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/namei.h>
10#include <linux/iversion.h>
11#include <linux/sched/mm.h>
12
13#include "ext4_jbd2.h"
14#include "ext4.h"
15#include "xattr.h"
16#include "truncate.h"
17
18#define EXT4_XATTR_SYSTEM_DATA "data"
19#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
20#define EXT4_INLINE_DOTDOT_OFFSET 2
21#define EXT4_INLINE_DOTDOT_SIZE 4
22
23static int ext4_get_inline_size(struct inode *inode)
24{
25 if (EXT4_I(inode)->i_inline_off)
26 return EXT4_I(inode)->i_inline_size;
27
28 return 0;
29}
30
31static int get_max_inline_xattr_value_size(struct inode *inode,
32 struct ext4_iloc *iloc)
33{
34 struct ext4_xattr_ibody_header *header;
35 struct ext4_xattr_entry *entry;
36 struct ext4_inode *raw_inode;
37 int free, min_offs;
38
39 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
40 return 0;
41
42 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
43 EXT4_GOOD_OLD_INODE_SIZE -
44 EXT4_I(inode)->i_extra_isize -
45 sizeof(struct ext4_xattr_ibody_header);
46
47 /*
48 * We need to subtract another sizeof(__u32) since an in-inode xattr
49 * needs an empty 4 bytes to indicate the gap between the xattr entry
50 * and the name/value pair.
51 */
52 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
53 return EXT4_XATTR_SIZE(min_offs -
54 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
55 EXT4_XATTR_ROUND - sizeof(__u32));
56
57 raw_inode = ext4_raw_inode(iloc);
58 header = IHDR(inode, raw_inode);
59 entry = IFIRST(header);
60
61 /* Compute min_offs. */
62 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
63 if (!entry->e_value_inum && entry->e_value_size) {
64 size_t offs = le16_to_cpu(entry->e_value_offs);
65 if (offs < min_offs)
66 min_offs = offs;
67 }
68 }
69 free = min_offs -
70 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
71
72 if (EXT4_I(inode)->i_inline_off) {
73 entry = (struct ext4_xattr_entry *)
74 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
75
76 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
77 goto out;
78 }
79
80 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
81
82 if (free > EXT4_XATTR_ROUND)
83 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
84 else
85 free = 0;
86
87out:
88 return free;
89}
90
91/*
92 * Get the maximum size we now can store in an inode.
93 * If we can't find the space for a xattr entry, don't use the space
94 * of the extents since we have no space to indicate the inline data.
95 */
96int ext4_get_max_inline_size(struct inode *inode)
97{
98 int error, max_inline_size;
99 struct ext4_iloc iloc;
100
101 if (EXT4_I(inode)->i_extra_isize == 0)
102 return 0;
103
104 error = ext4_get_inode_loc(inode, &iloc);
105 if (error) {
106 ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
107 "can't get inode location %lu",
108 inode->i_ino);
109 return 0;
110 }
111
112 down_read(&EXT4_I(inode)->xattr_sem);
113 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
114 up_read(&EXT4_I(inode)->xattr_sem);
115
116 brelse(iloc.bh);
117
118 if (!max_inline_size)
119 return 0;
120
121 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
122}
123
124/*
125 * this function does not take xattr_sem, which is OK because it is
126 * currently only used in a code path coming form ext4_iget, before
127 * the new inode has been unlocked
128 */
129int ext4_find_inline_data_nolock(struct inode *inode)
130{
131 struct ext4_xattr_ibody_find is = {
132 .s = { .not_found = -ENODATA, },
133 };
134 struct ext4_xattr_info i = {
135 .name_index = EXT4_XATTR_INDEX_SYSTEM,
136 .name = EXT4_XATTR_SYSTEM_DATA,
137 };
138 int error;
139
140 if (EXT4_I(inode)->i_extra_isize == 0)
141 return 0;
142
143 error = ext4_get_inode_loc(inode, &is.iloc);
144 if (error)
145 return error;
146
147 error = ext4_xattr_ibody_find(inode, &i, &is);
148 if (error)
149 goto out;
150
151 if (!is.s.not_found) {
152 if (is.s.here->e_value_inum) {
153 EXT4_ERROR_INODE(inode, "inline data xattr refers "
154 "to an external xattr inode");
155 error = -EFSCORRUPTED;
156 goto out;
157 }
158 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
159 (void *)ext4_raw_inode(&is.iloc));
160 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
161 le32_to_cpu(is.s.here->e_value_size);
162 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
163 }
164out:
165 brelse(is.iloc.bh);
166 return error;
167}
168
169static int ext4_read_inline_data(struct inode *inode, void *buffer,
170 unsigned int len,
171 struct ext4_iloc *iloc)
172{
173 struct ext4_xattr_entry *entry;
174 struct ext4_xattr_ibody_header *header;
175 int cp_len = 0;
176 struct ext4_inode *raw_inode;
177
178 if (!len)
179 return 0;
180
181 BUG_ON(len > EXT4_I(inode)->i_inline_size);
182
183 cp_len = min_t(unsigned int, 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_set.
212 * That saves us one memcpy.
213 */
214static void 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 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
223 return;
224
225 BUG_ON(!EXT4_I(inode)->i_inline_off);
226 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
227
228 raw_inode = ext4_raw_inode(iloc);
229 buffer += pos;
230
231 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
232 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
233 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
234 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
235
236 len -= cp_len;
237 buffer += cp_len;
238 pos += cp_len;
239 }
240
241 if (!len)
242 return;
243
244 pos -= EXT4_MIN_INLINE_DATA_SIZE;
245 header = IHDR(inode, raw_inode);
246 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
247 EXT4_I(inode)->i_inline_off);
248
249 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
250 buffer, len);
251}
252
253static int ext4_create_inline_data(handle_t *handle,
254 struct inode *inode, unsigned len)
255{
256 int error;
257 void *value = NULL;
258 struct ext4_xattr_ibody_find is = {
259 .s = { .not_found = -ENODATA, },
260 };
261 struct ext4_xattr_info i = {
262 .name_index = EXT4_XATTR_INDEX_SYSTEM,
263 .name = EXT4_XATTR_SYSTEM_DATA,
264 };
265
266 error = ext4_get_inode_loc(inode, &is.iloc);
267 if (error)
268 return error;
269
270 BUFFER_TRACE(is.iloc.bh, "get_write_access");
271 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
272 EXT4_JTR_NONE);
273 if (error)
274 goto out;
275
276 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
277 value = EXT4_ZERO_XATTR_VALUE;
278 len -= EXT4_MIN_INLINE_DATA_SIZE;
279 } else {
280 value = "";
281 len = 0;
282 }
283
284 /* Insert the xttr entry. */
285 i.value = value;
286 i.value_len = len;
287
288 error = ext4_xattr_ibody_find(inode, &i, &is);
289 if (error)
290 goto out;
291
292 BUG_ON(!is.s.not_found);
293
294 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
295 if (error) {
296 if (error == -ENOSPC)
297 ext4_clear_inode_state(inode,
298 EXT4_STATE_MAY_INLINE_DATA);
299 goto out;
300 }
301
302 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
303 0, EXT4_MIN_INLINE_DATA_SIZE);
304
305 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
306 (void *)ext4_raw_inode(&is.iloc));
307 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
308 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
309 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
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, inode->i_sb, is.iloc.bh,
359 EXT4_JTR_NONE);
360 if (error)
361 goto out;
362
363 /* Update the xattr entry. */
364 i.value = value;
365 i.value_len = len;
366
367 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
368 if (error)
369 goto out;
370
371 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
372 (void *)ext4_raw_inode(&is.iloc));
373 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
374 le32_to_cpu(is.s.here->e_value_size);
375 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
376 get_bh(is.iloc.bh);
377 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
378
379out:
380 kfree(value);
381 brelse(is.iloc.bh);
382 return error;
383}
384
385static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
386 unsigned int len)
387{
388 int ret, size, no_expand;
389 struct ext4_inode_info *ei = EXT4_I(inode);
390
391 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
392 return -ENOSPC;
393
394 size = ext4_get_max_inline_size(inode);
395 if (size < len)
396 return -ENOSPC;
397
398 ext4_write_lock_xattr(inode, &no_expand);
399
400 if (ei->i_inline_off)
401 ret = ext4_update_inline_data(handle, inode, len);
402 else
403 ret = ext4_create_inline_data(handle, inode, len);
404
405 ext4_write_unlock_xattr(inode, &no_expand);
406 return ret;
407}
408
409static int ext4_destroy_inline_data_nolock(handle_t *handle,
410 struct inode *inode)
411{
412 struct ext4_inode_info *ei = EXT4_I(inode);
413 struct ext4_xattr_ibody_find is = {
414 .s = { .not_found = 0, },
415 };
416 struct ext4_xattr_info i = {
417 .name_index = EXT4_XATTR_INDEX_SYSTEM,
418 .name = EXT4_XATTR_SYSTEM_DATA,
419 .value = NULL,
420 .value_len = 0,
421 };
422 int error;
423
424 if (!ei->i_inline_off)
425 return 0;
426
427 error = ext4_get_inode_loc(inode, &is.iloc);
428 if (error)
429 return error;
430
431 error = ext4_xattr_ibody_find(inode, &i, &is);
432 if (error)
433 goto out;
434
435 BUFFER_TRACE(is.iloc.bh, "get_write_access");
436 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
437 EXT4_JTR_NONE);
438 if (error)
439 goto out;
440
441 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
442 if (error)
443 goto out;
444
445 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
446 0, EXT4_MIN_INLINE_DATA_SIZE);
447 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
448
449 if (ext4_has_feature_extents(inode->i_sb)) {
450 if (S_ISDIR(inode->i_mode) ||
451 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
452 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
453 ext4_ext_tree_init(handle, inode);
454 }
455 }
456 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
457
458 get_bh(is.iloc.bh);
459 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
460
461 EXT4_I(inode)->i_inline_off = 0;
462 EXT4_I(inode)->i_inline_size = 0;
463 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
464out:
465 brelse(is.iloc.bh);
466 if (error == -ENODATA)
467 error = 0;
468 return error;
469}
470
471static int ext4_read_inline_page(struct inode *inode, struct page *page)
472{
473 void *kaddr;
474 int ret = 0;
475 size_t len;
476 struct ext4_iloc iloc;
477
478 BUG_ON(!PageLocked(page));
479 BUG_ON(!ext4_has_inline_data(inode));
480 BUG_ON(page->index);
481
482 if (!EXT4_I(inode)->i_inline_off) {
483 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
484 inode->i_ino);
485 goto out;
486 }
487
488 ret = ext4_get_inode_loc(inode, &iloc);
489 if (ret)
490 goto out;
491
492 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
493 kaddr = kmap_atomic(page);
494 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
495 flush_dcache_page(page);
496 kunmap_atomic(kaddr);
497 zero_user_segment(page, len, PAGE_SIZE);
498 SetPageUptodate(page);
499 brelse(iloc.bh);
500
501out:
502 return ret;
503}
504
505int ext4_readpage_inline(struct inode *inode, struct page *page)
506{
507 int ret = 0;
508
509 down_read(&EXT4_I(inode)->xattr_sem);
510 if (!ext4_has_inline_data(inode)) {
511 up_read(&EXT4_I(inode)->xattr_sem);
512 return -EAGAIN;
513 }
514
515 /*
516 * Current inline data can only exist in the 1st page,
517 * So for all the other pages, just set them uptodate.
518 */
519 if (!page->index)
520 ret = ext4_read_inline_page(inode, page);
521 else if (!PageUptodate(page)) {
522 zero_user_segment(page, 0, PAGE_SIZE);
523 SetPageUptodate(page);
524 }
525
526 up_read(&EXT4_I(inode)->xattr_sem);
527
528 unlock_page(page);
529 return ret >= 0 ? 0 : ret;
530}
531
532static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
533 struct inode *inode)
534{
535 int ret, needed_blocks, no_expand;
536 handle_t *handle = NULL;
537 int retries = 0, sem_held = 0;
538 struct page *page = NULL;
539 unsigned int flags;
540 unsigned from, to;
541 struct ext4_iloc iloc;
542
543 if (!ext4_has_inline_data(inode)) {
544 /*
545 * clear the flag so that no new write
546 * will trap here again.
547 */
548 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
549 return 0;
550 }
551
552 needed_blocks = ext4_writepage_trans_blocks(inode);
553
554 ret = ext4_get_inode_loc(inode, &iloc);
555 if (ret)
556 return ret;
557
558retry:
559 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
560 if (IS_ERR(handle)) {
561 ret = PTR_ERR(handle);
562 handle = NULL;
563 goto out;
564 }
565
566 /* We cannot recurse into the filesystem as the transaction is already
567 * started */
568 flags = memalloc_nofs_save();
569 page = grab_cache_page_write_begin(mapping, 0);
570 memalloc_nofs_restore(flags);
571 if (!page) {
572 ret = -ENOMEM;
573 goto out;
574 }
575
576 ext4_write_lock_xattr(inode, &no_expand);
577 sem_held = 1;
578 /* If some one has already done this for us, just exit. */
579 if (!ext4_has_inline_data(inode)) {
580 ret = 0;
581 goto out;
582 }
583
584 from = 0;
585 to = ext4_get_inline_size(inode);
586 if (!PageUptodate(page)) {
587 ret = ext4_read_inline_page(inode, page);
588 if (ret < 0)
589 goto out;
590 }
591
592 ret = ext4_destroy_inline_data_nolock(handle, inode);
593 if (ret)
594 goto out;
595
596 if (ext4_should_dioread_nolock(inode)) {
597 ret = __block_write_begin(page, from, to,
598 ext4_get_block_unwritten);
599 } else
600 ret = __block_write_begin(page, from, to, ext4_get_block);
601
602 if (!ret && ext4_should_journal_data(inode)) {
603 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
604 from, to, NULL,
605 do_journal_get_write_access);
606 }
607
608 if (ret) {
609 unlock_page(page);
610 put_page(page);
611 page = NULL;
612 ext4_orphan_add(handle, inode);
613 ext4_write_unlock_xattr(inode, &no_expand);
614 sem_held = 0;
615 ext4_journal_stop(handle);
616 handle = NULL;
617 ext4_truncate_failed_write(inode);
618 /*
619 * If truncate failed early the inode might
620 * still be on the orphan list; we need to
621 * make sure the inode is removed from the
622 * orphan list in that case.
623 */
624 if (inode->i_nlink)
625 ext4_orphan_del(NULL, inode);
626 }
627
628 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
629 goto retry;
630
631 if (page)
632 block_commit_write(page, from, to);
633out:
634 if (page) {
635 unlock_page(page);
636 put_page(page);
637 }
638 if (sem_held)
639 ext4_write_unlock_xattr(inode, &no_expand);
640 if (handle)
641 ext4_journal_stop(handle);
642 brelse(iloc.bh);
643 return ret;
644}
645
646/*
647 * Try to write data in the inode.
648 * If the inode has inline data, check whether the new write can be
649 * in the inode also. If not, create the page the handle, move the data
650 * to the page make it update and let the later codes create extent for it.
651 */
652int ext4_try_to_write_inline_data(struct address_space *mapping,
653 struct inode *inode,
654 loff_t pos, unsigned len,
655 struct page **pagep)
656{
657 int ret;
658 handle_t *handle;
659 unsigned int flags;
660 struct page *page;
661 struct ext4_iloc iloc;
662
663 if (pos + len > ext4_get_max_inline_size(inode))
664 goto convert;
665
666 ret = ext4_get_inode_loc(inode, &iloc);
667 if (ret)
668 return ret;
669
670 /*
671 * The possible write could happen in the inode,
672 * so try to reserve the space in inode first.
673 */
674 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
675 if (IS_ERR(handle)) {
676 ret = PTR_ERR(handle);
677 handle = NULL;
678 goto out;
679 }
680
681 ret = ext4_prepare_inline_data(handle, inode, pos + len);
682 if (ret && ret != -ENOSPC)
683 goto out;
684
685 /* We don't have space in inline inode, so convert it to extent. */
686 if (ret == -ENOSPC) {
687 ext4_journal_stop(handle);
688 brelse(iloc.bh);
689 goto convert;
690 }
691
692 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
693 EXT4_JTR_NONE);
694 if (ret)
695 goto out;
696
697 flags = memalloc_nofs_save();
698 page = grab_cache_page_write_begin(mapping, 0);
699 memalloc_nofs_restore(flags);
700 if (!page) {
701 ret = -ENOMEM;
702 goto out;
703 }
704
705 *pagep = page;
706 down_read(&EXT4_I(inode)->xattr_sem);
707 if (!ext4_has_inline_data(inode)) {
708 ret = 0;
709 unlock_page(page);
710 put_page(page);
711 goto out_up_read;
712 }
713
714 if (!PageUptodate(page)) {
715 ret = ext4_read_inline_page(inode, page);
716 if (ret < 0) {
717 unlock_page(page);
718 put_page(page);
719 goto out_up_read;
720 }
721 }
722
723 ret = 1;
724 handle = NULL;
725out_up_read:
726 up_read(&EXT4_I(inode)->xattr_sem);
727out:
728 if (handle && (ret != 1))
729 ext4_journal_stop(handle);
730 brelse(iloc.bh);
731 return ret;
732convert:
733 return ext4_convert_inline_data_to_extent(mapping, inode);
734}
735
736int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
737 unsigned copied, struct page *page)
738{
739 handle_t *handle = ext4_journal_current_handle();
740 int no_expand;
741 void *kaddr;
742 struct ext4_iloc iloc;
743 int ret = 0, ret2;
744
745 if (unlikely(copied < len) && !PageUptodate(page))
746 copied = 0;
747
748 if (likely(copied)) {
749 ret = ext4_get_inode_loc(inode, &iloc);
750 if (ret) {
751 unlock_page(page);
752 put_page(page);
753 ext4_std_error(inode->i_sb, ret);
754 goto out;
755 }
756 ext4_write_lock_xattr(inode, &no_expand);
757 BUG_ON(!ext4_has_inline_data(inode));
758
759 /*
760 * ei->i_inline_off may have changed since
761 * ext4_write_begin() called
762 * ext4_try_to_write_inline_data()
763 */
764 (void) ext4_find_inline_data_nolock(inode);
765
766 kaddr = kmap_atomic(page);
767 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
768 kunmap_atomic(kaddr);
769 SetPageUptodate(page);
770 /* clear page dirty so that writepages wouldn't work for us. */
771 ClearPageDirty(page);
772
773 ext4_write_unlock_xattr(inode, &no_expand);
774 brelse(iloc.bh);
775
776 /*
777 * It's important to update i_size while still holding page
778 * lock: page writeout could otherwise come in and zero
779 * beyond i_size.
780 */
781 ext4_update_inode_size(inode, pos + copied);
782 }
783 unlock_page(page);
784 put_page(page);
785
786 /*
787 * Don't mark the inode dirty under page lock. First, it unnecessarily
788 * makes the holding time of page lock longer. Second, it forces lock
789 * ordering of page lock and transaction start for journaling
790 * filesystems.
791 */
792 if (likely(copied))
793 mark_inode_dirty(inode);
794out:
795 /*
796 * If we didn't copy as much data as expected, we need to trim back
797 * size of xattr containing inline data.
798 */
799 if (pos + len > inode->i_size && ext4_can_truncate(inode))
800 ext4_orphan_add(handle, inode);
801
802 ret2 = ext4_journal_stop(handle);
803 if (!ret)
804 ret = ret2;
805 if (pos + len > inode->i_size) {
806 ext4_truncate_failed_write(inode);
807 /*
808 * If truncate failed early the inode might still be
809 * on the orphan list; we need to make sure the inode
810 * is removed from the orphan list in that case.
811 */
812 if (inode->i_nlink)
813 ext4_orphan_del(NULL, inode);
814 }
815 return ret ? ret : copied;
816}
817
818struct buffer_head *
819ext4_journalled_write_inline_data(struct inode *inode,
820 unsigned len,
821 struct page *page)
822{
823 int ret, no_expand;
824 void *kaddr;
825 struct ext4_iloc iloc;
826
827 ret = ext4_get_inode_loc(inode, &iloc);
828 if (ret) {
829 ext4_std_error(inode->i_sb, ret);
830 return NULL;
831 }
832
833 ext4_write_lock_xattr(inode, &no_expand);
834 kaddr = kmap_atomic(page);
835 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
836 kunmap_atomic(kaddr);
837 ext4_write_unlock_xattr(inode, &no_expand);
838
839 return iloc.bh;
840}
841
842/*
843 * Try to make the page cache and handle ready for the inline data case.
844 * We can call this function in 2 cases:
845 * 1. The inode is created and the first write exceeds inline size. We can
846 * clear the inode state safely.
847 * 2. The inode has inline data, then we need to read the data, make it
848 * update and dirty so that ext4_da_writepages can handle it. We don't
849 * need to start the journal since the file's metadata isn't changed now.
850 */
851static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
852 struct inode *inode,
853 void **fsdata)
854{
855 int ret = 0, inline_size;
856 struct page *page;
857
858 page = grab_cache_page_write_begin(mapping, 0);
859 if (!page)
860 return -ENOMEM;
861
862 down_read(&EXT4_I(inode)->xattr_sem);
863 if (!ext4_has_inline_data(inode)) {
864 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
865 goto out;
866 }
867
868 inline_size = ext4_get_inline_size(inode);
869
870 if (!PageUptodate(page)) {
871 ret = ext4_read_inline_page(inode, page);
872 if (ret < 0)
873 goto out;
874 }
875
876 ret = __block_write_begin(page, 0, inline_size,
877 ext4_da_get_block_prep);
878 if (ret) {
879 up_read(&EXT4_I(inode)->xattr_sem);
880 unlock_page(page);
881 put_page(page);
882 ext4_truncate_failed_write(inode);
883 return ret;
884 }
885
886 SetPageDirty(page);
887 SetPageUptodate(page);
888 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
889 *fsdata = (void *)CONVERT_INLINE_DATA;
890
891out:
892 up_read(&EXT4_I(inode)->xattr_sem);
893 if (page) {
894 unlock_page(page);
895 put_page(page);
896 }
897 return ret;
898}
899
900/*
901 * Prepare the write for the inline data.
902 * If the data can be written into the inode, we just read
903 * the page and make it uptodate, and start the journal.
904 * Otherwise read the page, makes it dirty so that it can be
905 * handle in writepages(the i_disksize update is left to the
906 * normal ext4_da_write_end).
907 */
908int ext4_da_write_inline_data_begin(struct address_space *mapping,
909 struct inode *inode,
910 loff_t pos, unsigned len,
911 struct page **pagep,
912 void **fsdata)
913{
914 int ret;
915 handle_t *handle;
916 struct page *page;
917 struct ext4_iloc iloc;
918 int retries = 0;
919 unsigned int flags;
920
921 ret = ext4_get_inode_loc(inode, &iloc);
922 if (ret)
923 return ret;
924
925retry_journal:
926 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
927 if (IS_ERR(handle)) {
928 ret = PTR_ERR(handle);
929 goto out;
930 }
931
932 ret = ext4_prepare_inline_data(handle, inode, pos + len);
933 if (ret && ret != -ENOSPC)
934 goto out_journal;
935
936 if (ret == -ENOSPC) {
937 ext4_journal_stop(handle);
938 ret = ext4_da_convert_inline_data_to_extent(mapping,
939 inode,
940 fsdata);
941 if (ret == -ENOSPC &&
942 ext4_should_retry_alloc(inode->i_sb, &retries))
943 goto retry_journal;
944 goto out;
945 }
946
947 /*
948 * We cannot recurse into the filesystem as the transaction
949 * is already started.
950 */
951 flags = memalloc_nofs_save();
952 page = grab_cache_page_write_begin(mapping, 0);
953 memalloc_nofs_restore(flags);
954 if (!page) {
955 ret = -ENOMEM;
956 goto out_journal;
957 }
958
959 down_read(&EXT4_I(inode)->xattr_sem);
960 if (!ext4_has_inline_data(inode)) {
961 ret = 0;
962 goto out_release_page;
963 }
964
965 if (!PageUptodate(page)) {
966 ret = ext4_read_inline_page(inode, page);
967 if (ret < 0)
968 goto out_release_page;
969 }
970 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
971 EXT4_JTR_NONE);
972 if (ret)
973 goto out_release_page;
974
975 up_read(&EXT4_I(inode)->xattr_sem);
976 *pagep = page;
977 brelse(iloc.bh);
978 return 1;
979out_release_page:
980 up_read(&EXT4_I(inode)->xattr_sem);
981 unlock_page(page);
982 put_page(page);
983out_journal:
984 ext4_journal_stop(handle);
985out:
986 brelse(iloc.bh);
987 return ret;
988}
989
990#ifdef INLINE_DIR_DEBUG
991void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
992 void *inline_start, int inline_size)
993{
994 int offset;
995 unsigned short de_len;
996 struct ext4_dir_entry_2 *de = inline_start;
997 void *dlimit = inline_start + inline_size;
998
999 trace_printk("inode %lu\n", dir->i_ino);
1000 offset = 0;
1001 while ((void *)de < dlimit) {
1002 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1003 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1004 offset, de_len, de->name_len, de->name,
1005 de->name_len, le32_to_cpu(de->inode));
1006 if (ext4_check_dir_entry(dir, NULL, de, bh,
1007 inline_start, inline_size, offset))
1008 BUG();
1009
1010 offset += de_len;
1011 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1012 }
1013}
1014#else
1015#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1016#endif
1017
1018/*
1019 * Add a new entry into a inline dir.
1020 * It will return -ENOSPC if no space is available, and -EIO
1021 * and -EEXIST if directory entry already exists.
1022 */
1023static int ext4_add_dirent_to_inline(handle_t *handle,
1024 struct ext4_filename *fname,
1025 struct inode *dir,
1026 struct inode *inode,
1027 struct ext4_iloc *iloc,
1028 void *inline_start, int inline_size)
1029{
1030 int err;
1031 struct ext4_dir_entry_2 *de;
1032
1033 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1034 inline_size, fname, &de);
1035 if (err)
1036 return err;
1037
1038 BUFFER_TRACE(iloc->bh, "get_write_access");
1039 err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1040 EXT4_JTR_NONE);
1041 if (err)
1042 return err;
1043 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1044
1045 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1046
1047 /*
1048 * XXX shouldn't update any times until successful
1049 * completion of syscall, but too many callers depend
1050 * on this.
1051 *
1052 * XXX similarly, too many callers depend on
1053 * ext4_new_inode() setting the times, but error
1054 * recovery deletes the inode, so the worst that can
1055 * happen is that the times are slightly out of date
1056 * and/or different from the directory change time.
1057 */
1058 dir->i_mtime = dir->i_ctime = current_time(dir);
1059 ext4_update_dx_flag(dir);
1060 inode_inc_iversion(dir);
1061 return 1;
1062}
1063
1064static void *ext4_get_inline_xattr_pos(struct inode *inode,
1065 struct ext4_iloc *iloc)
1066{
1067 struct ext4_xattr_entry *entry;
1068 struct ext4_xattr_ibody_header *header;
1069
1070 BUG_ON(!EXT4_I(inode)->i_inline_off);
1071
1072 header = IHDR(inode, ext4_raw_inode(iloc));
1073 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1074 EXT4_I(inode)->i_inline_off);
1075
1076 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1077}
1078
1079/* Set the final de to cover the whole block. */
1080static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1081{
1082 struct ext4_dir_entry_2 *de, *prev_de;
1083 void *limit;
1084 int de_len;
1085
1086 de = de_buf;
1087 if (old_size) {
1088 limit = de_buf + old_size;
1089 do {
1090 prev_de = de;
1091 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1092 de_buf += de_len;
1093 de = de_buf;
1094 } while (de_buf < limit);
1095
1096 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1097 old_size, new_size);
1098 } else {
1099 /* this is just created, so create an empty entry. */
1100 de->inode = 0;
1101 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1102 }
1103}
1104
1105static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1106 struct ext4_iloc *iloc)
1107{
1108 int ret;
1109 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1110 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1111
1112 if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1113 return -ENOSPC;
1114
1115 ret = ext4_update_inline_data(handle, dir,
1116 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1117 if (ret)
1118 return ret;
1119
1120 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1121 EXT4_I(dir)->i_inline_size -
1122 EXT4_MIN_INLINE_DATA_SIZE);
1123 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1124 return 0;
1125}
1126
1127static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1128 struct ext4_iloc *iloc,
1129 void *buf, int inline_size)
1130{
1131 int ret;
1132
1133 ret = ext4_create_inline_data(handle, inode, inline_size);
1134 if (ret) {
1135 ext4_msg(inode->i_sb, KERN_EMERG,
1136 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1137 inode->i_ino, ret);
1138 return;
1139 }
1140 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1141 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1142}
1143
1144static int ext4_finish_convert_inline_dir(handle_t *handle,
1145 struct inode *inode,
1146 struct buffer_head *dir_block,
1147 void *buf,
1148 int inline_size)
1149{
1150 int err, csum_size = 0, header_size = 0;
1151 struct ext4_dir_entry_2 *de;
1152 void *target = dir_block->b_data;
1153
1154 /*
1155 * First create "." and ".." and then copy the dir information
1156 * back to the block.
1157 */
1158 de = target;
1159 de = ext4_init_dot_dotdot(inode, de,
1160 inode->i_sb->s_blocksize, csum_size,
1161 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1162 header_size = (void *)de - target;
1163
1164 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1165 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1166
1167 if (ext4_has_metadata_csum(inode->i_sb))
1168 csum_size = sizeof(struct ext4_dir_entry_tail);
1169
1170 inode->i_size = inode->i_sb->s_blocksize;
1171 i_size_write(inode, inode->i_sb->s_blocksize);
1172 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1173 ext4_update_final_de(dir_block->b_data,
1174 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1175 inode->i_sb->s_blocksize - csum_size);
1176
1177 if (csum_size)
1178 ext4_initialize_dirent_tail(dir_block,
1179 inode->i_sb->s_blocksize);
1180 set_buffer_uptodate(dir_block);
1181 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1182 if (err)
1183 return err;
1184 set_buffer_verified(dir_block);
1185 return ext4_mark_inode_dirty(handle, inode);
1186}
1187
1188static int ext4_convert_inline_data_nolock(handle_t *handle,
1189 struct inode *inode,
1190 struct ext4_iloc *iloc)
1191{
1192 int error;
1193 void *buf = NULL;
1194 struct buffer_head *data_bh = NULL;
1195 struct ext4_map_blocks map;
1196 int inline_size;
1197
1198 inline_size = ext4_get_inline_size(inode);
1199 buf = kmalloc(inline_size, GFP_NOFS);
1200 if (!buf) {
1201 error = -ENOMEM;
1202 goto out;
1203 }
1204
1205 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1206 if (error < 0)
1207 goto out;
1208
1209 /*
1210 * Make sure the inline directory entries pass checks before we try to
1211 * convert them, so that we avoid touching stuff that needs fsck.
1212 */
1213 if (S_ISDIR(inode->i_mode)) {
1214 error = ext4_check_all_de(inode, iloc->bh,
1215 buf + EXT4_INLINE_DOTDOT_SIZE,
1216 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1217 if (error)
1218 goto out;
1219 }
1220
1221 error = ext4_destroy_inline_data_nolock(handle, inode);
1222 if (error)
1223 goto out;
1224
1225 map.m_lblk = 0;
1226 map.m_len = 1;
1227 map.m_flags = 0;
1228 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1229 if (error < 0)
1230 goto out_restore;
1231 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1232 error = -EIO;
1233 goto out_restore;
1234 }
1235
1236 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1237 if (!data_bh) {
1238 error = -ENOMEM;
1239 goto out_restore;
1240 }
1241
1242 lock_buffer(data_bh);
1243 error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1244 EXT4_JTR_NONE);
1245 if (error) {
1246 unlock_buffer(data_bh);
1247 error = -EIO;
1248 goto out_restore;
1249 }
1250 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1251
1252 if (!S_ISDIR(inode->i_mode)) {
1253 memcpy(data_bh->b_data, buf, inline_size);
1254 set_buffer_uptodate(data_bh);
1255 error = ext4_handle_dirty_metadata(handle,
1256 inode, data_bh);
1257 } else {
1258 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1259 buf, inline_size);
1260 }
1261
1262 unlock_buffer(data_bh);
1263out_restore:
1264 if (error)
1265 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1266
1267out:
1268 brelse(data_bh);
1269 kfree(buf);
1270 return error;
1271}
1272
1273/*
1274 * Try to add the new entry to the inline data.
1275 * If succeeds, return 0. If not, extended the inline dir and copied data to
1276 * the new created block.
1277 */
1278int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1279 struct inode *dir, struct inode *inode)
1280{
1281 int ret, ret2, inline_size, no_expand;
1282 void *inline_start;
1283 struct ext4_iloc iloc;
1284
1285 ret = ext4_get_inode_loc(dir, &iloc);
1286 if (ret)
1287 return ret;
1288
1289 ext4_write_lock_xattr(dir, &no_expand);
1290 if (!ext4_has_inline_data(dir))
1291 goto out;
1292
1293 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1294 EXT4_INLINE_DOTDOT_SIZE;
1295 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1296
1297 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1298 inline_start, inline_size);
1299 if (ret != -ENOSPC)
1300 goto out;
1301
1302 /* check whether it can be inserted to inline xattr space. */
1303 inline_size = EXT4_I(dir)->i_inline_size -
1304 EXT4_MIN_INLINE_DATA_SIZE;
1305 if (!inline_size) {
1306 /* Try to use the xattr space.*/
1307 ret = ext4_update_inline_dir(handle, dir, &iloc);
1308 if (ret && ret != -ENOSPC)
1309 goto out;
1310
1311 inline_size = EXT4_I(dir)->i_inline_size -
1312 EXT4_MIN_INLINE_DATA_SIZE;
1313 }
1314
1315 if (inline_size) {
1316 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1317
1318 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1319 inode, &iloc, inline_start,
1320 inline_size);
1321
1322 if (ret != -ENOSPC)
1323 goto out;
1324 }
1325
1326 /*
1327 * The inline space is filled up, so create a new block for it.
1328 * As the extent tree will be created, we have to save the inline
1329 * dir first.
1330 */
1331 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1332
1333out:
1334 ext4_write_unlock_xattr(dir, &no_expand);
1335 ret2 = ext4_mark_inode_dirty(handle, dir);
1336 if (unlikely(ret2 && !ret))
1337 ret = ret2;
1338 brelse(iloc.bh);
1339 return ret;
1340}
1341
1342/*
1343 * This function fills a red-black tree with information from an
1344 * inlined dir. It returns the number directory entries loaded
1345 * into the tree. If there is an error it is returned in err.
1346 */
1347int ext4_inlinedir_to_tree(struct file *dir_file,
1348 struct inode *dir, ext4_lblk_t block,
1349 struct dx_hash_info *hinfo,
1350 __u32 start_hash, __u32 start_minor_hash,
1351 int *has_inline_data)
1352{
1353 int err = 0, count = 0;
1354 unsigned int parent_ino;
1355 int pos;
1356 struct ext4_dir_entry_2 *de;
1357 struct inode *inode = file_inode(dir_file);
1358 int ret, inline_size = 0;
1359 struct ext4_iloc iloc;
1360 void *dir_buf = NULL;
1361 struct ext4_dir_entry_2 fake;
1362 struct fscrypt_str tmp_str;
1363
1364 ret = ext4_get_inode_loc(inode, &iloc);
1365 if (ret)
1366 return ret;
1367
1368 down_read(&EXT4_I(inode)->xattr_sem);
1369 if (!ext4_has_inline_data(inode)) {
1370 up_read(&EXT4_I(inode)->xattr_sem);
1371 *has_inline_data = 0;
1372 goto out;
1373 }
1374
1375 inline_size = ext4_get_inline_size(inode);
1376 dir_buf = kmalloc(inline_size, GFP_NOFS);
1377 if (!dir_buf) {
1378 ret = -ENOMEM;
1379 up_read(&EXT4_I(inode)->xattr_sem);
1380 goto out;
1381 }
1382
1383 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1384 up_read(&EXT4_I(inode)->xattr_sem);
1385 if (ret < 0)
1386 goto out;
1387
1388 pos = 0;
1389 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1390 while (pos < inline_size) {
1391 /*
1392 * As inlined dir doesn't store any information about '.' and
1393 * only the inode number of '..' is stored, we have to handle
1394 * them differently.
1395 */
1396 if (pos == 0) {
1397 fake.inode = cpu_to_le32(inode->i_ino);
1398 fake.name_len = 1;
1399 strcpy(fake.name, ".");
1400 fake.rec_len = ext4_rec_len_to_disk(
1401 ext4_dir_rec_len(fake.name_len, NULL),
1402 inline_size);
1403 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1404 de = &fake;
1405 pos = EXT4_INLINE_DOTDOT_OFFSET;
1406 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1407 fake.inode = cpu_to_le32(parent_ino);
1408 fake.name_len = 2;
1409 strcpy(fake.name, "..");
1410 fake.rec_len = ext4_rec_len_to_disk(
1411 ext4_dir_rec_len(fake.name_len, NULL),
1412 inline_size);
1413 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1414 de = &fake;
1415 pos = EXT4_INLINE_DOTDOT_SIZE;
1416 } else {
1417 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1418 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1419 if (ext4_check_dir_entry(inode, dir_file, de,
1420 iloc.bh, dir_buf,
1421 inline_size, pos)) {
1422 ret = count;
1423 goto out;
1424 }
1425 }
1426
1427 if (ext4_hash_in_dirent(dir)) {
1428 hinfo->hash = EXT4_DIRENT_HASH(de);
1429 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1430 } else {
1431 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1432 }
1433 if ((hinfo->hash < start_hash) ||
1434 ((hinfo->hash == start_hash) &&
1435 (hinfo->minor_hash < start_minor_hash)))
1436 continue;
1437 if (de->inode == 0)
1438 continue;
1439 tmp_str.name = de->name;
1440 tmp_str.len = de->name_len;
1441 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1442 hinfo->minor_hash, de, &tmp_str);
1443 if (err) {
1444 ret = err;
1445 goto out;
1446 }
1447 count++;
1448 }
1449 ret = count;
1450out:
1451 kfree(dir_buf);
1452 brelse(iloc.bh);
1453 return ret;
1454}
1455
1456/*
1457 * So this function is called when the volume is mkfsed with
1458 * dir_index disabled. In order to keep f_pos persistent
1459 * after we convert from an inlined dir to a blocked based,
1460 * we just pretend that we are a normal dir and return the
1461 * offset as if '.' and '..' really take place.
1462 *
1463 */
1464int ext4_read_inline_dir(struct file *file,
1465 struct dir_context *ctx,
1466 int *has_inline_data)
1467{
1468 unsigned int offset, parent_ino;
1469 int i;
1470 struct ext4_dir_entry_2 *de;
1471 struct super_block *sb;
1472 struct inode *inode = file_inode(file);
1473 int ret, inline_size = 0;
1474 struct ext4_iloc iloc;
1475 void *dir_buf = NULL;
1476 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1477
1478 ret = ext4_get_inode_loc(inode, &iloc);
1479 if (ret)
1480 return ret;
1481
1482 down_read(&EXT4_I(inode)->xattr_sem);
1483 if (!ext4_has_inline_data(inode)) {
1484 up_read(&EXT4_I(inode)->xattr_sem);
1485 *has_inline_data = 0;
1486 goto out;
1487 }
1488
1489 inline_size = ext4_get_inline_size(inode);
1490 dir_buf = kmalloc(inline_size, GFP_NOFS);
1491 if (!dir_buf) {
1492 ret = -ENOMEM;
1493 up_read(&EXT4_I(inode)->xattr_sem);
1494 goto out;
1495 }
1496
1497 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1498 up_read(&EXT4_I(inode)->xattr_sem);
1499 if (ret < 0)
1500 goto out;
1501
1502 ret = 0;
1503 sb = inode->i_sb;
1504 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1505 offset = ctx->pos;
1506
1507 /*
1508 * dotdot_offset and dotdot_size is the real offset and
1509 * size for ".." and "." if the dir is block based while
1510 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1511 * So we will use extra_offset and extra_size to indicate them
1512 * during the inline dir iteration.
1513 */
1514 dotdot_offset = ext4_dir_rec_len(1, NULL);
1515 dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1516 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1517 extra_size = extra_offset + inline_size;
1518
1519 /*
1520 * If the version has changed since the last call to
1521 * readdir(2), then we might be pointing to an invalid
1522 * dirent right now. Scan from the start of the inline
1523 * dir to make sure.
1524 */
1525 if (!inode_eq_iversion(inode, file->f_version)) {
1526 for (i = 0; i < extra_size && i < offset;) {
1527 /*
1528 * "." is with offset 0 and
1529 * ".." is dotdot_offset.
1530 */
1531 if (!i) {
1532 i = dotdot_offset;
1533 continue;
1534 } else if (i == dotdot_offset) {
1535 i = dotdot_size;
1536 continue;
1537 }
1538 /* for other entry, the real offset in
1539 * the buf has to be tuned accordingly.
1540 */
1541 de = (struct ext4_dir_entry_2 *)
1542 (dir_buf + i - extra_offset);
1543 /* It's too expensive to do a full
1544 * dirent test each time round this
1545 * loop, but we do have to test at
1546 * least that it is non-zero. A
1547 * failure will be detected in the
1548 * dirent test below. */
1549 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1550 < ext4_dir_rec_len(1, NULL))
1551 break;
1552 i += ext4_rec_len_from_disk(de->rec_len,
1553 extra_size);
1554 }
1555 offset = i;
1556 ctx->pos = offset;
1557 file->f_version = inode_query_iversion(inode);
1558 }
1559
1560 while (ctx->pos < extra_size) {
1561 if (ctx->pos == 0) {
1562 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1563 goto out;
1564 ctx->pos = dotdot_offset;
1565 continue;
1566 }
1567
1568 if (ctx->pos == dotdot_offset) {
1569 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1570 goto out;
1571 ctx->pos = dotdot_size;
1572 continue;
1573 }
1574
1575 de = (struct ext4_dir_entry_2 *)
1576 (dir_buf + ctx->pos - extra_offset);
1577 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1578 extra_size, ctx->pos))
1579 goto out;
1580 if (le32_to_cpu(de->inode)) {
1581 if (!dir_emit(ctx, de->name, de->name_len,
1582 le32_to_cpu(de->inode),
1583 get_dtype(sb, de->file_type)))
1584 goto out;
1585 }
1586 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1587 }
1588out:
1589 kfree(dir_buf);
1590 brelse(iloc.bh);
1591 return ret;
1592}
1593
1594void *ext4_read_inline_link(struct inode *inode)
1595{
1596 struct ext4_iloc iloc;
1597 int ret, inline_size;
1598 void *link;
1599
1600 ret = ext4_get_inode_loc(inode, &iloc);
1601 if (ret)
1602 return ERR_PTR(ret);
1603
1604 ret = -ENOMEM;
1605 inline_size = ext4_get_inline_size(inode);
1606 link = kmalloc(inline_size + 1, GFP_NOFS);
1607 if (!link)
1608 goto out;
1609
1610 ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1611 if (ret < 0) {
1612 kfree(link);
1613 goto out;
1614 }
1615 nd_terminate_link(link, inode->i_size, ret);
1616out:
1617 if (ret < 0)
1618 link = ERR_PTR(ret);
1619 brelse(iloc.bh);
1620 return link;
1621}
1622
1623struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1624 struct ext4_dir_entry_2 **parent_de,
1625 int *retval)
1626{
1627 struct ext4_iloc iloc;
1628
1629 *retval = ext4_get_inode_loc(inode, &iloc);
1630 if (*retval)
1631 return NULL;
1632
1633 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1634
1635 return iloc.bh;
1636}
1637
1638/*
1639 * Try to create the inline data for the new dir.
1640 * If it succeeds, return 0, otherwise return the error.
1641 * In case of ENOSPC, the caller should create the normal disk layout dir.
1642 */
1643int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1644 struct inode *inode)
1645{
1646 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1647 struct ext4_iloc iloc;
1648 struct ext4_dir_entry_2 *de;
1649
1650 ret = ext4_get_inode_loc(inode, &iloc);
1651 if (ret)
1652 return ret;
1653
1654 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1655 if (ret)
1656 goto out;
1657
1658 /*
1659 * For inline dir, we only save the inode information for the ".."
1660 * and create a fake dentry to cover the left space.
1661 */
1662 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1663 de->inode = cpu_to_le32(parent->i_ino);
1664 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1665 de->inode = 0;
1666 de->rec_len = ext4_rec_len_to_disk(
1667 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1668 inline_size);
1669 set_nlink(inode, 2);
1670 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1671out:
1672 brelse(iloc.bh);
1673 return ret;
1674}
1675
1676struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1677 struct ext4_filename *fname,
1678 struct ext4_dir_entry_2 **res_dir,
1679 int *has_inline_data)
1680{
1681 int ret;
1682 struct ext4_iloc iloc;
1683 void *inline_start;
1684 int inline_size;
1685
1686 if (ext4_get_inode_loc(dir, &iloc))
1687 return NULL;
1688
1689 down_read(&EXT4_I(dir)->xattr_sem);
1690 if (!ext4_has_inline_data(dir)) {
1691 *has_inline_data = 0;
1692 goto out;
1693 }
1694
1695 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1696 EXT4_INLINE_DOTDOT_SIZE;
1697 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1698 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1699 dir, fname, 0, res_dir);
1700 if (ret == 1)
1701 goto out_find;
1702 if (ret < 0)
1703 goto out;
1704
1705 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1706 goto out;
1707
1708 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1709 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1710
1711 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1712 dir, fname, 0, res_dir);
1713 if (ret == 1)
1714 goto out_find;
1715
1716out:
1717 brelse(iloc.bh);
1718 iloc.bh = NULL;
1719out_find:
1720 up_read(&EXT4_I(dir)->xattr_sem);
1721 return iloc.bh;
1722}
1723
1724int ext4_delete_inline_entry(handle_t *handle,
1725 struct inode *dir,
1726 struct ext4_dir_entry_2 *de_del,
1727 struct buffer_head *bh,
1728 int *has_inline_data)
1729{
1730 int err, inline_size, no_expand;
1731 struct ext4_iloc iloc;
1732 void *inline_start;
1733
1734 err = ext4_get_inode_loc(dir, &iloc);
1735 if (err)
1736 return err;
1737
1738 ext4_write_lock_xattr(dir, &no_expand);
1739 if (!ext4_has_inline_data(dir)) {
1740 *has_inline_data = 0;
1741 goto out;
1742 }
1743
1744 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1745 EXT4_MIN_INLINE_DATA_SIZE) {
1746 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1747 EXT4_INLINE_DOTDOT_SIZE;
1748 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1749 EXT4_INLINE_DOTDOT_SIZE;
1750 } else {
1751 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1752 inline_size = ext4_get_inline_size(dir) -
1753 EXT4_MIN_INLINE_DATA_SIZE;
1754 }
1755
1756 BUFFER_TRACE(bh, "get_write_access");
1757 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1758 EXT4_JTR_NONE);
1759 if (err)
1760 goto out;
1761
1762 err = ext4_generic_delete_entry(dir, de_del, bh,
1763 inline_start, inline_size, 0);
1764 if (err)
1765 goto out;
1766
1767 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1768out:
1769 ext4_write_unlock_xattr(dir, &no_expand);
1770 if (likely(err == 0))
1771 err = ext4_mark_inode_dirty(handle, dir);
1772 brelse(iloc.bh);
1773 if (err != -ENOENT)
1774 ext4_std_error(dir->i_sb, err);
1775 return err;
1776}
1777
1778/*
1779 * Get the inline dentry at offset.
1780 */
1781static inline struct ext4_dir_entry_2 *
1782ext4_get_inline_entry(struct inode *inode,
1783 struct ext4_iloc *iloc,
1784 unsigned int offset,
1785 void **inline_start,
1786 int *inline_size)
1787{
1788 void *inline_pos;
1789
1790 BUG_ON(offset > ext4_get_inline_size(inode));
1791
1792 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1793 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1794 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1795 } else {
1796 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1797 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1798 *inline_size = ext4_get_inline_size(inode) -
1799 EXT4_MIN_INLINE_DATA_SIZE;
1800 }
1801
1802 if (inline_start)
1803 *inline_start = inline_pos;
1804 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1805}
1806
1807bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1808{
1809 int err, inline_size;
1810 struct ext4_iloc iloc;
1811 size_t inline_len;
1812 void *inline_pos;
1813 unsigned int offset;
1814 struct ext4_dir_entry_2 *de;
1815 bool ret = false;
1816
1817 err = ext4_get_inode_loc(dir, &iloc);
1818 if (err) {
1819 EXT4_ERROR_INODE_ERR(dir, -err,
1820 "error %d getting inode %lu block",
1821 err, dir->i_ino);
1822 return false;
1823 }
1824
1825 down_read(&EXT4_I(dir)->xattr_sem);
1826 if (!ext4_has_inline_data(dir)) {
1827 *has_inline_data = 0;
1828 ret = true;
1829 goto out;
1830 }
1831
1832 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1833 if (!le32_to_cpu(de->inode)) {
1834 ext4_warning(dir->i_sb,
1835 "bad inline directory (dir #%lu) - no `..'",
1836 dir->i_ino);
1837 goto out;
1838 }
1839
1840 inline_len = ext4_get_inline_size(dir);
1841 offset = EXT4_INLINE_DOTDOT_SIZE;
1842 while (offset < inline_len) {
1843 de = ext4_get_inline_entry(dir, &iloc, offset,
1844 &inline_pos, &inline_size);
1845 if (ext4_check_dir_entry(dir, NULL, de,
1846 iloc.bh, inline_pos,
1847 inline_size, offset)) {
1848 ext4_warning(dir->i_sb,
1849 "bad inline directory (dir #%lu) - "
1850 "inode %u, rec_len %u, name_len %d"
1851 "inline size %d",
1852 dir->i_ino, le32_to_cpu(de->inode),
1853 le16_to_cpu(de->rec_len), de->name_len,
1854 inline_size);
1855 goto out;
1856 }
1857 if (le32_to_cpu(de->inode)) {
1858 goto out;
1859 }
1860 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1861 }
1862
1863 ret = true;
1864out:
1865 up_read(&EXT4_I(dir)->xattr_sem);
1866 brelse(iloc.bh);
1867 return ret;
1868}
1869
1870int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1871{
1872 int ret, no_expand;
1873
1874 ext4_write_lock_xattr(inode, &no_expand);
1875 ret = ext4_destroy_inline_data_nolock(handle, inode);
1876 ext4_write_unlock_xattr(inode, &no_expand);
1877
1878 return ret;
1879}
1880
1881int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1882{
1883 __u64 addr;
1884 int error = -EAGAIN;
1885 struct ext4_iloc iloc;
1886
1887 down_read(&EXT4_I(inode)->xattr_sem);
1888 if (!ext4_has_inline_data(inode))
1889 goto out;
1890
1891 error = ext4_get_inode_loc(inode, &iloc);
1892 if (error)
1893 goto out;
1894
1895 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1896 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1897 addr += offsetof(struct ext4_inode, i_block);
1898
1899 brelse(iloc.bh);
1900
1901 iomap->addr = addr;
1902 iomap->offset = 0;
1903 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1904 i_size_read(inode));
1905 iomap->type = IOMAP_INLINE;
1906 iomap->flags = 0;
1907
1908out:
1909 up_read(&EXT4_I(inode)->xattr_sem);
1910 return error;
1911}
1912
1913int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1914{
1915 handle_t *handle;
1916 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1917 size_t i_size;
1918 void *value = NULL;
1919 struct ext4_xattr_ibody_find is = {
1920 .s = { .not_found = -ENODATA, },
1921 };
1922 struct ext4_xattr_info i = {
1923 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1924 .name = EXT4_XATTR_SYSTEM_DATA,
1925 };
1926
1927
1928 needed_blocks = ext4_writepage_trans_blocks(inode);
1929 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1930 if (IS_ERR(handle))
1931 return PTR_ERR(handle);
1932
1933 ext4_write_lock_xattr(inode, &no_expand);
1934 if (!ext4_has_inline_data(inode)) {
1935 ext4_write_unlock_xattr(inode, &no_expand);
1936 *has_inline = 0;
1937 ext4_journal_stop(handle);
1938 return 0;
1939 }
1940
1941 if ((err = ext4_orphan_add(handle, inode)) != 0)
1942 goto out;
1943
1944 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1945 goto out;
1946
1947 down_write(&EXT4_I(inode)->i_data_sem);
1948 i_size = inode->i_size;
1949 inline_size = ext4_get_inline_size(inode);
1950 EXT4_I(inode)->i_disksize = i_size;
1951
1952 if (i_size < inline_size) {
1953 /*
1954 * if there's inline data to truncate and this file was
1955 * converted to extents after that inline data was written,
1956 * the extent status cache must be cleared to avoid leaving
1957 * behind stale delayed allocated extent entries
1958 */
1959 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1960retry:
1961 err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1962 if (err == -ENOMEM) {
1963 memalloc_retry_wait(GFP_ATOMIC);
1964 goto retry;
1965 }
1966 if (err)
1967 goto out_error;
1968 }
1969
1970 /* Clear the content in the xattr space. */
1971 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1972 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1973 goto out_error;
1974
1975 BUG_ON(is.s.not_found);
1976
1977 value_len = le32_to_cpu(is.s.here->e_value_size);
1978 value = kmalloc(value_len, GFP_NOFS);
1979 if (!value) {
1980 err = -ENOMEM;
1981 goto out_error;
1982 }
1983
1984 err = ext4_xattr_ibody_get(inode, i.name_index,
1985 i.name, value, value_len);
1986 if (err <= 0)
1987 goto out_error;
1988
1989 i.value = value;
1990 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1991 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1992 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1993 if (err)
1994 goto out_error;
1995 }
1996
1997 /* Clear the content within i_blocks. */
1998 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1999 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
2000 memset(p + i_size, 0,
2001 EXT4_MIN_INLINE_DATA_SIZE - i_size);
2002 }
2003
2004 EXT4_I(inode)->i_inline_size = i_size <
2005 EXT4_MIN_INLINE_DATA_SIZE ?
2006 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2007 }
2008
2009out_error:
2010 up_write(&EXT4_I(inode)->i_data_sem);
2011out:
2012 brelse(is.iloc.bh);
2013 ext4_write_unlock_xattr(inode, &no_expand);
2014 kfree(value);
2015 if (inode->i_nlink)
2016 ext4_orphan_del(handle, inode);
2017
2018 if (err == 0) {
2019 inode->i_mtime = inode->i_ctime = current_time(inode);
2020 err = ext4_mark_inode_dirty(handle, inode);
2021 if (IS_SYNC(inode))
2022 ext4_handle_sync(handle);
2023 }
2024 ext4_journal_stop(handle);
2025 return err;
2026}
2027
2028int ext4_convert_inline_data(struct inode *inode)
2029{
2030 int error, needed_blocks, no_expand;
2031 handle_t *handle;
2032 struct ext4_iloc iloc;
2033
2034 if (!ext4_has_inline_data(inode)) {
2035 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2036 return 0;
2037 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2038 /*
2039 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2040 * cleared. This means we are in the middle of moving of
2041 * inline data to delay allocated block. Just force writeout
2042 * here to finish conversion.
2043 */
2044 error = filemap_flush(inode->i_mapping);
2045 if (error)
2046 return error;
2047 if (!ext4_has_inline_data(inode))
2048 return 0;
2049 }
2050
2051 needed_blocks = ext4_writepage_trans_blocks(inode);
2052
2053 iloc.bh = NULL;
2054 error = ext4_get_inode_loc(inode, &iloc);
2055 if (error)
2056 return error;
2057
2058 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2059 if (IS_ERR(handle)) {
2060 error = PTR_ERR(handle);
2061 goto out_free;
2062 }
2063
2064 ext4_write_lock_xattr(inode, &no_expand);
2065 if (ext4_has_inline_data(inode))
2066 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2067 ext4_write_unlock_xattr(inode, &no_expand);
2068 ext4_journal_stop(handle);
2069out_free:
2070 brelse(iloc.bh);
2071 return error;
2072}
1/*
2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 */
14
15#include <linux/fiemap.h>
16
17#include "ext4_jbd2.h"
18#include "ext4.h"
19#include "xattr.h"
20#include "truncate.h"
21
22#define EXT4_XATTR_SYSTEM_DATA "data"
23#define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
24#define EXT4_INLINE_DOTDOT_OFFSET 2
25#define EXT4_INLINE_DOTDOT_SIZE 4
26
27static int ext4_get_inline_size(struct inode *inode)
28{
29 if (EXT4_I(inode)->i_inline_off)
30 return EXT4_I(inode)->i_inline_size;
31
32 return 0;
33}
34
35static int get_max_inline_xattr_value_size(struct inode *inode,
36 struct ext4_iloc *iloc)
37{
38 struct ext4_xattr_ibody_header *header;
39 struct ext4_xattr_entry *entry;
40 struct ext4_inode *raw_inode;
41 int free, min_offs;
42
43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 EXT4_GOOD_OLD_INODE_SIZE -
45 EXT4_I(inode)->i_extra_isize -
46 sizeof(struct ext4_xattr_ibody_header);
47
48 /*
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
52 */
53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 return EXT4_XATTR_SIZE(min_offs -
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 EXT4_XATTR_ROUND - sizeof(__u32));
57
58 raw_inode = ext4_raw_inode(iloc);
59 header = IHDR(inode, raw_inode);
60 entry = IFIRST(header);
61
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64 if (!entry->e_value_block && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
66 if (offs < min_offs)
67 min_offs = offs;
68 }
69 }
70 free = min_offs -
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78 goto out;
79 }
80
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85 else
86 free = 0;
87
88out:
89 return free;
90}
91
92/*
93 * Get the maximum size we now can store in an inode.
94 * If we can't find the space for a xattr entry, don't use the space
95 * of the extents since we have no space to indicate the inline data.
96 */
97int ext4_get_max_inline_size(struct inode *inode)
98{
99 int error, max_inline_size;
100 struct ext4_iloc iloc;
101
102 if (EXT4_I(inode)->i_extra_isize == 0)
103 return 0;
104
105 error = ext4_get_inode_loc(inode, &iloc);
106 if (error) {
107 ext4_error_inode(inode, __func__, __LINE__, 0,
108 "can't get inode location %lu",
109 inode->i_ino);
110 return 0;
111 }
112
113 down_read(&EXT4_I(inode)->xattr_sem);
114 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
115 up_read(&EXT4_I(inode)->xattr_sem);
116
117 brelse(iloc.bh);
118
119 if (!max_inline_size)
120 return 0;
121
122 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
123}
124
125/*
126 * this function does not take xattr_sem, which is OK because it is
127 * currently only used in a code path coming form ext4_iget, before
128 * the new inode has been unlocked
129 */
130int ext4_find_inline_data_nolock(struct inode *inode)
131{
132 struct ext4_xattr_ibody_find is = {
133 .s = { .not_found = -ENODATA, },
134 };
135 struct ext4_xattr_info i = {
136 .name_index = EXT4_XATTR_INDEX_SYSTEM,
137 .name = EXT4_XATTR_SYSTEM_DATA,
138 };
139 int error;
140
141 if (EXT4_I(inode)->i_extra_isize == 0)
142 return 0;
143
144 error = ext4_get_inode_loc(inode, &is.iloc);
145 if (error)
146 return error;
147
148 error = ext4_xattr_ibody_find(inode, &i, &is);
149 if (error)
150 goto out;
151
152 if (!is.s.not_found) {
153 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154 (void *)ext4_raw_inode(&is.iloc));
155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156 le32_to_cpu(is.s.here->e_value_size);
157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158 }
159out:
160 brelse(is.iloc.bh);
161 return error;
162}
163
164static int ext4_read_inline_data(struct inode *inode, void *buffer,
165 unsigned int len,
166 struct ext4_iloc *iloc)
167{
168 struct ext4_xattr_entry *entry;
169 struct ext4_xattr_ibody_header *header;
170 int cp_len = 0;
171 struct ext4_inode *raw_inode;
172
173 if (!len)
174 return 0;
175
176 BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179 len : EXT4_MIN_INLINE_DATA_SIZE;
180
181 raw_inode = ext4_raw_inode(iloc);
182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184 len -= cp_len;
185 buffer += cp_len;
186
187 if (!len)
188 goto out;
189
190 header = IHDR(inode, raw_inode);
191 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192 EXT4_I(inode)->i_inline_off);
193 len = min_t(unsigned int, len,
194 (unsigned int)le32_to_cpu(entry->e_value_size));
195
196 memcpy(buffer,
197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198 cp_len += len;
199
200out:
201 return cp_len;
202}
203
204/*
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
209 */
210static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211 void *buffer, loff_t pos, unsigned int len)
212{
213 struct ext4_xattr_entry *entry;
214 struct ext4_xattr_ibody_header *header;
215 struct ext4_inode *raw_inode;
216 int cp_len = 0;
217
218 BUG_ON(!EXT4_I(inode)->i_inline_off);
219 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
220
221 raw_inode = ext4_raw_inode(iloc);
222 buffer += pos;
223
224 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
225 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
226 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
227 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
228
229 len -= cp_len;
230 buffer += cp_len;
231 pos += cp_len;
232 }
233
234 if (!len)
235 return;
236
237 pos -= EXT4_MIN_INLINE_DATA_SIZE;
238 header = IHDR(inode, raw_inode);
239 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
240 EXT4_I(inode)->i_inline_off);
241
242 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
243 buffer, len);
244}
245
246static int ext4_create_inline_data(handle_t *handle,
247 struct inode *inode, unsigned len)
248{
249 int error;
250 void *value = NULL;
251 struct ext4_xattr_ibody_find is = {
252 .s = { .not_found = -ENODATA, },
253 };
254 struct ext4_xattr_info i = {
255 .name_index = EXT4_XATTR_INDEX_SYSTEM,
256 .name = EXT4_XATTR_SYSTEM_DATA,
257 };
258
259 error = ext4_get_inode_loc(inode, &is.iloc);
260 if (error)
261 return error;
262
263 BUFFER_TRACE(is.iloc.bh, "get_write_access");
264 error = ext4_journal_get_write_access(handle, is.iloc.bh);
265 if (error)
266 goto out;
267
268 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
269 value = EXT4_ZERO_XATTR_VALUE;
270 len -= EXT4_MIN_INLINE_DATA_SIZE;
271 } else {
272 value = "";
273 len = 0;
274 }
275
276 /* Insert the the xttr entry. */
277 i.value = value;
278 i.value_len = len;
279
280 error = ext4_xattr_ibody_find(inode, &i, &is);
281 if (error)
282 goto out;
283
284 BUG_ON(!is.s.not_found);
285
286 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
287 if (error) {
288 if (error == -ENOSPC)
289 ext4_clear_inode_state(inode,
290 EXT4_STATE_MAY_INLINE_DATA);
291 goto out;
292 }
293
294 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
295 0, EXT4_MIN_INLINE_DATA_SIZE);
296
297 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
298 (void *)ext4_raw_inode(&is.iloc));
299 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
300 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
301 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
302 /*
303 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
304 * get cleared
305 */
306 ext4_set_inode_flags(inode);
307 get_bh(is.iloc.bh);
308 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
309
310out:
311 brelse(is.iloc.bh);
312 return error;
313}
314
315static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
316 unsigned int len)
317{
318 int error;
319 void *value = NULL;
320 struct ext4_xattr_ibody_find is = {
321 .s = { .not_found = -ENODATA, },
322 };
323 struct ext4_xattr_info i = {
324 .name_index = EXT4_XATTR_INDEX_SYSTEM,
325 .name = EXT4_XATTR_SYSTEM_DATA,
326 };
327
328 /* If the old space is ok, write the data directly. */
329 if (len <= EXT4_I(inode)->i_inline_size)
330 return 0;
331
332 error = ext4_get_inode_loc(inode, &is.iloc);
333 if (error)
334 return error;
335
336 error = ext4_xattr_ibody_find(inode, &i, &is);
337 if (error)
338 goto out;
339
340 BUG_ON(is.s.not_found);
341
342 len -= EXT4_MIN_INLINE_DATA_SIZE;
343 value = kzalloc(len, GFP_NOFS);
344 if (!value) {
345 error = -ENOMEM;
346 goto out;
347 }
348
349 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
350 value, len);
351 if (error == -ENODATA)
352 goto out;
353
354 BUFFER_TRACE(is.iloc.bh, "get_write_access");
355 error = ext4_journal_get_write_access(handle, is.iloc.bh);
356 if (error)
357 goto out;
358
359 /* Update the xttr entry. */
360 i.value = value;
361 i.value_len = len;
362
363 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
364 if (error)
365 goto out;
366
367 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
368 (void *)ext4_raw_inode(&is.iloc));
369 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
370 le32_to_cpu(is.s.here->e_value_size);
371 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
372 get_bh(is.iloc.bh);
373 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
374
375out:
376 kfree(value);
377 brelse(is.iloc.bh);
378 return error;
379}
380
381static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
382 unsigned int len)
383{
384 int ret, size, no_expand;
385 struct ext4_inode_info *ei = EXT4_I(inode);
386
387 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
388 return -ENOSPC;
389
390 size = ext4_get_max_inline_size(inode);
391 if (size < len)
392 return -ENOSPC;
393
394 ext4_write_lock_xattr(inode, &no_expand);
395
396 if (ei->i_inline_off)
397 ret = ext4_update_inline_data(handle, inode, len);
398 else
399 ret = ext4_create_inline_data(handle, inode, len);
400
401 ext4_write_unlock_xattr(inode, &no_expand);
402 return ret;
403}
404
405static int ext4_destroy_inline_data_nolock(handle_t *handle,
406 struct inode *inode)
407{
408 struct ext4_inode_info *ei = EXT4_I(inode);
409 struct ext4_xattr_ibody_find is = {
410 .s = { .not_found = 0, },
411 };
412 struct ext4_xattr_info i = {
413 .name_index = EXT4_XATTR_INDEX_SYSTEM,
414 .name = EXT4_XATTR_SYSTEM_DATA,
415 .value = NULL,
416 .value_len = 0,
417 };
418 int error;
419
420 if (!ei->i_inline_off)
421 return 0;
422
423 error = ext4_get_inode_loc(inode, &is.iloc);
424 if (error)
425 return error;
426
427 error = ext4_xattr_ibody_find(inode, &i, &is);
428 if (error)
429 goto out;
430
431 BUFFER_TRACE(is.iloc.bh, "get_write_access");
432 error = ext4_journal_get_write_access(handle, is.iloc.bh);
433 if (error)
434 goto out;
435
436 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
437 if (error)
438 goto out;
439
440 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
441 0, EXT4_MIN_INLINE_DATA_SIZE);
442
443 if (ext4_has_feature_extents(inode->i_sb)) {
444 if (S_ISDIR(inode->i_mode) ||
445 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
446 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
447 ext4_ext_tree_init(handle, inode);
448 }
449 }
450 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
451 /*
452 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
453 * get set.
454 */
455 ext4_set_inode_flags(inode);
456
457 get_bh(is.iloc.bh);
458 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
459
460 EXT4_I(inode)->i_inline_off = 0;
461 EXT4_I(inode)->i_inline_size = 0;
462 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
463out:
464 brelse(is.iloc.bh);
465 if (error == -ENODATA)
466 error = 0;
467 return error;
468}
469
470static int ext4_read_inline_page(struct inode *inode, struct page *page)
471{
472 void *kaddr;
473 int ret = 0;
474 size_t len;
475 struct ext4_iloc iloc;
476
477 BUG_ON(!PageLocked(page));
478 BUG_ON(!ext4_has_inline_data(inode));
479 BUG_ON(page->index);
480
481 if (!EXT4_I(inode)->i_inline_off) {
482 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
483 inode->i_ino);
484 goto out;
485 }
486
487 ret = ext4_get_inode_loc(inode, &iloc);
488 if (ret)
489 goto out;
490
491 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
492 kaddr = kmap_atomic(page);
493 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
494 flush_dcache_page(page);
495 kunmap_atomic(kaddr);
496 zero_user_segment(page, len, PAGE_SIZE);
497 SetPageUptodate(page);
498 brelse(iloc.bh);
499
500out:
501 return ret;
502}
503
504int ext4_readpage_inline(struct inode *inode, struct page *page)
505{
506 int ret = 0;
507
508 down_read(&EXT4_I(inode)->xattr_sem);
509 if (!ext4_has_inline_data(inode)) {
510 up_read(&EXT4_I(inode)->xattr_sem);
511 return -EAGAIN;
512 }
513
514 /*
515 * Current inline data can only exist in the 1st page,
516 * So for all the other pages, just set them uptodate.
517 */
518 if (!page->index)
519 ret = ext4_read_inline_page(inode, page);
520 else if (!PageUptodate(page)) {
521 zero_user_segment(page, 0, PAGE_SIZE);
522 SetPageUptodate(page);
523 }
524
525 up_read(&EXT4_I(inode)->xattr_sem);
526
527 unlock_page(page);
528 return ret >= 0 ? 0 : ret;
529}
530
531static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
532 struct inode *inode,
533 unsigned flags)
534{
535 int ret, needed_blocks, no_expand;
536 handle_t *handle = NULL;
537 int retries = 0, sem_held = 0;
538 struct page *page = NULL;
539 unsigned from, to;
540 struct ext4_iloc iloc;
541
542 if (!ext4_has_inline_data(inode)) {
543 /*
544 * clear the flag so that no new write
545 * will trap here again.
546 */
547 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
548 return 0;
549 }
550
551 needed_blocks = ext4_writepage_trans_blocks(inode);
552
553 ret = ext4_get_inode_loc(inode, &iloc);
554 if (ret)
555 return ret;
556
557retry:
558 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
559 if (IS_ERR(handle)) {
560 ret = PTR_ERR(handle);
561 handle = NULL;
562 goto out;
563 }
564
565 /* We cannot recurse into the filesystem as the transaction is already
566 * started */
567 flags |= AOP_FLAG_NOFS;
568
569 page = grab_cache_page_write_begin(mapping, 0, flags);
570 if (!page) {
571 ret = -ENOMEM;
572 goto out;
573 }
574
575 ext4_write_lock_xattr(inode, &no_expand);
576 sem_held = 1;
577 /* If some one has already done this for us, just exit. */
578 if (!ext4_has_inline_data(inode)) {
579 ret = 0;
580 goto out;
581 }
582
583 from = 0;
584 to = ext4_get_inline_size(inode);
585 if (!PageUptodate(page)) {
586 ret = ext4_read_inline_page(inode, page);
587 if (ret < 0)
588 goto out;
589 }
590
591 ret = ext4_destroy_inline_data_nolock(handle, inode);
592 if (ret)
593 goto out;
594
595 if (ext4_should_dioread_nolock(inode)) {
596 ret = __block_write_begin(page, from, to,
597 ext4_get_block_unwritten);
598 } else
599 ret = __block_write_begin(page, from, to, ext4_get_block);
600
601 if (!ret && ext4_should_journal_data(inode)) {
602 ret = ext4_walk_page_buffers(handle, page_buffers(page),
603 from, to, NULL,
604 do_journal_get_write_access);
605 }
606
607 if (ret) {
608 unlock_page(page);
609 put_page(page);
610 page = NULL;
611 ext4_orphan_add(handle, inode);
612 ext4_write_unlock_xattr(inode, &no_expand);
613 sem_held = 0;
614 ext4_journal_stop(handle);
615 handle = NULL;
616 ext4_truncate_failed_write(inode);
617 /*
618 * If truncate failed early the inode might
619 * still be on the orphan list; we need to
620 * make sure the inode is removed from the
621 * orphan list in that case.
622 */
623 if (inode->i_nlink)
624 ext4_orphan_del(NULL, inode);
625 }
626
627 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
628 goto retry;
629
630 if (page)
631 block_commit_write(page, from, to);
632out:
633 if (page) {
634 unlock_page(page);
635 put_page(page);
636 }
637 if (sem_held)
638 ext4_write_unlock_xattr(inode, &no_expand);
639 if (handle)
640 ext4_journal_stop(handle);
641 brelse(iloc.bh);
642 return ret;
643}
644
645/*
646 * Try to write data in the inode.
647 * If the inode has inline data, check whether the new write can be
648 * in the inode also. If not, create the page the handle, move the data
649 * to the page make it update and let the later codes create extent for it.
650 */
651int ext4_try_to_write_inline_data(struct address_space *mapping,
652 struct inode *inode,
653 loff_t pos, unsigned len,
654 unsigned flags,
655 struct page **pagep)
656{
657 int ret;
658 handle_t *handle;
659 struct page *page;
660 struct ext4_iloc iloc;
661
662 if (pos + len > ext4_get_max_inline_size(inode))
663 goto convert;
664
665 ret = ext4_get_inode_loc(inode, &iloc);
666 if (ret)
667 return ret;
668
669 /*
670 * The possible write could happen in the inode,
671 * so try to reserve the space in inode first.
672 */
673 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
674 if (IS_ERR(handle)) {
675 ret = PTR_ERR(handle);
676 handle = NULL;
677 goto out;
678 }
679
680 ret = ext4_prepare_inline_data(handle, inode, pos + len);
681 if (ret && ret != -ENOSPC)
682 goto out;
683
684 /* We don't have space in inline inode, so convert it to extent. */
685 if (ret == -ENOSPC) {
686 ext4_journal_stop(handle);
687 brelse(iloc.bh);
688 goto convert;
689 }
690
691 flags |= AOP_FLAG_NOFS;
692
693 page = grab_cache_page_write_begin(mapping, 0, flags);
694 if (!page) {
695 ret = -ENOMEM;
696 goto out;
697 }
698
699 *pagep = page;
700 down_read(&EXT4_I(inode)->xattr_sem);
701 if (!ext4_has_inline_data(inode)) {
702 ret = 0;
703 unlock_page(page);
704 put_page(page);
705 goto out_up_read;
706 }
707
708 if (!PageUptodate(page)) {
709 ret = ext4_read_inline_page(inode, page);
710 if (ret < 0)
711 goto out_up_read;
712 }
713
714 ret = 1;
715 handle = NULL;
716out_up_read:
717 up_read(&EXT4_I(inode)->xattr_sem);
718out:
719 if (handle)
720 ext4_journal_stop(handle);
721 brelse(iloc.bh);
722 return ret;
723convert:
724 return ext4_convert_inline_data_to_extent(mapping,
725 inode, flags);
726}
727
728int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
729 unsigned copied, struct page *page)
730{
731 int ret, no_expand;
732 void *kaddr;
733 struct ext4_iloc iloc;
734
735 if (unlikely(copied < len)) {
736 if (!PageUptodate(page)) {
737 copied = 0;
738 goto out;
739 }
740 }
741
742 ret = ext4_get_inode_loc(inode, &iloc);
743 if (ret) {
744 ext4_std_error(inode->i_sb, ret);
745 copied = 0;
746 goto out;
747 }
748
749 ext4_write_lock_xattr(inode, &no_expand);
750 BUG_ON(!ext4_has_inline_data(inode));
751
752 kaddr = kmap_atomic(page);
753 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
754 kunmap_atomic(kaddr);
755 SetPageUptodate(page);
756 /* clear page dirty so that writepages wouldn't work for us. */
757 ClearPageDirty(page);
758
759 ext4_write_unlock_xattr(inode, &no_expand);
760 brelse(iloc.bh);
761out:
762 return copied;
763}
764
765struct buffer_head *
766ext4_journalled_write_inline_data(struct inode *inode,
767 unsigned len,
768 struct page *page)
769{
770 int ret, no_expand;
771 void *kaddr;
772 struct ext4_iloc iloc;
773
774 ret = ext4_get_inode_loc(inode, &iloc);
775 if (ret) {
776 ext4_std_error(inode->i_sb, ret);
777 return NULL;
778 }
779
780 ext4_write_lock_xattr(inode, &no_expand);
781 kaddr = kmap_atomic(page);
782 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
783 kunmap_atomic(kaddr);
784 ext4_write_unlock_xattr(inode, &no_expand);
785
786 return iloc.bh;
787}
788
789/*
790 * Try to make the page cache and handle ready for the inline data case.
791 * We can call this function in 2 cases:
792 * 1. The inode is created and the first write exceeds inline size. We can
793 * clear the inode state safely.
794 * 2. The inode has inline data, then we need to read the data, make it
795 * update and dirty so that ext4_da_writepages can handle it. We don't
796 * need to start the journal since the file's metatdata isn't changed now.
797 */
798static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
799 struct inode *inode,
800 unsigned flags,
801 void **fsdata)
802{
803 int ret = 0, inline_size;
804 struct page *page;
805
806 page = grab_cache_page_write_begin(mapping, 0, flags);
807 if (!page)
808 return -ENOMEM;
809
810 down_read(&EXT4_I(inode)->xattr_sem);
811 if (!ext4_has_inline_data(inode)) {
812 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
813 goto out;
814 }
815
816 inline_size = ext4_get_inline_size(inode);
817
818 if (!PageUptodate(page)) {
819 ret = ext4_read_inline_page(inode, page);
820 if (ret < 0)
821 goto out;
822 }
823
824 ret = __block_write_begin(page, 0, inline_size,
825 ext4_da_get_block_prep);
826 if (ret) {
827 up_read(&EXT4_I(inode)->xattr_sem);
828 unlock_page(page);
829 put_page(page);
830 ext4_truncate_failed_write(inode);
831 return ret;
832 }
833
834 SetPageDirty(page);
835 SetPageUptodate(page);
836 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
837 *fsdata = (void *)CONVERT_INLINE_DATA;
838
839out:
840 up_read(&EXT4_I(inode)->xattr_sem);
841 if (page) {
842 unlock_page(page);
843 put_page(page);
844 }
845 return ret;
846}
847
848/*
849 * Prepare the write for the inline data.
850 * If the the data can be written into the inode, we just read
851 * the page and make it uptodate, and start the journal.
852 * Otherwise read the page, makes it dirty so that it can be
853 * handle in writepages(the i_disksize update is left to the
854 * normal ext4_da_write_end).
855 */
856int ext4_da_write_inline_data_begin(struct address_space *mapping,
857 struct inode *inode,
858 loff_t pos, unsigned len,
859 unsigned flags,
860 struct page **pagep,
861 void **fsdata)
862{
863 int ret, inline_size;
864 handle_t *handle;
865 struct page *page;
866 struct ext4_iloc iloc;
867 int retries;
868
869 ret = ext4_get_inode_loc(inode, &iloc);
870 if (ret)
871 return ret;
872
873retry_journal:
874 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
875 if (IS_ERR(handle)) {
876 ret = PTR_ERR(handle);
877 goto out;
878 }
879
880 inline_size = ext4_get_max_inline_size(inode);
881
882 ret = -ENOSPC;
883 if (inline_size >= pos + len) {
884 ret = ext4_prepare_inline_data(handle, inode, pos + len);
885 if (ret && ret != -ENOSPC)
886 goto out_journal;
887 }
888
889 /*
890 * We cannot recurse into the filesystem as the transaction
891 * is already started.
892 */
893 flags |= AOP_FLAG_NOFS;
894
895 if (ret == -ENOSPC) {
896 ret = ext4_da_convert_inline_data_to_extent(mapping,
897 inode,
898 flags,
899 fsdata);
900 ext4_journal_stop(handle);
901 if (ret == -ENOSPC &&
902 ext4_should_retry_alloc(inode->i_sb, &retries))
903 goto retry_journal;
904 goto out;
905 }
906
907
908 page = grab_cache_page_write_begin(mapping, 0, flags);
909 if (!page) {
910 ret = -ENOMEM;
911 goto out_journal;
912 }
913
914 down_read(&EXT4_I(inode)->xattr_sem);
915 if (!ext4_has_inline_data(inode)) {
916 ret = 0;
917 goto out_release_page;
918 }
919
920 if (!PageUptodate(page)) {
921 ret = ext4_read_inline_page(inode, page);
922 if (ret < 0)
923 goto out_release_page;
924 }
925
926 up_read(&EXT4_I(inode)->xattr_sem);
927 *pagep = page;
928 brelse(iloc.bh);
929 return 1;
930out_release_page:
931 up_read(&EXT4_I(inode)->xattr_sem);
932 unlock_page(page);
933 put_page(page);
934out_journal:
935 ext4_journal_stop(handle);
936out:
937 brelse(iloc.bh);
938 return ret;
939}
940
941int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
942 unsigned len, unsigned copied,
943 struct page *page)
944{
945 int i_size_changed = 0;
946 int ret;
947
948 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
949 if (ret < 0) {
950 unlock_page(page);
951 put_page(page);
952 return ret;
953 }
954 copied = ret;
955
956 /*
957 * No need to use i_size_read() here, the i_size
958 * cannot change under us because we hold i_mutex.
959 *
960 * But it's important to update i_size while still holding page lock:
961 * page writeout could otherwise come in and zero beyond i_size.
962 */
963 if (pos+copied > inode->i_size) {
964 i_size_write(inode, pos+copied);
965 i_size_changed = 1;
966 }
967 unlock_page(page);
968 put_page(page);
969
970 /*
971 * Don't mark the inode dirty under page lock. First, it unnecessarily
972 * makes the holding time of page lock longer. Second, it forces lock
973 * ordering of page lock and transaction start for journaling
974 * filesystems.
975 */
976 if (i_size_changed)
977 mark_inode_dirty(inode);
978
979 return copied;
980}
981
982#ifdef INLINE_DIR_DEBUG
983void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
984 void *inline_start, int inline_size)
985{
986 int offset;
987 unsigned short de_len;
988 struct ext4_dir_entry_2 *de = inline_start;
989 void *dlimit = inline_start + inline_size;
990
991 trace_printk("inode %lu\n", dir->i_ino);
992 offset = 0;
993 while ((void *)de < dlimit) {
994 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
995 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
996 offset, de_len, de->name_len, de->name,
997 de->name_len, le32_to_cpu(de->inode));
998 if (ext4_check_dir_entry(dir, NULL, de, bh,
999 inline_start, inline_size, offset))
1000 BUG();
1001
1002 offset += de_len;
1003 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1004 }
1005}
1006#else
1007#define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1008#endif
1009
1010/*
1011 * Add a new entry into a inline dir.
1012 * It will return -ENOSPC if no space is available, and -EIO
1013 * and -EEXIST if directory entry already exists.
1014 */
1015static int ext4_add_dirent_to_inline(handle_t *handle,
1016 struct ext4_filename *fname,
1017 struct inode *dir,
1018 struct inode *inode,
1019 struct ext4_iloc *iloc,
1020 void *inline_start, int inline_size)
1021{
1022 int err;
1023 struct ext4_dir_entry_2 *de;
1024
1025 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1026 inline_size, fname, &de);
1027 if (err)
1028 return err;
1029
1030 BUFFER_TRACE(iloc->bh, "get_write_access");
1031 err = ext4_journal_get_write_access(handle, iloc->bh);
1032 if (err)
1033 return err;
1034 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1035
1036 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1037
1038 /*
1039 * XXX shouldn't update any times until successful
1040 * completion of syscall, but too many callers depend
1041 * on this.
1042 *
1043 * XXX similarly, too many callers depend on
1044 * ext4_new_inode() setting the times, but error
1045 * recovery deletes the inode, so the worst that can
1046 * happen is that the times are slightly out of date
1047 * and/or different from the directory change time.
1048 */
1049 dir->i_mtime = dir->i_ctime = current_time(dir);
1050 ext4_update_dx_flag(dir);
1051 dir->i_version++;
1052 ext4_mark_inode_dirty(handle, dir);
1053 return 1;
1054}
1055
1056static void *ext4_get_inline_xattr_pos(struct inode *inode,
1057 struct ext4_iloc *iloc)
1058{
1059 struct ext4_xattr_entry *entry;
1060 struct ext4_xattr_ibody_header *header;
1061
1062 BUG_ON(!EXT4_I(inode)->i_inline_off);
1063
1064 header = IHDR(inode, ext4_raw_inode(iloc));
1065 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1066 EXT4_I(inode)->i_inline_off);
1067
1068 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1069}
1070
1071/* Set the final de to cover the whole block. */
1072static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1073{
1074 struct ext4_dir_entry_2 *de, *prev_de;
1075 void *limit;
1076 int de_len;
1077
1078 de = (struct ext4_dir_entry_2 *)de_buf;
1079 if (old_size) {
1080 limit = de_buf + old_size;
1081 do {
1082 prev_de = de;
1083 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1084 de_buf += de_len;
1085 de = (struct ext4_dir_entry_2 *)de_buf;
1086 } while (de_buf < limit);
1087
1088 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1089 old_size, new_size);
1090 } else {
1091 /* this is just created, so create an empty entry. */
1092 de->inode = 0;
1093 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1094 }
1095}
1096
1097static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1098 struct ext4_iloc *iloc)
1099{
1100 int ret;
1101 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1102 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1103
1104 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1105 return -ENOSPC;
1106
1107 ret = ext4_update_inline_data(handle, dir,
1108 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1109 if (ret)
1110 return ret;
1111
1112 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1113 EXT4_I(dir)->i_inline_size -
1114 EXT4_MIN_INLINE_DATA_SIZE);
1115 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1116 return 0;
1117}
1118
1119static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1120 struct ext4_iloc *iloc,
1121 void *buf, int inline_size)
1122{
1123 ext4_create_inline_data(handle, inode, inline_size);
1124 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1125 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1126}
1127
1128static int ext4_finish_convert_inline_dir(handle_t *handle,
1129 struct inode *inode,
1130 struct buffer_head *dir_block,
1131 void *buf,
1132 int inline_size)
1133{
1134 int err, csum_size = 0, header_size = 0;
1135 struct ext4_dir_entry_2 *de;
1136 struct ext4_dir_entry_tail *t;
1137 void *target = dir_block->b_data;
1138
1139 /*
1140 * First create "." and ".." and then copy the dir information
1141 * back to the block.
1142 */
1143 de = (struct ext4_dir_entry_2 *)target;
1144 de = ext4_init_dot_dotdot(inode, de,
1145 inode->i_sb->s_blocksize, csum_size,
1146 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1147 header_size = (void *)de - target;
1148
1149 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1150 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1151
1152 if (ext4_has_metadata_csum(inode->i_sb))
1153 csum_size = sizeof(struct ext4_dir_entry_tail);
1154
1155 inode->i_size = inode->i_sb->s_blocksize;
1156 i_size_write(inode, inode->i_sb->s_blocksize);
1157 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1158 ext4_update_final_de(dir_block->b_data,
1159 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1160 inode->i_sb->s_blocksize - csum_size);
1161
1162 if (csum_size) {
1163 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1164 inode->i_sb->s_blocksize);
1165 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1166 }
1167 set_buffer_uptodate(dir_block);
1168 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1169 if (err)
1170 return err;
1171 set_buffer_verified(dir_block);
1172 return ext4_mark_inode_dirty(handle, inode);
1173}
1174
1175static int ext4_convert_inline_data_nolock(handle_t *handle,
1176 struct inode *inode,
1177 struct ext4_iloc *iloc)
1178{
1179 int error;
1180 void *buf = NULL;
1181 struct buffer_head *data_bh = NULL;
1182 struct ext4_map_blocks map;
1183 int inline_size;
1184
1185 inline_size = ext4_get_inline_size(inode);
1186 buf = kmalloc(inline_size, GFP_NOFS);
1187 if (!buf) {
1188 error = -ENOMEM;
1189 goto out;
1190 }
1191
1192 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1193 if (error < 0)
1194 goto out;
1195
1196 /*
1197 * Make sure the inline directory entries pass checks before we try to
1198 * convert them, so that we avoid touching stuff that needs fsck.
1199 */
1200 if (S_ISDIR(inode->i_mode)) {
1201 error = ext4_check_all_de(inode, iloc->bh,
1202 buf + EXT4_INLINE_DOTDOT_SIZE,
1203 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1204 if (error)
1205 goto out;
1206 }
1207
1208 error = ext4_destroy_inline_data_nolock(handle, inode);
1209 if (error)
1210 goto out;
1211
1212 map.m_lblk = 0;
1213 map.m_len = 1;
1214 map.m_flags = 0;
1215 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1216 if (error < 0)
1217 goto out_restore;
1218 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1219 error = -EIO;
1220 goto out_restore;
1221 }
1222
1223 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1224 if (!data_bh) {
1225 error = -ENOMEM;
1226 goto out_restore;
1227 }
1228
1229 lock_buffer(data_bh);
1230 error = ext4_journal_get_create_access(handle, data_bh);
1231 if (error) {
1232 unlock_buffer(data_bh);
1233 error = -EIO;
1234 goto out_restore;
1235 }
1236 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1237
1238 if (!S_ISDIR(inode->i_mode)) {
1239 memcpy(data_bh->b_data, buf, inline_size);
1240 set_buffer_uptodate(data_bh);
1241 error = ext4_handle_dirty_metadata(handle,
1242 inode, data_bh);
1243 } else {
1244 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1245 buf, inline_size);
1246 }
1247
1248 unlock_buffer(data_bh);
1249out_restore:
1250 if (error)
1251 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1252
1253out:
1254 brelse(data_bh);
1255 kfree(buf);
1256 return error;
1257}
1258
1259/*
1260 * Try to add the new entry to the inline data.
1261 * If succeeds, return 0. If not, extended the inline dir and copied data to
1262 * the new created block.
1263 */
1264int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1265 struct inode *dir, struct inode *inode)
1266{
1267 int ret, inline_size, no_expand;
1268 void *inline_start;
1269 struct ext4_iloc iloc;
1270
1271 ret = ext4_get_inode_loc(dir, &iloc);
1272 if (ret)
1273 return ret;
1274
1275 ext4_write_lock_xattr(dir, &no_expand);
1276 if (!ext4_has_inline_data(dir))
1277 goto out;
1278
1279 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1280 EXT4_INLINE_DOTDOT_SIZE;
1281 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1282
1283 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1284 inline_start, inline_size);
1285 if (ret != -ENOSPC)
1286 goto out;
1287
1288 /* check whether it can be inserted to inline xattr space. */
1289 inline_size = EXT4_I(dir)->i_inline_size -
1290 EXT4_MIN_INLINE_DATA_SIZE;
1291 if (!inline_size) {
1292 /* Try to use the xattr space.*/
1293 ret = ext4_update_inline_dir(handle, dir, &iloc);
1294 if (ret && ret != -ENOSPC)
1295 goto out;
1296
1297 inline_size = EXT4_I(dir)->i_inline_size -
1298 EXT4_MIN_INLINE_DATA_SIZE;
1299 }
1300
1301 if (inline_size) {
1302 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1303
1304 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1305 inode, &iloc, inline_start,
1306 inline_size);
1307
1308 if (ret != -ENOSPC)
1309 goto out;
1310 }
1311
1312 /*
1313 * The inline space is filled up, so create a new block for it.
1314 * As the extent tree will be created, we have to save the inline
1315 * dir first.
1316 */
1317 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1318
1319out:
1320 ext4_mark_inode_dirty(handle, dir);
1321 ext4_write_unlock_xattr(dir, &no_expand);
1322 brelse(iloc.bh);
1323 return ret;
1324}
1325
1326/*
1327 * This function fills a red-black tree with information from an
1328 * inlined dir. It returns the number directory entries loaded
1329 * into the tree. If there is an error it is returned in err.
1330 */
1331int htree_inlinedir_to_tree(struct file *dir_file,
1332 struct inode *dir, ext4_lblk_t block,
1333 struct dx_hash_info *hinfo,
1334 __u32 start_hash, __u32 start_minor_hash,
1335 int *has_inline_data)
1336{
1337 int err = 0, count = 0;
1338 unsigned int parent_ino;
1339 int pos;
1340 struct ext4_dir_entry_2 *de;
1341 struct inode *inode = file_inode(dir_file);
1342 int ret, inline_size = 0;
1343 struct ext4_iloc iloc;
1344 void *dir_buf = NULL;
1345 struct ext4_dir_entry_2 fake;
1346 struct fscrypt_str tmp_str;
1347
1348 ret = ext4_get_inode_loc(inode, &iloc);
1349 if (ret)
1350 return ret;
1351
1352 down_read(&EXT4_I(inode)->xattr_sem);
1353 if (!ext4_has_inline_data(inode)) {
1354 up_read(&EXT4_I(inode)->xattr_sem);
1355 *has_inline_data = 0;
1356 goto out;
1357 }
1358
1359 inline_size = ext4_get_inline_size(inode);
1360 dir_buf = kmalloc(inline_size, GFP_NOFS);
1361 if (!dir_buf) {
1362 ret = -ENOMEM;
1363 up_read(&EXT4_I(inode)->xattr_sem);
1364 goto out;
1365 }
1366
1367 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1368 up_read(&EXT4_I(inode)->xattr_sem);
1369 if (ret < 0)
1370 goto out;
1371
1372 pos = 0;
1373 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1374 while (pos < inline_size) {
1375 /*
1376 * As inlined dir doesn't store any information about '.' and
1377 * only the inode number of '..' is stored, we have to handle
1378 * them differently.
1379 */
1380 if (pos == 0) {
1381 fake.inode = cpu_to_le32(inode->i_ino);
1382 fake.name_len = 1;
1383 strcpy(fake.name, ".");
1384 fake.rec_len = ext4_rec_len_to_disk(
1385 EXT4_DIR_REC_LEN(fake.name_len),
1386 inline_size);
1387 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1388 de = &fake;
1389 pos = EXT4_INLINE_DOTDOT_OFFSET;
1390 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1391 fake.inode = cpu_to_le32(parent_ino);
1392 fake.name_len = 2;
1393 strcpy(fake.name, "..");
1394 fake.rec_len = ext4_rec_len_to_disk(
1395 EXT4_DIR_REC_LEN(fake.name_len),
1396 inline_size);
1397 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1398 de = &fake;
1399 pos = EXT4_INLINE_DOTDOT_SIZE;
1400 } else {
1401 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1402 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1403 if (ext4_check_dir_entry(inode, dir_file, de,
1404 iloc.bh, dir_buf,
1405 inline_size, pos)) {
1406 ret = count;
1407 goto out;
1408 }
1409 }
1410
1411 ext4fs_dirhash(de->name, de->name_len, hinfo);
1412 if ((hinfo->hash < start_hash) ||
1413 ((hinfo->hash == start_hash) &&
1414 (hinfo->minor_hash < start_minor_hash)))
1415 continue;
1416 if (de->inode == 0)
1417 continue;
1418 tmp_str.name = de->name;
1419 tmp_str.len = de->name_len;
1420 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1421 hinfo->minor_hash, de, &tmp_str);
1422 if (err) {
1423 count = err;
1424 goto out;
1425 }
1426 count++;
1427 }
1428 ret = count;
1429out:
1430 kfree(dir_buf);
1431 brelse(iloc.bh);
1432 return ret;
1433}
1434
1435/*
1436 * So this function is called when the volume is mkfsed with
1437 * dir_index disabled. In order to keep f_pos persistent
1438 * after we convert from an inlined dir to a blocked based,
1439 * we just pretend that we are a normal dir and return the
1440 * offset as if '.' and '..' really take place.
1441 *
1442 */
1443int ext4_read_inline_dir(struct file *file,
1444 struct dir_context *ctx,
1445 int *has_inline_data)
1446{
1447 unsigned int offset, parent_ino;
1448 int i;
1449 struct ext4_dir_entry_2 *de;
1450 struct super_block *sb;
1451 struct inode *inode = file_inode(file);
1452 int ret, inline_size = 0;
1453 struct ext4_iloc iloc;
1454 void *dir_buf = NULL;
1455 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1456
1457 ret = ext4_get_inode_loc(inode, &iloc);
1458 if (ret)
1459 return ret;
1460
1461 down_read(&EXT4_I(inode)->xattr_sem);
1462 if (!ext4_has_inline_data(inode)) {
1463 up_read(&EXT4_I(inode)->xattr_sem);
1464 *has_inline_data = 0;
1465 goto out;
1466 }
1467
1468 inline_size = ext4_get_inline_size(inode);
1469 dir_buf = kmalloc(inline_size, GFP_NOFS);
1470 if (!dir_buf) {
1471 ret = -ENOMEM;
1472 up_read(&EXT4_I(inode)->xattr_sem);
1473 goto out;
1474 }
1475
1476 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1477 up_read(&EXT4_I(inode)->xattr_sem);
1478 if (ret < 0)
1479 goto out;
1480
1481 ret = 0;
1482 sb = inode->i_sb;
1483 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1484 offset = ctx->pos;
1485
1486 /*
1487 * dotdot_offset and dotdot_size is the real offset and
1488 * size for ".." and "." if the dir is block based while
1489 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1490 * So we will use extra_offset and extra_size to indicate them
1491 * during the inline dir iteration.
1492 */
1493 dotdot_offset = EXT4_DIR_REC_LEN(1);
1494 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1495 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1496 extra_size = extra_offset + inline_size;
1497
1498 /*
1499 * If the version has changed since the last call to
1500 * readdir(2), then we might be pointing to an invalid
1501 * dirent right now. Scan from the start of the inline
1502 * dir to make sure.
1503 */
1504 if (file->f_version != inode->i_version) {
1505 for (i = 0; i < extra_size && i < offset;) {
1506 /*
1507 * "." is with offset 0 and
1508 * ".." is dotdot_offset.
1509 */
1510 if (!i) {
1511 i = dotdot_offset;
1512 continue;
1513 } else if (i == dotdot_offset) {
1514 i = dotdot_size;
1515 continue;
1516 }
1517 /* for other entry, the real offset in
1518 * the buf has to be tuned accordingly.
1519 */
1520 de = (struct ext4_dir_entry_2 *)
1521 (dir_buf + i - extra_offset);
1522 /* It's too expensive to do a full
1523 * dirent test each time round this
1524 * loop, but we do have to test at
1525 * least that it is non-zero. A
1526 * failure will be detected in the
1527 * dirent test below. */
1528 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1529 < EXT4_DIR_REC_LEN(1))
1530 break;
1531 i += ext4_rec_len_from_disk(de->rec_len,
1532 extra_size);
1533 }
1534 offset = i;
1535 ctx->pos = offset;
1536 file->f_version = inode->i_version;
1537 }
1538
1539 while (ctx->pos < extra_size) {
1540 if (ctx->pos == 0) {
1541 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1542 goto out;
1543 ctx->pos = dotdot_offset;
1544 continue;
1545 }
1546
1547 if (ctx->pos == dotdot_offset) {
1548 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1549 goto out;
1550 ctx->pos = dotdot_size;
1551 continue;
1552 }
1553
1554 de = (struct ext4_dir_entry_2 *)
1555 (dir_buf + ctx->pos - extra_offset);
1556 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1557 extra_size, ctx->pos))
1558 goto out;
1559 if (le32_to_cpu(de->inode)) {
1560 if (!dir_emit(ctx, de->name, de->name_len,
1561 le32_to_cpu(de->inode),
1562 get_dtype(sb, de->file_type)))
1563 goto out;
1564 }
1565 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1566 }
1567out:
1568 kfree(dir_buf);
1569 brelse(iloc.bh);
1570 return ret;
1571}
1572
1573struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1574 struct ext4_dir_entry_2 **parent_de,
1575 int *retval)
1576{
1577 struct ext4_iloc iloc;
1578
1579 *retval = ext4_get_inode_loc(inode, &iloc);
1580 if (*retval)
1581 return NULL;
1582
1583 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1584
1585 return iloc.bh;
1586}
1587
1588/*
1589 * Try to create the inline data for the new dir.
1590 * If it succeeds, return 0, otherwise return the error.
1591 * In case of ENOSPC, the caller should create the normal disk layout dir.
1592 */
1593int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1594 struct inode *inode)
1595{
1596 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1597 struct ext4_iloc iloc;
1598 struct ext4_dir_entry_2 *de;
1599
1600 ret = ext4_get_inode_loc(inode, &iloc);
1601 if (ret)
1602 return ret;
1603
1604 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1605 if (ret)
1606 goto out;
1607
1608 /*
1609 * For inline dir, we only save the inode information for the ".."
1610 * and create a fake dentry to cover the left space.
1611 */
1612 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1613 de->inode = cpu_to_le32(parent->i_ino);
1614 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1615 de->inode = 0;
1616 de->rec_len = ext4_rec_len_to_disk(
1617 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1618 inline_size);
1619 set_nlink(inode, 2);
1620 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1621out:
1622 brelse(iloc.bh);
1623 return ret;
1624}
1625
1626struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1627 struct ext4_filename *fname,
1628 const struct qstr *d_name,
1629 struct ext4_dir_entry_2 **res_dir,
1630 int *has_inline_data)
1631{
1632 int ret;
1633 struct ext4_iloc iloc;
1634 void *inline_start;
1635 int inline_size;
1636
1637 if (ext4_get_inode_loc(dir, &iloc))
1638 return NULL;
1639
1640 down_read(&EXT4_I(dir)->xattr_sem);
1641 if (!ext4_has_inline_data(dir)) {
1642 *has_inline_data = 0;
1643 goto out;
1644 }
1645
1646 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1647 EXT4_INLINE_DOTDOT_SIZE;
1648 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1649 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1650 dir, fname, d_name, 0, res_dir);
1651 if (ret == 1)
1652 goto out_find;
1653 if (ret < 0)
1654 goto out;
1655
1656 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1657 goto out;
1658
1659 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1660 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1661
1662 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1663 dir, fname, d_name, 0, res_dir);
1664 if (ret == 1)
1665 goto out_find;
1666
1667out:
1668 brelse(iloc.bh);
1669 iloc.bh = NULL;
1670out_find:
1671 up_read(&EXT4_I(dir)->xattr_sem);
1672 return iloc.bh;
1673}
1674
1675int ext4_delete_inline_entry(handle_t *handle,
1676 struct inode *dir,
1677 struct ext4_dir_entry_2 *de_del,
1678 struct buffer_head *bh,
1679 int *has_inline_data)
1680{
1681 int err, inline_size, no_expand;
1682 struct ext4_iloc iloc;
1683 void *inline_start;
1684
1685 err = ext4_get_inode_loc(dir, &iloc);
1686 if (err)
1687 return err;
1688
1689 ext4_write_lock_xattr(dir, &no_expand);
1690 if (!ext4_has_inline_data(dir)) {
1691 *has_inline_data = 0;
1692 goto out;
1693 }
1694
1695 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1696 EXT4_MIN_INLINE_DATA_SIZE) {
1697 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1698 EXT4_INLINE_DOTDOT_SIZE;
1699 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1700 EXT4_INLINE_DOTDOT_SIZE;
1701 } else {
1702 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1703 inline_size = ext4_get_inline_size(dir) -
1704 EXT4_MIN_INLINE_DATA_SIZE;
1705 }
1706
1707 BUFFER_TRACE(bh, "get_write_access");
1708 err = ext4_journal_get_write_access(handle, bh);
1709 if (err)
1710 goto out;
1711
1712 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1713 inline_start, inline_size, 0);
1714 if (err)
1715 goto out;
1716
1717 err = ext4_mark_inode_dirty(handle, dir);
1718 if (unlikely(err))
1719 goto out;
1720
1721 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1722out:
1723 ext4_write_unlock_xattr(dir, &no_expand);
1724 brelse(iloc.bh);
1725 if (err != -ENOENT)
1726 ext4_std_error(dir->i_sb, err);
1727 return err;
1728}
1729
1730/*
1731 * Get the inline dentry at offset.
1732 */
1733static inline struct ext4_dir_entry_2 *
1734ext4_get_inline_entry(struct inode *inode,
1735 struct ext4_iloc *iloc,
1736 unsigned int offset,
1737 void **inline_start,
1738 int *inline_size)
1739{
1740 void *inline_pos;
1741
1742 BUG_ON(offset > ext4_get_inline_size(inode));
1743
1744 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1745 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1746 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1747 } else {
1748 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1749 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1750 *inline_size = ext4_get_inline_size(inode) -
1751 EXT4_MIN_INLINE_DATA_SIZE;
1752 }
1753
1754 if (inline_start)
1755 *inline_start = inline_pos;
1756 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1757}
1758
1759bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1760{
1761 int err, inline_size;
1762 struct ext4_iloc iloc;
1763 void *inline_pos;
1764 unsigned int offset;
1765 struct ext4_dir_entry_2 *de;
1766 bool ret = true;
1767
1768 err = ext4_get_inode_loc(dir, &iloc);
1769 if (err) {
1770 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1771 err, dir->i_ino);
1772 return true;
1773 }
1774
1775 down_read(&EXT4_I(dir)->xattr_sem);
1776 if (!ext4_has_inline_data(dir)) {
1777 *has_inline_data = 0;
1778 goto out;
1779 }
1780
1781 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1782 if (!le32_to_cpu(de->inode)) {
1783 ext4_warning(dir->i_sb,
1784 "bad inline directory (dir #%lu) - no `..'",
1785 dir->i_ino);
1786 ret = true;
1787 goto out;
1788 }
1789
1790 offset = EXT4_INLINE_DOTDOT_SIZE;
1791 while (offset < dir->i_size) {
1792 de = ext4_get_inline_entry(dir, &iloc, offset,
1793 &inline_pos, &inline_size);
1794 if (ext4_check_dir_entry(dir, NULL, de,
1795 iloc.bh, inline_pos,
1796 inline_size, offset)) {
1797 ext4_warning(dir->i_sb,
1798 "bad inline directory (dir #%lu) - "
1799 "inode %u, rec_len %u, name_len %d"
1800 "inline size %d",
1801 dir->i_ino, le32_to_cpu(de->inode),
1802 le16_to_cpu(de->rec_len), de->name_len,
1803 inline_size);
1804 ret = true;
1805 goto out;
1806 }
1807 if (le32_to_cpu(de->inode)) {
1808 ret = false;
1809 goto out;
1810 }
1811 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1812 }
1813
1814out:
1815 up_read(&EXT4_I(dir)->xattr_sem);
1816 brelse(iloc.bh);
1817 return ret;
1818}
1819
1820int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1821{
1822 int ret, no_expand;
1823
1824 ext4_write_lock_xattr(inode, &no_expand);
1825 ret = ext4_destroy_inline_data_nolock(handle, inode);
1826 ext4_write_unlock_xattr(inode, &no_expand);
1827
1828 return ret;
1829}
1830
1831int ext4_inline_data_fiemap(struct inode *inode,
1832 struct fiemap_extent_info *fieinfo,
1833 int *has_inline, __u64 start, __u64 len)
1834{
1835 __u64 physical = 0;
1836 __u64 inline_len;
1837 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1838 FIEMAP_EXTENT_LAST;
1839 int error = 0;
1840 struct ext4_iloc iloc;
1841
1842 down_read(&EXT4_I(inode)->xattr_sem);
1843 if (!ext4_has_inline_data(inode)) {
1844 *has_inline = 0;
1845 goto out;
1846 }
1847 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1848 i_size_read(inode));
1849 if (start >= inline_len)
1850 goto out;
1851 if (start + len < inline_len)
1852 inline_len = start + len;
1853 inline_len -= start;
1854
1855 error = ext4_get_inode_loc(inode, &iloc);
1856 if (error)
1857 goto out;
1858
1859 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1860 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1861 physical += offsetof(struct ext4_inode, i_block);
1862
1863 if (physical)
1864 error = fiemap_fill_next_extent(fieinfo, start, physical,
1865 inline_len, flags);
1866 brelse(iloc.bh);
1867out:
1868 up_read(&EXT4_I(inode)->xattr_sem);
1869 return (error < 0 ? error : 0);
1870}
1871
1872/*
1873 * Called during xattr set, and if we can sparse space 'needed',
1874 * just create the extent tree evict the data to the outer block.
1875 *
1876 * We use jbd2 instead of page cache to move data to the 1st block
1877 * so that the whole transaction can be committed as a whole and
1878 * the data isn't lost because of the delayed page cache write.
1879 */
1880int ext4_try_to_evict_inline_data(handle_t *handle,
1881 struct inode *inode,
1882 int needed)
1883{
1884 int error;
1885 struct ext4_xattr_entry *entry;
1886 struct ext4_inode *raw_inode;
1887 struct ext4_iloc iloc;
1888
1889 error = ext4_get_inode_loc(inode, &iloc);
1890 if (error)
1891 return error;
1892
1893 raw_inode = ext4_raw_inode(&iloc);
1894 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1895 EXT4_I(inode)->i_inline_off);
1896 if (EXT4_XATTR_LEN(entry->e_name_len) +
1897 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1898 error = -ENOSPC;
1899 goto out;
1900 }
1901
1902 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1903out:
1904 brelse(iloc.bh);
1905 return error;
1906}
1907
1908void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1909{
1910 handle_t *handle;
1911 int inline_size, value_len, needed_blocks, no_expand;
1912 size_t i_size;
1913 void *value = NULL;
1914 struct ext4_xattr_ibody_find is = {
1915 .s = { .not_found = -ENODATA, },
1916 };
1917 struct ext4_xattr_info i = {
1918 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1919 .name = EXT4_XATTR_SYSTEM_DATA,
1920 };
1921
1922
1923 needed_blocks = ext4_writepage_trans_blocks(inode);
1924 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1925 if (IS_ERR(handle))
1926 return;
1927
1928 ext4_write_lock_xattr(inode, &no_expand);
1929 if (!ext4_has_inline_data(inode)) {
1930 *has_inline = 0;
1931 ext4_journal_stop(handle);
1932 return;
1933 }
1934
1935 if (ext4_orphan_add(handle, inode))
1936 goto out;
1937
1938 if (ext4_get_inode_loc(inode, &is.iloc))
1939 goto out;
1940
1941 down_write(&EXT4_I(inode)->i_data_sem);
1942 i_size = inode->i_size;
1943 inline_size = ext4_get_inline_size(inode);
1944 EXT4_I(inode)->i_disksize = i_size;
1945
1946 if (i_size < inline_size) {
1947 /* Clear the content in the xattr space. */
1948 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1949 if (ext4_xattr_ibody_find(inode, &i, &is))
1950 goto out_error;
1951
1952 BUG_ON(is.s.not_found);
1953
1954 value_len = le32_to_cpu(is.s.here->e_value_size);
1955 value = kmalloc(value_len, GFP_NOFS);
1956 if (!value)
1957 goto out_error;
1958
1959 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1960 value, value_len))
1961 goto out_error;
1962
1963 i.value = value;
1964 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1965 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1966 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1967 goto out_error;
1968 }
1969
1970 /* Clear the content within i_blocks. */
1971 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1972 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1973 memset(p + i_size, 0,
1974 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1975 }
1976
1977 EXT4_I(inode)->i_inline_size = i_size <
1978 EXT4_MIN_INLINE_DATA_SIZE ?
1979 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1980 }
1981
1982out_error:
1983 up_write(&EXT4_I(inode)->i_data_sem);
1984out:
1985 brelse(is.iloc.bh);
1986 ext4_write_unlock_xattr(inode, &no_expand);
1987 kfree(value);
1988 if (inode->i_nlink)
1989 ext4_orphan_del(handle, inode);
1990
1991 inode->i_mtime = inode->i_ctime = current_time(inode);
1992 ext4_mark_inode_dirty(handle, inode);
1993 if (IS_SYNC(inode))
1994 ext4_handle_sync(handle);
1995
1996 ext4_journal_stop(handle);
1997 return;
1998}
1999
2000int ext4_convert_inline_data(struct inode *inode)
2001{
2002 int error, needed_blocks, no_expand;
2003 handle_t *handle;
2004 struct ext4_iloc iloc;
2005
2006 if (!ext4_has_inline_data(inode)) {
2007 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2008 return 0;
2009 }
2010
2011 needed_blocks = ext4_writepage_trans_blocks(inode);
2012
2013 iloc.bh = NULL;
2014 error = ext4_get_inode_loc(inode, &iloc);
2015 if (error)
2016 return error;
2017
2018 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2019 if (IS_ERR(handle)) {
2020 error = PTR_ERR(handle);
2021 goto out_free;
2022 }
2023
2024 ext4_write_lock_xattr(inode, &no_expand);
2025 if (ext4_has_inline_data(inode))
2026 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2027 ext4_write_unlock_xattr(inode, &no_expand);
2028 ext4_journal_stop(handle);
2029out_free:
2030 brelse(iloc.bh);
2031 return error;
2032}