?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
thp, mm: avoid PageUnevictable on active/inactive lru lists
[linux-2.6.git] / fs / ext4 / inline.c
blobd9ecbf1113a75798f4d2a5903fd2fc522ce00dba
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 #include "ext4_jbd2.h"
15 #include "ext4.h"
16 #include "xattr.h"
17 #include "truncate.h"
18 #include <linux/fiemap.h>
19
20 #define EXT4_XATTR_SYSTEM_DATA "data"
21 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
22 #define EXT4_INLINE_DOTDOT_OFFSET 2
23 #define EXT4_INLINE_DOTDOT_SIZE 4
24
25 int ext4_get_inline_size(struct inode *inode)
26 {
27 if (EXT4_I(inode)->i_inline_off)
28 return EXT4_I(inode)->i_inline_size;
29
30 return 0;
31 }
32
33 static int get_max_inline_xattr_value_size(struct inode *inode,
34 struct ext4_iloc *iloc)
35 {
36 struct ext4_xattr_ibody_header *header;
37 struct ext4_xattr_entry *entry;
38 struct ext4_inode *raw_inode;
39 int free, min_offs;
40
41 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42 EXT4_GOOD_OLD_INODE_SIZE -
43 EXT4_I(inode)->i_extra_isize -
44 sizeof(struct ext4_xattr_ibody_header);
45
46 /*
47 * We need to subtract another sizeof(__u32) since an in-inode xattr
48 * needs an empty 4 bytes to indicate the gap between the xattr entry
49 * and the name/value pair.
50 */
51 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52 return EXT4_XATTR_SIZE(min_offs -
53 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54 EXT4_XATTR_ROUND - sizeof(__u32));
55
56 raw_inode = ext4_raw_inode(iloc);
57 header = IHDR(inode, raw_inode);
58 entry = IFIRST(header);
59
60 /* Compute min_offs. */
61 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62 if (!entry->e_value_block && entry->e_value_size) {
63 size_t offs = le16_to_cpu(entry->e_value_offs);
64 if (offs < min_offs)
65 min_offs = offs;
66 }
67 }
68 free = min_offs -
69 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71 if (EXT4_I(inode)->i_inline_off) {
72 entry = (struct ext4_xattr_entry *)
73 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76 goto out;
77 }
78
79 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81 if (free > EXT4_XATTR_ROUND)
82 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83 else
84 free = 0;
85
86 out:
87 return free;
88 }
89
90 /*
91 * Get the maximum size we now can store in an inode.
92 * If we can't find the space for a xattr entry, don't use the space
93 * of the extents since we have no space to indicate the inline data.
94 */
95 int ext4_get_max_inline_size(struct inode *inode)
96 {
97 int error, max_inline_size;
98 struct ext4_iloc iloc;
99
100 if (EXT4_I(inode)->i_extra_isize == 0)
101 return 0;
102
103 error = ext4_get_inode_loc(inode, &iloc);
104 if (error) {
105 ext4_error_inode(inode, __func__, __LINE__, 0,
106 "can't get inode location %lu",
107 inode->i_ino);
108 return 0;
109 }
110
111 down_read(&EXT4_I(inode)->xattr_sem);
112 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113 up_read(&EXT4_I(inode)->xattr_sem);
114
115 brelse(iloc.bh);
116
117 if (!max_inline_size)
118 return 0;
119
120 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121 }
122
123 int ext4_has_inline_data(struct inode *inode)
124 {
125 return ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) &&
126 EXT4_I(inode)->i_inline_off;
127 }
128
129 /*
130 * this function does not take xattr_sem, which is OK because it is
131 * currently only used in a code path coming form ext4_iget, before
132 * the new inode has been unlocked
133 */
134 int ext4_find_inline_data_nolock(struct inode *inode)
135 {
136 struct ext4_xattr_ibody_find is = {
137 .s = { .not_found = -ENODATA, },
138 };
139 struct ext4_xattr_info i = {
140 .name_index = EXT4_XATTR_INDEX_SYSTEM,
141 .name = EXT4_XATTR_SYSTEM_DATA,
142 };
143 int error;
144
145 if (EXT4_I(inode)->i_extra_isize == 0)
146 return 0;
147
148 error = ext4_get_inode_loc(inode, &is.iloc);
149 if (error)
150 return error;
151
152 error = ext4_xattr_ibody_find(inode, &i, &is);
153 if (error)
154 goto out;
155
156 if (!is.s.not_found) {
157 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
158 (void *)ext4_raw_inode(&is.iloc));
159 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
160 le32_to_cpu(is.s.here->e_value_size);
161 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
162 }
163 out:
164 brelse(is.iloc.bh);
165 return error;
166 }
167
168 static int ext4_read_inline_data(struct inode *inode, void *buffer,
169 unsigned int len,
170 struct ext4_iloc *iloc)
171 {
172 struct ext4_xattr_entry *entry;
173 struct ext4_xattr_ibody_header *header;
174 int cp_len = 0;
175 struct ext4_inode *raw_inode;
176
177 if (!len)
178 return 0;
179
180 BUG_ON(len > EXT4_I(inode)->i_inline_size);
181
182 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
183 len : EXT4_MIN_INLINE_DATA_SIZE;
184
185 raw_inode = ext4_raw_inode(iloc);
186 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
187
188 len -= cp_len;
189 buffer += cp_len;
190
191 if (!len)
192 goto out;
193
194 header = IHDR(inode, raw_inode);
195 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
196 EXT4_I(inode)->i_inline_off);
197 len = min_t(unsigned int, len,
198 (unsigned int)le32_to_cpu(entry->e_value_size));
199
200 memcpy(buffer,
201 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
202 cp_len += len;
203
204 out:
205 return cp_len;
206 }
207
208 /*
209 * write the buffer to the inline inode.
210 * If 'create' is set, we don't need to do the extra copy in the xattr
211 * value since it is already handled by ext4_xattr_ibody_inline_set.
212 * That saves us one memcpy.
213 */
214 void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
215 void *buffer, loff_t pos, unsigned int len)
216 {
217 struct ext4_xattr_entry *entry;
218 struct ext4_xattr_ibody_header *header;
219 struct ext4_inode *raw_inode;
220 int cp_len = 0;
221
222 BUG_ON(!EXT4_I(inode)->i_inline_off);
223 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
224
225 raw_inode = ext4_raw_inode(iloc);
226 buffer += pos;
227
228 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
229 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
230 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
231 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
232
233 len -= cp_len;
234 buffer += cp_len;
235 pos += cp_len;
236 }
237
238 if (!len)
239 return;
240
241 pos -= EXT4_MIN_INLINE_DATA_SIZE;
242 header = IHDR(inode, raw_inode);
243 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
244 EXT4_I(inode)->i_inline_off);
245
246 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
247 buffer, len);
248 }
249
250 static int ext4_create_inline_data(handle_t *handle,
251 struct inode *inode, unsigned len)
252 {
253 int error;
254 void *value = NULL;
255 struct ext4_xattr_ibody_find is = {
256 .s = { .not_found = -ENODATA, },
257 };
258 struct ext4_xattr_info i = {
259 .name_index = EXT4_XATTR_INDEX_SYSTEM,
260 .name = EXT4_XATTR_SYSTEM_DATA,
261 };
262
263 error = ext4_get_inode_loc(inode, &is.iloc);
264 if (error)
265 return error;
266
267 error = ext4_journal_get_write_access(handle, is.iloc.bh);
268 if (error)
269 goto out;
270
271 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272 value = EXT4_ZERO_XATTR_VALUE;
273 len -= EXT4_MIN_INLINE_DATA_SIZE;
274 } else {
275 value = "";
276 len = 0;
277 }
278
279 /* Insert the the xttr entry. */
280 i.value = value;
281 i.value_len = len;
282
283 error = ext4_xattr_ibody_find(inode, &i, &is);
284 if (error)
285 goto out;
286
287 BUG_ON(!is.s.not_found);
288
289 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290 if (error) {
291 if (error == -ENOSPC)
292 ext4_clear_inode_state(inode,
293 EXT4_STATE_MAY_INLINE_DATA);
294 goto out;
295 }
296
297 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301 (void *)ext4_raw_inode(&is.iloc));
302 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305 get_bh(is.iloc.bh);
306 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
307
308 out:
309 brelse(is.iloc.bh);
310 return error;
311 }
312
313 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
314 unsigned int len)
315 {
316 int error;
317 void *value = NULL;
318 struct ext4_xattr_ibody_find is = {
319 .s = { .not_found = -ENODATA, },
320 };
321 struct ext4_xattr_info i = {
322 .name_index = EXT4_XATTR_INDEX_SYSTEM,
323 .name = EXT4_XATTR_SYSTEM_DATA,
324 };
325
326 /* If the old space is ok, write the data directly. */
327 if (len <= EXT4_I(inode)->i_inline_size)
328 return 0;
329
330 error = ext4_get_inode_loc(inode, &is.iloc);
331 if (error)
332 return error;
333
334 error = ext4_xattr_ibody_find(inode, &i, &is);
335 if (error)
336 goto out;
337
338 BUG_ON(is.s.not_found);
339
340 len -= EXT4_MIN_INLINE_DATA_SIZE;
341 value = kzalloc(len, GFP_NOFS);
342 if (!value)
343 goto out;
344
345 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
346 value, len);
347 if (error == -ENODATA)
348 goto out;
349
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 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 error = ext4_journal_get_write_access(handle, is.iloc.bh);
428 if (error)
429 goto out;
430
431 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
432 if (error)
433 goto out;
434
435 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
436 0, EXT4_MIN_INLINE_DATA_SIZE);
437
438 if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
439 EXT4_FEATURE_INCOMPAT_EXTENTS)) {
440 if (S_ISDIR(inode->i_mode) ||
441 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
442 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
443 ext4_ext_tree_init(handle, inode);
444 }
445 }
446 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
447
448 get_bh(is.iloc.bh);
449 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
450
451 EXT4_I(inode)->i_inline_off = 0;
452 EXT4_I(inode)->i_inline_size = 0;
453 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
454 out:
455 brelse(is.iloc.bh);
456 if (error == -ENODATA)
457 error = 0;
458 return error;
459 }
460
461 static int ext4_read_inline_page(struct inode *inode, struct page *page)
462 {
463 void *kaddr;
464 int ret = 0;
465 size_t len;
466 struct ext4_iloc iloc;
467
468 BUG_ON(!PageLocked(page));
469 BUG_ON(!ext4_has_inline_data(inode));
470 BUG_ON(page->index);
471
472 if (!EXT4_I(inode)->i_inline_off) {
473 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
474 inode->i_ino);
475 goto out;
476 }
477
478 ret = ext4_get_inode_loc(inode, &iloc);
479 if (ret)
480 goto out;
481
482 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
483 kaddr = kmap_atomic(page);
484 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
485 flush_dcache_page(page);
486 kunmap_atomic(kaddr);
487 zero_user_segment(page, len, PAGE_CACHE_SIZE);
488 SetPageUptodate(page);
489 brelse(iloc.bh);
490
491 out:
492 return ret;
493 }
494
495 int ext4_readpage_inline(struct inode *inode, struct page *page)
496 {
497 int ret = 0;
498
499 down_read(&EXT4_I(inode)->xattr_sem);
500 if (!ext4_has_inline_data(inode)) {
501 up_read(&EXT4_I(inode)->xattr_sem);
502 return -EAGAIN;
503 }
504
505 /*
506 * Current inline data can only exist in the 1st page,
507 * So for all the other pages, just set them uptodate.
508 */
509 if (!page->index)
510 ret = ext4_read_inline_page(inode, page);
511 else if (!PageUptodate(page)) {
512 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
513 SetPageUptodate(page);
514 }
515
516 up_read(&EXT4_I(inode)->xattr_sem);
517
518 unlock_page(page);
519 return ret >= 0 ? 0 : ret;
520 }
521
522 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
523 struct inode *inode,
524 unsigned flags)
525 {
526 int ret, needed_blocks;
527 handle_t *handle = NULL;
528 int retries = 0, sem_held = 0;
529 struct page *page = NULL;
530 unsigned from, to;
531 struct ext4_iloc iloc;
532
533 if (!ext4_has_inline_data(inode)) {
534 /*
535 * clear the flag so that no new write
536 * will trap here again.
537 */
538 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
539 return 0;
540 }
541
542 needed_blocks = ext4_writepage_trans_blocks(inode);
543
544 ret = ext4_get_inode_loc(inode, &iloc);
545 if (ret)
546 return ret;
547
548 retry:
549 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
550 if (IS_ERR(handle)) {
551 ret = PTR_ERR(handle);
552 handle = NULL;
553 goto out;
554 }
555
556 /* We cannot recurse into the filesystem as the transaction is already
557 * started */
558 flags |= AOP_FLAG_NOFS;
559
560 page = grab_cache_page_write_begin(mapping, 0, flags);
561 if (!page) {
562 ret = -ENOMEM;
563 goto out;
564 }
565
566 down_write(&EXT4_I(inode)->xattr_sem);
567 sem_held = 1;
568 /* If some one has already done this for us, just exit. */
569 if (!ext4_has_inline_data(inode)) {
570 ret = 0;
571 goto out;
572 }
573
574 from = 0;
575 to = ext4_get_inline_size(inode);
576 if (!PageUptodate(page)) {
577 ret = ext4_read_inline_page(inode, page);
578 if (ret < 0)
579 goto out;
580 }
581
582 ret = ext4_destroy_inline_data_nolock(handle, inode);
583 if (ret)
584 goto out;
585
586 if (ext4_should_dioread_nolock(inode))
587 ret = __block_write_begin(page, from, to, ext4_get_block_write);
588 else
589 ret = __block_write_begin(page, from, to, ext4_get_block);
590
591 if (!ret && ext4_should_journal_data(inode)) {
592 ret = ext4_walk_page_buffers(handle, page_buffers(page),
593 from, to, NULL,
594 do_journal_get_write_access);
595 }
596
597 if (ret) {
598 unlock_page(page);
599 page_cache_release(page);
600 ext4_orphan_add(handle, inode);
601 up_write(&EXT4_I(inode)->xattr_sem);
602 sem_held = 0;
603 ext4_journal_stop(handle);
604 handle = NULL;
605 ext4_truncate_failed_write(inode);
606 /*
607 * If truncate failed early the inode might
608 * still be on the orphan list; we need to
609 * make sure the inode is removed from the
610 * orphan list in that case.
611 */
612 if (inode->i_nlink)
613 ext4_orphan_del(NULL, inode);
614 }
615
616 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
617 goto retry;
618
619 block_commit_write(page, from, to);
620 out:
621 if (page) {
622 unlock_page(page);
623 page_cache_release(page);
624 }
625 if (sem_held)
626 up_write(&EXT4_I(inode)->xattr_sem);
627 if (handle)
628 ext4_journal_stop(handle);
629 brelse(iloc.bh);
630 return ret;
631 }
632
633 /*
634 * Try to write data in the inode.
635 * If the inode has inline data, check whether the new write can be
636 * in the inode also. If not, create the page the handle, move the data
637 * to the page make it update and let the later codes create extent for it.
638 */
639 int ext4_try_to_write_inline_data(struct address_space *mapping,
640 struct inode *inode,
641 loff_t pos, unsigned len,
642 unsigned flags,
643 struct page **pagep)
644 {
645 int ret;
646 handle_t *handle;
647 struct page *page;
648 struct ext4_iloc iloc;
649
650 if (pos + len > ext4_get_max_inline_size(inode))
651 goto convert;
652
653 ret = ext4_get_inode_loc(inode, &iloc);
654 if (ret)
655 return ret;
656
657 /*
658 * The possible write could happen in the inode,
659 * so try to reserve the space in inode first.
660 */
661 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
662 if (IS_ERR(handle)) {
663 ret = PTR_ERR(handle);
664 handle = NULL;
665 goto out;
666 }
667
668 ret = ext4_prepare_inline_data(handle, inode, pos + len);
669 if (ret && ret != -ENOSPC)
670 goto out;
671
672 /* We don't have space in inline inode, so convert it to extent. */
673 if (ret == -ENOSPC) {
674 ext4_journal_stop(handle);
675 brelse(iloc.bh);
676 goto convert;
677 }
678
679 flags |= AOP_FLAG_NOFS;
680
681 page = grab_cache_page_write_begin(mapping, 0, flags);
682 if (!page) {
683 ret = -ENOMEM;
684 goto out;
685 }
686
687 *pagep = page;
688 down_read(&EXT4_I(inode)->xattr_sem);
689 if (!ext4_has_inline_data(inode)) {
690 ret = 0;
691 unlock_page(page);
692 page_cache_release(page);
693 goto out_up_read;
694 }
695
696 if (!PageUptodate(page)) {
697 ret = ext4_read_inline_page(inode, page);
698 if (ret < 0)
699 goto out_up_read;
700 }
701
702 ret = 1;
703 handle = NULL;
704 out_up_read:
705 up_read(&EXT4_I(inode)->xattr_sem);
706 out:
707 if (handle)
708 ext4_journal_stop(handle);
709 brelse(iloc.bh);
710 return ret;
711 convert:
712 return ext4_convert_inline_data_to_extent(mapping,
713 inode, flags);
714 }
715
716 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
717 unsigned copied, struct page *page)
718 {
719 int ret;
720 void *kaddr;
721 struct ext4_iloc iloc;
722
723 if (unlikely(copied < len)) {
724 if (!PageUptodate(page)) {
725 copied = 0;
726 goto out;
727 }
728 }
729
730 ret = ext4_get_inode_loc(inode, &iloc);
731 if (ret) {
732 ext4_std_error(inode->i_sb, ret);
733 copied = 0;
734 goto out;
735 }
736
737 down_write(&EXT4_I(inode)->xattr_sem);
738 BUG_ON(!ext4_has_inline_data(inode));
739
740 kaddr = kmap_atomic(page);
741 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
742 kunmap_atomic(kaddr);
743 SetPageUptodate(page);
744 /* clear page dirty so that writepages wouldn't work for us. */
745 ClearPageDirty(page);
746
747 up_write(&EXT4_I(inode)->xattr_sem);
748 brelse(iloc.bh);
749 out:
750 return copied;
751 }
752
753 struct buffer_head *
754 ext4_journalled_write_inline_data(struct inode *inode,
755 unsigned len,
756 struct page *page)
757 {
758 int ret;
759 void *kaddr;
760 struct ext4_iloc iloc;
761
762 ret = ext4_get_inode_loc(inode, &iloc);
763 if (ret) {
764 ext4_std_error(inode->i_sb, ret);
765 return NULL;
766 }
767
768 down_write(&EXT4_I(inode)->xattr_sem);
769 kaddr = kmap_atomic(page);
770 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
771 kunmap_atomic(kaddr);
772 up_write(&EXT4_I(inode)->xattr_sem);
773
774 return iloc.bh;
775 }
776
777 /*
778 * Try to make the page cache and handle ready for the inline data case.
779 * We can call this function in 2 cases:
780 * 1. The inode is created and the first write exceeds inline size. We can
781 * clear the inode state safely.
782 * 2. The inode has inline data, then we need to read the data, make it
783 * update and dirty so that ext4_da_writepages can handle it. We don't
784 * need to start the journal since the file's metatdata isn't changed now.
785 */
786 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
787 struct inode *inode,
788 unsigned flags,
789 void **fsdata)
790 {
791 int ret = 0, inline_size;
792 struct page *page;
793
794 page = grab_cache_page_write_begin(mapping, 0, flags);
795 if (!page)
796 return -ENOMEM;
797
798 down_read(&EXT4_I(inode)->xattr_sem);
799 if (!ext4_has_inline_data(inode)) {
800 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
801 goto out;
802 }
803
804 inline_size = ext4_get_inline_size(inode);
805
806 if (!PageUptodate(page)) {
807 ret = ext4_read_inline_page(inode, page);
808 if (ret < 0)
809 goto out;
810 }
811
812 ret = __block_write_begin(page, 0, inline_size,
813 ext4_da_get_block_prep);
814 if (ret) {
815 ext4_truncate_failed_write(inode);
816 goto out;
817 }
818
819 SetPageDirty(page);
820 SetPageUptodate(page);
821 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
822 *fsdata = (void *)CONVERT_INLINE_DATA;
823
824 out:
825 up_read(&EXT4_I(inode)->xattr_sem);
826 if (page) {
827 unlock_page(page);
828 page_cache_release(page);
829 }
830 return ret;
831 }
832
833 /*
834 * Prepare the write for the inline data.
835 * If the the data can be written into the inode, we just read
836 * the page and make it uptodate, and start the journal.
837 * Otherwise read the page, makes it dirty so that it can be
838 * handle in writepages(the i_disksize update is left to the
839 * normal ext4_da_write_end).
840 */
841 int ext4_da_write_inline_data_begin(struct address_space *mapping,
842 struct inode *inode,
843 loff_t pos, unsigned len,
844 unsigned flags,
845 struct page **pagep,
846 void **fsdata)
847 {
848 int ret, inline_size;
849 handle_t *handle;
850 struct page *page;
851 struct ext4_iloc iloc;
852
853 ret = ext4_get_inode_loc(inode, &iloc);
854 if (ret)
855 return ret;
856
857 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
858 if (IS_ERR(handle)) {
859 ret = PTR_ERR(handle);
860 handle = NULL;
861 goto out;
862 }
863
864 inline_size = ext4_get_max_inline_size(inode);
865
866 ret = -ENOSPC;
867 if (inline_size >= pos + len) {
868 ret = ext4_prepare_inline_data(handle, inode, pos + len);
869 if (ret && ret != -ENOSPC)
870 goto out;
871 }
872
873 if (ret == -ENOSPC) {
874 ret = ext4_da_convert_inline_data_to_extent(mapping,
875 inode,
876 flags,
877 fsdata);
878 goto out;
879 }
880
881 /*
882 * We cannot recurse into the filesystem as the transaction
883 * is already started.
884 */
885 flags |= AOP_FLAG_NOFS;
886
887 page = grab_cache_page_write_begin(mapping, 0, flags);
888 if (!page) {
889 ret = -ENOMEM;
890 goto out;
891 }
892
893 down_read(&EXT4_I(inode)->xattr_sem);
894 if (!ext4_has_inline_data(inode)) {
895 ret = 0;
896 goto out_release_page;
897 }
898
899 if (!PageUptodate(page)) {
900 ret = ext4_read_inline_page(inode, page);
901 if (ret < 0)
902 goto out_release_page;
903 }
904
905 up_read(&EXT4_I(inode)->xattr_sem);
906 *pagep = page;
907 handle = NULL;
908 brelse(iloc.bh);
909 return 1;
910 out_release_page:
911 up_read(&EXT4_I(inode)->xattr_sem);
912 unlock_page(page);
913 page_cache_release(page);
914 out:
915 if (handle)
916 ext4_journal_stop(handle);
917 brelse(iloc.bh);
918 return ret;
919 }
920
921 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
922 unsigned len, unsigned copied,
923 struct page *page)
924 {
925 int i_size_changed = 0;
926
927 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
928
929 /*
930 * No need to use i_size_read() here, the i_size
931 * cannot change under us because we hold i_mutex.
932 *
933 * But it's important to update i_size while still holding page lock:
934 * page writeout could otherwise come in and zero beyond i_size.
935 */
936 if (pos+copied > inode->i_size) {
937 i_size_write(inode, pos+copied);
938 i_size_changed = 1;
939 }
940 unlock_page(page);
941 page_cache_release(page);
942
943 /*
944 * Don't mark the inode dirty under page lock. First, it unnecessarily
945 * makes the holding time of page lock longer. Second, it forces lock
946 * ordering of page lock and transaction start for journaling
947 * filesystems.
948 */
949 if (i_size_changed)
950 mark_inode_dirty(inode);
951
952 return copied;
953 }
954
955 #ifdef INLINE_DIR_DEBUG
956 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
957 void *inline_start, int inline_size)
958 {
959 int offset;
960 unsigned short de_len;
961 struct ext4_dir_entry_2 *de = inline_start;
962 void *dlimit = inline_start + inline_size;
963
964 trace_printk("inode %lu\n", dir->i_ino);
965 offset = 0;
966 while ((void *)de < dlimit) {
967 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
968 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
969 offset, de_len, de->name_len, de->name,
970 de->name_len, le32_to_cpu(de->inode));
971 if (ext4_check_dir_entry(dir, NULL, de, bh,
972 inline_start, inline_size, offset))
973 BUG();
974
975 offset += de_len;
976 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
977 }
978 }
979 #else
980 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
981 #endif
982
983 /*
984 * Add a new entry into a inline dir.
985 * It will return -ENOSPC if no space is available, and -EIO
986 * and -EEXIST if directory entry already exists.
987 */
988 static int ext4_add_dirent_to_inline(handle_t *handle,
989 struct dentry *dentry,
990 struct inode *inode,
991 struct ext4_iloc *iloc,
992 void *inline_start, int inline_size)
993 {
994 struct inode *dir = dentry->d_parent->d_inode;
995 const char *name = dentry->d_name.name;
996 int namelen = dentry->d_name.len;
997 unsigned short reclen;
998 int err;
999 struct ext4_dir_entry_2 *de;
1000
1001 reclen = EXT4_DIR_REC_LEN(namelen);
1002 err = ext4_find_dest_de(dir, inode, iloc->bh,
1003 inline_start, inline_size,
1004 name, namelen, &de);
1005 if (err)
1006 return err;
1007
1008 err = ext4_journal_get_write_access(handle, iloc->bh);
1009 if (err)
1010 return err;
1011 ext4_insert_dentry(inode, de, inline_size, name, namelen);
1012
1013 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1014
1015 /*
1016 * XXX shouldn't update any times until successful
1017 * completion of syscall, but too many callers depend
1018 * on this.
1019 *
1020 * XXX similarly, too many callers depend on
1021 * ext4_new_inode() setting the times, but error
1022 * recovery deletes the inode, so the worst that can
1023 * happen is that the times are slightly out of date
1024 * and/or different from the directory change time.
1025 */
1026 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1027 ext4_update_dx_flag(dir);
1028 dir->i_version++;
1029 ext4_mark_inode_dirty(handle, dir);
1030 return 1;
1031 }
1032
1033 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1034 struct ext4_iloc *iloc)
1035 {
1036 struct ext4_xattr_entry *entry;
1037 struct ext4_xattr_ibody_header *header;
1038
1039 BUG_ON(!EXT4_I(inode)->i_inline_off);
1040
1041 header = IHDR(inode, ext4_raw_inode(iloc));
1042 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1043 EXT4_I(inode)->i_inline_off);
1044
1045 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1046 }
1047
1048 /* Set the final de to cover the whole block. */
1049 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1050 {
1051 struct ext4_dir_entry_2 *de, *prev_de;
1052 void *limit;
1053 int de_len;
1054
1055 de = (struct ext4_dir_entry_2 *)de_buf;
1056 if (old_size) {
1057 limit = de_buf + old_size;
1058 do {
1059 prev_de = de;
1060 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1061 de_buf += de_len;
1062 de = (struct ext4_dir_entry_2 *)de_buf;
1063 } while (de_buf < limit);
1064
1065 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1066 old_size, new_size);
1067 } else {
1068 /* this is just created, so create an empty entry. */
1069 de->inode = 0;
1070 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1071 }
1072 }
1073
1074 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1075 struct ext4_iloc *iloc)
1076 {
1077 int ret;
1078 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1079 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1080
1081 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1082 return -ENOSPC;
1083
1084 ret = ext4_update_inline_data(handle, dir,
1085 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1086 if (ret)
1087 return ret;
1088
1089 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1090 EXT4_I(dir)->i_inline_size -
1091 EXT4_MIN_INLINE_DATA_SIZE);
1092 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1093 return 0;
1094 }
1095
1096 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1097 struct ext4_iloc *iloc,
1098 void *buf, int inline_size)
1099 {
1100 ext4_create_inline_data(handle, inode, inline_size);
1101 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1102 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1103 }
1104
1105 static int ext4_finish_convert_inline_dir(handle_t *handle,
1106 struct inode *inode,
1107 struct buffer_head *dir_block,
1108 void *buf,
1109 int inline_size)
1110 {
1111 int err, csum_size = 0, header_size = 0;
1112 struct ext4_dir_entry_2 *de;
1113 struct ext4_dir_entry_tail *t;
1114 void *target = dir_block->b_data;
1115
1116 /*
1117 * First create "." and ".." and then copy the dir information
1118 * back to the block.
1119 */
1120 de = (struct ext4_dir_entry_2 *)target;
1121 de = ext4_init_dot_dotdot(inode, de,
1122 inode->i_sb->s_blocksize, csum_size,
1123 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1124 header_size = (void *)de - target;
1125
1126 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1127 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1128
1129 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1130 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1131 csum_size = sizeof(struct ext4_dir_entry_tail);
1132
1133 inode->i_size = inode->i_sb->s_blocksize;
1134 i_size_write(inode, inode->i_sb->s_blocksize);
1135 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1136 ext4_update_final_de(dir_block->b_data,
1137 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1138 inode->i_sb->s_blocksize - csum_size);
1139
1140 if (csum_size) {
1141 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1142 inode->i_sb->s_blocksize);
1143 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1144 }
1145 set_buffer_uptodate(dir_block);
1146 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1147 if (err)
1148 goto out;
1149 set_buffer_verified(dir_block);
1150 out:
1151 return err;
1152 }
1153
1154 static int ext4_convert_inline_data_nolock(handle_t *handle,
1155 struct inode *inode,
1156 struct ext4_iloc *iloc)
1157 {
1158 int error;
1159 void *buf = NULL;
1160 struct buffer_head *data_bh = NULL;
1161 struct ext4_map_blocks map;
1162 int inline_size;
1163
1164 inline_size = ext4_get_inline_size(inode);
1165 buf = kmalloc(inline_size, GFP_NOFS);
1166 if (!buf) {
1167 error = -ENOMEM;
1168 goto out;
1169 }
1170
1171 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1172 if (error < 0)
1173 goto out;
1174
1175 error = ext4_destroy_inline_data_nolock(handle, inode);
1176 if (error)
1177 goto out;
1178
1179 map.m_lblk = 0;
1180 map.m_len = 1;
1181 map.m_flags = 0;
1182 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1183 if (error < 0)
1184 goto out_restore;
1185 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1186 error = -EIO;
1187 goto out_restore;
1188 }
1189
1190 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1191 if (!data_bh) {
1192 error = -ENOMEM;
1193 goto out_restore;
1194 }
1195
1196 lock_buffer(data_bh);
1197 error = ext4_journal_get_create_access(handle, data_bh);
1198 if (error) {
1199 unlock_buffer(data_bh);
1200 error = -EIO;
1201 goto out_restore;
1202 }
1203 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1204
1205 if (!S_ISDIR(inode->i_mode)) {
1206 memcpy(data_bh->b_data, buf, inline_size);
1207 set_buffer_uptodate(data_bh);
1208 error = ext4_handle_dirty_metadata(handle,
1209 inode, data_bh);
1210 } else {
1211 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1212 buf, inline_size);
1213 }
1214
1215 unlock_buffer(data_bh);
1216 out_restore:
1217 if (error)
1218 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1219
1220 out:
1221 brelse(data_bh);
1222 kfree(buf);
1223 return error;
1224 }
1225
1226 /*
1227 * Try to add the new entry to the inline data.
1228 * If succeeds, return 0. If not, extended the inline dir and copied data to
1229 * the new created block.
1230 */
1231 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1232 struct inode *inode)
1233 {
1234 int ret, inline_size;
1235 void *inline_start;
1236 struct ext4_iloc iloc;
1237 struct inode *dir = dentry->d_parent->d_inode;
1238
1239 ret = ext4_get_inode_loc(dir, &iloc);
1240 if (ret)
1241 return ret;
1242
1243 down_write(&EXT4_I(dir)->xattr_sem);
1244 if (!ext4_has_inline_data(dir))
1245 goto out;
1246
1247 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1248 EXT4_INLINE_DOTDOT_SIZE;
1249 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1250
1251 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1252 inline_start, inline_size);
1253 if (ret != -ENOSPC)
1254 goto out;
1255
1256 /* check whether it can be inserted to inline xattr space. */
1257 inline_size = EXT4_I(dir)->i_inline_size -
1258 EXT4_MIN_INLINE_DATA_SIZE;
1259 if (!inline_size) {
1260 /* Try to use the xattr space.*/
1261 ret = ext4_update_inline_dir(handle, dir, &iloc);
1262 if (ret && ret != -ENOSPC)
1263 goto out;
1264
1265 inline_size = EXT4_I(dir)->i_inline_size -
1266 EXT4_MIN_INLINE_DATA_SIZE;
1267 }
1268
1269 if (inline_size) {
1270 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1271
1272 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1273 inline_start, inline_size);
1274
1275 if (ret != -ENOSPC)
1276 goto out;
1277 }
1278
1279 /*
1280 * The inline space is filled up, so create a new block for it.
1281 * As the extent tree will be created, we have to save the inline
1282 * dir first.
1283 */
1284 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1285
1286 out:
1287 ext4_mark_inode_dirty(handle, dir);
1288 up_write(&EXT4_I(dir)->xattr_sem);
1289 brelse(iloc.bh);
1290 return ret;
1291 }
1292
1293 /*
1294 * This function fills a red-black tree with information from an
1295 * inlined dir. It returns the number directory entries loaded
1296 * into the tree. If there is an error it is returned in err.
1297 */
1298 int htree_inlinedir_to_tree(struct file *dir_file,
1299 struct inode *dir, ext4_lblk_t block,
1300 struct dx_hash_info *hinfo,
1301 __u32 start_hash, __u32 start_minor_hash,
1302 int *has_inline_data)
1303 {
1304 int err = 0, count = 0;
1305 unsigned int parent_ino;
1306 int pos;
1307 struct ext4_dir_entry_2 *de;
1308 struct inode *inode = file_inode(dir_file);
1309 int ret, inline_size = 0;
1310 struct ext4_iloc iloc;
1311 void *dir_buf = NULL;
1312 struct ext4_dir_entry_2 fake;
1313
1314 ret = ext4_get_inode_loc(inode, &iloc);
1315 if (ret)
1316 return ret;
1317
1318 down_read(&EXT4_I(inode)->xattr_sem);
1319 if (!ext4_has_inline_data(inode)) {
1320 up_read(&EXT4_I(inode)->xattr_sem);
1321 *has_inline_data = 0;
1322 goto out;
1323 }
1324
1325 inline_size = ext4_get_inline_size(inode);
1326 dir_buf = kmalloc(inline_size, GFP_NOFS);
1327 if (!dir_buf) {
1328 ret = -ENOMEM;
1329 up_read(&EXT4_I(inode)->xattr_sem);
1330 goto out;
1331 }
1332
1333 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1334 up_read(&EXT4_I(inode)->xattr_sem);
1335 if (ret < 0)
1336 goto out;
1337
1338 pos = 0;
1339 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1340 while (pos < inline_size) {
1341 /*
1342 * As inlined dir doesn't store any information about '.' and
1343 * only the inode number of '..' is stored, we have to handle
1344 * them differently.
1345 */
1346 if (pos == 0) {
1347 fake.inode = cpu_to_le32(inode->i_ino);
1348 fake.name_len = 1;
1349 strcpy(fake.name, ".");
1350 fake.rec_len = ext4_rec_len_to_disk(
1351 EXT4_DIR_REC_LEN(fake.name_len),
1352 inline_size);
1353 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1354 de = &fake;
1355 pos = EXT4_INLINE_DOTDOT_OFFSET;
1356 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1357 fake.inode = cpu_to_le32(parent_ino);
1358 fake.name_len = 2;
1359 strcpy(fake.name, "..");
1360 fake.rec_len = ext4_rec_len_to_disk(
1361 EXT4_DIR_REC_LEN(fake.name_len),
1362 inline_size);
1363 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1364 de = &fake;
1365 pos = EXT4_INLINE_DOTDOT_SIZE;
1366 } else {
1367 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1368 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1369 if (ext4_check_dir_entry(inode, dir_file, de,
1370 iloc.bh, dir_buf,
1371 inline_size, pos)) {
1372 ret = count;
1373 goto out;
1374 }
1375 }
1376
1377 ext4fs_dirhash(de->name, de->name_len, hinfo);
1378 if ((hinfo->hash < start_hash) ||
1379 ((hinfo->hash == start_hash) &&
1380 (hinfo->minor_hash < start_minor_hash)))
1381 continue;
1382 if (de->inode == 0)
1383 continue;
1384 err = ext4_htree_store_dirent(dir_file,
1385 hinfo->hash, hinfo->minor_hash, de);
1386 if (err) {
1387 count = err;
1388 goto out;
1389 }
1390 count++;
1391 }
1392 ret = count;
1393 out:
1394 kfree(dir_buf);
1395 brelse(iloc.bh);
1396 return ret;
1397 }
1398
1399 /*
1400 * So this function is called when the volume is mkfsed with
1401 * dir_index disabled. In order to keep f_pos persistent
1402 * after we convert from an inlined dir to a blocked based,
1403 * we just pretend that we are a normal dir and return the
1404 * offset as if '.' and '..' really take place.
1405 *
1406 */
1407 int ext4_read_inline_dir(struct file *file,
1408 struct dir_context *ctx,
1409 int *has_inline_data)
1410 {
1411 unsigned int offset, parent_ino;
1412 int i;
1413 struct ext4_dir_entry_2 *de;
1414 struct super_block *sb;
1415 struct inode *inode = file_inode(file);
1416 int ret, inline_size = 0;
1417 struct ext4_iloc iloc;
1418 void *dir_buf = NULL;
1419 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1420
1421 ret = ext4_get_inode_loc(inode, &iloc);
1422 if (ret)
1423 return ret;
1424
1425 down_read(&EXT4_I(inode)->xattr_sem);
1426 if (!ext4_has_inline_data(inode)) {
1427 up_read(&EXT4_I(inode)->xattr_sem);
1428 *has_inline_data = 0;
1429 goto out;
1430 }
1431
1432 inline_size = ext4_get_inline_size(inode);
1433 dir_buf = kmalloc(inline_size, GFP_NOFS);
1434 if (!dir_buf) {
1435 ret = -ENOMEM;
1436 up_read(&EXT4_I(inode)->xattr_sem);
1437 goto out;
1438 }
1439
1440 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1441 up_read(&EXT4_I(inode)->xattr_sem);
1442 if (ret < 0)
1443 goto out;
1444
1445 sb = inode->i_sb;
1446 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1447 offset = ctx->pos;
1448
1449 /*
1450 * dotdot_offset and dotdot_size is the real offset and
1451 * size for ".." and "." if the dir is block based while
1452 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1453 * So we will use extra_offset and extra_size to indicate them
1454 * during the inline dir iteration.
1455 */
1456 dotdot_offset = EXT4_DIR_REC_LEN(1);
1457 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1458 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1459 extra_size = extra_offset + inline_size;
1460
1461 /*
1462 * If the version has changed since the last call to
1463 * readdir(2), then we might be pointing to an invalid
1464 * dirent right now. Scan from the start of the inline
1465 * dir to make sure.
1466 */
1467 if (file->f_version != inode->i_version) {
1468 for (i = 0; i < extra_size && i < offset;) {
1469 /*
1470 * "." is with offset 0 and
1471 * ".." is dotdot_offset.
1472 */
1473 if (!i) {
1474 i = dotdot_offset;
1475 continue;
1476 } else if (i == dotdot_offset) {
1477 i = dotdot_size;
1478 continue;
1479 }
1480 /* for other entry, the real offset in
1481 * the buf has to be tuned accordingly.
1482 */
1483 de = (struct ext4_dir_entry_2 *)
1484 (dir_buf + i - extra_offset);
1485 /* It's too expensive to do a full
1486 * dirent test each time round this
1487 * loop, but we do have to test at
1488 * least that it is non-zero. A
1489 * failure will be detected in the
1490 * dirent test below. */
1491 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1492 < EXT4_DIR_REC_LEN(1))
1493 break;
1494 i += ext4_rec_len_from_disk(de->rec_len,
1495 extra_size);
1496 }
1497 offset = i;
1498 ctx->pos = offset;
1499 file->f_version = inode->i_version;
1500 }
1501
1502 while (ctx->pos < extra_size) {
1503 if (ctx->pos == 0) {
1504 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1505 goto out;
1506 ctx->pos = dotdot_offset;
1507 continue;
1508 }
1509
1510 if (ctx->pos == dotdot_offset) {
1511 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1512 goto out;
1513 ctx->pos = dotdot_size;
1514 continue;
1515 }
1516
1517 de = (struct ext4_dir_entry_2 *)
1518 (dir_buf + ctx->pos - extra_offset);
1519 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1520 extra_size, ctx->pos))
1521 goto out;
1522 if (le32_to_cpu(de->inode)) {
1523 if (!dir_emit(ctx, de->name, de->name_len,
1524 le32_to_cpu(de->inode),
1525 get_dtype(sb, de->file_type)))
1526 goto out;
1527 }
1528 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1529 }
1530 out:
1531 kfree(dir_buf);
1532 brelse(iloc.bh);
1533 return ret;
1534 }
1535
1536 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1537 struct ext4_dir_entry_2 **parent_de,
1538 int *retval)
1539 {
1540 struct ext4_iloc iloc;
1541
1542 *retval = ext4_get_inode_loc(inode, &iloc);
1543 if (*retval)
1544 return NULL;
1545
1546 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1547
1548 return iloc.bh;
1549 }
1550
1551 /*
1552 * Try to create the inline data for the new dir.
1553 * If it succeeds, return 0, otherwise return the error.
1554 * In case of ENOSPC, the caller should create the normal disk layout dir.
1555 */
1556 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1557 struct inode *inode)
1558 {
1559 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1560 struct ext4_iloc iloc;
1561 struct ext4_dir_entry_2 *de;
1562
1563 ret = ext4_get_inode_loc(inode, &iloc);
1564 if (ret)
1565 return ret;
1566
1567 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1568 if (ret)
1569 goto out;
1570
1571 /*
1572 * For inline dir, we only save the inode information for the ".."
1573 * and create a fake dentry to cover the left space.
1574 */
1575 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1576 de->inode = cpu_to_le32(parent->i_ino);
1577 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1578 de->inode = 0;
1579 de->rec_len = ext4_rec_len_to_disk(
1580 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1581 inline_size);
1582 set_nlink(inode, 2);
1583 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1584 out:
1585 brelse(iloc.bh);
1586 return ret;
1587 }
1588
1589 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1590 const struct qstr *d_name,
1591 struct ext4_dir_entry_2 **res_dir,
1592 int *has_inline_data)
1593 {
1594 int ret;
1595 struct ext4_iloc iloc;
1596 void *inline_start;
1597 int inline_size;
1598
1599 if (ext4_get_inode_loc(dir, &iloc))
1600 return NULL;
1601
1602 down_read(&EXT4_I(dir)->xattr_sem);
1603 if (!ext4_has_inline_data(dir)) {
1604 *has_inline_data = 0;
1605 goto out;
1606 }
1607
1608 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1609 EXT4_INLINE_DOTDOT_SIZE;
1610 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1611 ret = search_dir(iloc.bh, inline_start, inline_size,
1612 dir, d_name, 0, res_dir);
1613 if (ret == 1)
1614 goto out_find;
1615 if (ret < 0)
1616 goto out;
1617
1618 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1619 goto out;
1620
1621 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1622 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1623
1624 ret = search_dir(iloc.bh, inline_start, inline_size,
1625 dir, d_name, 0, res_dir);
1626 if (ret == 1)
1627 goto out_find;
1628
1629 out:
1630 brelse(iloc.bh);
1631 iloc.bh = NULL;
1632 out_find:
1633 up_read(&EXT4_I(dir)->xattr_sem);
1634 return iloc.bh;
1635 }
1636
1637 int ext4_delete_inline_entry(handle_t *handle,
1638 struct inode *dir,
1639 struct ext4_dir_entry_2 *de_del,
1640 struct buffer_head *bh,
1641 int *has_inline_data)
1642 {
1643 int err, inline_size;
1644 struct ext4_iloc iloc;
1645 void *inline_start;
1646
1647 err = ext4_get_inode_loc(dir, &iloc);
1648 if (err)
1649 return err;
1650
1651 down_write(&EXT4_I(dir)->xattr_sem);
1652 if (!ext4_has_inline_data(dir)) {
1653 *has_inline_data = 0;
1654 goto out;
1655 }
1656
1657 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1658 EXT4_MIN_INLINE_DATA_SIZE) {
1659 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1660 EXT4_INLINE_DOTDOT_SIZE;
1661 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1662 EXT4_INLINE_DOTDOT_SIZE;
1663 } else {
1664 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1665 inline_size = ext4_get_inline_size(dir) -
1666 EXT4_MIN_INLINE_DATA_SIZE;
1667 }
1668
1669 err = ext4_journal_get_write_access(handle, bh);
1670 if (err)
1671 goto out;
1672
1673 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1674 inline_start, inline_size, 0);
1675 if (err)
1676 goto out;
1677
1678 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1679 err = ext4_mark_inode_dirty(handle, dir);
1680 if (unlikely(err))
1681 goto out;
1682
1683 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1684 out:
1685 up_write(&EXT4_I(dir)->xattr_sem);
1686 brelse(iloc.bh);
1687 if (err != -ENOENT)
1688 ext4_std_error(dir->i_sb, err);
1689 return err;
1690 }
1691
1692 /*
1693 * Get the inline dentry at offset.
1694 */
1695 static inline struct ext4_dir_entry_2 *
1696 ext4_get_inline_entry(struct inode *inode,
1697 struct ext4_iloc *iloc,
1698 unsigned int offset,
1699 void **inline_start,
1700 int *inline_size)
1701 {
1702 void *inline_pos;
1703
1704 BUG_ON(offset > ext4_get_inline_size(inode));
1705
1706 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1707 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1708 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1709 } else {
1710 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1711 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1712 *inline_size = ext4_get_inline_size(inode) -
1713 EXT4_MIN_INLINE_DATA_SIZE;
1714 }
1715
1716 if (inline_start)
1717 *inline_start = inline_pos;
1718 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1719 }
1720
1721 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1722 {
1723 int err, inline_size;
1724 struct ext4_iloc iloc;
1725 void *inline_pos;
1726 unsigned int offset;
1727 struct ext4_dir_entry_2 *de;
1728 int ret = 1;
1729
1730 err = ext4_get_inode_loc(dir, &iloc);
1731 if (err) {
1732 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1733 err, dir->i_ino);
1734 return 1;
1735 }
1736
1737 down_read(&EXT4_I(dir)->xattr_sem);
1738 if (!ext4_has_inline_data(dir)) {
1739 *has_inline_data = 0;
1740 goto out;
1741 }
1742
1743 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1744 if (!le32_to_cpu(de->inode)) {
1745 ext4_warning(dir->i_sb,
1746 "bad inline directory (dir #%lu) - no `..'",
1747 dir->i_ino);
1748 ret = 1;
1749 goto out;
1750 }
1751
1752 offset = EXT4_INLINE_DOTDOT_SIZE;
1753 while (offset < dir->i_size) {
1754 de = ext4_get_inline_entry(dir, &iloc, offset,
1755 &inline_pos, &inline_size);
1756 if (ext4_check_dir_entry(dir, NULL, de,
1757 iloc.bh, inline_pos,
1758 inline_size, offset)) {
1759 ext4_warning(dir->i_sb,
1760 "bad inline directory (dir #%lu) - "
1761 "inode %u, rec_len %u, name_len %d"
1762 "inline size %d\n",
1763 dir->i_ino, le32_to_cpu(de->inode),
1764 le16_to_cpu(de->rec_len), de->name_len,
1765 inline_size);
1766 ret = 1;
1767 goto out;
1768 }
1769 if (le32_to_cpu(de->inode)) {
1770 ret = 0;
1771 goto out;
1772 }
1773 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1774 }
1775
1776 out:
1777 up_read(&EXT4_I(dir)->xattr_sem);
1778 brelse(iloc.bh);
1779 return ret;
1780 }
1781
1782 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1783 {
1784 int ret;
1785
1786 down_write(&EXT4_I(inode)->xattr_sem);
1787 ret = ext4_destroy_inline_data_nolock(handle, inode);
1788 up_write(&EXT4_I(inode)->xattr_sem);
1789
1790 return ret;
1791 }
1792
1793 int ext4_inline_data_fiemap(struct inode *inode,
1794 struct fiemap_extent_info *fieinfo,
1795 int *has_inline)
1796 {
1797 __u64 physical = 0;
1798 __u64 length;
1799 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST;
1800 int error = 0;
1801 struct ext4_iloc iloc;
1802
1803 down_read(&EXT4_I(inode)->xattr_sem);
1804 if (!ext4_has_inline_data(inode)) {
1805 *has_inline = 0;
1806 goto out;
1807 }
1808
1809 error = ext4_get_inode_loc(inode, &iloc);
1810 if (error)
1811 goto out;
1812
1813 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1814 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1815 physical += offsetof(struct ext4_inode, i_block);
1816 length = i_size_read(inode);
1817
1818 if (physical)
1819 error = fiemap_fill_next_extent(fieinfo, 0, physical,
1820 length, flags);
1821 brelse(iloc.bh);
1822 out:
1823 up_read(&EXT4_I(inode)->xattr_sem);
1824 return (error < 0 ? error : 0);
1825 }
1826
1827 /*
1828 * Called during xattr set, and if we can sparse space 'needed',
1829 * just create the extent tree evict the data to the outer block.
1830 *
1831 * We use jbd2 instead of page cache to move data to the 1st block
1832 * so that the whole transaction can be committed as a whole and
1833 * the data isn't lost because of the delayed page cache write.
1834 */
1835 int ext4_try_to_evict_inline_data(handle_t *handle,
1836 struct inode *inode,
1837 int needed)
1838 {
1839 int error;
1840 struct ext4_xattr_entry *entry;
1841 struct ext4_xattr_ibody_header *header;
1842 struct ext4_inode *raw_inode;
1843 struct ext4_iloc iloc;
1844
1845 error = ext4_get_inode_loc(inode, &iloc);
1846 if (error)
1847 return error;
1848
1849 raw_inode = ext4_raw_inode(&iloc);
1850 header = IHDR(inode, raw_inode);
1851 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1852 EXT4_I(inode)->i_inline_off);
1853 if (EXT4_XATTR_LEN(entry->e_name_len) +
1854 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1855 error = -ENOSPC;
1856 goto out;
1857 }
1858
1859 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1860 out:
1861 brelse(iloc.bh);
1862 return error;
1863 }
1864
1865 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1866 {
1867 handle_t *handle;
1868 int inline_size, value_len, needed_blocks;
1869 size_t i_size;
1870 void *value = NULL;
1871 struct ext4_xattr_ibody_find is = {
1872 .s = { .not_found = -ENODATA, },
1873 };
1874 struct ext4_xattr_info i = {
1875 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1876 .name = EXT4_XATTR_SYSTEM_DATA,
1877 };
1878
1879
1880 needed_blocks = ext4_writepage_trans_blocks(inode);
1881 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1882 if (IS_ERR(handle))
1883 return;
1884
1885 down_write(&EXT4_I(inode)->xattr_sem);
1886 if (!ext4_has_inline_data(inode)) {
1887 *has_inline = 0;
1888 ext4_journal_stop(handle);
1889 return;
1890 }
1891
1892 if (ext4_orphan_add(handle, inode))
1893 goto out;
1894
1895 if (ext4_get_inode_loc(inode, &is.iloc))
1896 goto out;
1897
1898 down_write(&EXT4_I(inode)->i_data_sem);
1899 i_size = inode->i_size;
1900 inline_size = ext4_get_inline_size(inode);
1901 EXT4_I(inode)->i_disksize = i_size;
1902
1903 if (i_size < inline_size) {
1904 /* Clear the content in the xattr space. */
1905 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1906 if (ext4_xattr_ibody_find(inode, &i, &is))
1907 goto out_error;
1908
1909 BUG_ON(is.s.not_found);
1910
1911 value_len = le32_to_cpu(is.s.here->e_value_size);
1912 value = kmalloc(value_len, GFP_NOFS);
1913 if (!value)
1914 goto out_error;
1915
1916 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1917 value, value_len))
1918 goto out_error;
1919
1920 i.value = value;
1921 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1922 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1923 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1924 goto out_error;
1925 }
1926
1927 /* Clear the content within i_blocks. */
1928 if (i_size < EXT4_MIN_INLINE_DATA_SIZE)
1929 memset(ext4_raw_inode(&is.iloc)->i_block + i_size, 0,
1930 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1931
1932 EXT4_I(inode)->i_inline_size = i_size <
1933 EXT4_MIN_INLINE_DATA_SIZE ?
1934 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1935 }
1936
1937 out_error:
1938 up_write(&EXT4_I(inode)->i_data_sem);
1939 out:
1940 brelse(is.iloc.bh);
1941 up_write(&EXT4_I(inode)->xattr_sem);
1942 kfree(value);
1943 if (inode->i_nlink)
1944 ext4_orphan_del(handle, inode);
1945
1946 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1947 ext4_mark_inode_dirty(handle, inode);
1948 if (IS_SYNC(inode))
1949 ext4_handle_sync(handle);
1950
1951 ext4_journal_stop(handle);
1952 return;
1953 }
1954
1955 int ext4_convert_inline_data(struct inode *inode)
1956 {
1957 int error, needed_blocks;
1958 handle_t *handle;
1959 struct ext4_iloc iloc;
1960
1961 if (!ext4_has_inline_data(inode)) {
1962 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1963 return 0;
1964 }
1965
1966 needed_blocks = ext4_writepage_trans_blocks(inode);
1967
1968 iloc.bh = NULL;
1969 error = ext4_get_inode_loc(inode, &iloc);
1970 if (error)
1971 return error;
1972
1973 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1974 if (IS_ERR(handle)) {
1975 error = PTR_ERR(handle);
1976 goto out_free;
1977 }
1978
1979 down_write(&EXT4_I(inode)->xattr_sem);
1980 if (!ext4_has_inline_data(inode)) {
1981 up_write(&EXT4_I(inode)->xattr_sem);
1982 goto out;
1983 }
1984
1985 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1986 up_write(&EXT4_I(inode)->xattr_sem);
1987 out:
1988 ext4_journal_stop(handle);
1989 out_free:
1990 brelse(iloc.bh);
1991 return error;
1992 }
Linus' kernel tree