?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
gro: Allow tunnel stacking in the case of FOU/GUE
[linux/fpc-iii.git] / fs / ext4 / inline.c
blobd77d542c2ed50d139621656d97c2480544e5aade
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_block && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
66 if (offs < min_offs)
67 min_offs = offs;
68 }
69 }
70 free = min_offs -
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
78 goto out;
79 }
80
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
85 else
86 free = 0;
87
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 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154 (void *)ext4_raw_inode(&is.iloc));
155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156 le32_to_cpu(is.s.here->e_value_size);
157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
158 }
159 out:
160 brelse(is.iloc.bh);
161 return error;
162 }
163
164 static int ext4_read_inline_data(struct inode *inode, void *buffer,
165 unsigned int len,
166 struct ext4_iloc *iloc)
167 {
168 struct ext4_xattr_entry *entry;
169 struct ext4_xattr_ibody_header *header;
170 int cp_len = 0;
171 struct ext4_inode *raw_inode;
172
173 if (!len)
174 return 0;
175
176 BUG_ON(len > EXT4_I(inode)->i_inline_size);
177
178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179 len : EXT4_MIN_INLINE_DATA_SIZE;
180
181 raw_inode = ext4_raw_inode(iloc);
182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
183
184 len -= cp_len;
185 buffer += cp_len;
186
187 if (!len)
188 goto out;
189
190 header = IHDR(inode, raw_inode);
191 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192 EXT4_I(inode)->i_inline_off);
193 len = min_t(unsigned int, len,
194 (unsigned int)le32_to_cpu(entry->e_value_size));
195
196 memcpy(buffer,
197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
198 cp_len += len;
199
200 out:
201 return cp_len;
202 }
203
204 /*
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
209 */
210 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211 void *buffer, loff_t pos, unsigned int len)
212 {
213 struct ext4_xattr_entry *entry;
214 struct ext4_xattr_ibody_header *header;
215 struct ext4_inode *raw_inode;
216 int cp_len = 0;
217
218 BUG_ON(!EXT4_I(inode)->i_inline_off);
219 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
220
221 raw_inode = ext4_raw_inode(iloc);
222 buffer += pos;
223
224 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
225 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
226 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
227 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
228
229 len -= cp_len;
230 buffer += cp_len;
231 pos += cp_len;
232 }
233
234 if (!len)
235 return;
236
237 pos -= EXT4_MIN_INLINE_DATA_SIZE;
238 header = IHDR(inode, raw_inode);
239 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
240 EXT4_I(inode)->i_inline_off);
241
242 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
243 buffer, len);
244 }
245
246 static int ext4_create_inline_data(handle_t *handle,
247 struct inode *inode, unsigned len)
248 {
249 int error;
250 void *value = NULL;
251 struct ext4_xattr_ibody_find is = {
252 .s = { .not_found = -ENODATA, },
253 };
254 struct ext4_xattr_info i = {
255 .name_index = EXT4_XATTR_INDEX_SYSTEM,
256 .name = EXT4_XATTR_SYSTEM_DATA,
257 };
258
259 error = ext4_get_inode_loc(inode, &is.iloc);
260 if (error)
261 return error;
262
263 BUFFER_TRACE(is.iloc.bh, "get_write_access");
264 error = ext4_journal_get_write_access(handle, is.iloc.bh);
265 if (error)
266 goto out;
267
268 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
269 value = EXT4_ZERO_XATTR_VALUE;
270 len -= EXT4_MIN_INLINE_DATA_SIZE;
271 } else {
272 value = "";
273 len = 0;
274 }
275
276 /* Insert the the xttr entry. */
277 i.value = value;
278 i.value_len = len;
279
280 error = ext4_xattr_ibody_find(inode, &i, &is);
281 if (error)
282 goto out;
283
284 BUG_ON(!is.s.not_found);
285
286 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
287 if (error) {
288 if (error == -ENOSPC)
289 ext4_clear_inode_state(inode,
290 EXT4_STATE_MAY_INLINE_DATA);
291 goto out;
292 }
293
294 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
295 0, EXT4_MIN_INLINE_DATA_SIZE);
296
297 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
298 (void *)ext4_raw_inode(&is.iloc));
299 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
300 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
301 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
302 get_bh(is.iloc.bh);
303 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
304
305 out:
306 brelse(is.iloc.bh);
307 return error;
308 }
309
310 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
311 unsigned int len)
312 {
313 int error;
314 void *value = NULL;
315 struct ext4_xattr_ibody_find is = {
316 .s = { .not_found = -ENODATA, },
317 };
318 struct ext4_xattr_info i = {
319 .name_index = EXT4_XATTR_INDEX_SYSTEM,
320 .name = EXT4_XATTR_SYSTEM_DATA,
321 };
322
323 /* If the old space is ok, write the data directly. */
324 if (len <= EXT4_I(inode)->i_inline_size)
325 return 0;
326
327 error = ext4_get_inode_loc(inode, &is.iloc);
328 if (error)
329 return error;
330
331 error = ext4_xattr_ibody_find(inode, &i, &is);
332 if (error)
333 goto out;
334
335 BUG_ON(is.s.not_found);
336
337 len -= EXT4_MIN_INLINE_DATA_SIZE;
338 value = kzalloc(len, GFP_NOFS);
339 if (!value) {
340 error = -ENOMEM;
341 goto out;
342 }
343
344 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
345 value, len);
346 if (error == -ENODATA)
347 goto out;
348
349 BUFFER_TRACE(is.iloc.bh, "get_write_access");
350 error = ext4_journal_get_write_access(handle, is.iloc.bh);
351 if (error)
352 goto out;
353
354 /* Update the xttr entry. */
355 i.value = value;
356 i.value_len = len;
357
358 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
359 if (error)
360 goto out;
361
362 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
363 (void *)ext4_raw_inode(&is.iloc));
364 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
365 le32_to_cpu(is.s.here->e_value_size);
366 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
367 get_bh(is.iloc.bh);
368 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
369
370 out:
371 kfree(value);
372 brelse(is.iloc.bh);
373 return error;
374 }
375
376 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
377 unsigned int len)
378 {
379 int ret, size;
380 struct ext4_inode_info *ei = EXT4_I(inode);
381
382 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
383 return -ENOSPC;
384
385 size = ext4_get_max_inline_size(inode);
386 if (size < len)
387 return -ENOSPC;
388
389 down_write(&EXT4_I(inode)->xattr_sem);
390
391 if (ei->i_inline_off)
392 ret = ext4_update_inline_data(handle, inode, len);
393 else
394 ret = ext4_create_inline_data(handle, inode, len);
395
396 up_write(&EXT4_I(inode)->xattr_sem);
397
398 return ret;
399 }
400
401 static int ext4_destroy_inline_data_nolock(handle_t *handle,
402 struct inode *inode)
403 {
404 struct ext4_inode_info *ei = EXT4_I(inode);
405 struct ext4_xattr_ibody_find is = {
406 .s = { .not_found = 0, },
407 };
408 struct ext4_xattr_info i = {
409 .name_index = EXT4_XATTR_INDEX_SYSTEM,
410 .name = EXT4_XATTR_SYSTEM_DATA,
411 .value = NULL,
412 .value_len = 0,
413 };
414 int error;
415
416 if (!ei->i_inline_off)
417 return 0;
418
419 error = ext4_get_inode_loc(inode, &is.iloc);
420 if (error)
421 return error;
422
423 error = ext4_xattr_ibody_find(inode, &i, &is);
424 if (error)
425 goto out;
426
427 BUFFER_TRACE(is.iloc.bh, "get_write_access");
428 error = ext4_journal_get_write_access(handle, is.iloc.bh);
429 if (error)
430 goto out;
431
432 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
433 if (error)
434 goto out;
435
436 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
437 0, EXT4_MIN_INLINE_DATA_SIZE);
438
439 if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
440 EXT4_FEATURE_INCOMPAT_EXTENTS)) {
441 if (S_ISDIR(inode->i_mode) ||
442 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
443 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
444 ext4_ext_tree_init(handle, inode);
445 }
446 }
447 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
448
449 get_bh(is.iloc.bh);
450 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
451
452 EXT4_I(inode)->i_inline_off = 0;
453 EXT4_I(inode)->i_inline_size = 0;
454 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
455 out:
456 brelse(is.iloc.bh);
457 if (error == -ENODATA)
458 error = 0;
459 return error;
460 }
461
462 static int ext4_read_inline_page(struct inode *inode, struct page *page)
463 {
464 void *kaddr;
465 int ret = 0;
466 size_t len;
467 struct ext4_iloc iloc;
468
469 BUG_ON(!PageLocked(page));
470 BUG_ON(!ext4_has_inline_data(inode));
471 BUG_ON(page->index);
472
473 if (!EXT4_I(inode)->i_inline_off) {
474 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
475 inode->i_ino);
476 goto out;
477 }
478
479 ret = ext4_get_inode_loc(inode, &iloc);
480 if (ret)
481 goto out;
482
483 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
484 kaddr = kmap_atomic(page);
485 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
486 flush_dcache_page(page);
487 kunmap_atomic(kaddr);
488 zero_user_segment(page, len, PAGE_CACHE_SIZE);
489 SetPageUptodate(page);
490 brelse(iloc.bh);
491
492 out:
493 return ret;
494 }
495
496 int ext4_readpage_inline(struct inode *inode, struct page *page)
497 {
498 int ret = 0;
499
500 down_read(&EXT4_I(inode)->xattr_sem);
501 if (!ext4_has_inline_data(inode)) {
502 up_read(&EXT4_I(inode)->xattr_sem);
503 return -EAGAIN;
504 }
505
506 /*
507 * Current inline data can only exist in the 1st page,
508 * So for all the other pages, just set them uptodate.
509 */
510 if (!page->index)
511 ret = ext4_read_inline_page(inode, page);
512 else if (!PageUptodate(page)) {
513 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
514 SetPageUptodate(page);
515 }
516
517 up_read(&EXT4_I(inode)->xattr_sem);
518
519 unlock_page(page);
520 return ret >= 0 ? 0 : ret;
521 }
522
523 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
524 struct inode *inode,
525 unsigned flags)
526 {
527 int ret, needed_blocks;
528 handle_t *handle = NULL;
529 int retries = 0, sem_held = 0;
530 struct page *page = NULL;
531 unsigned from, to;
532 struct ext4_iloc iloc;
533
534 if (!ext4_has_inline_data(inode)) {
535 /*
536 * clear the flag so that no new write
537 * will trap here again.
538 */
539 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
540 return 0;
541 }
542
543 needed_blocks = ext4_writepage_trans_blocks(inode);
544
545 ret = ext4_get_inode_loc(inode, &iloc);
546 if (ret)
547 return ret;
548
549 retry:
550 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
551 if (IS_ERR(handle)) {
552 ret = PTR_ERR(handle);
553 handle = NULL;
554 goto out;
555 }
556
557 /* We cannot recurse into the filesystem as the transaction is already
558 * started */
559 flags |= AOP_FLAG_NOFS;
560
561 page = grab_cache_page_write_begin(mapping, 0, flags);
562 if (!page) {
563 ret = -ENOMEM;
564 goto out;
565 }
566
567 down_write(&EXT4_I(inode)->xattr_sem);
568 sem_held = 1;
569 /* If some one has already done this for us, just exit. */
570 if (!ext4_has_inline_data(inode)) {
571 ret = 0;
572 goto out;
573 }
574
575 from = 0;
576 to = ext4_get_inline_size(inode);
577 if (!PageUptodate(page)) {
578 ret = ext4_read_inline_page(inode, page);
579 if (ret < 0)
580 goto out;
581 }
582
583 ret = ext4_destroy_inline_data_nolock(handle, inode);
584 if (ret)
585 goto out;
586
587 if (ext4_should_dioread_nolock(inode))
588 ret = __block_write_begin(page, from, to, ext4_get_block_write);
589 else
590 ret = __block_write_begin(page, from, to, ext4_get_block);
591
592 if (!ret && ext4_should_journal_data(inode)) {
593 ret = ext4_walk_page_buffers(handle, page_buffers(page),
594 from, to, NULL,
595 do_journal_get_write_access);
596 }
597
598 if (ret) {
599 unlock_page(page);
600 page_cache_release(page);
601 page = NULL;
602 ext4_orphan_add(handle, inode);
603 up_write(&EXT4_I(inode)->xattr_sem);
604 sem_held = 0;
605 ext4_journal_stop(handle);
606 handle = NULL;
607 ext4_truncate_failed_write(inode);
608 /*
609 * If truncate failed early the inode might
610 * still be on the orphan list; we need to
611 * make sure the inode is removed from the
612 * orphan list in that case.
613 */
614 if (inode->i_nlink)
615 ext4_orphan_del(NULL, inode);
616 }
617
618 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
619 goto retry;
620
621 if (page)
622 block_commit_write(page, from, to);
623 out:
624 if (page) {
625 unlock_page(page);
626 page_cache_release(page);
627 }
628 if (sem_held)
629 up_write(&EXT4_I(inode)->xattr_sem);
630 if (handle)
631 ext4_journal_stop(handle);
632 brelse(iloc.bh);
633 return ret;
634 }
635
636 /*
637 * Try to write data in the inode.
638 * If the inode has inline data, check whether the new write can be
639 * in the inode also. If not, create the page the handle, move the data
640 * to the page make it update and let the later codes create extent for it.
641 */
642 int ext4_try_to_write_inline_data(struct address_space *mapping,
643 struct inode *inode,
644 loff_t pos, unsigned len,
645 unsigned flags,
646 struct page **pagep)
647 {
648 int ret;
649 handle_t *handle;
650 struct page *page;
651 struct ext4_iloc iloc;
652
653 if (pos + len > ext4_get_max_inline_size(inode))
654 goto convert;
655
656 ret = ext4_get_inode_loc(inode, &iloc);
657 if (ret)
658 return ret;
659
660 /*
661 * The possible write could happen in the inode,
662 * so try to reserve the space in inode first.
663 */
664 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
665 if (IS_ERR(handle)) {
666 ret = PTR_ERR(handle);
667 handle = NULL;
668 goto out;
669 }
670
671 ret = ext4_prepare_inline_data(handle, inode, pos + len);
672 if (ret && ret != -ENOSPC)
673 goto out;
674
675 /* We don't have space in inline inode, so convert it to extent. */
676 if (ret == -ENOSPC) {
677 ext4_journal_stop(handle);
678 brelse(iloc.bh);
679 goto convert;
680 }
681
682 flags |= AOP_FLAG_NOFS;
683
684 page = grab_cache_page_write_begin(mapping, 0, flags);
685 if (!page) {
686 ret = -ENOMEM;
687 goto out;
688 }
689
690 *pagep = page;
691 down_read(&EXT4_I(inode)->xattr_sem);
692 if (!ext4_has_inline_data(inode)) {
693 ret = 0;
694 unlock_page(page);
695 page_cache_release(page);
696 goto out_up_read;
697 }
698
699 if (!PageUptodate(page)) {
700 ret = ext4_read_inline_page(inode, page);
701 if (ret < 0)
702 goto out_up_read;
703 }
704
705 ret = 1;
706 handle = NULL;
707 out_up_read:
708 up_read(&EXT4_I(inode)->xattr_sem);
709 out:
710 if (handle)
711 ext4_journal_stop(handle);
712 brelse(iloc.bh);
713 return ret;
714 convert:
715 return ext4_convert_inline_data_to_extent(mapping,
716 inode, flags);
717 }
718
719 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
720 unsigned copied, struct page *page)
721 {
722 int ret;
723 void *kaddr;
724 struct ext4_iloc iloc;
725
726 if (unlikely(copied < len)) {
727 if (!PageUptodate(page)) {
728 copied = 0;
729 goto out;
730 }
731 }
732
733 ret = ext4_get_inode_loc(inode, &iloc);
734 if (ret) {
735 ext4_std_error(inode->i_sb, ret);
736 copied = 0;
737 goto out;
738 }
739
740 down_write(&EXT4_I(inode)->xattr_sem);
741 BUG_ON(!ext4_has_inline_data(inode));
742
743 kaddr = kmap_atomic(page);
744 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
745 kunmap_atomic(kaddr);
746 SetPageUptodate(page);
747 /* clear page dirty so that writepages wouldn't work for us. */
748 ClearPageDirty(page);
749
750 up_write(&EXT4_I(inode)->xattr_sem);
751 brelse(iloc.bh);
752 out:
753 return copied;
754 }
755
756 struct buffer_head *
757 ext4_journalled_write_inline_data(struct inode *inode,
758 unsigned len,
759 struct page *page)
760 {
761 int ret;
762 void *kaddr;
763 struct ext4_iloc iloc;
764
765 ret = ext4_get_inode_loc(inode, &iloc);
766 if (ret) {
767 ext4_std_error(inode->i_sb, ret);
768 return NULL;
769 }
770
771 down_write(&EXT4_I(inode)->xattr_sem);
772 kaddr = kmap_atomic(page);
773 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
774 kunmap_atomic(kaddr);
775 up_write(&EXT4_I(inode)->xattr_sem);
776
777 return iloc.bh;
778 }
779
780 /*
781 * Try to make the page cache and handle ready for the inline data case.
782 * We can call this function in 2 cases:
783 * 1. The inode is created and the first write exceeds inline size. We can
784 * clear the inode state safely.
785 * 2. The inode has inline data, then we need to read the data, make it
786 * update and dirty so that ext4_da_writepages can handle it. We don't
787 * need to start the journal since the file's metatdata isn't changed now.
788 */
789 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
790 struct inode *inode,
791 unsigned flags,
792 void **fsdata)
793 {
794 int ret = 0, inline_size;
795 struct page *page;
796
797 page = grab_cache_page_write_begin(mapping, 0, flags);
798 if (!page)
799 return -ENOMEM;
800
801 down_read(&EXT4_I(inode)->xattr_sem);
802 if (!ext4_has_inline_data(inode)) {
803 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
804 goto out;
805 }
806
807 inline_size = ext4_get_inline_size(inode);
808
809 if (!PageUptodate(page)) {
810 ret = ext4_read_inline_page(inode, page);
811 if (ret < 0)
812 goto out;
813 }
814
815 ret = __block_write_begin(page, 0, inline_size,
816 ext4_da_get_block_prep);
817 if (ret) {
818 up_read(&EXT4_I(inode)->xattr_sem);
819 unlock_page(page);
820 page_cache_release(page);
821 ext4_truncate_failed_write(inode);
822 return ret;
823 }
824
825 SetPageDirty(page);
826 SetPageUptodate(page);
827 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
828 *fsdata = (void *)CONVERT_INLINE_DATA;
829
830 out:
831 up_read(&EXT4_I(inode)->xattr_sem);
832 if (page) {
833 unlock_page(page);
834 page_cache_release(page);
835 }
836 return ret;
837 }
838
839 /*
840 * Prepare the write for the inline data.
841 * If the the data can be written into the inode, we just read
842 * the page and make it uptodate, and start the journal.
843 * Otherwise read the page, makes it dirty so that it can be
844 * handle in writepages(the i_disksize update is left to the
845 * normal ext4_da_write_end).
846 */
847 int ext4_da_write_inline_data_begin(struct address_space *mapping,
848 struct inode *inode,
849 loff_t pos, unsigned len,
850 unsigned flags,
851 struct page **pagep,
852 void **fsdata)
853 {
854 int ret, inline_size;
855 handle_t *handle;
856 struct page *page;
857 struct ext4_iloc iloc;
858 int retries;
859
860 ret = ext4_get_inode_loc(inode, &iloc);
861 if (ret)
862 return ret;
863
864 retry_journal:
865 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
866 if (IS_ERR(handle)) {
867 ret = PTR_ERR(handle);
868 goto out;
869 }
870
871 inline_size = ext4_get_max_inline_size(inode);
872
873 ret = -ENOSPC;
874 if (inline_size >= pos + len) {
875 ret = ext4_prepare_inline_data(handle, inode, pos + len);
876 if (ret && ret != -ENOSPC)
877 goto out_journal;
878 }
879
880 /*
881 * We cannot recurse into the filesystem as the transaction
882 * is already started.
883 */
884 flags |= AOP_FLAG_NOFS;
885
886 if (ret == -ENOSPC) {
887 ret = ext4_da_convert_inline_data_to_extent(mapping,
888 inode,
889 flags,
890 fsdata);
891 ext4_journal_stop(handle);
892 if (ret == -ENOSPC &&
893 ext4_should_retry_alloc(inode->i_sb, &retries))
894 goto retry_journal;
895 goto out;
896 }
897
898
899 page = grab_cache_page_write_begin(mapping, 0, flags);
900 if (!page) {
901 ret = -ENOMEM;
902 goto out_journal;
903 }
904
905 down_read(&EXT4_I(inode)->xattr_sem);
906 if (!ext4_has_inline_data(inode)) {
907 ret = 0;
908 goto out_release_page;
909 }
910
911 if (!PageUptodate(page)) {
912 ret = ext4_read_inline_page(inode, page);
913 if (ret < 0)
914 goto out_release_page;
915 }
916
917 up_read(&EXT4_I(inode)->xattr_sem);
918 *pagep = page;
919 brelse(iloc.bh);
920 return 1;
921 out_release_page:
922 up_read(&EXT4_I(inode)->xattr_sem);
923 unlock_page(page);
924 page_cache_release(page);
925 out_journal:
926 ext4_journal_stop(handle);
927 out:
928 brelse(iloc.bh);
929 return ret;
930 }
931
932 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
933 unsigned len, unsigned copied,
934 struct page *page)
935 {
936 int i_size_changed = 0;
937
938 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
939
940 /*
941 * No need to use i_size_read() here, the i_size
942 * cannot change under us because we hold i_mutex.
943 *
944 * But it's important to update i_size while still holding page lock:
945 * page writeout could otherwise come in and zero beyond i_size.
946 */
947 if (pos+copied > inode->i_size) {
948 i_size_write(inode, pos+copied);
949 i_size_changed = 1;
950 }
951 unlock_page(page);
952 page_cache_release(page);
953
954 /*
955 * Don't mark the inode dirty under page lock. First, it unnecessarily
956 * makes the holding time of page lock longer. Second, it forces lock
957 * ordering of page lock and transaction start for journaling
958 * filesystems.
959 */
960 if (i_size_changed)
961 mark_inode_dirty(inode);
962
963 return copied;
964 }
965
966 #ifdef INLINE_DIR_DEBUG
967 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
968 void *inline_start, int inline_size)
969 {
970 int offset;
971 unsigned short de_len;
972 struct ext4_dir_entry_2 *de = inline_start;
973 void *dlimit = inline_start + inline_size;
974
975 trace_printk("inode %lu\n", dir->i_ino);
976 offset = 0;
977 while ((void *)de < dlimit) {
978 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
979 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
980 offset, de_len, de->name_len, de->name,
981 de->name_len, le32_to_cpu(de->inode));
982 if (ext4_check_dir_entry(dir, NULL, de, bh,
983 inline_start, inline_size, offset))
984 BUG();
985
986 offset += de_len;
987 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
988 }
989 }
990 #else
991 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
992 #endif
993
994 /*
995 * Add a new entry into a inline dir.
996 * It will return -ENOSPC if no space is available, and -EIO
997 * and -EEXIST if directory entry already exists.
998 */
999 static int ext4_add_dirent_to_inline(handle_t *handle,
1000 struct dentry *dentry,
1001 struct inode *inode,
1002 struct ext4_iloc *iloc,
1003 void *inline_start, int inline_size)
1004 {
1005 struct inode *dir = d_inode(dentry->d_parent);
1006 const char *name = dentry->d_name.name;
1007 int namelen = dentry->d_name.len;
1008 int err;
1009 struct ext4_dir_entry_2 *de;
1010
1011 err = ext4_find_dest_de(dir, inode, iloc->bh,
1012 inline_start, inline_size,
1013 name, namelen, &de);
1014 if (err)
1015 return err;
1016
1017 BUFFER_TRACE(iloc->bh, "get_write_access");
1018 err = ext4_journal_get_write_access(handle, iloc->bh);
1019 if (err)
1020 return err;
1021 ext4_insert_dentry(dir, inode, de, inline_size, &dentry->d_name,
1022 name, namelen);
1023
1024 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1025
1026 /*
1027 * XXX shouldn't update any times until successful
1028 * completion of syscall, but too many callers depend
1029 * on this.
1030 *
1031 * XXX similarly, too many callers depend on
1032 * ext4_new_inode() setting the times, but error
1033 * recovery deletes the inode, so the worst that can
1034 * happen is that the times are slightly out of date
1035 * and/or different from the directory change time.
1036 */
1037 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1038 ext4_update_dx_flag(dir);
1039 dir->i_version++;
1040 ext4_mark_inode_dirty(handle, dir);
1041 return 1;
1042 }
1043
1044 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1045 struct ext4_iloc *iloc)
1046 {
1047 struct ext4_xattr_entry *entry;
1048 struct ext4_xattr_ibody_header *header;
1049
1050 BUG_ON(!EXT4_I(inode)->i_inline_off);
1051
1052 header = IHDR(inode, ext4_raw_inode(iloc));
1053 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1054 EXT4_I(inode)->i_inline_off);
1055
1056 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1057 }
1058
1059 /* Set the final de to cover the whole block. */
1060 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1061 {
1062 struct ext4_dir_entry_2 *de, *prev_de;
1063 void *limit;
1064 int de_len;
1065
1066 de = (struct ext4_dir_entry_2 *)de_buf;
1067 if (old_size) {
1068 limit = de_buf + old_size;
1069 do {
1070 prev_de = de;
1071 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1072 de_buf += de_len;
1073 de = (struct ext4_dir_entry_2 *)de_buf;
1074 } while (de_buf < limit);
1075
1076 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1077 old_size, new_size);
1078 } else {
1079 /* this is just created, so create an empty entry. */
1080 de->inode = 0;
1081 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1082 }
1083 }
1084
1085 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1086 struct ext4_iloc *iloc)
1087 {
1088 int ret;
1089 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1090 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1091
1092 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1093 return -ENOSPC;
1094
1095 ret = ext4_update_inline_data(handle, dir,
1096 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1097 if (ret)
1098 return ret;
1099
1100 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1101 EXT4_I(dir)->i_inline_size -
1102 EXT4_MIN_INLINE_DATA_SIZE);
1103 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1104 return 0;
1105 }
1106
1107 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1108 struct ext4_iloc *iloc,
1109 void *buf, int inline_size)
1110 {
1111 ext4_create_inline_data(handle, inode, inline_size);
1112 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1113 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1114 }
1115
1116 static int ext4_finish_convert_inline_dir(handle_t *handle,
1117 struct inode *inode,
1118 struct buffer_head *dir_block,
1119 void *buf,
1120 int inline_size)
1121 {
1122 int err, csum_size = 0, header_size = 0;
1123 struct ext4_dir_entry_2 *de;
1124 struct ext4_dir_entry_tail *t;
1125 void *target = dir_block->b_data;
1126
1127 /*
1128 * First create "." and ".." and then copy the dir information
1129 * back to the block.
1130 */
1131 de = (struct ext4_dir_entry_2 *)target;
1132 de = ext4_init_dot_dotdot(inode, de,
1133 inode->i_sb->s_blocksize, csum_size,
1134 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1135 header_size = (void *)de - target;
1136
1137 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1138 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1139
1140 if (ext4_has_metadata_csum(inode->i_sb))
1141 csum_size = sizeof(struct ext4_dir_entry_tail);
1142
1143 inode->i_size = inode->i_sb->s_blocksize;
1144 i_size_write(inode, inode->i_sb->s_blocksize);
1145 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1146 ext4_update_final_de(dir_block->b_data,
1147 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1148 inode->i_sb->s_blocksize - csum_size);
1149
1150 if (csum_size) {
1151 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1152 inode->i_sb->s_blocksize);
1153 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1154 }
1155 set_buffer_uptodate(dir_block);
1156 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1157 if (err)
1158 goto out;
1159 set_buffer_verified(dir_block);
1160 out:
1161 return err;
1162 }
1163
1164 static int ext4_convert_inline_data_nolock(handle_t *handle,
1165 struct inode *inode,
1166 struct ext4_iloc *iloc)
1167 {
1168 int error;
1169 void *buf = NULL;
1170 struct buffer_head *data_bh = NULL;
1171 struct ext4_map_blocks map;
1172 int inline_size;
1173
1174 inline_size = ext4_get_inline_size(inode);
1175 buf = kmalloc(inline_size, GFP_NOFS);
1176 if (!buf) {
1177 error = -ENOMEM;
1178 goto out;
1179 }
1180
1181 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1182 if (error < 0)
1183 goto out;
1184
1185 /*
1186 * Make sure the inline directory entries pass checks before we try to
1187 * convert them, so that we avoid touching stuff that needs fsck.
1188 */
1189 if (S_ISDIR(inode->i_mode)) {
1190 error = ext4_check_all_de(inode, iloc->bh,
1191 buf + EXT4_INLINE_DOTDOT_SIZE,
1192 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1193 if (error)
1194 goto out;
1195 }
1196
1197 error = ext4_destroy_inline_data_nolock(handle, inode);
1198 if (error)
1199 goto out;
1200
1201 map.m_lblk = 0;
1202 map.m_len = 1;
1203 map.m_flags = 0;
1204 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1205 if (error < 0)
1206 goto out_restore;
1207 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1208 error = -EIO;
1209 goto out_restore;
1210 }
1211
1212 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1213 if (!data_bh) {
1214 error = -ENOMEM;
1215 goto out_restore;
1216 }
1217
1218 lock_buffer(data_bh);
1219 error = ext4_journal_get_create_access(handle, data_bh);
1220 if (error) {
1221 unlock_buffer(data_bh);
1222 error = -EIO;
1223 goto out_restore;
1224 }
1225 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1226
1227 if (!S_ISDIR(inode->i_mode)) {
1228 memcpy(data_bh->b_data, buf, inline_size);
1229 set_buffer_uptodate(data_bh);
1230 error = ext4_handle_dirty_metadata(handle,
1231 inode, data_bh);
1232 } else {
1233 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1234 buf, inline_size);
1235 }
1236
1237 unlock_buffer(data_bh);
1238 out_restore:
1239 if (error)
1240 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1241
1242 out:
1243 brelse(data_bh);
1244 kfree(buf);
1245 return error;
1246 }
1247
1248 /*
1249 * Try to add the new entry to the inline data.
1250 * If succeeds, return 0. If not, extended the inline dir and copied data to
1251 * the new created block.
1252 */
1253 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1254 struct inode *inode)
1255 {
1256 int ret, inline_size;
1257 void *inline_start;
1258 struct ext4_iloc iloc;
1259 struct inode *dir = d_inode(dentry->d_parent);
1260
1261 ret = ext4_get_inode_loc(dir, &iloc);
1262 if (ret)
1263 return ret;
1264
1265 down_write(&EXT4_I(dir)->xattr_sem);
1266 if (!ext4_has_inline_data(dir))
1267 goto out;
1268
1269 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1270 EXT4_INLINE_DOTDOT_SIZE;
1271 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1272
1273 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1274 inline_start, inline_size);
1275 if (ret != -ENOSPC)
1276 goto out;
1277
1278 /* check whether it can be inserted to inline xattr space. */
1279 inline_size = EXT4_I(dir)->i_inline_size -
1280 EXT4_MIN_INLINE_DATA_SIZE;
1281 if (!inline_size) {
1282 /* Try to use the xattr space.*/
1283 ret = ext4_update_inline_dir(handle, dir, &iloc);
1284 if (ret && ret != -ENOSPC)
1285 goto out;
1286
1287 inline_size = EXT4_I(dir)->i_inline_size -
1288 EXT4_MIN_INLINE_DATA_SIZE;
1289 }
1290
1291 if (inline_size) {
1292 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1293
1294 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1295 inline_start, inline_size);
1296
1297 if (ret != -ENOSPC)
1298 goto out;
1299 }
1300
1301 /*
1302 * The inline space is filled up, so create a new block for it.
1303 * As the extent tree will be created, we have to save the inline
1304 * dir first.
1305 */
1306 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1307
1308 out:
1309 ext4_mark_inode_dirty(handle, dir);
1310 up_write(&EXT4_I(dir)->xattr_sem);
1311 brelse(iloc.bh);
1312 return ret;
1313 }
1314
1315 /*
1316 * This function fills a red-black tree with information from an
1317 * inlined dir. It returns the number directory entries loaded
1318 * into the tree. If there is an error it is returned in err.
1319 */
1320 int htree_inlinedir_to_tree(struct file *dir_file,
1321 struct inode *dir, ext4_lblk_t block,
1322 struct dx_hash_info *hinfo,
1323 __u32 start_hash, __u32 start_minor_hash,
1324 int *has_inline_data)
1325 {
1326 int err = 0, count = 0;
1327 unsigned int parent_ino;
1328 int pos;
1329 struct ext4_dir_entry_2 *de;
1330 struct inode *inode = file_inode(dir_file);
1331 int ret, inline_size = 0;
1332 struct ext4_iloc iloc;
1333 void *dir_buf = NULL;
1334 struct ext4_dir_entry_2 fake;
1335 struct ext4_str tmp_str;
1336
1337 ret = ext4_get_inode_loc(inode, &iloc);
1338 if (ret)
1339 return ret;
1340
1341 down_read(&EXT4_I(inode)->xattr_sem);
1342 if (!ext4_has_inline_data(inode)) {
1343 up_read(&EXT4_I(inode)->xattr_sem);
1344 *has_inline_data = 0;
1345 goto out;
1346 }
1347
1348 inline_size = ext4_get_inline_size(inode);
1349 dir_buf = kmalloc(inline_size, GFP_NOFS);
1350 if (!dir_buf) {
1351 ret = -ENOMEM;
1352 up_read(&EXT4_I(inode)->xattr_sem);
1353 goto out;
1354 }
1355
1356 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1357 up_read(&EXT4_I(inode)->xattr_sem);
1358 if (ret < 0)
1359 goto out;
1360
1361 pos = 0;
1362 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1363 while (pos < inline_size) {
1364 /*
1365 * As inlined dir doesn't store any information about '.' and
1366 * only the inode number of '..' is stored, we have to handle
1367 * them differently.
1368 */
1369 if (pos == 0) {
1370 fake.inode = cpu_to_le32(inode->i_ino);
1371 fake.name_len = 1;
1372 strcpy(fake.name, ".");
1373 fake.rec_len = ext4_rec_len_to_disk(
1374 EXT4_DIR_REC_LEN(fake.name_len),
1375 inline_size);
1376 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1377 de = &fake;
1378 pos = EXT4_INLINE_DOTDOT_OFFSET;
1379 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1380 fake.inode = cpu_to_le32(parent_ino);
1381 fake.name_len = 2;
1382 strcpy(fake.name, "..");
1383 fake.rec_len = ext4_rec_len_to_disk(
1384 EXT4_DIR_REC_LEN(fake.name_len),
1385 inline_size);
1386 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1387 de = &fake;
1388 pos = EXT4_INLINE_DOTDOT_SIZE;
1389 } else {
1390 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1391 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1392 if (ext4_check_dir_entry(inode, dir_file, de,
1393 iloc.bh, dir_buf,
1394 inline_size, pos)) {
1395 ret = count;
1396 goto out;
1397 }
1398 }
1399
1400 ext4fs_dirhash(de->name, de->name_len, hinfo);
1401 if ((hinfo->hash < start_hash) ||
1402 ((hinfo->hash == start_hash) &&
1403 (hinfo->minor_hash < start_minor_hash)))
1404 continue;
1405 if (de->inode == 0)
1406 continue;
1407 tmp_str.name = de->name;
1408 tmp_str.len = de->name_len;
1409 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1410 hinfo->minor_hash, de, &tmp_str);
1411 if (err) {
1412 count = err;
1413 goto out;
1414 }
1415 count++;
1416 }
1417 ret = count;
1418 out:
1419 kfree(dir_buf);
1420 brelse(iloc.bh);
1421 return ret;
1422 }
1423
1424 /*
1425 * So this function is called when the volume is mkfsed with
1426 * dir_index disabled. In order to keep f_pos persistent
1427 * after we convert from an inlined dir to a blocked based,
1428 * we just pretend that we are a normal dir and return the
1429 * offset as if '.' and '..' really take place.
1430 *
1431 */
1432 int ext4_read_inline_dir(struct file *file,
1433 struct dir_context *ctx,
1434 int *has_inline_data)
1435 {
1436 unsigned int offset, parent_ino;
1437 int i;
1438 struct ext4_dir_entry_2 *de;
1439 struct super_block *sb;
1440 struct inode *inode = file_inode(file);
1441 int ret, inline_size = 0;
1442 struct ext4_iloc iloc;
1443 void *dir_buf = NULL;
1444 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1445
1446 ret = ext4_get_inode_loc(inode, &iloc);
1447 if (ret)
1448 return ret;
1449
1450 down_read(&EXT4_I(inode)->xattr_sem);
1451 if (!ext4_has_inline_data(inode)) {
1452 up_read(&EXT4_I(inode)->xattr_sem);
1453 *has_inline_data = 0;
1454 goto out;
1455 }
1456
1457 inline_size = ext4_get_inline_size(inode);
1458 dir_buf = kmalloc(inline_size, GFP_NOFS);
1459 if (!dir_buf) {
1460 ret = -ENOMEM;
1461 up_read(&EXT4_I(inode)->xattr_sem);
1462 goto out;
1463 }
1464
1465 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1466 up_read(&EXT4_I(inode)->xattr_sem);
1467 if (ret < 0)
1468 goto out;
1469
1470 ret = 0;
1471 sb = inode->i_sb;
1472 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1473 offset = ctx->pos;
1474
1475 /*
1476 * dotdot_offset and dotdot_size is the real offset and
1477 * size for ".." and "." if the dir is block based while
1478 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1479 * So we will use extra_offset and extra_size to indicate them
1480 * during the inline dir iteration.
1481 */
1482 dotdot_offset = EXT4_DIR_REC_LEN(1);
1483 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1484 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1485 extra_size = extra_offset + inline_size;
1486
1487 /*
1488 * If the version has changed since the last call to
1489 * readdir(2), then we might be pointing to an invalid
1490 * dirent right now. Scan from the start of the inline
1491 * dir to make sure.
1492 */
1493 if (file->f_version != inode->i_version) {
1494 for (i = 0; i < extra_size && i < offset;) {
1495 /*
1496 * "." is with offset 0 and
1497 * ".." is dotdot_offset.
1498 */
1499 if (!i) {
1500 i = dotdot_offset;
1501 continue;
1502 } else if (i == dotdot_offset) {
1503 i = dotdot_size;
1504 continue;
1505 }
1506 /* for other entry, the real offset in
1507 * the buf has to be tuned accordingly.
1508 */
1509 de = (struct ext4_dir_entry_2 *)
1510 (dir_buf + i - extra_offset);
1511 /* It's too expensive to do a full
1512 * dirent test each time round this
1513 * loop, but we do have to test at
1514 * least that it is non-zero. A
1515 * failure will be detected in the
1516 * dirent test below. */
1517 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1518 < EXT4_DIR_REC_LEN(1))
1519 break;
1520 i += ext4_rec_len_from_disk(de->rec_len,
1521 extra_size);
1522 }
1523 offset = i;
1524 ctx->pos = offset;
1525 file->f_version = inode->i_version;
1526 }
1527
1528 while (ctx->pos < extra_size) {
1529 if (ctx->pos == 0) {
1530 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1531 goto out;
1532 ctx->pos = dotdot_offset;
1533 continue;
1534 }
1535
1536 if (ctx->pos == dotdot_offset) {
1537 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1538 goto out;
1539 ctx->pos = dotdot_size;
1540 continue;
1541 }
1542
1543 de = (struct ext4_dir_entry_2 *)
1544 (dir_buf + ctx->pos - extra_offset);
1545 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1546 extra_size, ctx->pos))
1547 goto out;
1548 if (le32_to_cpu(de->inode)) {
1549 if (!dir_emit(ctx, de->name, de->name_len,
1550 le32_to_cpu(de->inode),
1551 get_dtype(sb, de->file_type)))
1552 goto out;
1553 }
1554 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1555 }
1556 out:
1557 kfree(dir_buf);
1558 brelse(iloc.bh);
1559 return ret;
1560 }
1561
1562 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1563 struct ext4_dir_entry_2 **parent_de,
1564 int *retval)
1565 {
1566 struct ext4_iloc iloc;
1567
1568 *retval = ext4_get_inode_loc(inode, &iloc);
1569 if (*retval)
1570 return NULL;
1571
1572 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1573
1574 return iloc.bh;
1575 }
1576
1577 /*
1578 * Try to create the inline data for the new dir.
1579 * If it succeeds, return 0, otherwise return the error.
1580 * In case of ENOSPC, the caller should create the normal disk layout dir.
1581 */
1582 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1583 struct inode *inode)
1584 {
1585 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1586 struct ext4_iloc iloc;
1587 struct ext4_dir_entry_2 *de;
1588
1589 ret = ext4_get_inode_loc(inode, &iloc);
1590 if (ret)
1591 return ret;
1592
1593 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1594 if (ret)
1595 goto out;
1596
1597 /*
1598 * For inline dir, we only save the inode information for the ".."
1599 * and create a fake dentry to cover the left space.
1600 */
1601 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1602 de->inode = cpu_to_le32(parent->i_ino);
1603 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1604 de->inode = 0;
1605 de->rec_len = ext4_rec_len_to_disk(
1606 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1607 inline_size);
1608 set_nlink(inode, 2);
1609 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1610 out:
1611 brelse(iloc.bh);
1612 return ret;
1613 }
1614
1615 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1616 const struct qstr *d_name,
1617 struct ext4_dir_entry_2 **res_dir,
1618 int *has_inline_data)
1619 {
1620 int ret;
1621 struct ext4_iloc iloc;
1622 void *inline_start;
1623 int inline_size;
1624
1625 if (ext4_get_inode_loc(dir, &iloc))
1626 return NULL;
1627
1628 down_read(&EXT4_I(dir)->xattr_sem);
1629 if (!ext4_has_inline_data(dir)) {
1630 *has_inline_data = 0;
1631 goto out;
1632 }
1633
1634 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1635 EXT4_INLINE_DOTDOT_SIZE;
1636 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1637 ret = search_dir(iloc.bh, inline_start, inline_size,
1638 dir, d_name, 0, res_dir);
1639 if (ret == 1)
1640 goto out_find;
1641 if (ret < 0)
1642 goto out;
1643
1644 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1645 goto out;
1646
1647 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1648 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1649
1650 ret = search_dir(iloc.bh, inline_start, inline_size,
1651 dir, d_name, 0, res_dir);
1652 if (ret == 1)
1653 goto out_find;
1654
1655 out:
1656 brelse(iloc.bh);
1657 iloc.bh = NULL;
1658 out_find:
1659 up_read(&EXT4_I(dir)->xattr_sem);
1660 return iloc.bh;
1661 }
1662
1663 int ext4_delete_inline_entry(handle_t *handle,
1664 struct inode *dir,
1665 struct ext4_dir_entry_2 *de_del,
1666 struct buffer_head *bh,
1667 int *has_inline_data)
1668 {
1669 int err, inline_size;
1670 struct ext4_iloc iloc;
1671 void *inline_start;
1672
1673 err = ext4_get_inode_loc(dir, &iloc);
1674 if (err)
1675 return err;
1676
1677 down_write(&EXT4_I(dir)->xattr_sem);
1678 if (!ext4_has_inline_data(dir)) {
1679 *has_inline_data = 0;
1680 goto out;
1681 }
1682
1683 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1684 EXT4_MIN_INLINE_DATA_SIZE) {
1685 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1686 EXT4_INLINE_DOTDOT_SIZE;
1687 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1688 EXT4_INLINE_DOTDOT_SIZE;
1689 } else {
1690 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1691 inline_size = ext4_get_inline_size(dir) -
1692 EXT4_MIN_INLINE_DATA_SIZE;
1693 }
1694
1695 BUFFER_TRACE(bh, "get_write_access");
1696 err = ext4_journal_get_write_access(handle, bh);
1697 if (err)
1698 goto out;
1699
1700 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1701 inline_start, inline_size, 0);
1702 if (err)
1703 goto out;
1704
1705 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1706 err = ext4_mark_inode_dirty(handle, dir);
1707 if (unlikely(err))
1708 goto out;
1709
1710 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1711 out:
1712 up_write(&EXT4_I(dir)->xattr_sem);
1713 brelse(iloc.bh);
1714 if (err != -ENOENT)
1715 ext4_std_error(dir->i_sb, err);
1716 return err;
1717 }
1718
1719 /*
1720 * Get the inline dentry at offset.
1721 */
1722 static inline struct ext4_dir_entry_2 *
1723 ext4_get_inline_entry(struct inode *inode,
1724 struct ext4_iloc *iloc,
1725 unsigned int offset,
1726 void **inline_start,
1727 int *inline_size)
1728 {
1729 void *inline_pos;
1730
1731 BUG_ON(offset > ext4_get_inline_size(inode));
1732
1733 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1734 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1735 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1736 } else {
1737 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1738 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1739 *inline_size = ext4_get_inline_size(inode) -
1740 EXT4_MIN_INLINE_DATA_SIZE;
1741 }
1742
1743 if (inline_start)
1744 *inline_start = inline_pos;
1745 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1746 }
1747
1748 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1749 {
1750 int err, inline_size;
1751 struct ext4_iloc iloc;
1752 void *inline_pos;
1753 unsigned int offset;
1754 struct ext4_dir_entry_2 *de;
1755 int ret = 1;
1756
1757 err = ext4_get_inode_loc(dir, &iloc);
1758 if (err) {
1759 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1760 err, dir->i_ino);
1761 return 1;
1762 }
1763
1764 down_read(&EXT4_I(dir)->xattr_sem);
1765 if (!ext4_has_inline_data(dir)) {
1766 *has_inline_data = 0;
1767 goto out;
1768 }
1769
1770 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1771 if (!le32_to_cpu(de->inode)) {
1772 ext4_warning(dir->i_sb,
1773 "bad inline directory (dir #%lu) - no `..'",
1774 dir->i_ino);
1775 ret = 1;
1776 goto out;
1777 }
1778
1779 offset = EXT4_INLINE_DOTDOT_SIZE;
1780 while (offset < dir->i_size) {
1781 de = ext4_get_inline_entry(dir, &iloc, offset,
1782 &inline_pos, &inline_size);
1783 if (ext4_check_dir_entry(dir, NULL, de,
1784 iloc.bh, inline_pos,
1785 inline_size, offset)) {
1786 ext4_warning(dir->i_sb,
1787 "bad inline directory (dir #%lu) - "
1788 "inode %u, rec_len %u, name_len %d"
1789 "inline size %d\n",
1790 dir->i_ino, le32_to_cpu(de->inode),
1791 le16_to_cpu(de->rec_len), de->name_len,
1792 inline_size);
1793 ret = 1;
1794 goto out;
1795 }
1796 if (le32_to_cpu(de->inode)) {
1797 ret = 0;
1798 goto out;
1799 }
1800 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1801 }
1802
1803 out:
1804 up_read(&EXT4_I(dir)->xattr_sem);
1805 brelse(iloc.bh);
1806 return ret;
1807 }
1808
1809 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1810 {
1811 int ret;
1812
1813 down_write(&EXT4_I(inode)->xattr_sem);
1814 ret = ext4_destroy_inline_data_nolock(handle, inode);
1815 up_write(&EXT4_I(inode)->xattr_sem);
1816
1817 return ret;
1818 }
1819
1820 int ext4_inline_data_fiemap(struct inode *inode,
1821 struct fiemap_extent_info *fieinfo,
1822 int *has_inline, __u64 start, __u64 len)
1823 {
1824 __u64 physical = 0;
1825 __u64 inline_len;
1826 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1827 FIEMAP_EXTENT_LAST;
1828 int error = 0;
1829 struct ext4_iloc iloc;
1830
1831 down_read(&EXT4_I(inode)->xattr_sem);
1832 if (!ext4_has_inline_data(inode)) {
1833 *has_inline = 0;
1834 goto out;
1835 }
1836 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1837 i_size_read(inode));
1838 if (start >= inline_len)
1839 goto out;
1840 if (start + len < inline_len)
1841 inline_len = start + len;
1842 inline_len -= start;
1843
1844 error = ext4_get_inode_loc(inode, &iloc);
1845 if (error)
1846 goto out;
1847
1848 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1849 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1850 physical += offsetof(struct ext4_inode, i_block);
1851
1852 if (physical)
1853 error = fiemap_fill_next_extent(fieinfo, start, physical,
1854 inline_len, flags);
1855 brelse(iloc.bh);
1856 out:
1857 up_read(&EXT4_I(inode)->xattr_sem);
1858 return (error < 0 ? error : 0);
1859 }
1860
1861 /*
1862 * Called during xattr set, and if we can sparse space 'needed',
1863 * just create the extent tree evict the data to the outer block.
1864 *
1865 * We use jbd2 instead of page cache to move data to the 1st block
1866 * so that the whole transaction can be committed as a whole and
1867 * the data isn't lost because of the delayed page cache write.
1868 */
1869 int ext4_try_to_evict_inline_data(handle_t *handle,
1870 struct inode *inode,
1871 int needed)
1872 {
1873 int error;
1874 struct ext4_xattr_entry *entry;
1875 struct ext4_inode *raw_inode;
1876 struct ext4_iloc iloc;
1877
1878 error = ext4_get_inode_loc(inode, &iloc);
1879 if (error)
1880 return error;
1881
1882 raw_inode = ext4_raw_inode(&iloc);
1883 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1884 EXT4_I(inode)->i_inline_off);
1885 if (EXT4_XATTR_LEN(entry->e_name_len) +
1886 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1887 error = -ENOSPC;
1888 goto out;
1889 }
1890
1891 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1892 out:
1893 brelse(iloc.bh);
1894 return error;
1895 }
1896
1897 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1898 {
1899 handle_t *handle;
1900 int inline_size, value_len, needed_blocks;
1901 size_t i_size;
1902 void *value = NULL;
1903 struct ext4_xattr_ibody_find is = {
1904 .s = { .not_found = -ENODATA, },
1905 };
1906 struct ext4_xattr_info i = {
1907 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1908 .name = EXT4_XATTR_SYSTEM_DATA,
1909 };
1910
1911
1912 needed_blocks = ext4_writepage_trans_blocks(inode);
1913 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1914 if (IS_ERR(handle))
1915 return;
1916
1917 down_write(&EXT4_I(inode)->xattr_sem);
1918 if (!ext4_has_inline_data(inode)) {
1919 *has_inline = 0;
1920 ext4_journal_stop(handle);
1921 return;
1922 }
1923
1924 if (ext4_orphan_add(handle, inode))
1925 goto out;
1926
1927 if (ext4_get_inode_loc(inode, &is.iloc))
1928 goto out;
1929
1930 down_write(&EXT4_I(inode)->i_data_sem);
1931 i_size = inode->i_size;
1932 inline_size = ext4_get_inline_size(inode);
1933 EXT4_I(inode)->i_disksize = i_size;
1934
1935 if (i_size < inline_size) {
1936 /* Clear the content in the xattr space. */
1937 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1938 if (ext4_xattr_ibody_find(inode, &i, &is))
1939 goto out_error;
1940
1941 BUG_ON(is.s.not_found);
1942
1943 value_len = le32_to_cpu(is.s.here->e_value_size);
1944 value = kmalloc(value_len, GFP_NOFS);
1945 if (!value)
1946 goto out_error;
1947
1948 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1949 value, value_len))
1950 goto out_error;
1951
1952 i.value = value;
1953 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1954 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1955 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1956 goto out_error;
1957 }
1958
1959 /* Clear the content within i_blocks. */
1960 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1961 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1962 memset(p + i_size, 0,
1963 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1964 }
1965
1966 EXT4_I(inode)->i_inline_size = i_size <
1967 EXT4_MIN_INLINE_DATA_SIZE ?
1968 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1969 }
1970
1971 out_error:
1972 up_write(&EXT4_I(inode)->i_data_sem);
1973 out:
1974 brelse(is.iloc.bh);
1975 up_write(&EXT4_I(inode)->xattr_sem);
1976 kfree(value);
1977 if (inode->i_nlink)
1978 ext4_orphan_del(handle, inode);
1979
1980 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1981 ext4_mark_inode_dirty(handle, inode);
1982 if (IS_SYNC(inode))
1983 ext4_handle_sync(handle);
1984
1985 ext4_journal_stop(handle);
1986 return;
1987 }
1988
1989 int ext4_convert_inline_data(struct inode *inode)
1990 {
1991 int error, needed_blocks;
1992 handle_t *handle;
1993 struct ext4_iloc iloc;
1994
1995 if (!ext4_has_inline_data(inode)) {
1996 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1997 return 0;
1998 }
1999
2000 needed_blocks = ext4_writepage_trans_blocks(inode);
2001
2002 iloc.bh = NULL;
2003 error = ext4_get_inode_loc(inode, &iloc);
2004 if (error)
2005 return error;
2006
2007 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2008 if (IS_ERR(handle)) {
2009 error = PTR_ERR(handle);
2010 goto out_free;
2011 }
2012
2013 down_write(&EXT4_I(inode)->xattr_sem);
2014 if (!ext4_has_inline_data(inode)) {
2015 up_write(&EXT4_I(inode)->xattr_sem);
2016 goto out;
2017 }
2018
2019 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2020 up_write(&EXT4_I(inode)->xattr_sem);
2021 out:
2022 ext4_journal_stop(handle);
2023 out_free:
2024 brelse(iloc.bh);
2025 return error;
2026 }
Linux with added FPC-III support