?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / ext4 / inline.c
blob89a4ac83cbcd9d85dee38abc18987b2a0ddb1c74
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
27 static 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
35 static int get_max_inline_xattr_value_size(struct inode *inode,
36 struct ext4_iloc *iloc)
37 {
38 struct ext4_xattr_ibody_header *header;
39 struct ext4_xattr_entry *entry;
40 struct ext4_inode *raw_inode;
41 int free, min_offs;
42
43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 EXT4_GOOD_OLD_INODE_SIZE -
45 EXT4_I(inode)->i_extra_isize -
46 sizeof(struct ext4_xattr_ibody_header);
47
48 /*
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
52 */
53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 return EXT4_XATTR_SIZE(min_offs -
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 EXT4_XATTR_ROUND - sizeof(__u32));
57
58 raw_inode = ext4_raw_inode(iloc);
59 header = IHDR(inode, raw_inode);
60 entry = IFIRST(header);
61
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64 if (!entry->e_value_inum && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
66 if (offs < min_offs)
67 min_offs = offs;
68 }
69 }
70 free = min_offs -
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78 goto out;
79 }
80
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85 else
86 free = 0;
87
88 out:
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 */
97 int 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 */
130 int 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 if (is.s.here->e_value_inum) {
154 EXT4_ERROR_INODE(inode, "inline data xattr refers "
155 "to an external xattr inode");
156 error = -EFSCORRUPTED;
157 goto out;
158 }
159 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
160 (void *)ext4_raw_inode(&is.iloc));
161 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
162 le32_to_cpu(is.s.here->e_value_size);
163 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
164 }
165 out:
166 brelse(is.iloc.bh);
167 return error;
168 }
169
170 static int ext4_read_inline_data(struct inode *inode, void *buffer,
171 unsigned int len,
172 struct ext4_iloc *iloc)
173 {
174 struct ext4_xattr_entry *entry;
175 struct ext4_xattr_ibody_header *header;
176 int cp_len = 0;
177 struct ext4_inode *raw_inode;
178
179 if (!len)
180 return 0;
181
182 BUG_ON(len > EXT4_I(inode)->i_inline_size);
183
184 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
185 len : EXT4_MIN_INLINE_DATA_SIZE;
186
187 raw_inode = ext4_raw_inode(iloc);
188 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
189
190 len -= cp_len;
191 buffer += cp_len;
192
193 if (!len)
194 goto out;
195
196 header = IHDR(inode, raw_inode);
197 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
198 EXT4_I(inode)->i_inline_off);
199 len = min_t(unsigned int, len,
200 (unsigned int)le32_to_cpu(entry->e_value_size));
201
202 memcpy(buffer,
203 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
204 cp_len += len;
205
206 out:
207 return cp_len;
208 }
209
210 /*
211 * write the buffer to the inline inode.
212 * If 'create' is set, we don't need to do the extra copy in the xattr
213 * value since it is already handled by ext4_xattr_ibody_inline_set.
214 * That saves us one memcpy.
215 */
216 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
217 void *buffer, loff_t pos, unsigned int len)
218 {
219 struct ext4_xattr_entry *entry;
220 struct ext4_xattr_ibody_header *header;
221 struct ext4_inode *raw_inode;
222 int cp_len = 0;
223
224 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
225 return;
226
227 BUG_ON(!EXT4_I(inode)->i_inline_off);
228 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
229
230 raw_inode = ext4_raw_inode(iloc);
231 buffer += pos;
232
233 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
234 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
235 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
236 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
237
238 len -= cp_len;
239 buffer += cp_len;
240 pos += cp_len;
241 }
242
243 if (!len)
244 return;
245
246 pos -= EXT4_MIN_INLINE_DATA_SIZE;
247 header = IHDR(inode, raw_inode);
248 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
249 EXT4_I(inode)->i_inline_off);
250
251 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
252 buffer, len);
253 }
254
255 static int ext4_create_inline_data(handle_t *handle,
256 struct inode *inode, unsigned len)
257 {
258 int error;
259 void *value = NULL;
260 struct ext4_xattr_ibody_find is = {
261 .s = { .not_found = -ENODATA, },
262 };
263 struct ext4_xattr_info i = {
264 .name_index = EXT4_XATTR_INDEX_SYSTEM,
265 .name = EXT4_XATTR_SYSTEM_DATA,
266 };
267
268 error = ext4_get_inode_loc(inode, &is.iloc);
269 if (error)
270 return error;
271
272 BUFFER_TRACE(is.iloc.bh, "get_write_access");
273 error = ext4_journal_get_write_access(handle, is.iloc.bh);
274 if (error)
275 goto out;
276
277 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
278 value = EXT4_ZERO_XATTR_VALUE;
279 len -= EXT4_MIN_INLINE_DATA_SIZE;
280 } else {
281 value = "";
282 len = 0;
283 }
284
285 /* Insert the the xttr entry. */
286 i.value = value;
287 i.value_len = len;
288
289 error = ext4_xattr_ibody_find(inode, &i, &is);
290 if (error)
291 goto out;
292
293 BUG_ON(!is.s.not_found);
294
295 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
296 if (error) {
297 if (error == -ENOSPC)
298 ext4_clear_inode_state(inode,
299 EXT4_STATE_MAY_INLINE_DATA);
300 goto out;
301 }
302
303 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
304 0, EXT4_MIN_INLINE_DATA_SIZE);
305
306 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
307 (void *)ext4_raw_inode(&is.iloc));
308 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
309 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
310 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
311 get_bh(is.iloc.bh);
312 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
313
314 out:
315 brelse(is.iloc.bh);
316 return error;
317 }
318
319 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
320 unsigned int len)
321 {
322 int error;
323 void *value = NULL;
324 struct ext4_xattr_ibody_find is = {
325 .s = { .not_found = -ENODATA, },
326 };
327 struct ext4_xattr_info i = {
328 .name_index = EXT4_XATTR_INDEX_SYSTEM,
329 .name = EXT4_XATTR_SYSTEM_DATA,
330 };
331
332 /* If the old space is ok, write the data directly. */
333 if (len <= EXT4_I(inode)->i_inline_size)
334 return 0;
335
336 error = ext4_get_inode_loc(inode, &is.iloc);
337 if (error)
338 return error;
339
340 error = ext4_xattr_ibody_find(inode, &i, &is);
341 if (error)
342 goto out;
343
344 BUG_ON(is.s.not_found);
345
346 len -= EXT4_MIN_INLINE_DATA_SIZE;
347 value = kzalloc(len, GFP_NOFS);
348 if (!value) {
349 error = -ENOMEM;
350 goto out;
351 }
352
353 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
354 value, len);
355 if (error == -ENODATA)
356 goto out;
357
358 BUFFER_TRACE(is.iloc.bh, "get_write_access");
359 error = ext4_journal_get_write_access(handle, is.iloc.bh);
360 if (error)
361 goto out;
362
363 /* Update the xttr entry. */
364 i.value = value;
365 i.value_len = len;
366
367 error = ext4_xattr_ibody_inline_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
379 out:
380 kfree(value);
381 brelse(is.iloc.bh);
382 return error;
383 }
384
385 static 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
409 static 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, is.iloc.bh);
437 if (error)
438 goto out;
439
440 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
441 if (error)
442 goto out;
443
444 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
445 0, EXT4_MIN_INLINE_DATA_SIZE);
446 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
447
448 if (ext4_has_feature_extents(inode->i_sb)) {
449 if (S_ISDIR(inode->i_mode) ||
450 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
451 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
452 ext4_ext_tree_init(handle, inode);
453 }
454 }
455 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
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);
463 out:
464 brelse(is.iloc.bh);
465 if (error == -ENODATA)
466 error = 0;
467 return error;
468 }
469
470 static 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
500 out:
501 return ret;
502 }
503
504 int 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
531 static 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
557 retry:
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);
632 out:
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 */
651 int 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 ret = ext4_journal_get_write_access(handle, iloc.bh);
692 if (ret)
693 goto out;
694
695 flags |= AOP_FLAG_NOFS;
696
697 page = grab_cache_page_write_begin(mapping, 0, flags);
698 if (!page) {
699 ret = -ENOMEM;
700 goto out;
701 }
702
703 *pagep = page;
704 down_read(&EXT4_I(inode)->xattr_sem);
705 if (!ext4_has_inline_data(inode)) {
706 ret = 0;
707 unlock_page(page);
708 put_page(page);
709 goto out_up_read;
710 }
711
712 if (!PageUptodate(page)) {
713 ret = ext4_read_inline_page(inode, page);
714 if (ret < 0) {
715 unlock_page(page);
716 put_page(page);
717 goto out_up_read;
718 }
719 }
720
721 ret = 1;
722 handle = NULL;
723 out_up_read:
724 up_read(&EXT4_I(inode)->xattr_sem);
725 out:
726 if (handle && (ret != 1))
727 ext4_journal_stop(handle);
728 brelse(iloc.bh);
729 return ret;
730 convert:
731 return ext4_convert_inline_data_to_extent(mapping,
732 inode, flags);
733 }
734
735 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
736 unsigned copied, struct page *page)
737 {
738 int ret, no_expand;
739 void *kaddr;
740 struct ext4_iloc iloc;
741
742 if (unlikely(copied < len)) {
743 if (!PageUptodate(page)) {
744 copied = 0;
745 goto out;
746 }
747 }
748
749 ret = ext4_get_inode_loc(inode, &iloc);
750 if (ret) {
751 ext4_std_error(inode->i_sb, ret);
752 copied = 0;
753 goto out;
754 }
755
756 ext4_write_lock_xattr(inode, &no_expand);
757 BUG_ON(!ext4_has_inline_data(inode));
758
759 kaddr = kmap_atomic(page);
760 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
761 kunmap_atomic(kaddr);
762 SetPageUptodate(page);
763 /* clear page dirty so that writepages wouldn't work for us. */
764 ClearPageDirty(page);
765
766 ext4_write_unlock_xattr(inode, &no_expand);
767 brelse(iloc.bh);
768 mark_inode_dirty(inode);
769 out:
770 return copied;
771 }
772
773 struct buffer_head *
774 ext4_journalled_write_inline_data(struct inode *inode,
775 unsigned len,
776 struct page *page)
777 {
778 int ret, no_expand;
779 void *kaddr;
780 struct ext4_iloc iloc;
781
782 ret = ext4_get_inode_loc(inode, &iloc);
783 if (ret) {
784 ext4_std_error(inode->i_sb, ret);
785 return NULL;
786 }
787
788 ext4_write_lock_xattr(inode, &no_expand);
789 kaddr = kmap_atomic(page);
790 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
791 kunmap_atomic(kaddr);
792 ext4_write_unlock_xattr(inode, &no_expand);
793
794 return iloc.bh;
795 }
796
797 /*
798 * Try to make the page cache and handle ready for the inline data case.
799 * We can call this function in 2 cases:
800 * 1. The inode is created and the first write exceeds inline size. We can
801 * clear the inode state safely.
802 * 2. The inode has inline data, then we need to read the data, make it
803 * update and dirty so that ext4_da_writepages can handle it. We don't
804 * need to start the journal since the file's metatdata isn't changed now.
805 */
806 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
807 struct inode *inode,
808 unsigned flags,
809 void **fsdata)
810 {
811 int ret = 0, inline_size;
812 struct page *page;
813
814 page = grab_cache_page_write_begin(mapping, 0, flags);
815 if (!page)
816 return -ENOMEM;
817
818 down_read(&EXT4_I(inode)->xattr_sem);
819 if (!ext4_has_inline_data(inode)) {
820 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
821 goto out;
822 }
823
824 inline_size = ext4_get_inline_size(inode);
825
826 if (!PageUptodate(page)) {
827 ret = ext4_read_inline_page(inode, page);
828 if (ret < 0)
829 goto out;
830 }
831
832 ret = __block_write_begin(page, 0, inline_size,
833 ext4_da_get_block_prep);
834 if (ret) {
835 up_read(&EXT4_I(inode)->xattr_sem);
836 unlock_page(page);
837 put_page(page);
838 ext4_truncate_failed_write(inode);
839 return ret;
840 }
841
842 SetPageDirty(page);
843 SetPageUptodate(page);
844 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
845 *fsdata = (void *)CONVERT_INLINE_DATA;
846
847 out:
848 up_read(&EXT4_I(inode)->xattr_sem);
849 if (page) {
850 unlock_page(page);
851 put_page(page);
852 }
853 return ret;
854 }
855
856 /*
857 * Prepare the write for the inline data.
858 * If the the data can be written into the inode, we just read
859 * the page and make it uptodate, and start the journal.
860 * Otherwise read the page, makes it dirty so that it can be
861 * handle in writepages(the i_disksize update is left to the
862 * normal ext4_da_write_end).
863 */
864 int ext4_da_write_inline_data_begin(struct address_space *mapping,
865 struct inode *inode,
866 loff_t pos, unsigned len,
867 unsigned flags,
868 struct page **pagep,
869 void **fsdata)
870 {
871 int ret, inline_size;
872 handle_t *handle;
873 struct page *page;
874 struct ext4_iloc iloc;
875 int retries = 0;
876
877 ret = ext4_get_inode_loc(inode, &iloc);
878 if (ret)
879 return ret;
880
881 retry_journal:
882 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
883 if (IS_ERR(handle)) {
884 ret = PTR_ERR(handle);
885 goto out;
886 }
887
888 inline_size = ext4_get_max_inline_size(inode);
889
890 ret = -ENOSPC;
891 if (inline_size >= pos + len) {
892 ret = ext4_prepare_inline_data(handle, inode, pos + len);
893 if (ret && ret != -ENOSPC)
894 goto out_journal;
895 }
896
897 /*
898 * We cannot recurse into the filesystem as the transaction
899 * is already started.
900 */
901 flags |= AOP_FLAG_NOFS;
902
903 if (ret == -ENOSPC) {
904 ext4_journal_stop(handle);
905 ret = ext4_da_convert_inline_data_to_extent(mapping,
906 inode,
907 flags,
908 fsdata);
909 if (ret == -ENOSPC &&
910 ext4_should_retry_alloc(inode->i_sb, &retries))
911 goto retry_journal;
912 goto out;
913 }
914
915 page = grab_cache_page_write_begin(mapping, 0, flags);
916 if (!page) {
917 ret = -ENOMEM;
918 goto out_journal;
919 }
920
921 down_read(&EXT4_I(inode)->xattr_sem);
922 if (!ext4_has_inline_data(inode)) {
923 ret = 0;
924 goto out_release_page;
925 }
926
927 if (!PageUptodate(page)) {
928 ret = ext4_read_inline_page(inode, page);
929 if (ret < 0)
930 goto out_release_page;
931 }
932 ret = ext4_journal_get_write_access(handle, iloc.bh);
933 if (ret)
934 goto out_release_page;
935
936 up_read(&EXT4_I(inode)->xattr_sem);
937 *pagep = page;
938 brelse(iloc.bh);
939 return 1;
940 out_release_page:
941 up_read(&EXT4_I(inode)->xattr_sem);
942 unlock_page(page);
943 put_page(page);
944 out_journal:
945 ext4_journal_stop(handle);
946 out:
947 brelse(iloc.bh);
948 return ret;
949 }
950
951 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
952 unsigned len, unsigned copied,
953 struct page *page)
954 {
955 int ret;
956
957 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
958 if (ret < 0) {
959 unlock_page(page);
960 put_page(page);
961 return ret;
962 }
963 copied = ret;
964
965 /*
966 * No need to use i_size_read() here, the i_size
967 * cannot change under us because we hold i_mutex.
968 *
969 * But it's important to update i_size while still holding page lock:
970 * page writeout could otherwise come in and zero beyond i_size.
971 */
972 if (pos+copied > inode->i_size)
973 i_size_write(inode, pos+copied);
974 unlock_page(page);
975 put_page(page);
976
977 /*
978 * Don't mark the inode dirty under page lock. First, it unnecessarily
979 * makes the holding time of page lock longer. Second, it forces lock
980 * ordering of page lock and transaction start for journaling
981 * filesystems.
982 */
983 mark_inode_dirty(inode);
984
985 return copied;
986 }
987
988 #ifdef INLINE_DIR_DEBUG
989 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
990 void *inline_start, int inline_size)
991 {
992 int offset;
993 unsigned short de_len;
994 struct ext4_dir_entry_2 *de = inline_start;
995 void *dlimit = inline_start + inline_size;
996
997 trace_printk("inode %lu\n", dir->i_ino);
998 offset = 0;
999 while ((void *)de < dlimit) {
1000 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1001 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1002 offset, de_len, de->name_len, de->name,
1003 de->name_len, le32_to_cpu(de->inode));
1004 if (ext4_check_dir_entry(dir, NULL, de, bh,
1005 inline_start, inline_size, offset))
1006 BUG();
1007
1008 offset += de_len;
1009 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1010 }
1011 }
1012 #else
1013 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1014 #endif
1015
1016 /*
1017 * Add a new entry into a inline dir.
1018 * It will return -ENOSPC if no space is available, and -EIO
1019 * and -EEXIST if directory entry already exists.
1020 */
1021 static int ext4_add_dirent_to_inline(handle_t *handle,
1022 struct ext4_filename *fname,
1023 struct inode *dir,
1024 struct inode *inode,
1025 struct ext4_iloc *iloc,
1026 void *inline_start, int inline_size)
1027 {
1028 int err;
1029 struct ext4_dir_entry_2 *de;
1030
1031 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1032 inline_size, fname, &de);
1033 if (err)
1034 return err;
1035
1036 BUFFER_TRACE(iloc->bh, "get_write_access");
1037 err = ext4_journal_get_write_access(handle, iloc->bh);
1038 if (err)
1039 return err;
1040 ext4_insert_dentry(inode, de, inline_size, fname);
1041
1042 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1043
1044 /*
1045 * XXX shouldn't update any times until successful
1046 * completion of syscall, but too many callers depend
1047 * on this.
1048 *
1049 * XXX similarly, too many callers depend on
1050 * ext4_new_inode() setting the times, but error
1051 * recovery deletes the inode, so the worst that can
1052 * happen is that the times are slightly out of date
1053 * and/or different from the directory change time.
1054 */
1055 dir->i_mtime = dir->i_ctime = current_time(dir);
1056 ext4_update_dx_flag(dir);
1057 dir->i_version++;
1058 return 1;
1059 }
1060
1061 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1062 struct ext4_iloc *iloc)
1063 {
1064 struct ext4_xattr_entry *entry;
1065 struct ext4_xattr_ibody_header *header;
1066
1067 BUG_ON(!EXT4_I(inode)->i_inline_off);
1068
1069 header = IHDR(inode, ext4_raw_inode(iloc));
1070 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1071 EXT4_I(inode)->i_inline_off);
1072
1073 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1074 }
1075
1076 /* Set the final de to cover the whole block. */
1077 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1078 {
1079 struct ext4_dir_entry_2 *de, *prev_de;
1080 void *limit;
1081 int de_len;
1082
1083 de = (struct ext4_dir_entry_2 *)de_buf;
1084 if (old_size) {
1085 limit = de_buf + old_size;
1086 do {
1087 prev_de = de;
1088 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1089 de_buf += de_len;
1090 de = (struct ext4_dir_entry_2 *)de_buf;
1091 } while (de_buf < limit);
1092
1093 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1094 old_size, new_size);
1095 } else {
1096 /* this is just created, so create an empty entry. */
1097 de->inode = 0;
1098 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1099 }
1100 }
1101
1102 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1103 struct ext4_iloc *iloc)
1104 {
1105 int ret;
1106 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1107 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1108
1109 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1110 return -ENOSPC;
1111
1112 ret = ext4_update_inline_data(handle, dir,
1113 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1114 if (ret)
1115 return ret;
1116
1117 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1118 EXT4_I(dir)->i_inline_size -
1119 EXT4_MIN_INLINE_DATA_SIZE);
1120 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1121 return 0;
1122 }
1123
1124 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1125 struct ext4_iloc *iloc,
1126 void *buf, int inline_size)
1127 {
1128 ext4_create_inline_data(handle, inode, inline_size);
1129 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1130 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1131 }
1132
1133 static int ext4_finish_convert_inline_dir(handle_t *handle,
1134 struct inode *inode,
1135 struct buffer_head *dir_block,
1136 void *buf,
1137 int inline_size)
1138 {
1139 int err, csum_size = 0, header_size = 0;
1140 struct ext4_dir_entry_2 *de;
1141 struct ext4_dir_entry_tail *t;
1142 void *target = dir_block->b_data;
1143
1144 /*
1145 * First create "." and ".." and then copy the dir information
1146 * back to the block.
1147 */
1148 de = (struct ext4_dir_entry_2 *)target;
1149 de = ext4_init_dot_dotdot(inode, de,
1150 inode->i_sb->s_blocksize, csum_size,
1151 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1152 header_size = (void *)de - target;
1153
1154 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1155 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1156
1157 if (ext4_has_metadata_csum(inode->i_sb))
1158 csum_size = sizeof(struct ext4_dir_entry_tail);
1159
1160 inode->i_size = inode->i_sb->s_blocksize;
1161 i_size_write(inode, inode->i_sb->s_blocksize);
1162 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1163 ext4_update_final_de(dir_block->b_data,
1164 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1165 inode->i_sb->s_blocksize - csum_size);
1166
1167 if (csum_size) {
1168 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1169 inode->i_sb->s_blocksize);
1170 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1171 }
1172 set_buffer_uptodate(dir_block);
1173 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1174 if (err)
1175 return err;
1176 set_buffer_verified(dir_block);
1177 return ext4_mark_inode_dirty(handle, inode);
1178 }
1179
1180 static int ext4_convert_inline_data_nolock(handle_t *handle,
1181 struct inode *inode,
1182 struct ext4_iloc *iloc)
1183 {
1184 int error;
1185 void *buf = NULL;
1186 struct buffer_head *data_bh = NULL;
1187 struct ext4_map_blocks map;
1188 int inline_size;
1189
1190 inline_size = ext4_get_inline_size(inode);
1191 buf = kmalloc(inline_size, GFP_NOFS);
1192 if (!buf) {
1193 error = -ENOMEM;
1194 goto out;
1195 }
1196
1197 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1198 if (error < 0)
1199 goto out;
1200
1201 /*
1202 * Make sure the inline directory entries pass checks before we try to
1203 * convert them, so that we avoid touching stuff that needs fsck.
1204 */
1205 if (S_ISDIR(inode->i_mode)) {
1206 error = ext4_check_all_de(inode, iloc->bh,
1207 buf + EXT4_INLINE_DOTDOT_SIZE,
1208 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1209 if (error)
1210 goto out;
1211 }
1212
1213 error = ext4_destroy_inline_data_nolock(handle, inode);
1214 if (error)
1215 goto out;
1216
1217 map.m_lblk = 0;
1218 map.m_len = 1;
1219 map.m_flags = 0;
1220 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1221 if (error < 0)
1222 goto out_restore;
1223 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1224 error = -EIO;
1225 goto out_restore;
1226 }
1227
1228 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1229 if (!data_bh) {
1230 error = -ENOMEM;
1231 goto out_restore;
1232 }
1233
1234 lock_buffer(data_bh);
1235 error = ext4_journal_get_create_access(handle, data_bh);
1236 if (error) {
1237 unlock_buffer(data_bh);
1238 error = -EIO;
1239 goto out_restore;
1240 }
1241 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1242
1243 if (!S_ISDIR(inode->i_mode)) {
1244 memcpy(data_bh->b_data, buf, inline_size);
1245 set_buffer_uptodate(data_bh);
1246 error = ext4_handle_dirty_metadata(handle,
1247 inode, data_bh);
1248 } else {
1249 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1250 buf, inline_size);
1251 }
1252
1253 unlock_buffer(data_bh);
1254 out_restore:
1255 if (error)
1256 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1257
1258 out:
1259 brelse(data_bh);
1260 kfree(buf);
1261 return error;
1262 }
1263
1264 /*
1265 * Try to add the new entry to the inline data.
1266 * If succeeds, return 0. If not, extended the inline dir and copied data to
1267 * the new created block.
1268 */
1269 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1270 struct inode *dir, struct inode *inode)
1271 {
1272 int ret, inline_size, no_expand;
1273 void *inline_start;
1274 struct ext4_iloc iloc;
1275
1276 ret = ext4_get_inode_loc(dir, &iloc);
1277 if (ret)
1278 return ret;
1279
1280 ext4_write_lock_xattr(dir, &no_expand);
1281 if (!ext4_has_inline_data(dir))
1282 goto out;
1283
1284 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1285 EXT4_INLINE_DOTDOT_SIZE;
1286 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1287
1288 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1289 inline_start, inline_size);
1290 if (ret != -ENOSPC)
1291 goto out;
1292
1293 /* check whether it can be inserted to inline xattr space. */
1294 inline_size = EXT4_I(dir)->i_inline_size -
1295 EXT4_MIN_INLINE_DATA_SIZE;
1296 if (!inline_size) {
1297 /* Try to use the xattr space.*/
1298 ret = ext4_update_inline_dir(handle, dir, &iloc);
1299 if (ret && ret != -ENOSPC)
1300 goto out;
1301
1302 inline_size = EXT4_I(dir)->i_inline_size -
1303 EXT4_MIN_INLINE_DATA_SIZE;
1304 }
1305
1306 if (inline_size) {
1307 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1308
1309 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1310 inode, &iloc, inline_start,
1311 inline_size);
1312
1313 if (ret != -ENOSPC)
1314 goto out;
1315 }
1316
1317 /*
1318 * The inline space is filled up, so create a new block for it.
1319 * As the extent tree will be created, we have to save the inline
1320 * dir first.
1321 */
1322 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1323
1324 out:
1325 ext4_write_unlock_xattr(dir, &no_expand);
1326 ext4_mark_inode_dirty(handle, dir);
1327 brelse(iloc.bh);
1328 return ret;
1329 }
1330
1331 /*
1332 * This function fills a red-black tree with information from an
1333 * inlined dir. It returns the number directory entries loaded
1334 * into the tree. If there is an error it is returned in err.
1335 */
1336 int htree_inlinedir_to_tree(struct file *dir_file,
1337 struct inode *dir, ext4_lblk_t block,
1338 struct dx_hash_info *hinfo,
1339 __u32 start_hash, __u32 start_minor_hash,
1340 int *has_inline_data)
1341 {
1342 int err = 0, count = 0;
1343 unsigned int parent_ino;
1344 int pos;
1345 struct ext4_dir_entry_2 *de;
1346 struct inode *inode = file_inode(dir_file);
1347 int ret, inline_size = 0;
1348 struct ext4_iloc iloc;
1349 void *dir_buf = NULL;
1350 struct ext4_dir_entry_2 fake;
1351 struct fscrypt_str tmp_str;
1352
1353 ret = ext4_get_inode_loc(inode, &iloc);
1354 if (ret)
1355 return ret;
1356
1357 down_read(&EXT4_I(inode)->xattr_sem);
1358 if (!ext4_has_inline_data(inode)) {
1359 up_read(&EXT4_I(inode)->xattr_sem);
1360 *has_inline_data = 0;
1361 goto out;
1362 }
1363
1364 inline_size = ext4_get_inline_size(inode);
1365 dir_buf = kmalloc(inline_size, GFP_NOFS);
1366 if (!dir_buf) {
1367 ret = -ENOMEM;
1368 up_read(&EXT4_I(inode)->xattr_sem);
1369 goto out;
1370 }
1371
1372 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1373 up_read(&EXT4_I(inode)->xattr_sem);
1374 if (ret < 0)
1375 goto out;
1376
1377 pos = 0;
1378 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1379 while (pos < inline_size) {
1380 /*
1381 * As inlined dir doesn't store any information about '.' and
1382 * only the inode number of '..' is stored, we have to handle
1383 * them differently.
1384 */
1385 if (pos == 0) {
1386 fake.inode = cpu_to_le32(inode->i_ino);
1387 fake.name_len = 1;
1388 strcpy(fake.name, ".");
1389 fake.rec_len = ext4_rec_len_to_disk(
1390 EXT4_DIR_REC_LEN(fake.name_len),
1391 inline_size);
1392 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1393 de = &fake;
1394 pos = EXT4_INLINE_DOTDOT_OFFSET;
1395 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1396 fake.inode = cpu_to_le32(parent_ino);
1397 fake.name_len = 2;
1398 strcpy(fake.name, "..");
1399 fake.rec_len = ext4_rec_len_to_disk(
1400 EXT4_DIR_REC_LEN(fake.name_len),
1401 inline_size);
1402 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1403 de = &fake;
1404 pos = EXT4_INLINE_DOTDOT_SIZE;
1405 } else {
1406 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1407 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1408 if (ext4_check_dir_entry(inode, dir_file, de,
1409 iloc.bh, dir_buf,
1410 inline_size, pos)) {
1411 ret = count;
1412 goto out;
1413 }
1414 }
1415
1416 ext4fs_dirhash(de->name, de->name_len, hinfo);
1417 if ((hinfo->hash < start_hash) ||
1418 ((hinfo->hash == start_hash) &&
1419 (hinfo->minor_hash < start_minor_hash)))
1420 continue;
1421 if (de->inode == 0)
1422 continue;
1423 tmp_str.name = de->name;
1424 tmp_str.len = de->name_len;
1425 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1426 hinfo->minor_hash, de, &tmp_str);
1427 if (err) {
1428 ret = err;
1429 goto out;
1430 }
1431 count++;
1432 }
1433 ret = count;
1434 out:
1435 kfree(dir_buf);
1436 brelse(iloc.bh);
1437 return ret;
1438 }
1439
1440 /*
1441 * So this function is called when the volume is mkfsed with
1442 * dir_index disabled. In order to keep f_pos persistent
1443 * after we convert from an inlined dir to a blocked based,
1444 * we just pretend that we are a normal dir and return the
1445 * offset as if '.' and '..' really take place.
1446 *
1447 */
1448 int ext4_read_inline_dir(struct file *file,
1449 struct dir_context *ctx,
1450 int *has_inline_data)
1451 {
1452 unsigned int offset, parent_ino;
1453 int i;
1454 struct ext4_dir_entry_2 *de;
1455 struct super_block *sb;
1456 struct inode *inode = file_inode(file);
1457 int ret, inline_size = 0;
1458 struct ext4_iloc iloc;
1459 void *dir_buf = NULL;
1460 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1461
1462 ret = ext4_get_inode_loc(inode, &iloc);
1463 if (ret)
1464 return ret;
1465
1466 down_read(&EXT4_I(inode)->xattr_sem);
1467 if (!ext4_has_inline_data(inode)) {
1468 up_read(&EXT4_I(inode)->xattr_sem);
1469 *has_inline_data = 0;
1470 goto out;
1471 }
1472
1473 inline_size = ext4_get_inline_size(inode);
1474 dir_buf = kmalloc(inline_size, GFP_NOFS);
1475 if (!dir_buf) {
1476 ret = -ENOMEM;
1477 up_read(&EXT4_I(inode)->xattr_sem);
1478 goto out;
1479 }
1480
1481 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1482 up_read(&EXT4_I(inode)->xattr_sem);
1483 if (ret < 0)
1484 goto out;
1485
1486 ret = 0;
1487 sb = inode->i_sb;
1488 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1489 offset = ctx->pos;
1490
1491 /*
1492 * dotdot_offset and dotdot_size is the real offset and
1493 * size for ".." and "." if the dir is block based while
1494 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1495 * So we will use extra_offset and extra_size to indicate them
1496 * during the inline dir iteration.
1497 */
1498 dotdot_offset = EXT4_DIR_REC_LEN(1);
1499 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1500 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1501 extra_size = extra_offset + inline_size;
1502
1503 /*
1504 * If the version has changed since the last call to
1505 * readdir(2), then we might be pointing to an invalid
1506 * dirent right now. Scan from the start of the inline
1507 * dir to make sure.
1508 */
1509 if (file->f_version != inode->i_version) {
1510 for (i = 0; i < extra_size && i < offset;) {
1511 /*
1512 * "." is with offset 0 and
1513 * ".." is dotdot_offset.
1514 */
1515 if (!i) {
1516 i = dotdot_offset;
1517 continue;
1518 } else if (i == dotdot_offset) {
1519 i = dotdot_size;
1520 continue;
1521 }
1522 /* for other entry, the real offset in
1523 * the buf has to be tuned accordingly.
1524 */
1525 de = (struct ext4_dir_entry_2 *)
1526 (dir_buf + i - extra_offset);
1527 /* It's too expensive to do a full
1528 * dirent test each time round this
1529 * loop, but we do have to test at
1530 * least that it is non-zero. A
1531 * failure will be detected in the
1532 * dirent test below. */
1533 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1534 < EXT4_DIR_REC_LEN(1))
1535 break;
1536 i += ext4_rec_len_from_disk(de->rec_len,
1537 extra_size);
1538 }
1539 offset = i;
1540 ctx->pos = offset;
1541 file->f_version = inode->i_version;
1542 }
1543
1544 while (ctx->pos < extra_size) {
1545 if (ctx->pos == 0) {
1546 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1547 goto out;
1548 ctx->pos = dotdot_offset;
1549 continue;
1550 }
1551
1552 if (ctx->pos == dotdot_offset) {
1553 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1554 goto out;
1555 ctx->pos = dotdot_size;
1556 continue;
1557 }
1558
1559 de = (struct ext4_dir_entry_2 *)
1560 (dir_buf + ctx->pos - extra_offset);
1561 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1562 extra_size, ctx->pos))
1563 goto out;
1564 if (le32_to_cpu(de->inode)) {
1565 if (!dir_emit(ctx, de->name, de->name_len,
1566 le32_to_cpu(de->inode),
1567 get_dtype(sb, de->file_type)))
1568 goto out;
1569 }
1570 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1571 }
1572 out:
1573 kfree(dir_buf);
1574 brelse(iloc.bh);
1575 return ret;
1576 }
1577
1578 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1579 struct ext4_dir_entry_2 **parent_de,
1580 int *retval)
1581 {
1582 struct ext4_iloc iloc;
1583
1584 *retval = ext4_get_inode_loc(inode, &iloc);
1585 if (*retval)
1586 return NULL;
1587
1588 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1589
1590 return iloc.bh;
1591 }
1592
1593 /*
1594 * Try to create the inline data for the new dir.
1595 * If it succeeds, return 0, otherwise return the error.
1596 * In case of ENOSPC, the caller should create the normal disk layout dir.
1597 */
1598 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1599 struct inode *inode)
1600 {
1601 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1602 struct ext4_iloc iloc;
1603 struct ext4_dir_entry_2 *de;
1604
1605 ret = ext4_get_inode_loc(inode, &iloc);
1606 if (ret)
1607 return ret;
1608
1609 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1610 if (ret)
1611 goto out;
1612
1613 /*
1614 * For inline dir, we only save the inode information for the ".."
1615 * and create a fake dentry to cover the left space.
1616 */
1617 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1618 de->inode = cpu_to_le32(parent->i_ino);
1619 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1620 de->inode = 0;
1621 de->rec_len = ext4_rec_len_to_disk(
1622 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1623 inline_size);
1624 set_nlink(inode, 2);
1625 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1626 out:
1627 brelse(iloc.bh);
1628 return ret;
1629 }
1630
1631 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1632 struct ext4_filename *fname,
1633 struct ext4_dir_entry_2 **res_dir,
1634 int *has_inline_data)
1635 {
1636 int ret;
1637 struct ext4_iloc iloc;
1638 void *inline_start;
1639 int inline_size;
1640
1641 if (ext4_get_inode_loc(dir, &iloc))
1642 return NULL;
1643
1644 down_read(&EXT4_I(dir)->xattr_sem);
1645 if (!ext4_has_inline_data(dir)) {
1646 *has_inline_data = 0;
1647 goto out;
1648 }
1649
1650 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1651 EXT4_INLINE_DOTDOT_SIZE;
1652 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1653 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1654 dir, fname, 0, res_dir);
1655 if (ret == 1)
1656 goto out_find;
1657 if (ret < 0)
1658 goto out;
1659
1660 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1661 goto out;
1662
1663 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1664 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1665
1666 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1667 dir, fname, 0, res_dir);
1668 if (ret == 1)
1669 goto out_find;
1670
1671 out:
1672 brelse(iloc.bh);
1673 iloc.bh = NULL;
1674 out_find:
1675 up_read(&EXT4_I(dir)->xattr_sem);
1676 return iloc.bh;
1677 }
1678
1679 int ext4_delete_inline_entry(handle_t *handle,
1680 struct inode *dir,
1681 struct ext4_dir_entry_2 *de_del,
1682 struct buffer_head *bh,
1683 int *has_inline_data)
1684 {
1685 int err, inline_size, no_expand;
1686 struct ext4_iloc iloc;
1687 void *inline_start;
1688
1689 err = ext4_get_inode_loc(dir, &iloc);
1690 if (err)
1691 return err;
1692
1693 ext4_write_lock_xattr(dir, &no_expand);
1694 if (!ext4_has_inline_data(dir)) {
1695 *has_inline_data = 0;
1696 goto out;
1697 }
1698
1699 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1700 EXT4_MIN_INLINE_DATA_SIZE) {
1701 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1702 EXT4_INLINE_DOTDOT_SIZE;
1703 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1704 EXT4_INLINE_DOTDOT_SIZE;
1705 } else {
1706 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1707 inline_size = ext4_get_inline_size(dir) -
1708 EXT4_MIN_INLINE_DATA_SIZE;
1709 }
1710
1711 BUFFER_TRACE(bh, "get_write_access");
1712 err = ext4_journal_get_write_access(handle, bh);
1713 if (err)
1714 goto out;
1715
1716 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1717 inline_start, inline_size, 0);
1718 if (err)
1719 goto out;
1720
1721 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1722 out:
1723 ext4_write_unlock_xattr(dir, &no_expand);
1724 if (likely(err == 0))
1725 err = ext4_mark_inode_dirty(handle, dir);
1726 brelse(iloc.bh);
1727 if (err != -ENOENT)
1728 ext4_std_error(dir->i_sb, err);
1729 return err;
1730 }
1731
1732 /*
1733 * Get the inline dentry at offset.
1734 */
1735 static inline struct ext4_dir_entry_2 *
1736 ext4_get_inline_entry(struct inode *inode,
1737 struct ext4_iloc *iloc,
1738 unsigned int offset,
1739 void **inline_start,
1740 int *inline_size)
1741 {
1742 void *inline_pos;
1743
1744 BUG_ON(offset > ext4_get_inline_size(inode));
1745
1746 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1747 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1748 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1749 } else {
1750 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1751 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1752 *inline_size = ext4_get_inline_size(inode) -
1753 EXT4_MIN_INLINE_DATA_SIZE;
1754 }
1755
1756 if (inline_start)
1757 *inline_start = inline_pos;
1758 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1759 }
1760
1761 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1762 {
1763 int err, inline_size;
1764 struct ext4_iloc iloc;
1765 size_t inline_len;
1766 void *inline_pos;
1767 unsigned int offset;
1768 struct ext4_dir_entry_2 *de;
1769 bool ret = true;
1770
1771 err = ext4_get_inode_loc(dir, &iloc);
1772 if (err) {
1773 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1774 err, dir->i_ino);
1775 return true;
1776 }
1777
1778 down_read(&EXT4_I(dir)->xattr_sem);
1779 if (!ext4_has_inline_data(dir)) {
1780 *has_inline_data = 0;
1781 goto out;
1782 }
1783
1784 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1785 if (!le32_to_cpu(de->inode)) {
1786 ext4_warning(dir->i_sb,
1787 "bad inline directory (dir #%lu) - no `..'",
1788 dir->i_ino);
1789 ret = true;
1790 goto out;
1791 }
1792
1793 inline_len = ext4_get_inline_size(dir);
1794 offset = EXT4_INLINE_DOTDOT_SIZE;
1795 while (offset < inline_len) {
1796 de = ext4_get_inline_entry(dir, &iloc, offset,
1797 &inline_pos, &inline_size);
1798 if (ext4_check_dir_entry(dir, NULL, de,
1799 iloc.bh, inline_pos,
1800 inline_size, offset)) {
1801 ext4_warning(dir->i_sb,
1802 "bad inline directory (dir #%lu) - "
1803 "inode %u, rec_len %u, name_len %d"
1804 "inline size %d",
1805 dir->i_ino, le32_to_cpu(de->inode),
1806 le16_to_cpu(de->rec_len), de->name_len,
1807 inline_size);
1808 ret = true;
1809 goto out;
1810 }
1811 if (le32_to_cpu(de->inode)) {
1812 ret = false;
1813 goto out;
1814 }
1815 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1816 }
1817
1818 out:
1819 up_read(&EXT4_I(dir)->xattr_sem);
1820 brelse(iloc.bh);
1821 return ret;
1822 }
1823
1824 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1825 {
1826 int ret, no_expand;
1827
1828 ext4_write_lock_xattr(inode, &no_expand);
1829 ret = ext4_destroy_inline_data_nolock(handle, inode);
1830 ext4_write_unlock_xattr(inode, &no_expand);
1831
1832 return ret;
1833 }
1834
1835 int ext4_inline_data_fiemap(struct inode *inode,
1836 struct fiemap_extent_info *fieinfo,
1837 int *has_inline, __u64 start, __u64 len)
1838 {
1839 __u64 physical = 0;
1840 __u64 inline_len;
1841 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1842 FIEMAP_EXTENT_LAST;
1843 int error = 0;
1844 struct ext4_iloc iloc;
1845
1846 down_read(&EXT4_I(inode)->xattr_sem);
1847 if (!ext4_has_inline_data(inode)) {
1848 *has_inline = 0;
1849 goto out;
1850 }
1851 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1852 i_size_read(inode));
1853 if (start >= inline_len)
1854 goto out;
1855 if (start + len < inline_len)
1856 inline_len = start + len;
1857 inline_len -= start;
1858
1859 error = ext4_get_inode_loc(inode, &iloc);
1860 if (error)
1861 goto out;
1862
1863 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1864 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1865 physical += offsetof(struct ext4_inode, i_block);
1866
1867 brelse(iloc.bh);
1868 out:
1869 up_read(&EXT4_I(inode)->xattr_sem);
1870 if (physical)
1871 error = fiemap_fill_next_extent(fieinfo, start, physical,
1872 inline_len, flags);
1873 return (error < 0 ? error : 0);
1874 }
1875
1876 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1877 {
1878 handle_t *handle;
1879 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1880 size_t i_size;
1881 void *value = NULL;
1882 struct ext4_xattr_ibody_find is = {
1883 .s = { .not_found = -ENODATA, },
1884 };
1885 struct ext4_xattr_info i = {
1886 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1887 .name = EXT4_XATTR_SYSTEM_DATA,
1888 };
1889
1890
1891 needed_blocks = ext4_writepage_trans_blocks(inode);
1892 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1893 if (IS_ERR(handle))
1894 return PTR_ERR(handle);
1895
1896 ext4_write_lock_xattr(inode, &no_expand);
1897 if (!ext4_has_inline_data(inode)) {
1898 ext4_write_unlock_xattr(inode, &no_expand);
1899 *has_inline = 0;
1900 ext4_journal_stop(handle);
1901 return 0;
1902 }
1903
1904 if ((err = ext4_orphan_add(handle, inode)) != 0)
1905 goto out;
1906
1907 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1908 goto out;
1909
1910 down_write(&EXT4_I(inode)->i_data_sem);
1911 i_size = inode->i_size;
1912 inline_size = ext4_get_inline_size(inode);
1913 EXT4_I(inode)->i_disksize = i_size;
1914
1915 if (i_size < inline_size) {
1916 /* Clear the content in the xattr space. */
1917 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1918 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1919 goto out_error;
1920
1921 BUG_ON(is.s.not_found);
1922
1923 value_len = le32_to_cpu(is.s.here->e_value_size);
1924 value = kmalloc(value_len, GFP_NOFS);
1925 if (!value) {
1926 err = -ENOMEM;
1927 goto out_error;
1928 }
1929
1930 err = ext4_xattr_ibody_get(inode, i.name_index,
1931 i.name, value, value_len);
1932 if (err <= 0)
1933 goto out_error;
1934
1935 i.value = value;
1936 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1937 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1938 err = ext4_xattr_ibody_inline_set(handle, inode,
1939 &i, &is);
1940 if (err)
1941 goto out_error;
1942 }
1943
1944 /* Clear the content within i_blocks. */
1945 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1946 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1947 memset(p + i_size, 0,
1948 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1949 }
1950
1951 EXT4_I(inode)->i_inline_size = i_size <
1952 EXT4_MIN_INLINE_DATA_SIZE ?
1953 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1954 }
1955
1956 out_error:
1957 up_write(&EXT4_I(inode)->i_data_sem);
1958 out:
1959 brelse(is.iloc.bh);
1960 ext4_write_unlock_xattr(inode, &no_expand);
1961 kfree(value);
1962 if (inode->i_nlink)
1963 ext4_orphan_del(handle, inode);
1964
1965 if (err == 0) {
1966 inode->i_mtime = inode->i_ctime = current_time(inode);
1967 err = ext4_mark_inode_dirty(handle, inode);
1968 if (IS_SYNC(inode))
1969 ext4_handle_sync(handle);
1970 }
1971 ext4_journal_stop(handle);
1972 return err;
1973 }
1974
1975 int ext4_convert_inline_data(struct inode *inode)
1976 {
1977 int error, needed_blocks, no_expand;
1978 handle_t *handle;
1979 struct ext4_iloc iloc;
1980
1981 if (!ext4_has_inline_data(inode)) {
1982 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1983 return 0;
1984 }
1985
1986 needed_blocks = ext4_writepage_trans_blocks(inode);
1987
1988 iloc.bh = NULL;
1989 error = ext4_get_inode_loc(inode, &iloc);
1990 if (error)
1991 return error;
1992
1993 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1994 if (IS_ERR(handle)) {
1995 error = PTR_ERR(handle);
1996 goto out_free;
1997 }
1998
1999 ext4_write_lock_xattr(inode, &no_expand);
2000 if (ext4_has_inline_data(inode))
2001 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2002 ext4_write_unlock_xattr(inode, &no_expand);
2003 ext4_journal_stop(handle);
2004 out_free:
2005 brelse(iloc.bh);
2006 return error;
2007 }
Linux with added FPC-III support