file_type)))
1529 goto out;
1530 }
1531 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1532 }
1533out:
1534 kfree(dir_buf);
1535 brelse(iloc.bh);
1536 return ret;
1537}
1538
1539struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1540 struct ext4_dir_entry_2 **parent_de,
1541 int *retval)
1542{
1543 struct ext4_iloc iloc;
1544
1545 *retval = ext4_get_inode_loc(inode, &iloc);
1546 if (*retval)
1547 return NULL;
1548
1549 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1550
1551 return iloc.bh;
1552}
1553
1554/*
1555 * Try to create the inline data for the new dir.
1556 * If it succeeds, return 0, otherwise return the error.
1557 * In case of ENOSPC, the caller should create the normal disk layout dir.
1558 */
1559int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1560 struct inode *inode)
1561{
1562 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1563 struct ext4_iloc iloc;
1564 struct ext4_dir_entry_2 *de;
1565
1566 ret = ext4_get_inode_loc(inode, &iloc);
1567 if (ret)
1568 return ret;
1569
1570 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1571 if (ret)
1572 goto out;
1573
1574 /*
1575 * For inline dir, we only save the inode information for the ".."
1576 * and create a fake dentry to cover the left space.
1577 */
1578 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1579 de->inode = cpu_to_le32(parent->i_ino);
1580 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1581 de->inode = 0;
1582 de->rec_len = ext4_rec_len_to_disk(
1583 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1584 inline_size);
1585 set_nlink(inode, 2);
1586 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1587out:
1588 brelse(iloc.bh);
1589 return ret;
1590}
1591
1592struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1593 const struct qstr *d_name,
1594 struct ext4_dir_entry_2 **res_dir,
1595 int *has_inline_data)
1596{
1597 int ret;
1598 struct ext4_iloc iloc;
1599 void *inline_start;
1600 int inline_size;
1601
1602 if (ext4_get_inode_loc(dir, &iloc))
1603 return NULL;
1604
1605 down_read(&EXT4_I(dir)->xattr_sem);
1606 if (!ext4_has_inline_data(dir)) {
1607 *has_inline_data = 0;
1608 goto out;
1609 }
1610
1611 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1612 EXT4_INLINE_DOTDOT_SIZE;
1613 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1614 ret = search_dir(iloc.bh, inline_start, inline_size,
1615 dir, d_name, 0, res_dir);
1616 if (ret == 1)
1617 goto out_find;
1618 if (ret < 0)
1619 goto out;
1620
1621 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1622 goto out;
1623
1624 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1625 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1626
1627 ret = search_dir(iloc.bh, inline_start, inline_size,
1628 dir, d_name, 0, res_dir);
1629 if (ret == 1)
1630 goto out_find;
1631
1632out:
1633 brelse(iloc.bh);
1634 iloc.bh = NULL;
1635out_find:
1636 up_read(&EXT4_I(dir)->xattr_sem);
1637 return iloc.bh;
1638}
1639
1640int ext4_delete_inline_entry(handle_t *handle,
1641 struct inode *dir,
1642 struct ext4_dir_entry_2 *de_del,
1643 struct buffer_head *bh,
1644 int *has_inline_data)
1645{
1646 int err, inline_size;
1647 struct ext4_iloc iloc;
1648 void *inline_start;
1649
1650 err = ext4_get_inode_loc(dir, &iloc);
1651 if (err)
1652 return err;
1653
1654 down_write(&EXT4_I(dir)->xattr_sem);
1655 if (!ext4_has_inline_data(dir)) {
1656 *has_inline_data = 0;
1657 goto out;
1658 }
1659
1660 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1661 EXT4_MIN_INLINE_DATA_SIZE) {
1662 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1663 EXT4_INLINE_DOTDOT_SIZE;
1664 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1665 EXT4_INLINE_DOTDOT_SIZE;
1666 } else {
1667 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1668 inline_size = ext4_get_inline_size(dir) -
1669 EXT4_MIN_INLINE_DATA_SIZE;
1670 }
1671
1672 err = ext4_journal_get_write_access(handle, bh);
1673 if (err)
1674 goto out;
1675
1676 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1677 inline_start, inline_size, 0);
1678 if (err)
1679 goto out;
1680
1681 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1682 err = ext4_mark_inode_dirty(handle, dir);
1683 if (unlikely(err))
1684 goto out;
1685
1686 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1687out:
1688 up_write(&EXT4_I(dir)->xattr_sem);
1689 brelse(iloc.bh);
1690 if (err != -ENOENT)
1691 ext4_std_error(dir->i_sb, err);
1692 return err;
1693}
1694
1695/*
1696 * Get the inline dentry at offset.
1697 */
1698static inline struct ext4_dir_entry_2 *
1699ext4_get_inline_entry(struct inode *inode,
1700 struct ext4_iloc *iloc,
1701 unsigned int offset,
1702 void **inline_start,
1703 int *inline_size)
1704{
1705 void *inline_pos;
1706
1707 BUG_ON(offset > ext4_get_inline_size(inode));
1708
1709 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1710 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1711 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1712 } else {
1713 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1714 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1715 *inline_size = ext4_get_inline_size(inode) -
1716 EXT4_MIN_INLINE_DATA_SIZE;
1717 }
1718
1719 if (inline_start)
1720 *inline_start = inline_pos;
1721 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1722}
1723
1724int empty_inline_dir(struct inode *dir, int *has_inline_data)
1725{
1726 int err, inline_size;
1727 struct ext4_iloc iloc;
1728 void *inline_pos;
1729 unsigned int offset;
1730 struct ext4_dir_entry_2 *de;
1731 int ret = 1;
1732
1733 err = ext4_get_inode_loc(dir, &iloc);
1734 if (err) {
1735 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1736 err, dir->i_ino);
1737 return 1;
1738 }
1739
1740 down_read(&EXT4_I(dir)->xattr_sem);
1741 if (!ext4_has_inline_data(dir)) {
1742 *has_inline_data = 0;
1743 goto out;
1744 }
1745
1746 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1747 if (!le32_to_cpu(de->inode)) {
1748 ext4_warning(dir->i_sb,
1749 "bad inline directory (dir #%lu) - no `..'",
1750 dir->i_ino);
1751 ret = 1;
1752 goto out;
1753 }
1754
1755 offset = EXT4_INLINE_DOTDOT_SIZE;
1756 while (offset < dir->i_size) {
1757 de = ext4_get_inline_entry(dir, &iloc, offset,
1758 &inline_pos, &inline_size);
1759 if (ext4_check_dir_entry(dir, NULL, de,
1760 iloc.bh, inline_pos,
1761 inline_size, offset)) {
1762 ext4_warning(dir->i_sb,
1763 "bad inline directory (dir #%lu) - "
1764 "inode %u, rec_len %u, name_len %d"
1765 "inline size %d\n",
1766 dir->i_ino, le32_to_cpu(de->inode),
1767 le16_to_cpu(de->rec_len), de->name_len,
1768 inline_size);
1769 ret = 1;
1770 goto out;
1771 }
1772 if (le32_to_cpu(de->inode)) {
1773 ret = 0;
1774 goto out;
1775 }
1776 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1777 }
1778
1779out:
1780 up_read(&EXT4_I(dir)->xattr_sem);
1781 brelse(iloc.bh);
1782 return ret;
1783}
1784
1785int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1786{
1787 int ret;
1788
1789 down_write(&EXT4_I(inode)->xattr_sem);
1790 ret = ext4_destroy_inline_data_nolock(handle, inode);
1791 up_write(&EXT4_I(inode)->xattr_sem);
1792
1793 return ret;
1794}
1795
1796int ext4_inline_data_fiemap(struct inode *inode,
1797 struct fiemap_extent_info *fieinfo,
1798 int *has_inline)
1799{
1800 __u64 physical = 0;
1801 __u64 length;
1802 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1803 int error = 0;
1804 struct ext4_iloc iloc;
1805
1806 down_read(&EXT4_I(inode)->xattr_sem);
1807 if (!ext4_has_inline_data(inode)) {
1808 *has_inline = 0;
1809 goto out;
1810 }
1811
1812 error = ext4_get_inode_loc(inode, &iloc);
1813 if (error)
1814 goto out;
1815
1816 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1817 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1818 physical += offsetof(struct ext4_inode, i_block);
1819 length = i_size_read(inode);
1820
1821 if (physical)
1822 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1823 length, flags);
1824 brelse(iloc.bh);
1825out:
1826 up_read(&EXT4_I(inode)->xattr_sem);
1827 return (error < 0 ? error : 0);
1828}
1829
1830/*
1831 * Called during xattr set, and if we can sparse space 'needed',
1832 * just create the extent tree evict the data to the outer block.
1833 *
1834 * We use jbd2 instead of page cache to move data to the 1st block
1835 * so that the whole transaction can be committed as a whole and
1836 * the data isn't lost because of the delayed page cache write.
1837 */
1838int ext4_try_to_evict_inline_data(handle_t *handle,
1839 struct inode *inode,
1840 int needed)
1841{
1842 int error;
1843 struct ext4_xattr_entry *entry;
1844 struct ext4_inode *raw_inode;
1845 struct ext4_iloc iloc;
1846
1847 error = ext4_get_inode_loc(inode, &iloc);
1848 if (error)
1849 return error;
1850
1851 raw_inode = ext4_raw_inode(&iloc);
1852 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1853 EXT4_I(inode)->i_inline_off);
1854 if (EXT4_XATTR_LEN(entry->e_name_len) +
1855 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1856 error = -ENOSPC;
1857 goto out;
1858 }
1859
1860 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1861out:
1862 brelse(iloc.bh);
1863 return error;
1864}
1865
1866void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1867{
1868 handle_t *handle;
1869 int inline_size, value_len, needed_blocks;
1870 size_t i_size;
1871 void *value = NULL;
1872 struct ext4_xattr_ibody_find is = {
1873 .s = { .not_found = -ENODATA, },
1874 };
1875 struct ext4_xattr_info i = {
1876 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1877 .name = EXT4_XATTR_SYSTEM_DATA,
1878 };
1879
1880
1881 needed_blocks = ext4_writepage_trans_blocks(inode);
1882 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1883 if (IS_ERR(handle))
1884 return;
1885
1886 down_write(&EXT4_I(inode)->xattr_sem);
1887 if (!ext4_has_inline_data(inode)) {
1888 *has_inline = 0;
1889 ext4_journal_stop(handle);
1890 return;
1891 }
1892
1893 if (ext4_orphan_add(handle, inode))
1894 goto out;
1895
1896 if (ext4_get_inode_loc(inode, &is.iloc))
1897 goto out;
1898
1899 down_write(&EXT4_I(inode)->i_data_sem);
1900 i_size = inode->i_size;
1901 inline_size = ext4_get_inline_size(inode);
1902 EXT4_I(inode)->i_disksize = i_size;
1903
1904 if (i_size < inline_size) {
1905 /* Clear the content in the xattr space. */
1906 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1907 if (ext4_xattr_ibody_find(inode, &i, &is))
1908 goto out_error;
1909
1910 BUG_ON(is.s.not_found);
1911
1912 value_len = le32_to_cpu(is.s.here->e_value_size);
1913 value = kmalloc(value_len, GFP_NOFS);
1914 if (!value)
1915 goto out_error;
1916
1917 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1918 value, value_len))
1919 goto out_error;
1920
1921 i.value = value;
1922 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1923 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1924 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1925 goto out_error;
1926 }
1927
1928 /* Clear the content within i_blocks. */
1929 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1930 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1931 memset(p + i_size, 0,
1932 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1933 }
1934
1935 EXT4_I(inode)->i_inline_size = i_size <
1936 EXT4_MIN_INLINE_DATA_SIZE ?
1937 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1938 }
1939
1940out_error:
1941 up_write(&EXT4_I(inode)->i_data_sem);
1942out:
1943 brelse(is.iloc.bh);
1944 up_write(&EXT4_I(inode)->xattr_sem);
1945 kfree(value);
1946 if (inode->i_nlink)
1947 ext4_orphan_del(handle, inode);
1948
1949 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1950 ext4_mark_inode_dirty(handle, inode);
1951 if (IS_SYNC(inode))
1952 ext4_handle_sync(handle);
1953
1954 ext4_journal_stop(handle);
1955 return;
1956}
1957
1958int ext4_convert_inline_data(struct inode *inode)
1959{
1960 int error, needed_blocks;
1961 handle_t *handle;
1962 struct ext4_iloc iloc;
1963
1964 if (!ext4_has_inline_data(inode)) {
1965 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1966 return 0;
1967 }
1968
1969 needed_blocks = ext4_writepage_trans_blocks(inode);
1970
1971 iloc.bh = NULL;
1972 error = ext4_get_inode_loc(inode, &iloc);
1973 if (error)
1974 return error;
1975
1976 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1977 if (IS_ERR(handle)) {
1978 error = PTR_ERR(handle);
1979 goto out_free;
1980 }
1981
1982 down_write(&EXT4_I(inode)->xattr_sem);
1983 if (!ext4_has_inline_data(inode)) {
1984 up_write(&EXT4_I(inode)->xattr_sem);
1985 goto out;
1986 }
1987
1988 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1989 up_write(&EXT4_I(inode)->xattr_sem);
1990out:
1991 ext4_journal_stop(handle);
1992out_free:
1993 brelse(iloc.bh);
1994 return error;
1995}