projects/linux·blob v4.13 · refs

2037 lines · 49.6 KiB

1/*
2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <[email protected]>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 */
14
15#include <linux/fiemap.h>
16
17#include "ext4_jbd2.h"
18#include "ext4.h"
19#include "xattr.h"
20#include "truncate.h"
21
22#define EXT4_XATTR_SYSTEM_DATA "data"
23#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
24#define EXT4_INLINE_DOTDOT_OFFSET 2
25#define EXT4_INLINE_DOTDOT_SIZE 4
26
27static int ext4_get_inline_size(struct inode *inode)
28{
29 if (EXT4_I(inode)->i_inline_off)
30 return EXT4_I(inode)->i_inline_size;
31
32 return 0;
33}
34
35static int get_max_inline_xattr_value_size(struct inode *inode,
36 struct ext4_iloc *iloc)
37{
38 struct ext4_xattr_ibody_header *header;
39 struct ext4_xattr_entry *entry;
40 struct ext4_inode *raw_inode;
41 int free, min_offs;
42
43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 EXT4_GOOD_OLD_INODE_SIZE -
45 EXT4_I(inode)->i_extra_isize -
46 sizeof(struct ext4_xattr_ibody_header);
47
48 /*
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
52 */
53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 return EXT4_XATTR_SIZE(min_offs -
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 EXT4_XATTR_ROUND - sizeof(__u32));
57
58 raw_inode = ext4_raw_inode(iloc);
59 header = IHDR(inode, raw_inode);
60 entry = IFIRST(header);
61
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64 if (!entry->e_value_inum && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
66 if (offs < min_offs)
67 min_offs = offs;
68 }
69 }
70 free = min_offs -
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78 goto out;
79 }
80
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85 else
86 free = 0;
87
88out:
89 return free;
90}
91
92/*
93 * Get the maximum size we now can store in an inode.
94 * If we can't find the space for a xattr entry, don't use the space
95 * of the extents since we have no space to indicate the inline data.
96 */
97int ext4_get_max_inline_size(struct inode *inode)
98{
99 int error, max_inline_size;
100 struct ext4_iloc iloc;
101
102 if (EXT4_I(inode)->i_extra_isize == 0)
103 return 0;
104
105 error = ext4_get_inode_loc(inode, &iloc);
106 if (error) {
107 ext4_error_inode(inode, __func__, __LINE__, 0,
108 "can't get inode location %lu",
109 inode->i_ino);
110 return 0;
111 }
112
113 down_read(&EXT4_I(inode)->xattr_sem);
114 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
115 up_read(&EXT4_I(inode)->xattr_sem);
116
117 brelse(iloc.bh);
118
119 if (!max_inline_size)
120 return 0;
121
122 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
123}
124
125/*
126 * this function does not take xattr_sem, which is OK because it is
127 * currently only used in a code path coming form ext4_iget, before
128 * the new inode has been unlocked
129 */
130int ext4_find_inline_data_nolock(struct inode *inode)
131{
132 struct ext4_xattr_ibody_find is = {
133 .s = { .not_found = -ENODATA, },
134 };
135 struct ext4_xattr_info i = {
136 .name_index = EXT4_XATTR_INDEX_SYSTEM,
137 .name = EXT4_XATTR_SYSTEM_DATA,
138 };
139 int error;
140
141 if (EXT4_I(inode)->i_extra_isize == 0)
142 return 0;
143
144 error = ext4_get_inode_loc(inode, &is.iloc);
145 if (error)
146 return error;
147
148 error = ext4_xattr_ibody_find(inode, &i, &is);
149 if (error)
150 goto out;
151
152 if (!is.s.not_found) {
153 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154 (void *)ext4_raw_inode(&is.iloc));
155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156 le32_to_cpu(is.s.here->e_value_size);
157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158 }
159out:
160 brelse(is.iloc.bh);
161 return error;
162}
163
164static int ext4_read_inline_data(struct inode *inode, void *buffer,
165 unsigned int len,
166 struct ext4_iloc *iloc)
167{
168 struct ext4_xattr_entry *entry;
169 struct ext4_xattr_ibody_header *header;
170 int cp_len = 0;
171 struct ext4_inode *raw_inode;
172
173 if (!len)
174 return 0;
175
176 BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179 len : EXT4_MIN_INLINE_DATA_SIZE;
180
181 raw_inode = ext4_raw_inode(iloc);
182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184 len -= cp_len;
185 buffer += cp_len;
186
187 if (!len)
188 goto out;
189
190 header = IHDR(inode, raw_inode);
191 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192 EXT4_I(inode)->i_inline_off);
193 len = min_t(unsigned int, len,
194 (unsigned int)le32_to_cpu(entry->e_value_size));
195
196 memcpy(buffer,
197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198 cp_len += len;
199
200out:
201 return cp_len;
202}
203
204/*
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
209 */
210static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211 void *buffer, loff_t pos, unsigned int len)
212{
213 struct ext4_xattr_entry *entry;
214 struct ext4_xattr_ibody_header *header;
215 struct ext4_inode *raw_inode;
216 int cp_len = 0;
217
218 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
219 return;
220
221 BUG_ON(!EXT4_I(inode)->i_inline_off);
222 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
223
224 raw_inode = ext4_raw_inode(iloc);
225 buffer += pos;
226
227 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
228 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
229 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
230 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
231
232 len -= cp_len;
233 buffer += cp_len;
234 pos += cp_len;
235 }
236
237 if (!len)
238 return;
239
240 pos -= EXT4_MIN_INLINE_DATA_SIZE;
241 header = IHDR(inode, raw_inode);
242 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
243 EXT4_I(inode)->i_inline_off);
244
245 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
246 buffer, len);
247}
248
249static int ext4_create_inline_data(handle_t *handle,
250 struct inode *inode, unsigned len)
251{
252 int error;
253 void *value = NULL;
254 struct ext4_xattr_ibody_find is = {
255 .s = { .not_found = -ENODATA, },
256 };
257 struct ext4_xattr_info i = {
258 .name_index = EXT4_XATTR_INDEX_SYSTEM,
259 .name = EXT4_XATTR_SYSTEM_DATA,
260 };
261
262 error = ext4_get_inode_loc(inode, &is.iloc);
263 if (error)
264 return error;
265
266 BUFFER_TRACE(is.iloc.bh, "get_write_access");
267 error = ext4_journal_get_write_access(handle, is.iloc.bh);
268 if (error)
269 goto out;
270
271 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272 value = EXT4_ZERO_XATTR_VALUE;
273 len -= EXT4_MIN_INLINE_DATA_SIZE;
274 } else {
275 value = "";
276 len = 0;
277 }
278
279 /* Insert the the xttr entry. */
280 i.value = value;
281 i.value_len = len;
282
283 error = ext4_xattr_ibody_find(inode, &i, &is);
284 if (error)
285 goto out;
286
287 BUG_ON(!is.s.not_found);
288
289 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290 if (error) {
291 if (error == -ENOSPC)
292 ext4_clear_inode_state(inode,
293 EXT4_STATE_MAY_INLINE_DATA);
294 goto out;
295 }
296
297 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301 (void *)ext4_raw_inode(&is.iloc));
302 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305 /*
306 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
307 * get cleared
308 */
309 ext4_set_inode_flags(inode);
310 get_bh(is.iloc.bh);
311 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
312
313out:
314 brelse(is.iloc.bh);
315 return error;
316}
317
318static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
319 unsigned int len)
320{
321 int error;
322 void *value = NULL;
323 struct ext4_xattr_ibody_find is = {
324 .s = { .not_found = -ENODATA, },
325 };
326 struct ext4_xattr_info i = {
327 .name_index = EXT4_XATTR_INDEX_SYSTEM,
328 .name = EXT4_XATTR_SYSTEM_DATA,
329 };
330
331 /* If the old space is ok, write the data directly. */
332 if (len <= EXT4_I(inode)->i_inline_size)
333 return 0;
334
335 error = ext4_get_inode_loc(inode, &is.iloc);
336 if (error)
337 return error;
338
339 error = ext4_xattr_ibody_find(inode, &i, &is);
340 if (error)
341 goto out;
342
343 BUG_ON(is.s.not_found);
344
345 len -= EXT4_MIN_INLINE_DATA_SIZE;
346 value = kzalloc(len, GFP_NOFS);
347 if (!value) {
348 error = -ENOMEM;
349 goto out;
350 }
351
352 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
353 value, len);
354 if (error == -ENODATA)
355 goto out;
356
357 BUFFER_TRACE(is.iloc.bh, "get_write_access");
358 error = ext4_journal_get_write_access(handle, is.iloc.bh);
359 if (error)
360 goto out;
361
362 /* Update the xttr entry. */
363 i.value = value;
364 i.value_len = len;
365
366 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
367 if (error)
368 goto out;
369
370 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
371 (void *)ext4_raw_inode(&is.iloc));
372 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
373 le32_to_cpu(is.s.here->e_value_size);
374 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
375 get_bh(is.iloc.bh);
376 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
377
378out:
379 kfree(value);
380 brelse(is.iloc.bh);
381 return error;
382}
383
384static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
385 unsigned int len)
386{
387 int ret, size, no_expand;
388 struct ext4_inode_info *ei = EXT4_I(inode);
389
390 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
391 return -ENOSPC;
392
393 size = ext4_get_max_inline_size(inode);
394 if (size < len)
395 return -ENOSPC;
396
397 ext4_write_lock_xattr(inode, &no_expand);
398
399 if (ei->i_inline_off)
400 ret = ext4_update_inline_data(handle, inode, len);
401 else
402 ret = ext4_create_inline_data(handle, inode, len);
403
404 ext4_write_unlock_xattr(inode, &no_expand);
405 return ret;
406}
407
408static int ext4_destroy_inline_data_nolock(handle_t *handle,
409 struct inode *inode)
410{
411 struct ext4_inode_info *ei = EXT4_I(inode);
412 struct ext4_xattr_ibody_find is = {
413 .s = { .not_found = 0, },
414 };
415 struct ext4_xattr_info i = {
416 .name_index = EXT4_XATTR_INDEX_SYSTEM,
417 .name = EXT4_XATTR_SYSTEM_DATA,
418 .value = NULL,
419 .value_len = 0,
420 };
421 int error;
422
423 if (!ei->i_inline_off)
424 return 0;
425
426 error = ext4_get_inode_loc(inode, &is.iloc);
427 if (error)
428 return error;
429
430 error = ext4_xattr_ibody_find(inode, &i, &is);
431 if (error)
432 goto out;
433
434 BUFFER_TRACE(is.iloc.bh, "get_write_access");
435 error = ext4_journal_get_write_access(handle, is.iloc.bh);
436 if (error)
437 goto out;
438
439 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
440 if (error)
441 goto out;
442
443 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
444 0, EXT4_MIN_INLINE_DATA_SIZE);
445
446 if (ext4_has_feature_extents(inode->i_sb)) {
447 if (S_ISDIR(inode->i_mode) ||
448 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
449 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
450 ext4_ext_tree_init(handle, inode);
451 }
452 }
453 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
454 /*
455 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
456 * get set.
457 */
458 ext4_set_inode_flags(inode);
459
460 get_bh(is.iloc.bh);
461 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
462
463 EXT4_I(inode)->i_inline_off = 0;
464 EXT4_I(inode)->i_inline_size = 0;
465 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
466out:
467 brelse(is.iloc.bh);
468 if (error == -ENODATA)
469 error = 0;
470 return error;
471}
472
473static int ext4_read_inline_page(struct inode *inode, struct page *page)
474{
475 void *kaddr;
476 int ret = 0;
477 size_t len;
478 struct ext4_iloc iloc;
479
480 BUG_ON(!PageLocked(page));
481 BUG_ON(!ext4_has_inline_data(inode));
482 BUG_ON(page->index);
483
484 if (!EXT4_I(inode)->i_inline_off) {
485 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
486 inode->i_ino);
487 goto out;
488 }
489
490 ret = ext4_get_inode_loc(inode, &iloc);
491 if (ret)
492 goto out;
493
494 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
495 kaddr = kmap_atomic(page);
496 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
497 flush_dcache_page(page);
498 kunmap_atomic(kaddr);
499 zero_user_segment(page, len, PAGE_SIZE);
500 SetPageUptodate(page);
501 brelse(iloc.bh);
502
503out:
504 return ret;
505}
506
507int ext4_readpage_inline(struct inode *inode, struct page *page)
508{
509 int ret = 0;
510
511 down_read(&EXT4_I(inode)->xattr_sem);
512 if (!ext4_has_inline_data(inode)) {
513 up_read(&EXT4_I(inode)->xattr_sem);
514 return -EAGAIN;
515 }
516
517 /*
518 * Current inline data can only exist in the 1st page,
519 * So for all the other pages, just set them uptodate.
520 */
521 if (!page->index)
522 ret = ext4_read_inline_page(inode, page);
523 else if (!PageUptodate(page)) {
524 zero_user_segment(page, 0, PAGE_SIZE);
525 SetPageUptodate(page);
526 }
527
528 up_read(&EXT4_I(inode)->xattr_sem);
529
530 unlock_page(page);
531 return ret >= 0 ? 0 : ret;
532}
533
534static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
535 struct inode *inode,
536 unsigned flags)
537{
538 int ret, needed_blocks, no_expand;
539 handle_t *handle = NULL;
540 int retries = 0, sem_held = 0;
541 struct page *page = NULL;
542 unsigned from, to;
543 struct ext4_iloc iloc;
544
545 if (!ext4_has_inline_data(inode)) {
546 /*
547 * clear the flag so that no new write
548 * will trap here again.
549 */
550 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
551 return 0;
552 }
553
554 needed_blocks = ext4_writepage_trans_blocks(inode);
555
556 ret = ext4_get_inode_loc(inode, &iloc);
557 if (ret)
558 return ret;
559
560retry:
561 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
562 if (IS_ERR(handle)) {
563 ret = PTR_ERR(handle);
564 handle = NULL;
565 goto out;
566 }
567
568 /* We cannot recurse into the filesystem as the transaction is already
569 * started */
570 flags |= AOP_FLAG_NOFS;
571
572 page = grab_cache_page_write_begin(mapping, 0, flags);
573 if (!page) {
574 ret = -ENOMEM;
575 goto out;
576 }
577
578 ext4_write_lock_xattr(inode, &no_expand);
579 sem_held = 1;
580 /* If some one has already done this for us, just exit. */
581 if (!ext4_has_inline_data(inode)) {
582 ret = 0;
583 goto out;
584 }
585
586 from = 0;
587 to = ext4_get_inline_size(inode);
588 if (!PageUptodate(page)) {
589 ret = ext4_read_inline_page(inode, page);
590 if (ret < 0)
591 goto out;
592 }
593
594 ret = ext4_destroy_inline_data_nolock(handle, inode);
595 if (ret)
596 goto out;
597
598 if (ext4_should_dioread_nolock(inode)) {
599 ret = __block_write_begin(page, from, to,
600 ext4_get_block_unwritten);
601 } else
602 ret = __block_write_begin(page, from, to, ext4_get_block);
603
604 if (!ret && ext4_should_journal_data(inode)) {
605 ret = ext4_walk_page_buffers(handle, page_buffers(page),
606 from, to, NULL,
607 do_journal_get_write_access);
608 }
609
610 if (ret) {
611 unlock_page(page);
612 put_page(page);
613 page = NULL;
614 ext4_orphan_add(handle, inode);
615 ext4_write_unlock_xattr(inode, &no_expand);
616 sem_held = 0;
617 ext4_journal_stop(handle);
618 handle = NULL;
619 ext4_truncate_failed_write(inode);
620 /*
621 * If truncate failed early the inode might
622 * still be on the orphan list; we need to
623 * make sure the inode is removed from the
624 * orphan list in that case.
625 */
626 if (inode->i_nlink)
627 ext4_orphan_del(NULL, inode);
628 }
629
630 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
631 goto retry;
632
633 if (page)
634 block_commit_write(page, from, to);
635out:
636 if (page) {
637 unlock_page(page);
638 put_page(page);
639 }
640 if (sem_held)
641 ext4_write_unlock_xattr(inode, &no_expand);
642 if (handle)
643 ext4_journal_stop(handle);
644 brelse(iloc.bh);
645 return ret;
646}
647
648/*
649 * Try to write data in the inode.
650 * If the inode has inline data, check whether the new write can be
651 * in the inode also. If not, create the page the handle, move the data
652 * to the page make it update and let the later codes create extent for it.
653 */
654int ext4_try_to_write_inline_data(struct address_space *mapping,
655 struct inode *inode,
656 loff_t pos, unsigned len,
657 unsigned flags,
658 struct page **pagep)
659{
660 int ret;
661 handle_t *handle;
662 struct page *page;
663 struct ext4_iloc iloc;
664
665 if (pos + len > ext4_get_max_inline_size(inode))
666 goto convert;
667
668 ret = ext4_get_inode_loc(inode, &iloc);
669 if (ret)
670 return ret;
671
672 /*
673 * The possible write could happen in the inode,
674 * so try to reserve the space in inode first.
675 */
676 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
677 if (IS_ERR(handle)) {
678 ret = PTR_ERR(handle);
679 handle = NULL;
680 goto out;
681 }
682
683 ret = ext4_prepare_inline_data(handle, inode, pos + len);
684 if (ret && ret != -ENOSPC)
685 goto out;
686
687 /* We don't have space in inline inode, so convert it to extent. */
688 if (ret == -ENOSPC) {
689 ext4_journal_stop(handle);
690 brelse(iloc.bh);
691 goto convert;
692 }
693
694 flags |= AOP_FLAG_NOFS;
695
696 page = grab_cache_page_write_begin(mapping, 0, flags);
697 if (!page) {
698 ret = -ENOMEM;
699 goto out;
700 }
701
702 *pagep = page;
703 down_read(&EXT4_I(inode)->xattr_sem);
704 if (!ext4_has_inline_data(inode)) {
705 ret = 0;
706 unlock_page(page);
707 put_page(page);
708 goto out_up_read;
709 }
710
711 if (!PageUptodate(page)) {
712 ret = ext4_read_inline_page(inode, page);
713 if (ret < 0)
714 goto out_up_read;
715 }
716
717 ret = 1;
718 handle = NULL;
719out_up_read:
720 up_read(&EXT4_I(inode)->xattr_sem);
721out:
722 if (handle)
723 ext4_journal_stop(handle);
724 brelse(iloc.bh);
725 return ret;
726convert:
727 return ext4_convert_inline_data_to_extent(mapping,
728 inode, flags);
729}
730
731int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
732 unsigned copied, struct page *page)
733{
734 int ret, no_expand;
735 void *kaddr;
736 struct ext4_iloc iloc;
737
738 if (unlikely(copied < len)) {
739 if (!PageUptodate(page)) {
740 copied = 0;
741 goto out;
742 }
743 }
744
745 ret = ext4_get_inode_loc(inode, &iloc);
746 if (ret) {
747 ext4_std_error(inode->i_sb, ret);
748 copied = 0;
749 goto out;
750 }
751
752 ext4_write_lock_xattr(inode, &no_expand);
753 BUG_ON(!ext4_has_inline_data(inode));
754
755 kaddr = kmap_atomic(page);
756 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
757 kunmap_atomic(kaddr);
758 SetPageUptodate(page);
759 /* clear page dirty so that writepages wouldn't work for us. */
760 ClearPageDirty(page);
761
762 ext4_write_unlock_xattr(inode, &no_expand);
763 brelse(iloc.bh);
764out:
765 return copied;
766}
767
768struct buffer_head *
769ext4_journalled_write_inline_data(struct inode *inode,
770 unsigned len,
771 struct page *page)
772{
773 int ret, no_expand;
774 void *kaddr;
775 struct ext4_iloc iloc;
776
777 ret = ext4_get_inode_loc(inode, &iloc);
778 if (ret) {
779 ext4_std_error(inode->i_sb, ret);
780 return NULL;
781 }
782
783 ext4_write_lock_xattr(inode, &no_expand);
784 kaddr = kmap_atomic(page);
785 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
786 kunmap_atomic(kaddr);
787 ext4_write_unlock_xattr(inode, &no_expand);
788
789 return iloc.bh;
790}
791
792/*
793 * Try to make the page cache and handle ready for the inline data case.
794 * We can call this function in 2 cases:
795 * 1. The inode is created and the first write exceeds inline size. We can
796 * clear the inode state safely.
797 * 2. The inode has inline data, then we need to read the data, make it
798 * update and dirty so that ext4_da_writepages can handle it. We don't
799 * need to start the journal since the file's metatdata isn't changed now.
800 */
801static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
802 struct inode *inode,
803 unsigned flags,
804 void **fsdata)
805{
806 int ret = 0, inline_size;
807 struct page *page;
808
809 page = grab_cache_page_write_begin(mapping, 0, flags);
810 if (!page)
811 return -ENOMEM;
812
813 down_read(&EXT4_I(inode)->xattr_sem);
814 if (!ext4_has_inline_data(inode)) {
815 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
816 goto out;
817 }
818
819 inline_size = ext4_get_inline_size(inode);
820
821 if (!PageUptodate(page)) {
822 ret = ext4_read_inline_page(inode, page);
823 if (ret < 0)
824 goto out;
825 }
826
827 ret = __block_write_begin(page, 0, inline_size,
828 ext4_da_get_block_prep);
829 if (ret) {
830 up_read(&EXT4_I(inode)->xattr_sem);
831 unlock_page(page);
832 put_page(page);
833 ext4_truncate_failed_write(inode);
834 return ret;
835 }
836
837 SetPageDirty(page);
838 SetPageUptodate(page);
839 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
840 *fsdata = (void *)CONVERT_INLINE_DATA;
841
842out:
843 up_read(&EXT4_I(inode)->xattr_sem);
844 if (page) {
845 unlock_page(page);
846 put_page(page);
847 }
848 return ret;
849}
850
851/*
852 * Prepare the write for the inline data.
853 * If the the data can be written into the inode, we just read
854 * the page and make it uptodate, and start the journal.
855 * Otherwise read the page, makes it dirty so that it can be
856 * handle in writepages(the i_disksize update is left to the
857 * normal ext4_da_write_end).
858 */
859int ext4_da_write_inline_data_begin(struct address_space *mapping,
860 struct inode *inode,
861 loff_t pos, unsigned len,
862 unsigned flags,
863 struct page **pagep,
864 void **fsdata)
865{
866 int ret, inline_size;
867 handle_t *handle;
868 struct page *page;
869 struct ext4_iloc iloc;
870 int retries;
871
872 ret = ext4_get_inode_loc(inode, &iloc);
873 if (ret)
874 return ret;
875
876retry_journal:
877 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
878 if (IS_ERR(handle)) {
879 ret = PTR_ERR(handle);
880 goto out;
881 }
882
883 inline_size = ext4_get_max_inline_size(inode);
884
885 ret = -ENOSPC;
886 if (inline_size >= pos + len) {
887 ret = ext4_prepare_inline_data(handle, inode, pos + len);
888 if (ret && ret != -ENOSPC)
889 goto out_journal;
890 }
891
892 /*
893 * We cannot recurse into the filesystem as the transaction
894 * is already started.
895 */
896 flags |= AOP_FLAG_NOFS;
897
898 if (ret == -ENOSPC) {
899 ret = ext4_da_convert_inline_data_to_extent(mapping,
900 inode,
901 flags,
902 fsdata);
903 ext4_journal_stop(handle);
904 if (ret == -ENOSPC &&
905 ext4_should_retry_alloc(inode->i_sb, &retries))
906 goto retry_journal;
907 goto out;
908 }
909
910
911 page = grab_cache_page_write_begin(mapping, 0, flags);
912 if (!page) {
913 ret = -ENOMEM;
914 goto out_journal;
915 }
916
917 down_read(&EXT4_I(inode)->xattr_sem);
918 if (!ext4_has_inline_data(inode)) {
919 ret = 0;
920 goto out_release_page;
921 }
922
923 if (!PageUptodate(page)) {
924 ret = ext4_read_inline_page(inode, page);
925 if (ret < 0)
926 goto out_release_page;
927 }
928
929 up_read(&EXT4_I(inode)->xattr_sem);
930 *pagep = page;
931 brelse(iloc.bh);
932 return 1;
933out_release_page:
934 up_read(&EXT4_I(inode)->xattr_sem);
935 unlock_page(page);
936 put_page(page);
937out_journal:
938 ext4_journal_stop(handle);
939out:
940 brelse(iloc.bh);
941 return ret;
942}
943
944int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
945 unsigned len, unsigned copied,
946 struct page *page)
947{
948 int i_size_changed = 0;
949 int ret;
950
951 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
952 if (ret < 0) {
953 unlock_page(page);
954 put_page(page);
955 return ret;
956 }
957 copied = ret;
958
959 /*
960 * No need to use i_size_read() here, the i_size
961 * cannot change under us because we hold i_mutex.
962 *
963 * But it's important to update i_size while still holding page lock:
964 * page writeout could otherwise come in and zero beyond i_size.
965 */
966 if (pos+copied > inode->i_size) {
967 i_size_write(inode, pos+copied);
968 i_size_changed = 1;
969 }
970 unlock_page(page);
971 put_page(page);
972
973 /*
974 * Don't mark the inode dirty under page lock. First, it unnecessarily
975 * makes the holding time of page lock longer. Second, it forces lock
976 * ordering of page lock and transaction start for journaling
977 * filesystems.
978 */
979 if (i_size_changed)
980 mark_inode_dirty(inode);
981
982 return copied;
983}
984
985#ifdef INLINE_DIR_DEBUG
986void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
987 void *inline_start, int inline_size)
988{
989 int offset;
990 unsigned short de_len;
991 struct ext4_dir_entry_2 *de = inline_start;
992 void *dlimit = inline_start + inline_size;
993
994 trace_printk("inode %lu\n", dir->i_ino);
995 offset = 0;
996 while ((void *)de < dlimit) {
997 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
998 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
999 offset, de_len, de->name_len, de->name,
1000 de->name_len, le32_to_cpu(de->inode));
1001 if (ext4_check_dir_entry(dir, NULL, de, bh,
1002 inline_start, inline_size, offset))
1003 BUG();
1004
1005 offset += de_len;
1006 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1007 }
1008}
1009#else
1010#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1011#endif
1012
1013/*
1014 * Add a new entry into a inline dir.
1015 * It will return -ENOSPC if no space is available, and -EIO
1016 * and -EEXIST if directory entry already exists.
1017 */
1018static int ext4_add_dirent_to_inline(handle_t *handle,
1019 struct ext4_filename *fname,
1020 struct inode *dir,
1021 struct inode *inode,
1022 struct ext4_iloc *iloc,
1023 void *inline_start, int inline_size)
1024{
1025 int err;
1026 struct ext4_dir_entry_2 *de;
1027
1028 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1029 inline_size, fname, &de);
1030 if (err)
1031 return err;
1032
1033 BUFFER_TRACE(iloc->bh, "get_write_access");
1034 err = ext4_journal_get_write_access(handle, iloc->bh);
1035 if (err)
1036 return err;
1037 ext4_insert_dentry(inode, de, inline_size, fname);
1038
1039 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1040
1041 /*
1042 * XXX shouldn't update any times until successful
1043 * completion of syscall, but too many callers depend
1044 * on this.
1045 *
1046 * XXX similarly, too many callers depend on
1047 * ext4_new_inode() setting the times, but error
1048 * recovery deletes the inode, so the worst that can
1049 * happen is that the times are slightly out of date
1050 * and/or different from the directory change time.
1051 */
1052 dir->i_mtime = dir->i_ctime = current_time(dir);
1053 ext4_update_dx_flag(dir);
1054 dir->i_version++;
1055 return 1;
1056}
1057
1058static void *ext4_get_inline_xattr_pos(struct inode *inode,
1059 struct ext4_iloc *iloc)
1060{
1061 struct ext4_xattr_entry *entry;
1062 struct ext4_xattr_ibody_header *header;
1063
1064 BUG_ON(!EXT4_I(inode)->i_inline_off);
1065
1066 header = IHDR(inode, ext4_raw_inode(iloc));
1067 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1068 EXT4_I(inode)->i_inline_off);
1069
1070 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1071}
1072
1073/* Set the final de to cover the whole block. */
1074static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1075{
1076 struct ext4_dir_entry_2 *de, *prev_de;
1077 void *limit;
1078 int de_len;
1079
1080 de = (struct ext4_dir_entry_2 *)de_buf;
1081 if (old_size) {
1082 limit = de_buf + old_size;
1083 do {
1084 prev_de = de;
1085 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1086 de_buf += de_len;
1087 de = (struct ext4_dir_entry_2 *)de_buf;
1088 } while (de_buf < limit);
1089
1090 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1091 old_size, new_size);
1092 } else {
1093 /* this is just created, so create an empty entry. */
1094 de->inode = 0;
1095 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1096 }
1097}
1098
1099static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1100 struct ext4_iloc *iloc)
1101{
1102 int ret;
1103 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1104 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1105
1106 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1107 return -ENOSPC;
1108
1109 ret = ext4_update_inline_data(handle, dir,
1110 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1111 if (ret)
1112 return ret;
1113
1114 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1115 EXT4_I(dir)->i_inline_size -
1116 EXT4_MIN_INLINE_DATA_SIZE);
1117 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1118 return 0;
1119}
1120
1121static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1122 struct ext4_iloc *iloc,
1123 void *buf, int inline_size)
1124{
1125 ext4_create_inline_data(handle, inode, inline_size);
1126 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1127 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1128}
1129
1130static int ext4_finish_convert_inline_dir(handle_t *handle,
1131 struct inode *inode,
1132 struct buffer_head *dir_block,
1133 void *buf,
1134 int inline_size)
1135{
1136 int err, csum_size = 0, header_size = 0;
1137 struct ext4_dir_entry_2 *de;
1138 struct ext4_dir_entry_tail *t;
1139 void *target = dir_block->b_data;
1140
1141 /*
1142 * First create "." and ".." and then copy the dir information
1143 * back to the block.
1144 */
1145 de = (struct ext4_dir_entry_2 *)target;
1146 de = ext4_init_dot_dotdot(inode, de,
1147 inode->i_sb->s_blocksize, csum_size,
1148 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1149 header_size = (void *)de - target;
1150
1151 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1152 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1153
1154 if (ext4_has_metadata_csum(inode->i_sb))
1155 csum_size = sizeof(struct ext4_dir_entry_tail);
1156
1157 inode->i_size = inode->i_sb->s_blocksize;
1158 i_size_write(inode, inode->i_sb->s_blocksize);
1159 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1160 ext4_update_final_de(dir_block->b_data,
1161 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1162 inode->i_sb->s_blocksize - csum_size);
1163
1164 if (csum_size) {
1165 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1166 inode->i_sb->s_blocksize);
1167 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1168 }
1169 set_buffer_uptodate(dir_block);
1170 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1171 if (err)
1172 return err;
1173 set_buffer_verified(dir_block);
1174 return ext4_mark_inode_dirty(handle, inode);
1175}
1176
1177static int ext4_convert_inline_data_nolock(handle_t *handle,
1178 struct inode *inode,
1179 struct ext4_iloc *iloc)
1180{
1181 int error;
1182 void *buf = NULL;
1183 struct buffer_head *data_bh = NULL;
1184 struct ext4_map_blocks map;
1185 int inline_size;
1186
1187 inline_size = ext4_get_inline_size(inode);
1188 buf = kmalloc(inline_size, GFP_NOFS);
1189 if (!buf) {
1190 error = -ENOMEM;
1191 goto out;
1192 }
1193
1194 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1195 if (error < 0)
1196 goto out;
1197
1198 /*
1199 * Make sure the inline directory entries pass checks before we try to
1200 * convert them, so that we avoid touching stuff that needs fsck.
1201 */
1202 if (S_ISDIR(inode->i_mode)) {
1203 error = ext4_check_all_de(inode, iloc->bh,
1204 buf + EXT4_INLINE_DOTDOT_SIZE,
1205 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1206 if (error)
1207 goto out;
1208 }
1209
1210 error = ext4_destroy_inline_data_nolock(handle, inode);
1211 if (error)
1212 goto out;
1213
1214 map.m_lblk = 0;
1215 map.m_len = 1;
1216 map.m_flags = 0;
1217 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1218 if (error < 0)
1219 goto out_restore;
1220 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1221 error = -EIO;
1222 goto out_restore;
1223 }
1224
1225 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1226 if (!data_bh) {
1227 error = -ENOMEM;
1228 goto out_restore;
1229 }
1230
1231 lock_buffer(data_bh);
1232 error = ext4_journal_get_create_access(handle, data_bh);
1233 if (error) {
1234 unlock_buffer(data_bh);
1235 error = -EIO;
1236 goto out_restore;
1237 }
1238 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1239
1240 if (!S_ISDIR(inode->i_mode)) {
1241 memcpy(data_bh->b_data, buf, inline_size);
1242 set_buffer_uptodate(data_bh);
1243 error = ext4_handle_dirty_metadata(handle,
1244 inode, data_bh);
1245 } else {
1246 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1247 buf, inline_size);
1248 }
1249
1250 unlock_buffer(data_bh);
1251out_restore:
1252 if (error)
1253 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1254
1255out:
1256 brelse(data_bh);
1257 kfree(buf);
1258 return error;
1259}
1260
1261/*
1262 * Try to add the new entry to the inline data.
1263 * If succeeds, return 0. If not, extended the inline dir and copied data to
1264 * the new created block.
1265 */
1266int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1267 struct inode *dir, struct inode *inode)
1268{
1269 int ret, inline_size, no_expand;
1270 void *inline_start;
1271 struct ext4_iloc iloc;
1272
1273 ret = ext4_get_inode_loc(dir, &iloc);
1274 if (ret)
1275 return ret;
1276
1277 ext4_write_lock_xattr(dir, &no_expand);
1278 if (!ext4_has_inline_data(dir))
1279 goto out;
1280
1281 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1282 EXT4_INLINE_DOTDOT_SIZE;
1283 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1284
1285 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1286 inline_start, inline_size);
1287 if (ret != -ENOSPC)
1288 goto out;
1289
1290 /* check whether it can be inserted to inline xattr space. */
1291 inline_size = EXT4_I(dir)->i_inline_size -
1292 EXT4_MIN_INLINE_DATA_SIZE;
1293 if (!inline_size) {
1294 /* Try to use the xattr space.*/
1295 ret = ext4_update_inline_dir(handle, dir, &iloc);
1296 if (ret && ret != -ENOSPC)
1297 goto out;
1298
1299 inline_size = EXT4_I(dir)->i_inline_size -
1300 EXT4_MIN_INLINE_DATA_SIZE;
1301 }
1302
1303 if (inline_size) {
1304 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1305
1306 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1307 inode, &iloc, inline_start,
1308 inline_size);
1309
1310 if (ret != -ENOSPC)
1311 goto out;
1312 }
1313
1314 /*
1315 * The inline space is filled up, so create a new block for it.
1316 * As the extent tree will be created, we have to save the inline
1317 * dir first.
1318 */
1319 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1320
1321out:
1322 ext4_write_unlock_xattr(dir, &no_expand);
1323 ext4_mark_inode_dirty(handle, dir);
1324 brelse(iloc.bh);
1325 return ret;
1326}
1327
1328/*
1329 * This function fills a red-black tree with information from an
1330 * inlined dir. It returns the number directory entries loaded
1331 * into the tree. If there is an error it is returned in err.
1332 */
1333int htree_inlinedir_to_tree(struct file *dir_file,
1334 struct inode *dir, ext4_lblk_t block,
1335 struct dx_hash_info *hinfo,
1336 __u32 start_hash, __u32 start_minor_hash,
1337 int *has_inline_data)
1338{
1339 int err = 0, count = 0;
1340 unsigned int parent_ino;
1341 int pos;
1342 struct ext4_dir_entry_2 *de;
1343 struct inode *inode = file_inode(dir_file);
1344 int ret, inline_size = 0;
1345 struct ext4_iloc iloc;
1346 void *dir_buf = NULL;
1347 struct ext4_dir_entry_2 fake;
1348 struct fscrypt_str tmp_str;
1349
1350 ret = ext4_get_inode_loc(inode, &iloc);
1351 if (ret)
1352 return ret;
1353
1354 down_read(&EXT4_I(inode)->xattr_sem);
1355 if (!ext4_has_inline_data(inode)) {
1356 up_read(&EXT4_I(inode)->xattr_sem);
1357 *has_inline_data = 0;
1358 goto out;
1359 }
1360
1361 inline_size = ext4_get_inline_size(inode);
1362 dir_buf = kmalloc(inline_size, GFP_NOFS);
1363 if (!dir_buf) {
1364 ret = -ENOMEM;
1365 up_read(&EXT4_I(inode)->xattr_sem);
1366 goto out;
1367 }
1368
1369 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1370 up_read(&EXT4_I(inode)->xattr_sem);
1371 if (ret < 0)
1372 goto out;
1373
1374 pos = 0;
1375 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1376 while (pos < inline_size) {
1377 /*
1378 * As inlined dir doesn't store any information about '.' and
1379 * only the inode number of '..' is stored, we have to handle
1380 * them differently.
1381 */
1382 if (pos == 0) {
1383 fake.inode = cpu_to_le32(inode->i_ino);
1384 fake.name_len = 1;
1385 strcpy(fake.name, ".");
1386 fake.rec_len = ext4_rec_len_to_disk(
1387 EXT4_DIR_REC_LEN(fake.name_len),
1388 inline_size);
1389 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1390 de = &fake;
1391 pos = EXT4_INLINE_DOTDOT_OFFSET;
1392 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1393 fake.inode = cpu_to_le32(parent_ino);
1394 fake.name_len = 2;
1395 strcpy(fake.name, "..");
1396 fake.rec_len = ext4_rec_len_to_disk(
1397 EXT4_DIR_REC_LEN(fake.name_len),
1398 inline_size);
1399 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1400 de = &fake;
1401 pos = EXT4_INLINE_DOTDOT_SIZE;
1402 } else {
1403 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1404 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1405 if (ext4_check_dir_entry(inode, dir_file, de,
1406 iloc.bh, dir_buf,
1407 inline_size, pos)) {
1408 ret = count;
1409 goto out;
1410 }
1411 }
1412
1413 ext4fs_dirhash(de->name, de->name_len, hinfo);
1414 if ((hinfo->hash < start_hash) ||
1415 ((hinfo->hash == start_hash) &&
1416 (hinfo->minor_hash < start_minor_hash)))
1417 continue;
1418 if (de->inode == 0)
1419 continue;
1420 tmp_str.name = de->name;
1421 tmp_str.len = de->name_len;
1422 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1423 hinfo->minor_hash, de, &tmp_str);
1424 if (err) {
1425 count = err;
1426 goto out;
1427 }
1428 count++;
1429 }
1430 ret = count;
1431out:
1432 kfree(dir_buf);
1433 brelse(iloc.bh);
1434 return ret;
1435}
1436
1437/*
1438 * So this function is called when the volume is mkfsed with
1439 * dir_index disabled. In order to keep f_pos persistent
1440 * after we convert from an inlined dir to a blocked based,
1441 * we just pretend that we are a normal dir and return the
1442 * offset as if '.' and '..' really take place.
1443 *
1444 */
1445int ext4_read_inline_dir(struct file *file,
1446 struct dir_context *ctx,
1447 int *has_inline_data)
1448{
1449 unsigned int offset, parent_ino;
1450 int i;
1451 struct ext4_dir_entry_2 *de;
1452 struct super_block *sb;
1453 struct inode *inode = file_inode(file);
1454 int ret, inline_size = 0;
1455 struct ext4_iloc iloc;
1456 void *dir_buf = NULL;
1457 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1458
1459 ret = ext4_get_inode_loc(inode, &iloc);
1460 if (ret)
1461 return ret;
1462
1463 down_read(&EXT4_I(inode)->xattr_sem);
1464 if (!ext4_has_inline_data(inode)) {
1465 up_read(&EXT4_I(inode)->xattr_sem);
1466 *has_inline_data = 0;
1467 goto out;
1468 }
1469
1470 inline_size = ext4_get_inline_size(inode);
1471 dir_buf = kmalloc(inline_size, GFP_NOFS);
1472 if (!dir_buf) {
1473 ret = -ENOMEM;
1474 up_read(&EXT4_I(inode)->xattr_sem);
1475 goto out;
1476 }
1477
1478 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1479 up_read(&EXT4_I(inode)->xattr_sem);
1480 if (ret < 0)
1481 goto out;
1482
1483 ret = 0;
1484 sb = inode->i_sb;
1485 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1486 offset = ctx->pos;
1487
1488 /*
1489 * dotdot_offset and dotdot_size is the real offset and
1490 * size for ".." and "." if the dir is block based while
1491 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1492 * So we will use extra_offset and extra_size to indicate them
1493 * during the inline dir iteration.
1494 */
1495 dotdot_offset = EXT4_DIR_REC_LEN(1);
1496 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1497 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1498 extra_size = extra_offset + inline_size;
1499
1500 /*
1501 * If the version has changed since the last call to
1502 * readdir(2), then we might be pointing to an invalid
1503 * dirent right now. Scan from the start of the inline
1504 * dir to make sure.
1505 */
1506 if (file->f_version != inode->i_version) {
1507 for (i = 0; i < extra_size && i < offset;) {
1508 /*
1509 * "." is with offset 0 and
1510 * ".." is dotdot_offset.
1511 */
1512 if (!i) {
1513 i = dotdot_offset;
1514 continue;
1515 } else if (i == dotdot_offset) {
1516 i = dotdot_size;
1517 continue;
1518 }
1519 /* for other entry, the real offset in
1520 * the buf has to be tuned accordingly.
1521 */
1522 de = (struct ext4_dir_entry_2 *)
1523 (dir_buf + i - extra_offset);
1524 /* It's too expensive to do a full
1525 * dirent test each time round this
1526 * loop, but we do have to test at
1527 * least that it is non-zero. A
1528 * failure will be detected in the
1529 * dirent test below. */
1530 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1531 < EXT4_DIR_REC_LEN(1))
1532 break;
1533 i += ext4_rec_len_from_disk(de->rec_len,
1534 extra_size);
1535 }
1536 offset = i;
1537 ctx->pos = offset;
1538 file->f_version = inode->i_version;
1539 }
1540
1541 while (ctx->pos < extra_size) {
1542 if (ctx->pos == 0) {
1543 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1544 goto out;
1545 ctx->pos = dotdot_offset;
1546 continue;
1547 }
1548
1549 if (ctx->pos == dotdot_offset) {
1550 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1551 goto out;
1552 ctx->pos = dotdot_size;
1553 continue;
1554 }
1555
1556 de = (struct ext4_dir_entry_2 *)
1557 (dir_buf + ctx->pos - extra_offset);
1558 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1559 extra_size, ctx->pos))
1560 goto out;
1561 if (le32_to_cpu(de->inode)) {
1562 if (!dir_emit(ctx, de->name, de->name_len,
1563 le32_to_cpu(de->inode),
1564 get_dtype(sb, de->file_type)))
1565 goto out;
1566 }
1567 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1568 }
1569out:
1570 kfree(dir_buf);
1571 brelse(iloc.bh);
1572 return ret;
1573}
1574
1575struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1576 struct ext4_dir_entry_2 **parent_de,
1577 int *retval)
1578{
1579 struct ext4_iloc iloc;
1580
1581 *retval = ext4_get_inode_loc(inode, &iloc);
1582 if (*retval)
1583 return NULL;
1584
1585 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1586
1587 return iloc.bh;
1588}
1589
1590/*
1591 * Try to create the inline data for the new dir.
1592 * If it succeeds, return 0, otherwise return the error.
1593 * In case of ENOSPC, the caller should create the normal disk layout dir.
1594 */
1595int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1596 struct inode *inode)
1597{
1598 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1599 struct ext4_iloc iloc;
1600 struct ext4_dir_entry_2 *de;
1601
1602 ret = ext4_get_inode_loc(inode, &iloc);
1603 if (ret)
1604 return ret;
1605
1606 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1607 if (ret)
1608 goto out;
1609
1610 /*
1611 * For inline dir, we only save the inode information for the ".."
1612 * and create a fake dentry to cover the left space.
1613 */
1614 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1615 de->inode = cpu_to_le32(parent->i_ino);
1616 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1617 de->inode = 0;
1618 de->rec_len = ext4_rec_len_to_disk(
1619 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1620 inline_size);
1621 set_nlink(inode, 2);
1622 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1623out:
1624 brelse(iloc.bh);
1625 return ret;
1626}
1627
1628struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1629 struct ext4_filename *fname,
1630 struct ext4_dir_entry_2 **res_dir,
1631 int *has_inline_data)
1632{
1633 int ret;
1634 struct ext4_iloc iloc;
1635 void *inline_start;
1636 int inline_size;
1637
1638 if (ext4_get_inode_loc(dir, &iloc))
1639 return NULL;
1640
1641 down_read(&EXT4_I(dir)->xattr_sem);
1642 if (!ext4_has_inline_data(dir)) {
1643 *has_inline_data = 0;
1644 goto out;
1645 }
1646
1647 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1648 EXT4_INLINE_DOTDOT_SIZE;
1649 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1650 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1651 dir, fname, 0, res_dir);
1652 if (ret == 1)
1653 goto out_find;
1654 if (ret < 0)
1655 goto out;
1656
1657 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1658 goto out;
1659
1660 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1661 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1662
1663 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1664 dir, fname, 0, res_dir);
1665 if (ret == 1)
1666 goto out_find;
1667
1668out:
1669 brelse(iloc.bh);
1670 iloc.bh = NULL;
1671out_find:
1672 up_read(&EXT4_I(dir)->xattr_sem);
1673 return iloc.bh;
1674}
1675
1676int ext4_delete_inline_entry(handle_t *handle,
1677 struct inode *dir,
1678 struct ext4_dir_entry_2 *de_del,
1679 struct buffer_head *bh,
1680 int *has_inline_data)
1681{
1682 int err, inline_size, no_expand;
1683 struct ext4_iloc iloc;
1684 void *inline_start;
1685
1686 err = ext4_get_inode_loc(dir, &iloc);
1687 if (err)
1688 return err;
1689
1690 ext4_write_lock_xattr(dir, &no_expand);
1691 if (!ext4_has_inline_data(dir)) {
1692 *has_inline_data = 0;
1693 goto out;
1694 }
1695
1696 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1697 EXT4_MIN_INLINE_DATA_SIZE) {
1698 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1699 EXT4_INLINE_DOTDOT_SIZE;
1700 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1701 EXT4_INLINE_DOTDOT_SIZE;
1702 } else {
1703 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1704 inline_size = ext4_get_inline_size(dir) -
1705 EXT4_MIN_INLINE_DATA_SIZE;
1706 }
1707
1708 BUFFER_TRACE(bh, "get_write_access");
1709 err = ext4_journal_get_write_access(handle, bh);
1710 if (err)
1711 goto out;
1712
1713 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1714 inline_start, inline_size, 0);
1715 if (err)
1716 goto out;
1717
1718 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1719out:
1720 ext4_write_unlock_xattr(dir, &no_expand);
1721 if (likely(err == 0))
1722 err = ext4_mark_inode_dirty(handle, dir);
1723 brelse(iloc.bh);
1724 if (err != -ENOENT)
1725 ext4_std_error(dir->i_sb, err);
1726 return err;
1727}
1728
1729/*
1730 * Get the inline dentry at offset.
1731 */
1732static inline struct ext4_dir_entry_2 *
1733ext4_get_inline_entry(struct inode *inode,
1734 struct ext4_iloc *iloc,
1735 unsigned int offset,
1736 void **inline_start,
1737 int *inline_size)
1738{
1739 void *inline_pos;
1740
1741 BUG_ON(offset > ext4_get_inline_size(inode));
1742
1743 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1744 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1745 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1746 } else {
1747 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1748 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1749 *inline_size = ext4_get_inline_size(inode) -
1750 EXT4_MIN_INLINE_DATA_SIZE;
1751 }
1752
1753 if (inline_start)
1754 *inline_start = inline_pos;
1755 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1756}
1757
1758bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1759{
1760 int err, inline_size;
1761 struct ext4_iloc iloc;
1762 void *inline_pos;
1763 unsigned int offset;
1764 struct ext4_dir_entry_2 *de;
1765 bool ret = true;
1766
1767 err = ext4_get_inode_loc(dir, &iloc);
1768 if (err) {
1769 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1770 err, dir->i_ino);
1771 return true;
1772 }
1773
1774 down_read(&EXT4_I(dir)->xattr_sem);
1775 if (!ext4_has_inline_data(dir)) {
1776 *has_inline_data = 0;
1777 goto out;
1778 }
1779
1780 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1781 if (!le32_to_cpu(de->inode)) {
1782 ext4_warning(dir->i_sb,
1783 "bad inline directory (dir #%lu) - no `..'",
1784 dir->i_ino);
1785 ret = true;
1786 goto out;
1787 }
1788
1789 offset = EXT4_INLINE_DOTDOT_SIZE;
1790 while (offset < dir->i_size) {
1791 de = ext4_get_inline_entry(dir, &iloc, offset,
1792 &inline_pos, &inline_size);
1793 if (ext4_check_dir_entry(dir, NULL, de,
1794 iloc.bh, inline_pos,
1795 inline_size, offset)) {
1796 ext4_warning(dir->i_sb,
1797 "bad inline directory (dir #%lu) - "
1798 "inode %u, rec_len %u, name_len %d"
1799 "inline size %d",
1800 dir->i_ino, le32_to_cpu(de->inode),
1801 le16_to_cpu(de->rec_len), de->name_len,
1802 inline_size);
1803 ret = true;
1804 goto out;
1805 }
1806 if (le32_to_cpu(de->inode)) {
1807 ret = false;
1808 goto out;
1809 }
1810 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1811 }
1812
1813out:
1814 up_read(&EXT4_I(dir)->xattr_sem);
1815 brelse(iloc.bh);
1816 return ret;
1817}
1818
1819int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1820{
1821 int ret, no_expand;
1822
1823 ext4_write_lock_xattr(inode, &no_expand);
1824 ret = ext4_destroy_inline_data_nolock(handle, inode);
1825 ext4_write_unlock_xattr(inode, &no_expand);
1826
1827 return ret;
1828}
1829
1830int ext4_inline_data_fiemap(struct inode *inode,
1831 struct fiemap_extent_info *fieinfo,
1832 int *has_inline, __u64 start, __u64 len)
1833{
1834 __u64 physical = 0;
1835 __u64 inline_len;
1836 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1837 FIEMAP_EXTENT_LAST;
1838 int error = 0;
1839 struct ext4_iloc iloc;
1840
1841 down_read(&EXT4_I(inode)->xattr_sem);
1842 if (!ext4_has_inline_data(inode)) {
1843 *has_inline = 0;
1844 goto out;
1845 }
1846 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1847 i_size_read(inode));
1848 if (start >= inline_len)
1849 goto out;
1850 if (start + len < inline_len)
1851 inline_len = start + len;
1852 inline_len -= start;
1853
1854 error = ext4_get_inode_loc(inode, &iloc);
1855 if (error)
1856 goto out;
1857
1858 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1859 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1860 physical += offsetof(struct ext4_inode, i_block);
1861
1862 if (physical)
1863 error = fiemap_fill_next_extent(fieinfo, start, physical,
1864 inline_len, flags);
1865 brelse(iloc.bh);
1866out:
1867 up_read(&EXT4_I(inode)->xattr_sem);
1868 return (error < 0 ? error : 0);
1869}
1870
1871/*
1872 * Called during xattr set, and if we can sparse space 'needed',
1873 * just create the extent tree evict the data to the outer block.
1874 *
1875 * We use jbd2 instead of page cache to move data to the 1st block
1876 * so that the whole transaction can be committed as a whole and
1877 * the data isn't lost because of the delayed page cache write.
1878 */
1879int ext4_try_to_evict_inline_data(handle_t *handle,
1880 struct inode *inode,
1881 int needed)
1882{
1883 int error;
1884 struct ext4_xattr_entry *entry;
1885 struct ext4_inode *raw_inode;
1886 struct ext4_iloc iloc;
1887
1888 error = ext4_get_inode_loc(inode, &iloc);
1889 if (error)
1890 return error;
1891
1892 raw_inode = ext4_raw_inode(&iloc);
1893 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1894 EXT4_I(inode)->i_inline_off);
1895 if (EXT4_XATTR_LEN(entry->e_name_len) +
1896 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1897 error = -ENOSPC;
1898 goto out;
1899 }
1900
1901 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1902out:
1903 brelse(iloc.bh);
1904 return error;
1905}
1906
1907int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1908{
1909 handle_t *handle;
1910 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1911 size_t i_size;
1912 void *value = NULL;
1913 struct ext4_xattr_ibody_find is = {
1914 .s = { .not_found = -ENODATA, },
1915 };
1916 struct ext4_xattr_info i = {
1917 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1918 .name = EXT4_XATTR_SYSTEM_DATA,
1919 };
1920
1921
1922 needed_blocks = ext4_writepage_trans_blocks(inode);
1923 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1924 if (IS_ERR(handle))
1925 return PTR_ERR(handle);
1926
1927 ext4_write_lock_xattr(inode, &no_expand);
1928 if (!ext4_has_inline_data(inode)) {
1929 *has_inline = 0;
1930 ext4_journal_stop(handle);
1931 return 0;
1932 }
1933
1934 if ((err = ext4_orphan_add(handle, inode)) != 0)
1935 goto out;
1936
1937 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1938 goto out;
1939
1940 down_write(&EXT4_I(inode)->i_data_sem);
1941 i_size = inode->i_size;
1942 inline_size = ext4_get_inline_size(inode);
1943 EXT4_I(inode)->i_disksize = i_size;
1944
1945 if (i_size < inline_size) {
1946 /* Clear the content in the xattr space. */
1947 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1948 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1949 goto out_error;
1950
1951 BUG_ON(is.s.not_found);
1952
1953 value_len = le32_to_cpu(is.s.here->e_value_size);
1954 value = kmalloc(value_len, GFP_NOFS);
1955 if (!value) {
1956 err = -ENOMEM;
1957 goto out_error;
1958 }
1959
1960 err = ext4_xattr_ibody_get(inode, i.name_index,
1961 i.name, value, value_len);
1962 if (err <= 0)
1963 goto out_error;
1964
1965 i.value = value;
1966 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1967 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1968 err = ext4_xattr_ibody_inline_set(handle, inode,
1969 &i, &is);
1970 if (err)
1971 goto out_error;
1972 }
1973
1974 /* Clear the content within i_blocks. */
1975 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1976 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1977 memset(p + i_size, 0,
1978 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1979 }
1980
1981 EXT4_I(inode)->i_inline_size = i_size <
1982 EXT4_MIN_INLINE_DATA_SIZE ?
1983 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1984 }
1985
1986out_error:
1987 up_write(&EXT4_I(inode)->i_data_sem);
1988out:
1989 brelse(is.iloc.bh);
1990 ext4_write_unlock_xattr(inode, &no_expand);
1991 kfree(value);
1992 if (inode->i_nlink)
1993 ext4_orphan_del(handle, inode);
1994
1995 if (err == 0) {
1996 inode->i_mtime = inode->i_ctime = current_time(inode);
1997 err = ext4_mark_inode_dirty(handle, inode);
1998 if (IS_SYNC(inode))
1999 ext4_handle_sync(handle);
2000 }
2001 ext4_journal_stop(handle);
2002 return err;
2003}
2004
2005int ext4_convert_inline_data(struct inode *inode)
2006{
2007 int error, needed_blocks, no_expand;
2008 handle_t *handle;
2009 struct ext4_iloc iloc;
2010
2011 if (!ext4_has_inline_data(inode)) {
2012 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2013 return 0;
2014 }
2015
2016 needed_blocks = ext4_writepage_trans_blocks(inode);
2017
2018 iloc.bh = NULL;
2019 error = ext4_get_inode_loc(inode, &iloc);
2020 if (error)
2021 return error;
2022
2023 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2024 if (IS_ERR(handle)) {
2025 error = PTR_ERR(handle);
2026 goto out_free;
2027 }
2028
2029 ext4_write_lock_xattr(inode, &no_expand);
2030 if (ext4_has_inline_data(inode))
2031 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2032 ext4_write_unlock_xattr(inode, &no_expand);
2033 ext4_journal_stop(handle);
2034out_free:
2035 brelse(iloc.bh);
2036 return error;
2037}