projects / GitHub / LineageOS / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge 4.14.94 into android-4.14-p
[GitHub/LineageOS/android_kernel_motorola_exynos9610.git] / fs / ext4 / inline.c
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 #include <trace/events/android_fs.h>
22
23 #define EXT4_XATTR_SYSTEM_DATA "data"
24 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
25 #define EXT4_INLINE_DOTDOT_OFFSET 2
26 #define EXT4_INLINE_DOTDOT_SIZE 4
27
28 static int ext4_get_inline_size(struct inode *inode)
29 {
30 if (EXT4_I(inode)->i_inline_off)
31 return EXT4_I(inode)->i_inline_size;
32
33 return 0;
34 }
35
36 static int get_max_inline_xattr_value_size(struct inode *inode,
37 struct ext4_iloc *iloc)
38 {
39 struct ext4_xattr_ibody_header *header;
40 struct ext4_xattr_entry *entry;
41 struct ext4_inode *raw_inode;
42 int free, min_offs;
43
44 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
45 EXT4_GOOD_OLD_INODE_SIZE -
46 EXT4_I(inode)->i_extra_isize -
47 sizeof(struct ext4_xattr_ibody_header);
48
49 /*
50 * We need to subtract another sizeof(__u32) since an in-inode xattr
51 * needs an empty 4 bytes to indicate the gap between the xattr entry
52 * and the name/value pair.
53 */
54 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
55 return EXT4_XATTR_SIZE(min_offs -
56 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
57 EXT4_XATTR_ROUND - sizeof(__u32));
58
59 raw_inode = ext4_raw_inode(iloc);
60 header = IHDR(inode, raw_inode);
61 entry = IFIRST(header);
62
63 /* Compute min_offs. */
64 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
65 if (!entry->e_value_inum && entry->e_value_size) {
66 size_t offs = le16_to_cpu(entry->e_value_offs);
67 if (offs < min_offs)
68 min_offs = offs;
69 }
70 }
71 free = min_offs -
72 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
73
74 if (EXT4_I(inode)->i_inline_off) {
75 entry = (struct ext4_xattr_entry *)
76 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
77
78 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
79 goto out;
80 }
81
82 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
83
84 if (free > EXT4_XATTR_ROUND)
85 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
86 else
87 free = 0;
88
89 out:
90 return free;
91 }
92
93 /*
94 * Get the maximum size we now can store in an inode.
95 * If we can't find the space for a xattr entry, don't use the space
96 * of the extents since we have no space to indicate the inline data.
97 */
98 int ext4_get_max_inline_size(struct inode *inode)
99 {
100 int error, max_inline_size;
101 struct ext4_iloc iloc;
102
103 if (EXT4_I(inode)->i_extra_isize == 0)
104 return 0;
105
106 error = ext4_get_inode_loc(inode, &iloc);
107 if (error) {
108 ext4_error_inode(inode, __func__, __LINE__, 0,
109 "can't get inode location %lu",
110 inode->i_ino);
111 return 0;
112 }
113
114 down_read(&EXT4_I(inode)->xattr_sem);
115 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
116 up_read(&EXT4_I(inode)->xattr_sem);
117
118 brelse(iloc.bh);
119
120 if (!max_inline_size)
121 return 0;
122
123 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
124 }
125
126 /*
127 * this function does not take xattr_sem, which is OK because it is
128 * currently only used in a code path coming form ext4_iget, before
129 * the new inode has been unlocked
130 */
131 int ext4_find_inline_data_nolock(struct inode *inode)
132 {
133 struct ext4_xattr_ibody_find is = {
134 .s = { .not_found = -ENODATA, },
135 };
136 struct ext4_xattr_info i = {
137 .name_index = EXT4_XATTR_INDEX_SYSTEM,
138 .name = EXT4_XATTR_SYSTEM_DATA,
139 };
140 int error;
141
142 if (EXT4_I(inode)->i_extra_isize == 0)
143 return 0;
144
145 error = ext4_get_inode_loc(inode, &is.iloc);
146 if (error)
147 return error;
148
149 error = ext4_xattr_ibody_find(inode, &i, &is);
150 if (error)
151 goto out;
152
153 if (!is.s.not_found) {
154 if (is.s.here->e_value_inum) {
155 EXT4_ERROR_INODE(inode, "inline data xattr refers "
156 "to an external xattr inode");
157 error = -EFSCORRUPTED;
158 goto out;
159 }
160 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
161 (void *)ext4_raw_inode(&is.iloc));
162 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
163 le32_to_cpu(is.s.here->e_value_size);
164 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
165 }
166 out:
167 brelse(is.iloc.bh);
168 return error;
169 }
170
171 static int ext4_read_inline_data(struct inode *inode, void *buffer,
172 unsigned int len,
173 struct ext4_iloc *iloc)
174 {
175 struct ext4_xattr_entry *entry;
176 struct ext4_xattr_ibody_header *header;
177 int cp_len = 0;
178 struct ext4_inode *raw_inode;
179
180 if (!len)
181 return 0;
182
183 BUG_ON(len > EXT4_I(inode)->i_inline_size);
184
185 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
186 len : EXT4_MIN_INLINE_DATA_SIZE;
187
188 raw_inode = ext4_raw_inode(iloc);
189 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
190
191 len -= cp_len;
192 buffer += cp_len;
193
194 if (!len)
195 goto out;
196
197 header = IHDR(inode, raw_inode);
198 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
199 EXT4_I(inode)->i_inline_off);
200 len = min_t(unsigned int, len,
201 (unsigned int)le32_to_cpu(entry->e_value_size));
202
203 memcpy(buffer,
204 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
205 cp_len += len;
206
207 out:
208 return cp_len;
209 }
210
211 /*
212 * write the buffer to the inline inode.
213 * If 'create' is set, we don't need to do the extra copy in the xattr
214 * value since it is already handled by ext4_xattr_ibody_inline_set.
215 * That saves us one memcpy.
216 */
217 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
218 void *buffer, loff_t pos, unsigned int len)
219 {
220 struct ext4_xattr_entry *entry;
221 struct ext4_xattr_ibody_header *header;
222 struct ext4_inode *raw_inode;
223 int cp_len = 0;
224
225 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
226 return;
227
228 BUG_ON(!EXT4_I(inode)->i_inline_off);
229 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
230
231 raw_inode = ext4_raw_inode(iloc);
232 buffer += pos;
233
234 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
235 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
236 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
237 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
238
239 len -= cp_len;
240 buffer += cp_len;
241 pos += cp_len;
242 }
243
244 if (!len)
245 return;
246
247 pos -= EXT4_MIN_INLINE_DATA_SIZE;
248 header = IHDR(inode, raw_inode);
249 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
250 EXT4_I(inode)->i_inline_off);
251
252 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
253 buffer, len);
254 }
255
256 static int ext4_create_inline_data(handle_t *handle,
257 struct inode *inode, unsigned len)
258 {
259 int error;
260 void *value = NULL;
261 struct ext4_xattr_ibody_find is = {
262 .s = { .not_found = -ENODATA, },
263 };
264 struct ext4_xattr_info i = {
265 .name_index = EXT4_XATTR_INDEX_SYSTEM,
266 .name = EXT4_XATTR_SYSTEM_DATA,
267 };
268
269 error = ext4_get_inode_loc(inode, &is.iloc);
270 if (error)
271 return error;
272
273 BUFFER_TRACE(is.iloc.bh, "get_write_access");
274 error = ext4_journal_get_write_access(handle, is.iloc.bh);
275 if (error)
276 goto out;
277
278 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
279 value = EXT4_ZERO_XATTR_VALUE;
280 len -= EXT4_MIN_INLINE_DATA_SIZE;
281 } else {
282 value = "";
283 len = 0;
284 }
285
286 /* Insert the the xttr entry. */
287 i.value = value;
288 i.value_len = len;
289
290 error = ext4_xattr_ibody_find(inode, &i, &is);
291 if (error)
292 goto out;
293
294 BUG_ON(!is.s.not_found);
295
296 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
297 if (error) {
298 if (error == -ENOSPC)
299 ext4_clear_inode_state(inode,
300 EXT4_STATE_MAY_INLINE_DATA);
301 goto out;
302 }
303
304 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
305 0, EXT4_MIN_INLINE_DATA_SIZE);
306
307 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
308 (void *)ext4_raw_inode(&is.iloc));
309 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
310 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
311 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
312 get_bh(is.iloc.bh);
313 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
314
315 out:
316 brelse(is.iloc.bh);
317 return error;
318 }
319
320 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
321 unsigned int len)
322 {
323 int error;
324 void *value = NULL;
325 struct ext4_xattr_ibody_find is = {
326 .s = { .not_found = -ENODATA, },
327 };
328 struct ext4_xattr_info i = {
329 .name_index = EXT4_XATTR_INDEX_SYSTEM,
330 .name = EXT4_XATTR_SYSTEM_DATA,
331 };
332
333 /* If the old space is ok, write the data directly. */
334 if (len <= EXT4_I(inode)->i_inline_size)
335 return 0;
336
337 error = ext4_get_inode_loc(inode, &is.iloc);
338 if (error)
339 return error;
340
341 error = ext4_xattr_ibody_find(inode, &i, &is);
342 if (error)
343 goto out;
344
345 BUG_ON(is.s.not_found);
346
347 len -= EXT4_MIN_INLINE_DATA_SIZE;
348 value = kzalloc(len, GFP_NOFS);
349 if (!value) {
350 error = -ENOMEM;
351 goto out;
352 }
353
354 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
355 value, len);
356 if (error == -ENODATA)
357 goto out;
358
359 BUFFER_TRACE(is.iloc.bh, "get_write_access");
360 error = ext4_journal_get_write_access(handle, is.iloc.bh);
361 if (error)
362 goto out;
363
364 /* Update the xttr entry. */
365 i.value = value;
366 i.value_len = len;
367
368 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
369 if (error)
370 goto out;
371
372 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
373 (void *)ext4_raw_inode(&is.iloc));
374 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
375 le32_to_cpu(is.s.here->e_value_size);
376 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
377 get_bh(is.iloc.bh);
378 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
379
380 out:
381 kfree(value);
382 brelse(is.iloc.bh);
383 return error;
384 }
385
386 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
387 unsigned int len)
388 {
389 int ret, size, no_expand;
390 struct ext4_inode_info *ei = EXT4_I(inode);
391
392 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
393 return -ENOSPC;
394
395 size = ext4_get_max_inline_size(inode);
396 if (size < len)
397 return -ENOSPC;
398
399 ext4_write_lock_xattr(inode, &no_expand);
400
401 if (ei->i_inline_off)
402 ret = ext4_update_inline_data(handle, inode, len);
403 else
404 ret = ext4_create_inline_data(handle, inode, len);
405
406 ext4_write_unlock_xattr(inode, &no_expand);
407 return ret;
408 }
409
410 static int ext4_destroy_inline_data_nolock(handle_t *handle,
411 struct inode *inode)
412 {
413 struct ext4_inode_info *ei = EXT4_I(inode);
414 struct ext4_xattr_ibody_find is = {
415 .s = { .not_found = 0, },
416 };
417 struct ext4_xattr_info i = {
418 .name_index = EXT4_XATTR_INDEX_SYSTEM,
419 .name = EXT4_XATTR_SYSTEM_DATA,
420 .value = NULL,
421 .value_len = 0,
422 };
423 int error;
424
425 if (!ei->i_inline_off)
426 return 0;
427
428 error = ext4_get_inode_loc(inode, &is.iloc);
429 if (error)
430 return error;
431
432 error = ext4_xattr_ibody_find(inode, &i, &is);
433 if (error)
434 goto out;
435
436 BUFFER_TRACE(is.iloc.bh, "get_write_access");
437 error = ext4_journal_get_write_access(handle, is.iloc.bh);
438 if (error)
439 goto out;
440
441 error = ext4_xattr_ibody_inline_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);
464 out:
465 brelse(is.iloc.bh);
466 if (error == -ENODATA)
467 error = 0;
468 return error;
469 }
470
471 static 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
501 out:
502 return ret;
503 }
504
505 int 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 if (trace_android_fs_dataread_start_enabled()) {
516 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
517
518 path = android_fstrace_get_pathname(pathbuf,
519 MAX_TRACE_PATHBUF_LEN,
520 inode);
521 trace_android_fs_dataread_start(inode, page_offset(page),
522 PAGE_SIZE, current->pid,
523 path, current->comm);
524 }
525
526 /*
527 * Current inline data can only exist in the 1st page,
528 * So for all the other pages, just set them uptodate.
529 */
530 if (!page->index)
531 ret = ext4_read_inline_page(inode, page);
532 else if (!PageUptodate(page)) {
533 zero_user_segment(page, 0, PAGE_SIZE);
534 SetPageUptodate(page);
535 }
536
537 trace_android_fs_dataread_end(inode, page_offset(page), PAGE_SIZE);
538
539 up_read(&EXT4_I(inode)->xattr_sem);
540
541 unlock_page(page);
542 return ret >= 0 ? 0 : ret;
543 }
544
545 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
546 struct inode *inode,
547 unsigned flags)
548 {
549 int ret, needed_blocks, no_expand;
550 handle_t *handle = NULL;
551 int retries = 0, sem_held = 0;
552 struct page *page = NULL;
553 unsigned from, to;
554 struct ext4_iloc iloc;
555
556 if (!ext4_has_inline_data(inode)) {
557 /*
558 * clear the flag so that no new write
559 * will trap here again.
560 */
561 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
562 return 0;
563 }
564
565 needed_blocks = ext4_writepage_trans_blocks(inode);
566
567 ret = ext4_get_inode_loc(inode, &iloc);
568 if (ret)
569 return ret;
570
571 retry:
572 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
573 if (IS_ERR(handle)) {
574 ret = PTR_ERR(handle);
575 handle = NULL;
576 goto out;
577 }
578
579 /* We cannot recurse into the filesystem as the transaction is already
580 * started */
581 flags |= AOP_FLAG_NOFS;
582
583 page = grab_cache_page_write_begin(mapping, 0, flags);
584 if (!page) {
585 ret = -ENOMEM;
586 goto out;
587 }
588
589 ext4_write_lock_xattr(inode, &no_expand);
590 sem_held = 1;
591 /* If some one has already done this for us, just exit. */
592 if (!ext4_has_inline_data(inode)) {
593 ret = 0;
594 goto out;
595 }
596
597 from = 0;
598 to = ext4_get_inline_size(inode);
599 if (!PageUptodate(page)) {
600 ret = ext4_read_inline_page(inode, page);
601 if (ret < 0)
602 goto out;
603 }
604
605 ret = ext4_destroy_inline_data_nolock(handle, inode);
606 if (ret)
607 goto out;
608
609 if (ext4_should_dioread_nolock(inode)) {
610 ret = __block_write_begin(page, from, to,
611 ext4_get_block_unwritten);
612 } else
613 ret = __block_write_begin(page, from, to, ext4_get_block);
614
615 if (!ret && ext4_should_journal_data(inode)) {
616 ret = ext4_walk_page_buffers(handle, page_buffers(page),
617 from, to, NULL,
618 do_journal_get_write_access);
619 }
620
621 if (ret) {
622 unlock_page(page);
623 put_page(page);
624 page = NULL;
625 ext4_orphan_add(handle, inode);
626 ext4_write_unlock_xattr(inode, &no_expand);
627 sem_held = 0;
628 ext4_journal_stop(handle);
629 handle = NULL;
630 ext4_truncate_failed_write(inode);
631 /*
632 * If truncate failed early the inode might
633 * still be on the orphan list; we need to
634 * make sure the inode is removed from the
635 * orphan list in that case.
636 */
637 if (inode->i_nlink)
638 ext4_orphan_del(NULL, inode);
639 }
640
641 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
642 goto retry;
643
644 if (page)
645 block_commit_write(page, from, to);
646 out:
647 if (page) {
648 unlock_page(page);
649 put_page(page);
650 }
651 if (sem_held)
652 ext4_write_unlock_xattr(inode, &no_expand);
653 if (handle)
654 ext4_journal_stop(handle);
655 brelse(iloc.bh);
656 return ret;
657 }
658
659 /*
660 * Try to write data in the inode.
661 * If the inode has inline data, check whether the new write can be
662 * in the inode also. If not, create the page the handle, move the data
663 * to the page make it update and let the later codes create extent for it.
664 */
665 int ext4_try_to_write_inline_data(struct address_space *mapping,
666 struct inode *inode,
667 loff_t pos, unsigned len,
668 unsigned flags,
669 struct page **pagep)
670 {
671 int ret;
672 handle_t *handle;
673 struct page *page;
674 struct ext4_iloc iloc;
675
676 if (pos + len > ext4_get_max_inline_size(inode))
677 goto convert;
678
679 ret = ext4_get_inode_loc(inode, &iloc);
680 if (ret)
681 return ret;
682
683 /*
684 * The possible write could happen in the inode,
685 * so try to reserve the space in inode first.
686 */
687 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
688 if (IS_ERR(handle)) {
689 ret = PTR_ERR(handle);
690 handle = NULL;
691 goto out;
692 }
693
694 ret = ext4_prepare_inline_data(handle, inode, pos + len);
695 if (ret && ret != -ENOSPC)
696 goto out;
697
698 /* We don't have space in inline inode, so convert it to extent. */
699 if (ret == -ENOSPC) {
700 ext4_journal_stop(handle);
701 brelse(iloc.bh);
702 goto convert;
703 }
704
705 ret = ext4_journal_get_write_access(handle, iloc.bh);
706 if (ret)
707 goto out;
708
709 flags |= AOP_FLAG_NOFS;
710
711 page = grab_cache_page_write_begin(mapping, 0, flags);
712 if (!page) {
713 ret = -ENOMEM;
714 goto out;
715 }
716
717 *pagep = page;
718 down_read(&EXT4_I(inode)->xattr_sem);
719 if (!ext4_has_inline_data(inode)) {
720 ret = 0;
721 unlock_page(page);
722 put_page(page);
723 goto out_up_read;
724 }
725
726 if (!PageUptodate(page)) {
727 ret = ext4_read_inline_page(inode, page);
728 if (ret < 0) {
729 unlock_page(page);
730 put_page(page);
731 goto out_up_read;
732 }
733 }
734
735 ret = 1;
736 handle = NULL;
737 out_up_read:
738 up_read(&EXT4_I(inode)->xattr_sem);
739 out:
740 if (handle && (ret != 1))
741 ext4_journal_stop(handle);
742 brelse(iloc.bh);
743 return ret;
744 convert:
745 return ext4_convert_inline_data_to_extent(mapping,
746 inode, flags);
747 }
748
749 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
750 unsigned copied, struct page *page)
751 {
752 int ret, no_expand;
753 void *kaddr;
754 struct ext4_iloc iloc;
755
756 if (unlikely(copied < len)) {
757 if (!PageUptodate(page)) {
758 copied = 0;
759 goto out;
760 }
761 }
762
763 ret = ext4_get_inode_loc(inode, &iloc);
764 if (ret) {
765 ext4_std_error(inode->i_sb, ret);
766 copied = 0;
767 goto out;
768 }
769
770 ext4_write_lock_xattr(inode, &no_expand);
771 BUG_ON(!ext4_has_inline_data(inode));
772
773 kaddr = kmap_atomic(page);
774 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
775 kunmap_atomic(kaddr);
776 SetPageUptodate(page);
777 /* clear page dirty so that writepages wouldn't work for us. */
778 ClearPageDirty(page);
779
780 ext4_write_unlock_xattr(inode, &no_expand);
781 brelse(iloc.bh);
782 mark_inode_dirty(inode);
783 out:
784 return copied;
785 }
786
787 struct buffer_head *
788 ext4_journalled_write_inline_data(struct inode *inode,
789 unsigned len,
790 struct page *page)
791 {
792 int ret, no_expand;
793 void *kaddr;
794 struct ext4_iloc iloc;
795
796 ret = ext4_get_inode_loc(inode, &iloc);
797 if (ret) {
798 ext4_std_error(inode->i_sb, ret);
799 return NULL;
800 }
801
802 ext4_write_lock_xattr(inode, &no_expand);
803 kaddr = kmap_atomic(page);
804 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
805 kunmap_atomic(kaddr);
806 ext4_write_unlock_xattr(inode, &no_expand);
807
808 return iloc.bh;
809 }
810
811 /*
812 * Try to make the page cache and handle ready for the inline data case.
813 * We can call this function in 2 cases:
814 * 1. The inode is created and the first write exceeds inline size. We can
815 * clear the inode state safely.
816 * 2. The inode has inline data, then we need to read the data, make it
817 * update and dirty so that ext4_da_writepages can handle it. We don't
818 * need to start the journal since the file's metatdata isn't changed now.
819 */
820 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
821 struct inode *inode,
822 unsigned flags,
823 void **fsdata)
824 {
825 int ret = 0, inline_size;
826 struct page *page;
827
828 page = grab_cache_page_write_begin(mapping, 0, flags);
829 if (!page)
830 return -ENOMEM;
831
832 down_read(&EXT4_I(inode)->xattr_sem);
833 if (!ext4_has_inline_data(inode)) {
834 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
835 goto out;
836 }
837
838 inline_size = ext4_get_inline_size(inode);
839
840 if (!PageUptodate(page)) {
841 ret = ext4_read_inline_page(inode, page);
842 if (ret < 0)
843 goto out;
844 }
845
846 ret = __block_write_begin(page, 0, inline_size,
847 ext4_da_get_block_prep);
848 if (ret) {
849 up_read(&EXT4_I(inode)->xattr_sem);
850 unlock_page(page);
851 put_page(page);
852 ext4_truncate_failed_write(inode);
853 return ret;
854 }
855
856 SetPageDirty(page);
857 SetPageUptodate(page);
858 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
859 *fsdata = (void *)CONVERT_INLINE_DATA;
860
861 out:
862 up_read(&EXT4_I(inode)->xattr_sem);
863 if (page) {
864 unlock_page(page);
865 put_page(page);
866 }
867 return ret;
868 }
869
870 /*
871 * Prepare the write for the inline data.
872 * If the the data can be written into the inode, we just read
873 * the page and make it uptodate, and start the journal.
874 * Otherwise read the page, makes it dirty so that it can be
875 * handle in writepages(the i_disksize update is left to the
876 * normal ext4_da_write_end).
877 */
878 int ext4_da_write_inline_data_begin(struct address_space *mapping,
879 struct inode *inode,
880 loff_t pos, unsigned len,
881 unsigned flags,
882 struct page **pagep,
883 void **fsdata)
884 {
885 int ret, inline_size;
886 handle_t *handle;
887 struct page *page;
888 struct ext4_iloc iloc;
889 int retries = 0;
890
891 ret = ext4_get_inode_loc(inode, &iloc);
892 if (ret)
893 return ret;
894
895 retry_journal:
896 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
897 if (IS_ERR(handle)) {
898 ret = PTR_ERR(handle);
899 goto out;
900 }
901
902 inline_size = ext4_get_max_inline_size(inode);
903
904 ret = -ENOSPC;
905 if (inline_size >= pos + len) {
906 ret = ext4_prepare_inline_data(handle, inode, pos + len);
907 if (ret && ret != -ENOSPC)
908 goto out_journal;
909 }
910
911 /*
912 * We cannot recurse into the filesystem as the transaction
913 * is already started.
914 */
915 flags |= AOP_FLAG_NOFS;
916
917 if (ret == -ENOSPC) {
918 ext4_journal_stop(handle);
919 ret = ext4_da_convert_inline_data_to_extent(mapping,
920 inode,
921 flags,
922 fsdata);
923 if (ret == -ENOSPC &&
924 ext4_should_retry_alloc(inode->i_sb, &retries))
925 goto retry_journal;
926 goto out;
927 }
928
929 page = grab_cache_page_write_begin(mapping, 0, flags);
930 if (!page) {
931 ret = -ENOMEM;
932 goto out_journal;
933 }
934
935 down_read(&EXT4_I(inode)->xattr_sem);
936 if (!ext4_has_inline_data(inode)) {
937 ret = 0;
938 goto out_release_page;
939 }
940
941 if (!PageUptodate(page)) {
942 ret = ext4_read_inline_page(inode, page);
943 if (ret < 0)
944 goto out_release_page;
945 }
946 ret = ext4_journal_get_write_access(handle, iloc.bh);
947 if (ret)
948 goto out_release_page;
949
950 up_read(&EXT4_I(inode)->xattr_sem);
951 *pagep = page;
952 brelse(iloc.bh);
953 return 1;
954 out_release_page:
955 up_read(&EXT4_I(inode)->xattr_sem);
956 unlock_page(page);
957 put_page(page);
958 out_journal:
959 ext4_journal_stop(handle);
960 out:
961 brelse(iloc.bh);
962 return ret;
963 }
964
965 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
966 unsigned len, unsigned copied,
967 struct page *page)
968 {
969 int ret;
970
971 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
972 if (ret < 0) {
973 unlock_page(page);
974 put_page(page);
975 return ret;
976 }
977 copied = ret;
978
979 /*
980 * No need to use i_size_read() here, the i_size
981 * cannot change under us because we hold i_mutex.
982 *
983 * But it's important to update i_size while still holding page lock:
984 * page writeout could otherwise come in and zero beyond i_size.
985 */
986 if (pos+copied > inode->i_size)
987 i_size_write(inode, pos+copied);
988 unlock_page(page);
989 put_page(page);
990
991 /*
992 * Don't mark the inode dirty under page lock. First, it unnecessarily
993 * makes the holding time of page lock longer. Second, it forces lock
994 * ordering of page lock and transaction start for journaling
995 * filesystems.
996 */
997 mark_inode_dirty(inode);
998
999 return copied;
1000 }
1001
1002 #ifdef INLINE_DIR_DEBUG
1003 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
1004 void *inline_start, int inline_size)
1005 {
1006 int offset;
1007 unsigned short de_len;
1008 struct ext4_dir_entry_2 *de = inline_start;
1009 void *dlimit = inline_start + inline_size;
1010
1011 trace_printk("inode %lu\n", dir->i_ino);
1012 offset = 0;
1013 while ((void *)de < dlimit) {
1014 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1015 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1016 offset, de_len, de->name_len, de->name,
1017 de->name_len, le32_to_cpu(de->inode));
1018 if (ext4_check_dir_entry(dir, NULL, de, bh,
1019 inline_start, inline_size, offset))
1020 BUG();
1021
1022 offset += de_len;
1023 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1024 }
1025 }
1026 #else
1027 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1028 #endif
1029
1030 /*
1031 * Add a new entry into a inline dir.
1032 * It will return -ENOSPC if no space is available, and -EIO
1033 * and -EEXIST if directory entry already exists.
1034 */
1035 static int ext4_add_dirent_to_inline(handle_t *handle,
1036 struct ext4_filename *fname,
1037 struct inode *dir,
1038 struct inode *inode,
1039 struct ext4_iloc *iloc,
1040 void *inline_start, int inline_size)
1041 {
1042 int err;
1043 struct ext4_dir_entry_2 *de;
1044
1045 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1046 inline_size, fname, &de);
1047 if (err)
1048 return err;
1049
1050 BUFFER_TRACE(iloc->bh, "get_write_access");
1051 err = ext4_journal_get_write_access(handle, iloc->bh);
1052 if (err)
1053 return err;
1054 ext4_insert_dentry(inode, de, inline_size, fname);
1055
1056 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1057
1058 /*
1059 * XXX shouldn't update any times until successful
1060 * completion of syscall, but too many callers depend
1061 * on this.
1062 *
1063 * XXX similarly, too many callers depend on
1064 * ext4_new_inode() setting the times, but error
1065 * recovery deletes the inode, so the worst that can
1066 * happen is that the times are slightly out of date
1067 * and/or different from the directory change time.
1068 */
1069 dir->i_mtime = dir->i_ctime = current_time(dir);
1070 ext4_update_dx_flag(dir);
1071 dir->i_version++;
1072 return 1;
1073 }
1074
1075 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1076 struct ext4_iloc *iloc)
1077 {
1078 struct ext4_xattr_entry *entry;
1079 struct ext4_xattr_ibody_header *header;
1080
1081 BUG_ON(!EXT4_I(inode)->i_inline_off);
1082
1083 header = IHDR(inode, ext4_raw_inode(iloc));
1084 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1085 EXT4_I(inode)->i_inline_off);
1086
1087 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1088 }
1089
1090 /* Set the final de to cover the whole block. */
1091 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1092 {
1093 struct ext4_dir_entry_2 *de, *prev_de;
1094 void *limit;
1095 int de_len;
1096
1097 de = (struct ext4_dir_entry_2 *)de_buf;
1098 if (old_size) {
1099 limit = de_buf + old_size;
1100 do {
1101 prev_de = de;
1102 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1103 de_buf += de_len;
1104 de = (struct ext4_dir_entry_2 *)de_buf;
1105 } while (de_buf < limit);
1106
1107 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1108 old_size, new_size);
1109 } else {
1110 /* this is just created, so create an empty entry. */
1111 de->inode = 0;
1112 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1113 }
1114 }
1115
1116 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1117 struct ext4_iloc *iloc)
1118 {
1119 int ret;
1120 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1121 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1122
1123 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1124 return -ENOSPC;
1125
1126 ret = ext4_update_inline_data(handle, dir,
1127 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1128 if (ret)
1129 return ret;
1130
1131 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1132 EXT4_I(dir)->i_inline_size -
1133 EXT4_MIN_INLINE_DATA_SIZE);
1134 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1135 return 0;
1136 }
1137
1138 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1139 struct ext4_iloc *iloc,
1140 void *buf, int inline_size)
1141 {
1142 ext4_create_inline_data(handle, inode, inline_size);
1143 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1144 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1145 }
1146
1147 static int ext4_finish_convert_inline_dir(handle_t *handle,
1148 struct inode *inode,
1149 struct buffer_head *dir_block,
1150 void *buf,
1151 int inline_size)
1152 {
1153 int err, csum_size = 0, header_size = 0;
1154 struct ext4_dir_entry_2 *de;
1155 struct ext4_dir_entry_tail *t;
1156 void *target = dir_block->b_data;
1157
1158 /*
1159 * First create "." and ".." and then copy the dir information
1160 * back to the block.
1161 */
1162 de = (struct ext4_dir_entry_2 *)target;
1163 de = ext4_init_dot_dotdot(inode, de,
1164 inode->i_sb->s_blocksize, csum_size,
1165 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1166 header_size = (void *)de - target;
1167
1168 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1169 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1170
1171 if (ext4_has_metadata_csum(inode->i_sb))
1172 csum_size = sizeof(struct ext4_dir_entry_tail);
1173
1174 inode->i_size = inode->i_sb->s_blocksize;
1175 i_size_write(inode, inode->i_sb->s_blocksize);
1176 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1177 ext4_update_final_de(dir_block->b_data,
1178 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1179 inode->i_sb->s_blocksize - csum_size);
1180
1181 if (csum_size) {
1182 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1183 inode->i_sb->s_blocksize);
1184 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1185 }
1186 set_buffer_uptodate(dir_block);
1187 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1188 if (err)
1189 return err;
1190 set_buffer_verified(dir_block);
1191 return ext4_mark_inode_dirty(handle, inode);
1192 }
1193
1194 static int ext4_convert_inline_data_nolock(handle_t *handle,
1195 struct inode *inode,
1196 struct ext4_iloc *iloc)
1197 {
1198 int error;
1199 void *buf = NULL;
1200 struct buffer_head *data_bh = NULL;
1201 struct ext4_map_blocks map;
1202 int inline_size;
1203
1204 inline_size = ext4_get_inline_size(inode);
1205 buf = kmalloc(inline_size, GFP_NOFS);
1206 if (!buf) {
1207 error = -ENOMEM;
1208 goto out;
1209 }
1210
1211 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1212 if (error < 0)
1213 goto out;
1214
1215 /*
1216 * Make sure the inline directory entries pass checks before we try to
1217 * convert them, so that we avoid touching stuff that needs fsck.
1218 */
1219 if (S_ISDIR(inode->i_mode)) {
1220 error = ext4_check_all_de(inode, iloc->bh,
1221 buf + EXT4_INLINE_DOTDOT_SIZE,
1222 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1223 if (error)
1224 goto out;
1225 }
1226
1227 error = ext4_destroy_inline_data_nolock(handle, inode);
1228 if (error)
1229 goto out;
1230
1231 map.m_lblk = 0;
1232 map.m_len = 1;
1233 map.m_flags = 0;
1234 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1235 if (error < 0)
1236 goto out_restore;
1237 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1238 error = -EIO;
1239 goto out_restore;
1240 }
1241
1242 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1243 if (!data_bh) {
1244 error = -ENOMEM;
1245 goto out_restore;
1246 }
1247
1248 lock_buffer(data_bh);
1249 error = ext4_journal_get_create_access(handle, data_bh);
1250 if (error) {
1251 unlock_buffer(data_bh);
1252 error = -EIO;
1253 goto out_restore;
1254 }
1255 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1256
1257 if (!S_ISDIR(inode->i_mode)) {
1258 memcpy(data_bh->b_data, buf, inline_size);
1259 set_buffer_uptodate(data_bh);
1260 error = ext4_handle_dirty_metadata(handle,
1261 inode, data_bh);
1262 } else {
1263 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1264 buf, inline_size);
1265 }
1266
1267 unlock_buffer(data_bh);
1268 out_restore:
1269 if (error)
1270 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1271
1272 out:
1273 brelse(data_bh);
1274 kfree(buf);
1275 return error;
1276 }
1277
1278 /*
1279 * Try to add the new entry to the inline data.
1280 * If succeeds, return 0. If not, extended the inline dir and copied data to
1281 * the new created block.
1282 */
1283 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1284 struct inode *dir, struct inode *inode)
1285 {
1286 int ret, inline_size, no_expand;
1287 void *inline_start;
1288 struct ext4_iloc iloc;
1289
1290 ret = ext4_get_inode_loc(dir, &iloc);
1291 if (ret)
1292 return ret;
1293
1294 ext4_write_lock_xattr(dir, &no_expand);
1295 if (!ext4_has_inline_data(dir))
1296 goto out;
1297
1298 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1299 EXT4_INLINE_DOTDOT_SIZE;
1300 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1301
1302 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1303 inline_start, inline_size);
1304 if (ret != -ENOSPC)
1305 goto out;
1306
1307 /* check whether it can be inserted to inline xattr space. */
1308 inline_size = EXT4_I(dir)->i_inline_size -
1309 EXT4_MIN_INLINE_DATA_SIZE;
1310 if (!inline_size) {
1311 /* Try to use the xattr space.*/
1312 ret = ext4_update_inline_dir(handle, dir, &iloc);
1313 if (ret && ret != -ENOSPC)
1314 goto out;
1315
1316 inline_size = EXT4_I(dir)->i_inline_size -
1317 EXT4_MIN_INLINE_DATA_SIZE;
1318 }
1319
1320 if (inline_size) {
1321 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1322
1323 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1324 inode, &iloc, inline_start,
1325 inline_size);
1326
1327 if (ret != -ENOSPC)
1328 goto out;
1329 }
1330
1331 /*
1332 * The inline space is filled up, so create a new block for it.
1333 * As the extent tree will be created, we have to save the inline
1334 * dir first.
1335 */
1336 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1337
1338 out:
1339 ext4_write_unlock_xattr(dir, &no_expand);
1340 ext4_mark_inode_dirty(handle, dir);
1341 brelse(iloc.bh);
1342 return ret;
1343 }
1344
1345 /*
1346 * This function fills a red-black tree with information from an
1347 * inlined dir. It returns the number directory entries loaded
1348 * into the tree. If there is an error it is returned in err.
1349 */
1350 int htree_inlinedir_to_tree(struct file *dir_file,
1351 struct inode *dir, ext4_lblk_t block,
1352 struct dx_hash_info *hinfo,
1353 __u32 start_hash, __u32 start_minor_hash,
1354 int *has_inline_data)
1355 {
1356 int err = 0, count = 0;
1357 unsigned int parent_ino;
1358 int pos;
1359 struct ext4_dir_entry_2 *de;
1360 struct inode *inode = file_inode(dir_file);
1361 int ret, inline_size = 0;
1362 struct ext4_iloc iloc;
1363 void *dir_buf = NULL;
1364 struct ext4_dir_entry_2 fake;
1365 struct fscrypt_str tmp_str;
1366
1367 ret = ext4_get_inode_loc(inode, &iloc);
1368 if (ret)
1369 return ret;
1370
1371 down_read(&EXT4_I(inode)->xattr_sem);
1372 if (!ext4_has_inline_data(inode)) {
1373 up_read(&EXT4_I(inode)->xattr_sem);
1374 *has_inline_data = 0;
1375 goto out;
1376 }
1377
1378 inline_size = ext4_get_inline_size(inode);
1379 dir_buf = kmalloc(inline_size, GFP_NOFS);
1380 if (!dir_buf) {
1381 ret = -ENOMEM;
1382 up_read(&EXT4_I(inode)->xattr_sem);
1383 goto out;
1384 }
1385
1386 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1387 up_read(&EXT4_I(inode)->xattr_sem);
1388 if (ret < 0)
1389 goto out;
1390
1391 pos = 0;
1392 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1393 while (pos < inline_size) {
1394 /*
1395 * As inlined dir doesn't store any information about '.' and
1396 * only the inode number of '..' is stored, we have to handle
1397 * them differently.
1398 */
1399 if (pos == 0) {
1400 fake.inode = cpu_to_le32(inode->i_ino);
1401 fake.name_len = 1;
1402 strcpy(fake.name, ".");
1403 fake.rec_len = ext4_rec_len_to_disk(
1404 EXT4_DIR_REC_LEN(fake.name_len),
1405 inline_size);
1406 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1407 de = &fake;
1408 pos = EXT4_INLINE_DOTDOT_OFFSET;
1409 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1410 fake.inode = cpu_to_le32(parent_ino);
1411 fake.name_len = 2;
1412 strcpy(fake.name, "..");
1413 fake.rec_len = ext4_rec_len_to_disk(
1414 EXT4_DIR_REC_LEN(fake.name_len),
1415 inline_size);
1416 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1417 de = &fake;
1418 pos = EXT4_INLINE_DOTDOT_SIZE;
1419 } else {
1420 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1421 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1422 if (ext4_check_dir_entry(inode, dir_file, de,
1423 iloc.bh, dir_buf,
1424 inline_size, pos)) {
1425 ret = count;
1426 goto out;
1427 }
1428 }
1429
1430 ext4fs_dirhash(de->name, de->name_len, hinfo);
1431 if ((hinfo->hash < start_hash) ||
1432 ((hinfo->hash == start_hash) &&
1433 (hinfo->minor_hash < start_minor_hash)))
1434 continue;
1435 if (de->inode == 0)
1436 continue;
1437 tmp_str.name = de->name;
1438 tmp_str.len = de->name_len;
1439 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1440 hinfo->minor_hash, de, &tmp_str);
1441 if (err) {
1442 count = err;
1443 goto out;
1444 }
1445 count++;
1446 }
1447 ret = count;
1448 out:
1449 kfree(dir_buf);
1450 brelse(iloc.bh);
1451 return ret;
1452 }
1453
1454 /*
1455 * So this function is called when the volume is mkfsed with
1456 * dir_index disabled. In order to keep f_pos persistent
1457 * after we convert from an inlined dir to a blocked based,
1458 * we just pretend that we are a normal dir and return the
1459 * offset as if '.' and '..' really take place.
1460 *
1461 */
1462 int ext4_read_inline_dir(struct file *file,
1463 struct dir_context *ctx,
1464 int *has_inline_data)
1465 {
1466 unsigned int offset, parent_ino;
1467 int i;
1468 struct ext4_dir_entry_2 *de;
1469 struct super_block *sb;
1470 struct inode *inode = file_inode(file);
1471 int ret, inline_size = 0;
1472 struct ext4_iloc iloc;
1473 void *dir_buf = NULL;
1474 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1475
1476 ret = ext4_get_inode_loc(inode, &iloc);
1477 if (ret)
1478 return ret;
1479
1480 down_read(&EXT4_I(inode)->xattr_sem);
1481 if (!ext4_has_inline_data(inode)) {
1482 up_read(&EXT4_I(inode)->xattr_sem);
1483 *has_inline_data = 0;
1484 goto out;
1485 }
1486
1487 inline_size = ext4_get_inline_size(inode);
1488 dir_buf = kmalloc(inline_size, GFP_NOFS);
1489 if (!dir_buf) {
1490 ret = -ENOMEM;
1491 up_read(&EXT4_I(inode)->xattr_sem);
1492 goto out;
1493 }
1494
1495 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1496 up_read(&EXT4_I(inode)->xattr_sem);
1497 if (ret < 0)
1498 goto out;
1499
1500 ret = 0;
1501 sb = inode->i_sb;
1502 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1503 offset = ctx->pos;
1504
1505 /*
1506 * dotdot_offset and dotdot_size is the real offset and
1507 * size for ".." and "." if the dir is block based while
1508 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1509 * So we will use extra_offset and extra_size to indicate them
1510 * during the inline dir iteration.
1511 */
1512 dotdot_offset = EXT4_DIR_REC_LEN(1);
1513 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1514 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1515 extra_size = extra_offset + inline_size;
1516
1517 /*
1518 * If the version has changed since the last call to
1519 * readdir(2), then we might be pointing to an invalid
1520 * dirent right now. Scan from the start of the inline
1521 * dir to make sure.
1522 */
1523 if (file->f_version != inode->i_version) {
1524 for (i = 0; i < extra_size && i < offset;) {
1525 /*
1526 * "." is with offset 0 and
1527 * ".." is dotdot_offset.
1528 */
1529 if (!i) {
1530 i = dotdot_offset;
1531 continue;
1532 } else if (i == dotdot_offset) {
1533 i = dotdot_size;
1534 continue;
1535 }
1536 /* for other entry, the real offset in
1537 * the buf has to be tuned accordingly.
1538 */
1539 de = (struct ext4_dir_entry_2 *)
1540 (dir_buf + i - extra_offset);
1541 /* It's too expensive to do a full
1542 * dirent test each time round this
1543 * loop, but we do have to test at
1544 * least that it is non-zero. A
1545 * failure will be detected in the
1546 * dirent test below. */
1547 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1548 < EXT4_DIR_REC_LEN(1))
1549 break;
1550 i += ext4_rec_len_from_disk(de->rec_len,
1551 extra_size);
1552 }
1553 offset = i;
1554 ctx->pos = offset;
1555 file->f_version = inode->i_version;
1556 }
1557
1558 while (ctx->pos < extra_size) {
1559 if (ctx->pos == 0) {
1560 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1561 goto out;
1562 ctx->pos = dotdot_offset;
1563 continue;
1564 }
1565
1566 if (ctx->pos == dotdot_offset) {
1567 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1568 goto out;
1569 ctx->pos = dotdot_size;
1570 continue;
1571 }
1572
1573 de = (struct ext4_dir_entry_2 *)
1574 (dir_buf + ctx->pos - extra_offset);
1575 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1576 extra_size, ctx->pos))
1577 goto out;
1578 if (le32_to_cpu(de->inode)) {
1579 if (!dir_emit(ctx, de->name, de->name_len,
1580 le32_to_cpu(de->inode),
1581 get_dtype(sb, de->file_type)))
1582 goto out;
1583 }
1584 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1585 }
1586 out:
1587 kfree(dir_buf);
1588 brelse(iloc.bh);
1589 return ret;
1590 }
1591
1592 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1593 struct ext4_dir_entry_2 **parent_de,
1594 int *retval)
1595 {
1596 struct ext4_iloc iloc;
1597
1598 *retval = ext4_get_inode_loc(inode, &iloc);
1599 if (*retval)
1600 return NULL;
1601
1602 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1603
1604 return iloc.bh;
1605 }
1606
1607 /*
1608 * Try to create the inline data for the new dir.
1609 * If it succeeds, return 0, otherwise return the error.
1610 * In case of ENOSPC, the caller should create the normal disk layout dir.
1611 */
1612 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1613 struct inode *inode)
1614 {
1615 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1616 struct ext4_iloc iloc;
1617 struct ext4_dir_entry_2 *de;
1618
1619 ret = ext4_get_inode_loc(inode, &iloc);
1620 if (ret)
1621 return ret;
1622
1623 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1624 if (ret)
1625 goto out;
1626
1627 /*
1628 * For inline dir, we only save the inode information for the ".."
1629 * and create a fake dentry to cover the left space.
1630 */
1631 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1632 de->inode = cpu_to_le32(parent->i_ino);
1633 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1634 de->inode = 0;
1635 de->rec_len = ext4_rec_len_to_disk(
1636 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1637 inline_size);
1638 set_nlink(inode, 2);
1639 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1640 out:
1641 brelse(iloc.bh);
1642 return ret;
1643 }
1644
1645 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1646 struct ext4_filename *fname,
1647 struct ext4_dir_entry_2 **res_dir,
1648 int *has_inline_data)
1649 {
1650 int ret;
1651 struct ext4_iloc iloc;
1652 void *inline_start;
1653 int inline_size;
1654
1655 if (ext4_get_inode_loc(dir, &iloc))
1656 return NULL;
1657
1658 down_read(&EXT4_I(dir)->xattr_sem);
1659 if (!ext4_has_inline_data(dir)) {
1660 *has_inline_data = 0;
1661 goto out;
1662 }
1663
1664 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1665 EXT4_INLINE_DOTDOT_SIZE;
1666 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1667 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1668 dir, fname, 0, res_dir);
1669 if (ret == 1)
1670 goto out_find;
1671 if (ret < 0)
1672 goto out;
1673
1674 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1675 goto out;
1676
1677 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1678 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1679
1680 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1681 dir, fname, 0, res_dir);
1682 if (ret == 1)
1683 goto out_find;
1684
1685 out:
1686 brelse(iloc.bh);
1687 iloc.bh = NULL;
1688 out_find:
1689 up_read(&EXT4_I(dir)->xattr_sem);
1690 return iloc.bh;
1691 }
1692
1693 int ext4_delete_inline_entry(handle_t *handle,
1694 struct inode *dir,
1695 struct ext4_dir_entry_2 *de_del,
1696 struct buffer_head *bh,
1697 int *has_inline_data)
1698 {
1699 int err, inline_size, no_expand;
1700 struct ext4_iloc iloc;
1701 void *inline_start;
1702
1703 err = ext4_get_inode_loc(dir, &iloc);
1704 if (err)
1705 return err;
1706
1707 ext4_write_lock_xattr(dir, &no_expand);
1708 if (!ext4_has_inline_data(dir)) {
1709 *has_inline_data = 0;
1710 goto out;
1711 }
1712
1713 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1714 EXT4_MIN_INLINE_DATA_SIZE) {
1715 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1716 EXT4_INLINE_DOTDOT_SIZE;
1717 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1718 EXT4_INLINE_DOTDOT_SIZE;
1719 } else {
1720 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1721 inline_size = ext4_get_inline_size(dir) -
1722 EXT4_MIN_INLINE_DATA_SIZE;
1723 }
1724
1725 BUFFER_TRACE(bh, "get_write_access");
1726 err = ext4_journal_get_write_access(handle, bh);
1727 if (err)
1728 goto out;
1729
1730 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1731 inline_start, inline_size, 0);
1732 if (err)
1733 goto out;
1734
1735 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1736 out:
1737 ext4_write_unlock_xattr(dir, &no_expand);
1738 if (likely(err == 0))
1739 err = ext4_mark_inode_dirty(handle, dir);
1740 brelse(iloc.bh);
1741 if (err != -ENOENT)
1742 ext4_std_error(dir->i_sb, err);
1743 return err;
1744 }
1745
1746 /*
1747 * Get the inline dentry at offset.
1748 */
1749 static inline struct ext4_dir_entry_2 *
1750 ext4_get_inline_entry(struct inode *inode,
1751 struct ext4_iloc *iloc,
1752 unsigned int offset,
1753 void **inline_start,
1754 int *inline_size)
1755 {
1756 void *inline_pos;
1757
1758 BUG_ON(offset > ext4_get_inline_size(inode));
1759
1760 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1761 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1762 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1763 } else {
1764 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1765 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1766 *inline_size = ext4_get_inline_size(inode) -
1767 EXT4_MIN_INLINE_DATA_SIZE;
1768 }
1769
1770 if (inline_start)
1771 *inline_start = inline_pos;
1772 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1773 }
1774
1775 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1776 {
1777 int err, inline_size;
1778 struct ext4_iloc iloc;
1779 size_t inline_len;
1780 void *inline_pos;
1781 unsigned int offset;
1782 struct ext4_dir_entry_2 *de;
1783 bool ret = true;
1784
1785 err = ext4_get_inode_loc(dir, &iloc);
1786 if (err) {
1787 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1788 err, dir->i_ino);
1789 return true;
1790 }
1791
1792 down_read(&EXT4_I(dir)->xattr_sem);
1793 if (!ext4_has_inline_data(dir)) {
1794 *has_inline_data = 0;
1795 goto out;
1796 }
1797
1798 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1799 if (!le32_to_cpu(de->inode)) {
1800 ext4_warning(dir->i_sb,
1801 "bad inline directory (dir #%lu) - no `..'",
1802 dir->i_ino);
1803 ret = true;
1804 goto out;
1805 }
1806
1807 inline_len = ext4_get_inline_size(dir);
1808 offset = EXT4_INLINE_DOTDOT_SIZE;
1809 while (offset < inline_len) {
1810 de = ext4_get_inline_entry(dir, &iloc, offset,
1811 &inline_pos, &inline_size);
1812 if (ext4_check_dir_entry(dir, NULL, de,
1813 iloc.bh, inline_pos,
1814 inline_size, offset)) {
1815 ext4_warning(dir->i_sb,
1816 "bad inline directory (dir #%lu) - "
1817 "inode %u, rec_len %u, name_len %d"
1818 "inline size %d",
1819 dir->i_ino, le32_to_cpu(de->inode),
1820 le16_to_cpu(de->rec_len), de->name_len,
1821 inline_size);
1822 ret = true;
1823 goto out;
1824 }
1825 if (le32_to_cpu(de->inode)) {
1826 ret = false;
1827 goto out;
1828 }
1829 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1830 }
1831
1832 out:
1833 up_read(&EXT4_I(dir)->xattr_sem);
1834 brelse(iloc.bh);
1835 return ret;
1836 }
1837
1838 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1839 {
1840 int ret, no_expand;
1841
1842 ext4_write_lock_xattr(inode, &no_expand);
1843 ret = ext4_destroy_inline_data_nolock(handle, inode);
1844 ext4_write_unlock_xattr(inode, &no_expand);
1845
1846 return ret;
1847 }
1848
1849 int ext4_inline_data_fiemap(struct inode *inode,
1850 struct fiemap_extent_info *fieinfo,
1851 int *has_inline, __u64 start, __u64 len)
1852 {
1853 __u64 physical = 0;
1854 __u64 inline_len;
1855 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1856 FIEMAP_EXTENT_LAST;
1857 int error = 0;
1858 struct ext4_iloc iloc;
1859
1860 down_read(&EXT4_I(inode)->xattr_sem);
1861 if (!ext4_has_inline_data(inode)) {
1862 *has_inline = 0;
1863 goto out;
1864 }
1865 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1866 i_size_read(inode));
1867 if (start >= inline_len)
1868 goto out;
1869 if (start + len < inline_len)
1870 inline_len = start + len;
1871 inline_len -= start;
1872
1873 error = ext4_get_inode_loc(inode, &iloc);
1874 if (error)
1875 goto out;
1876
1877 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1878 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1879 physical += offsetof(struct ext4_inode, i_block);
1880
1881 brelse(iloc.bh);
1882 out:
1883 up_read(&EXT4_I(inode)->xattr_sem);
1884 if (physical)
1885 error = fiemap_fill_next_extent(fieinfo, start, physical,
1886 inline_len, flags);
1887 return (error < 0 ? error : 0);
1888 }
1889
1890 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1891 {
1892 handle_t *handle;
1893 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1894 size_t i_size;
1895 void *value = NULL;
1896 struct ext4_xattr_ibody_find is = {
1897 .s = { .not_found = -ENODATA, },
1898 };
1899 struct ext4_xattr_info i = {
1900 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1901 .name = EXT4_XATTR_SYSTEM_DATA,
1902 };
1903
1904
1905 needed_blocks = ext4_writepage_trans_blocks(inode);
1906 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1907 if (IS_ERR(handle))
1908 return PTR_ERR(handle);
1909
1910 ext4_write_lock_xattr(inode, &no_expand);
1911 if (!ext4_has_inline_data(inode)) {
1912 *has_inline = 0;
1913 ext4_journal_stop(handle);
1914 return 0;
1915 }
1916
1917 if ((err = ext4_orphan_add(handle, inode)) != 0)
1918 goto out;
1919
1920 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1921 goto out;
1922
1923 down_write(&EXT4_I(inode)->i_data_sem);
1924 i_size = inode->i_size;
1925 inline_size = ext4_get_inline_size(inode);
1926 EXT4_I(inode)->i_disksize = i_size;
1927
1928 if (i_size < inline_size) {
1929 /* Clear the content in the xattr space. */
1930 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1931 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1932 goto out_error;
1933
1934 BUG_ON(is.s.not_found);
1935
1936 value_len = le32_to_cpu(is.s.here->e_value_size);
1937 value = kmalloc(value_len, GFP_NOFS);
1938 if (!value) {
1939 err = -ENOMEM;
1940 goto out_error;
1941 }
1942
1943 err = ext4_xattr_ibody_get(inode, i.name_index,
1944 i.name, value, value_len);
1945 if (err <= 0)
1946 goto out_error;
1947
1948 i.value = value;
1949 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1950 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1951 err = ext4_xattr_ibody_inline_set(handle, inode,
1952 &i, &is);
1953 if (err)
1954 goto out_error;
1955 }
1956
1957 /* Clear the content within i_blocks. */
1958 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1959 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1960 memset(p + i_size, 0,
1961 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1962 }
1963
1964 EXT4_I(inode)->i_inline_size = i_size <
1965 EXT4_MIN_INLINE_DATA_SIZE ?
1966 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1967 }
1968
1969 out_error:
1970 up_write(&EXT4_I(inode)->i_data_sem);
1971 out:
1972 brelse(is.iloc.bh);
1973 ext4_write_unlock_xattr(inode, &no_expand);
1974 kfree(value);
1975 if (inode->i_nlink)
1976 ext4_orphan_del(handle, inode);
1977
1978 if (err == 0) {
1979 inode->i_mtime = inode->i_ctime = current_time(inode);
1980 err = ext4_mark_inode_dirty(handle, inode);
1981 if (IS_SYNC(inode))
1982 ext4_handle_sync(handle);
1983 }
1984 ext4_journal_stop(handle);
1985 return err;
1986 }
1987
1988 int ext4_convert_inline_data(struct inode *inode)
1989 {
1990 int error, needed_blocks, no_expand;
1991 handle_t *handle;
1992 struct ext4_iloc iloc;
1993
1994 if (!ext4_has_inline_data(inode)) {
1995 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1996 return 0;
1997 }
1998
1999 needed_blocks = ext4_writepage_trans_blocks(inode);
2000
2001 iloc.bh = NULL;
2002 error = ext4_get_inode_loc(inode, &iloc);
2003 if (error)
2004 return error;
2005
2006 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2007 if (IS_ERR(handle)) {
2008 error = PTR_ERR(handle);
2009 goto out_free;
2010 }
2011
2012 ext4_write_lock_xattr(inode, &no_expand);
2013 if (ext4_has_inline_data(inode))
2014 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2015 ext4_write_unlock_xattr(inode, &no_expand);
2016 ext4_journal_stop(handle);
2017 out_free:
2018 brelse(iloc.bh);
2019 return error;
2020 }