?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
slub/memcg: cure the brainless abuse of sysfs attributes
[linux/fpc-iii.git] / fs / ext4 / inline.c
blobdad8e7bdf0a61c19c0afe352d1252ffa09cfdf84
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_feature_extents(inode->i_sb)) {
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 page = NULL;
601 ext4_orphan_add(handle, inode);
602 up_write(&EXT4_I(inode)->xattr_sem);
603 sem_held = 0;
604 ext4_journal_stop(handle);
605 handle = NULL;
606 ext4_truncate_failed_write(inode);
607 /*
608 * If truncate failed early the inode might
609 * still be on the orphan list; we need to
610 * make sure the inode is removed from the
611 * orphan list in that case.
612 */
613 if (inode->i_nlink)
614 ext4_orphan_del(NULL, inode);
615 }
616
617 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
618 goto retry;
619
620 if (page)
621 block_commit_write(page, from, to);
622 out:
623 if (page) {
624 unlock_page(page);
625 page_cache_release(page);
626 }
627 if (sem_held)
628 up_write(&EXT4_I(inode)->xattr_sem);
629 if (handle)
630 ext4_journal_stop(handle);
631 brelse(iloc.bh);
632 return ret;
633 }
634
635 /*
636 * Try to write data in the inode.
637 * If the inode has inline data, check whether the new write can be
638 * in the inode also. If not, create the page the handle, move the data
639 * to the page make it update and let the later codes create extent for it.
640 */
641 int ext4_try_to_write_inline_data(struct address_space *mapping,
642 struct inode *inode,
643 loff_t pos, unsigned len,
644 unsigned flags,
645 struct page **pagep)
646 {
647 int ret;
648 handle_t *handle;
649 struct page *page;
650 struct ext4_iloc iloc;
651
652 if (pos + len > ext4_get_max_inline_size(inode))
653 goto convert;
654
655 ret = ext4_get_inode_loc(inode, &iloc);
656 if (ret)
657 return ret;
658
659 /*
660 * The possible write could happen in the inode,
661 * so try to reserve the space in inode first.
662 */
663 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
664 if (IS_ERR(handle)) {
665 ret = PTR_ERR(handle);
666 handle = NULL;
667 goto out;
668 }
669
670 ret = ext4_prepare_inline_data(handle, inode, pos + len);
671 if (ret && ret != -ENOSPC)
672 goto out;
673
674 /* We don't have space in inline inode, so convert it to extent. */
675 if (ret == -ENOSPC) {
676 ext4_journal_stop(handle);
677 brelse(iloc.bh);
678 goto convert;
679 }
680
681 flags |= AOP_FLAG_NOFS;
682
683 page = grab_cache_page_write_begin(mapping, 0, flags);
684 if (!page) {
685 ret = -ENOMEM;
686 goto out;
687 }
688
689 *pagep = page;
690 down_read(&EXT4_I(inode)->xattr_sem);
691 if (!ext4_has_inline_data(inode)) {
692 ret = 0;
693 unlock_page(page);
694 page_cache_release(page);
695 goto out_up_read;
696 }
697
698 if (!PageUptodate(page)) {
699 ret = ext4_read_inline_page(inode, page);
700 if (ret < 0)
701 goto out_up_read;
702 }
703
704 ret = 1;
705 handle = NULL;
706 out_up_read:
707 up_read(&EXT4_I(inode)->xattr_sem);
708 out:
709 if (handle)
710 ext4_journal_stop(handle);
711 brelse(iloc.bh);
712 return ret;
713 convert:
714 return ext4_convert_inline_data_to_extent(mapping,
715 inode, flags);
716 }
717
718 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
719 unsigned copied, struct page *page)
720 {
721 int ret;
722 void *kaddr;
723 struct ext4_iloc iloc;
724
725 if (unlikely(copied < len)) {
726 if (!PageUptodate(page)) {
727 copied = 0;
728 goto out;
729 }
730 }
731
732 ret = ext4_get_inode_loc(inode, &iloc);
733 if (ret) {
734 ext4_std_error(inode->i_sb, ret);
735 copied = 0;
736 goto out;
737 }
738
739 down_write(&EXT4_I(inode)->xattr_sem);
740 BUG_ON(!ext4_has_inline_data(inode));
741
742 kaddr = kmap_atomic(page);
743 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
744 kunmap_atomic(kaddr);
745 SetPageUptodate(page);
746 /* clear page dirty so that writepages wouldn't work for us. */
747 ClearPageDirty(page);
748
749 up_write(&EXT4_I(inode)->xattr_sem);
750 brelse(iloc.bh);
751 out:
752 return copied;
753 }
754
755 struct buffer_head *
756 ext4_journalled_write_inline_data(struct inode *inode,
757 unsigned len,
758 struct page *page)
759 {
760 int ret;
761 void *kaddr;
762 struct ext4_iloc iloc;
763
764 ret = ext4_get_inode_loc(inode, &iloc);
765 if (ret) {
766 ext4_std_error(inode->i_sb, ret);
767 return NULL;
768 }
769
770 down_write(&EXT4_I(inode)->xattr_sem);
771 kaddr = kmap_atomic(page);
772 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
773 kunmap_atomic(kaddr);
774 up_write(&EXT4_I(inode)->xattr_sem);
775
776 return iloc.bh;
777 }
778
779 /*
780 * Try to make the page cache and handle ready for the inline data case.
781 * We can call this function in 2 cases:
782 * 1. The inode is created and the first write exceeds inline size. We can
783 * clear the inode state safely.
784 * 2. The inode has inline data, then we need to read the data, make it
785 * update and dirty so that ext4_da_writepages can handle it. We don't
786 * need to start the journal since the file's metatdata isn't changed now.
787 */
788 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
789 struct inode *inode,
790 unsigned flags,
791 void **fsdata)
792 {
793 int ret = 0, inline_size;
794 struct page *page;
795
796 page = grab_cache_page_write_begin(mapping, 0, flags);
797 if (!page)
798 return -ENOMEM;
799
800 down_read(&EXT4_I(inode)->xattr_sem);
801 if (!ext4_has_inline_data(inode)) {
802 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
803 goto out;
804 }
805
806 inline_size = ext4_get_inline_size(inode);
807
808 if (!PageUptodate(page)) {
809 ret = ext4_read_inline_page(inode, page);
810 if (ret < 0)
811 goto out;
812 }
813
814 ret = __block_write_begin(page, 0, inline_size,
815 ext4_da_get_block_prep);
816 if (ret) {
817 up_read(&EXT4_I(inode)->xattr_sem);
818 unlock_page(page);
819 page_cache_release(page);
820 ext4_truncate_failed_write(inode);
821 return ret;
822 }
823
824 SetPageDirty(page);
825 SetPageUptodate(page);
826 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
827 *fsdata = (void *)CONVERT_INLINE_DATA;
828
829 out:
830 up_read(&EXT4_I(inode)->xattr_sem);
831 if (page) {
832 unlock_page(page);
833 page_cache_release(page);
834 }
835 return ret;
836 }
837
838 /*
839 * Prepare the write for the inline data.
840 * If the the data can be written into the inode, we just read
841 * the page and make it uptodate, and start the journal.
842 * Otherwise read the page, makes it dirty so that it can be
843 * handle in writepages(the i_disksize update is left to the
844 * normal ext4_da_write_end).
845 */
846 int ext4_da_write_inline_data_begin(struct address_space *mapping,
847 struct inode *inode,
848 loff_t pos, unsigned len,
849 unsigned flags,
850 struct page **pagep,
851 void **fsdata)
852 {
853 int ret, inline_size;
854 handle_t *handle;
855 struct page *page;
856 struct ext4_iloc iloc;
857 int retries;
858
859 ret = ext4_get_inode_loc(inode, &iloc);
860 if (ret)
861 return ret;
862
863 retry_journal:
864 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
865 if (IS_ERR(handle)) {
866 ret = PTR_ERR(handle);
867 goto out;
868 }
869
870 inline_size = ext4_get_max_inline_size(inode);
871
872 ret = -ENOSPC;
873 if (inline_size >= pos + len) {
874 ret = ext4_prepare_inline_data(handle, inode, pos + len);
875 if (ret && ret != -ENOSPC)
876 goto out_journal;
877 }
878
879 /*
880 * We cannot recurse into the filesystem as the transaction
881 * is already started.
882 */
883 flags |= AOP_FLAG_NOFS;
884
885 if (ret == -ENOSPC) {
886 ret = ext4_da_convert_inline_data_to_extent(mapping,
887 inode,
888 flags,
889 fsdata);
890 ext4_journal_stop(handle);
891 if (ret == -ENOSPC &&
892 ext4_should_retry_alloc(inode->i_sb, &retries))
893 goto retry_journal;
894 goto out;
895 }
896
897
898 page = grab_cache_page_write_begin(mapping, 0, flags);
899 if (!page) {
900 ret = -ENOMEM;
901 goto out_journal;
902 }
903
904 down_read(&EXT4_I(inode)->xattr_sem);
905 if (!ext4_has_inline_data(inode)) {
906 ret = 0;
907 goto out_release_page;
908 }
909
910 if (!PageUptodate(page)) {
911 ret = ext4_read_inline_page(inode, page);
912 if (ret < 0)
913 goto out_release_page;
914 }
915
916 up_read(&EXT4_I(inode)->xattr_sem);
917 *pagep = page;
918 brelse(iloc.bh);
919 return 1;
920 out_release_page:
921 up_read(&EXT4_I(inode)->xattr_sem);
922 unlock_page(page);
923 page_cache_release(page);
924 out_journal:
925 ext4_journal_stop(handle);
926 out:
927 brelse(iloc.bh);
928 return ret;
929 }
930
931 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
932 unsigned len, unsigned copied,
933 struct page *page)
934 {
935 int i_size_changed = 0;
936 int ret;
937
938 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
939 if (ret < 0) {
940 unlock_page(page);
941 put_page(page);
942 return ret;
943 }
944 copied = ret;
945
946 /*
947 * No need to use i_size_read() here, the i_size
948 * cannot change under us because we hold i_mutex.
949 *
950 * But it's important to update i_size while still holding page lock:
951 * page writeout could otherwise come in and zero beyond i_size.
952 */
953 if (pos+copied > inode->i_size) {
954 i_size_write(inode, pos+copied);
955 i_size_changed = 1;
956 }
957 unlock_page(page);
958 page_cache_release(page);
959
960 /*
961 * Don't mark the inode dirty under page lock. First, it unnecessarily
962 * makes the holding time of page lock longer. Second, it forces lock
963 * ordering of page lock and transaction start for journaling
964 * filesystems.
965 */
966 if (i_size_changed)
967 mark_inode_dirty(inode);
968
969 return copied;
970 }
971
972 #ifdef INLINE_DIR_DEBUG
973 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
974 void *inline_start, int inline_size)
975 {
976 int offset;
977 unsigned short de_len;
978 struct ext4_dir_entry_2 *de = inline_start;
979 void *dlimit = inline_start + inline_size;
980
981 trace_printk("inode %lu\n", dir->i_ino);
982 offset = 0;
983 while ((void *)de < dlimit) {
984 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
985 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
986 offset, de_len, de->name_len, de->name,
987 de->name_len, le32_to_cpu(de->inode));
988 if (ext4_check_dir_entry(dir, NULL, de, bh,
989 inline_start, inline_size, offset))
990 BUG();
991
992 offset += de_len;
993 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
994 }
995 }
996 #else
997 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
998 #endif
999
1000 /*
1001 * Add a new entry into a inline dir.
1002 * It will return -ENOSPC if no space is available, and -EIO
1003 * and -EEXIST if directory entry already exists.
1004 */
1005 static int ext4_add_dirent_to_inline(handle_t *handle,
1006 struct ext4_filename *fname,
1007 struct dentry *dentry,
1008 struct inode *inode,
1009 struct ext4_iloc *iloc,
1010 void *inline_start, int inline_size)
1011 {
1012 struct inode *dir = d_inode(dentry->d_parent);
1013 int err;
1014 struct ext4_dir_entry_2 *de;
1015
1016 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1017 inline_size, fname, &de);
1018 if (err)
1019 return err;
1020
1021 BUFFER_TRACE(iloc->bh, "get_write_access");
1022 err = ext4_journal_get_write_access(handle, iloc->bh);
1023 if (err)
1024 return err;
1025 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1026
1027 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1028
1029 /*
1030 * XXX shouldn't update any times until successful
1031 * completion of syscall, but too many callers depend
1032 * on this.
1033 *
1034 * XXX similarly, too many callers depend on
1035 * ext4_new_inode() setting the times, but error
1036 * recovery deletes the inode, so the worst that can
1037 * happen is that the times are slightly out of date
1038 * and/or different from the directory change time.
1039 */
1040 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1041 ext4_update_dx_flag(dir);
1042 dir->i_version++;
1043 ext4_mark_inode_dirty(handle, dir);
1044 return 1;
1045 }
1046
1047 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1048 struct ext4_iloc *iloc)
1049 {
1050 struct ext4_xattr_entry *entry;
1051 struct ext4_xattr_ibody_header *header;
1052
1053 BUG_ON(!EXT4_I(inode)->i_inline_off);
1054
1055 header = IHDR(inode, ext4_raw_inode(iloc));
1056 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1057 EXT4_I(inode)->i_inline_off);
1058
1059 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1060 }
1061
1062 /* Set the final de to cover the whole block. */
1063 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1064 {
1065 struct ext4_dir_entry_2 *de, *prev_de;
1066 void *limit;
1067 int de_len;
1068
1069 de = (struct ext4_dir_entry_2 *)de_buf;
1070 if (old_size) {
1071 limit = de_buf + old_size;
1072 do {
1073 prev_de = de;
1074 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1075 de_buf += de_len;
1076 de = (struct ext4_dir_entry_2 *)de_buf;
1077 } while (de_buf < limit);
1078
1079 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1080 old_size, new_size);
1081 } else {
1082 /* this is just created, so create an empty entry. */
1083 de->inode = 0;
1084 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1085 }
1086 }
1087
1088 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1089 struct ext4_iloc *iloc)
1090 {
1091 int ret;
1092 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1093 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1094
1095 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1096 return -ENOSPC;
1097
1098 ret = ext4_update_inline_data(handle, dir,
1099 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1100 if (ret)
1101 return ret;
1102
1103 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1104 EXT4_I(dir)->i_inline_size -
1105 EXT4_MIN_INLINE_DATA_SIZE);
1106 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1107 return 0;
1108 }
1109
1110 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1111 struct ext4_iloc *iloc,
1112 void *buf, int inline_size)
1113 {
1114 ext4_create_inline_data(handle, inode, inline_size);
1115 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1116 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1117 }
1118
1119 static int ext4_finish_convert_inline_dir(handle_t *handle,
1120 struct inode *inode,
1121 struct buffer_head *dir_block,
1122 void *buf,
1123 int inline_size)
1124 {
1125 int err, csum_size = 0, header_size = 0;
1126 struct ext4_dir_entry_2 *de;
1127 struct ext4_dir_entry_tail *t;
1128 void *target = dir_block->b_data;
1129
1130 /*
1131 * First create "." and ".." and then copy the dir information
1132 * back to the block.
1133 */
1134 de = (struct ext4_dir_entry_2 *)target;
1135 de = ext4_init_dot_dotdot(inode, de,
1136 inode->i_sb->s_blocksize, csum_size,
1137 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1138 header_size = (void *)de - target;
1139
1140 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1141 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1142
1143 if (ext4_has_metadata_csum(inode->i_sb))
1144 csum_size = sizeof(struct ext4_dir_entry_tail);
1145
1146 inode->i_size = inode->i_sb->s_blocksize;
1147 i_size_write(inode, inode->i_sb->s_blocksize);
1148 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1149 ext4_update_final_de(dir_block->b_data,
1150 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1151 inode->i_sb->s_blocksize - csum_size);
1152
1153 if (csum_size) {
1154 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1155 inode->i_sb->s_blocksize);
1156 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1157 }
1158 set_buffer_uptodate(dir_block);
1159 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1160 if (err)
1161 return err;
1162 set_buffer_verified(dir_block);
1163 return ext4_mark_inode_dirty(handle, inode);
1164 }
1165
1166 static int ext4_convert_inline_data_nolock(handle_t *handle,
1167 struct inode *inode,
1168 struct ext4_iloc *iloc)
1169 {
1170 int error;
1171 void *buf = NULL;
1172 struct buffer_head *data_bh = NULL;
1173 struct ext4_map_blocks map;
1174 int inline_size;
1175
1176 inline_size = ext4_get_inline_size(inode);
1177 buf = kmalloc(inline_size, GFP_NOFS);
1178 if (!buf) {
1179 error = -ENOMEM;
1180 goto out;
1181 }
1182
1183 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1184 if (error < 0)
1185 goto out;
1186
1187 /*
1188 * Make sure the inline directory entries pass checks before we try to
1189 * convert them, so that we avoid touching stuff that needs fsck.
1190 */
1191 if (S_ISDIR(inode->i_mode)) {
1192 error = ext4_check_all_de(inode, iloc->bh,
1193 buf + EXT4_INLINE_DOTDOT_SIZE,
1194 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1195 if (error)
1196 goto out;
1197 }
1198
1199 error = ext4_destroy_inline_data_nolock(handle, inode);
1200 if (error)
1201 goto out;
1202
1203 map.m_lblk = 0;
1204 map.m_len = 1;
1205 map.m_flags = 0;
1206 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1207 if (error < 0)
1208 goto out_restore;
1209 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1210 error = -EIO;
1211 goto out_restore;
1212 }
1213
1214 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1215 if (!data_bh) {
1216 error = -ENOMEM;
1217 goto out_restore;
1218 }
1219
1220 lock_buffer(data_bh);
1221 error = ext4_journal_get_create_access(handle, data_bh);
1222 if (error) {
1223 unlock_buffer(data_bh);
1224 error = -EIO;
1225 goto out_restore;
1226 }
1227 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1228
1229 if (!S_ISDIR(inode->i_mode)) {
1230 memcpy(data_bh->b_data, buf, inline_size);
1231 set_buffer_uptodate(data_bh);
1232 error = ext4_handle_dirty_metadata(handle,
1233 inode, data_bh);
1234 } else {
1235 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1236 buf, inline_size);
1237 }
1238
1239 unlock_buffer(data_bh);
1240 out_restore:
1241 if (error)
1242 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1243
1244 out:
1245 brelse(data_bh);
1246 kfree(buf);
1247 return error;
1248 }
1249
1250 /*
1251 * Try to add the new entry to the inline data.
1252 * If succeeds, return 0. If not, extended the inline dir and copied data to
1253 * the new created block.
1254 */
1255 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1256 struct dentry *dentry, struct inode *inode)
1257 {
1258 int ret, inline_size;
1259 void *inline_start;
1260 struct ext4_iloc iloc;
1261 struct inode *dir = d_inode(dentry->d_parent);
1262
1263 ret = ext4_get_inode_loc(dir, &iloc);
1264 if (ret)
1265 return ret;
1266
1267 down_write(&EXT4_I(dir)->xattr_sem);
1268 if (!ext4_has_inline_data(dir))
1269 goto out;
1270
1271 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1272 EXT4_INLINE_DOTDOT_SIZE;
1273 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1274
1275 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1276 inline_start, inline_size);
1277 if (ret != -ENOSPC)
1278 goto out;
1279
1280 /* check whether it can be inserted to inline xattr space. */
1281 inline_size = EXT4_I(dir)->i_inline_size -
1282 EXT4_MIN_INLINE_DATA_SIZE;
1283 if (!inline_size) {
1284 /* Try to use the xattr space.*/
1285 ret = ext4_update_inline_dir(handle, dir, &iloc);
1286 if (ret && ret != -ENOSPC)
1287 goto out;
1288
1289 inline_size = EXT4_I(dir)->i_inline_size -
1290 EXT4_MIN_INLINE_DATA_SIZE;
1291 }
1292
1293 if (inline_size) {
1294 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1295
1296 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1297 inode, &iloc, inline_start,
1298 inline_size);
1299
1300 if (ret != -ENOSPC)
1301 goto out;
1302 }
1303
1304 /*
1305 * The inline space is filled up, so create a new block for it.
1306 * As the extent tree will be created, we have to save the inline
1307 * dir first.
1308 */
1309 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1310
1311 out:
1312 ext4_mark_inode_dirty(handle, dir);
1313 up_write(&EXT4_I(dir)->xattr_sem);
1314 brelse(iloc.bh);
1315 return ret;
1316 }
1317
1318 /*
1319 * This function fills a red-black tree with information from an
1320 * inlined dir. It returns the number directory entries loaded
1321 * into the tree. If there is an error it is returned in err.
1322 */
1323 int htree_inlinedir_to_tree(struct file *dir_file,
1324 struct inode *dir, ext4_lblk_t block,
1325 struct dx_hash_info *hinfo,
1326 __u32 start_hash, __u32 start_minor_hash,
1327 int *has_inline_data)
1328 {
1329 int err = 0, count = 0;
1330 unsigned int parent_ino;
1331 int pos;
1332 struct ext4_dir_entry_2 *de;
1333 struct inode *inode = file_inode(dir_file);
1334 int ret, inline_size = 0;
1335 struct ext4_iloc iloc;
1336 void *dir_buf = NULL;
1337 struct ext4_dir_entry_2 fake;
1338 struct ext4_str tmp_str;
1339
1340 ret = ext4_get_inode_loc(inode, &iloc);
1341 if (ret)
1342 return ret;
1343
1344 down_read(&EXT4_I(inode)->xattr_sem);
1345 if (!ext4_has_inline_data(inode)) {
1346 up_read(&EXT4_I(inode)->xattr_sem);
1347 *has_inline_data = 0;
1348 goto out;
1349 }
1350
1351 inline_size = ext4_get_inline_size(inode);
1352 dir_buf = kmalloc(inline_size, GFP_NOFS);
1353 if (!dir_buf) {
1354 ret = -ENOMEM;
1355 up_read(&EXT4_I(inode)->xattr_sem);
1356 goto out;
1357 }
1358
1359 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1360 up_read(&EXT4_I(inode)->xattr_sem);
1361 if (ret < 0)
1362 goto out;
1363
1364 pos = 0;
1365 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1366 while (pos < inline_size) {
1367 /*
1368 * As inlined dir doesn't store any information about '.' and
1369 * only the inode number of '..' is stored, we have to handle
1370 * them differently.
1371 */
1372 if (pos == 0) {
1373 fake.inode = cpu_to_le32(inode->i_ino);
1374 fake.name_len = 1;
1375 strcpy(fake.name, ".");
1376 fake.rec_len = ext4_rec_len_to_disk(
1377 EXT4_DIR_REC_LEN(fake.name_len),
1378 inline_size);
1379 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1380 de = &fake;
1381 pos = EXT4_INLINE_DOTDOT_OFFSET;
1382 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1383 fake.inode = cpu_to_le32(parent_ino);
1384 fake.name_len = 2;
1385 strcpy(fake.name, "..");
1386 fake.rec_len = ext4_rec_len_to_disk(
1387 EXT4_DIR_REC_LEN(fake.name_len),
1388 inline_size);
1389 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1390 de = &fake;
1391 pos = EXT4_INLINE_DOTDOT_SIZE;
1392 } else {
1393 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1394 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1395 if (ext4_check_dir_entry(inode, dir_file, de,
1396 iloc.bh, dir_buf,
1397 inline_size, pos)) {
1398 ret = count;
1399 goto out;
1400 }
1401 }
1402
1403 ext4fs_dirhash(de->name, de->name_len, hinfo);
1404 if ((hinfo->hash < start_hash) ||
1405 ((hinfo->hash == start_hash) &&
1406 (hinfo->minor_hash < start_minor_hash)))
1407 continue;
1408 if (de->inode == 0)
1409 continue;
1410 tmp_str.name = de->name;
1411 tmp_str.len = de->name_len;
1412 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1413 hinfo->minor_hash, de, &tmp_str);
1414 if (err) {
1415 count = err;
1416 goto out;
1417 }
1418 count++;
1419 }
1420 ret = count;
1421 out:
1422 kfree(dir_buf);
1423 brelse(iloc.bh);
1424 return ret;
1425 }
1426
1427 /*
1428 * So this function is called when the volume is mkfsed with
1429 * dir_index disabled. In order to keep f_pos persistent
1430 * after we convert from an inlined dir to a blocked based,
1431 * we just pretend that we are a normal dir and return the
1432 * offset as if '.' and '..' really take place.
1433 *
1434 */
1435 int ext4_read_inline_dir(struct file *file,
1436 struct dir_context *ctx,
1437 int *has_inline_data)
1438 {
1439 unsigned int offset, parent_ino;
1440 int i;
1441 struct ext4_dir_entry_2 *de;
1442 struct super_block *sb;
1443 struct inode *inode = file_inode(file);
1444 int ret, inline_size = 0;
1445 struct ext4_iloc iloc;
1446 void *dir_buf = NULL;
1447 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1448
1449 ret = ext4_get_inode_loc(inode, &iloc);
1450 if (ret)
1451 return ret;
1452
1453 down_read(&EXT4_I(inode)->xattr_sem);
1454 if (!ext4_has_inline_data(inode)) {
1455 up_read(&EXT4_I(inode)->xattr_sem);
1456 *has_inline_data = 0;
1457 goto out;
1458 }
1459
1460 inline_size = ext4_get_inline_size(inode);
1461 dir_buf = kmalloc(inline_size, GFP_NOFS);
1462 if (!dir_buf) {
1463 ret = -ENOMEM;
1464 up_read(&EXT4_I(inode)->xattr_sem);
1465 goto out;
1466 }
1467
1468 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1469 up_read(&EXT4_I(inode)->xattr_sem);
1470 if (ret < 0)
1471 goto out;
1472
1473 ret = 0;
1474 sb = inode->i_sb;
1475 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1476 offset = ctx->pos;
1477
1478 /*
1479 * dotdot_offset and dotdot_size is the real offset and
1480 * size for ".." and "." if the dir is block based while
1481 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1482 * So we will use extra_offset and extra_size to indicate them
1483 * during the inline dir iteration.
1484 */
1485 dotdot_offset = EXT4_DIR_REC_LEN(1);
1486 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1487 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1488 extra_size = extra_offset + inline_size;
1489
1490 /*
1491 * If the version has changed since the last call to
1492 * readdir(2), then we might be pointing to an invalid
1493 * dirent right now. Scan from the start of the inline
1494 * dir to make sure.
1495 */
1496 if (file->f_version != inode->i_version) {
1497 for (i = 0; i < extra_size && i < offset;) {
1498 /*
1499 * "." is with offset 0 and
1500 * ".." is dotdot_offset.
1501 */
1502 if (!i) {
1503 i = dotdot_offset;
1504 continue;
1505 } else if (i == dotdot_offset) {
1506 i = dotdot_size;
1507 continue;
1508 }
1509 /* for other entry, the real offset in
1510 * the buf has to be tuned accordingly.
1511 */
1512 de = (struct ext4_dir_entry_2 *)
1513 (dir_buf + i - extra_offset);
1514 /* It's too expensive to do a full
1515 * dirent test each time round this
1516 * loop, but we do have to test at
1517 * least that it is non-zero. A
1518 * failure will be detected in the
1519 * dirent test below. */
1520 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1521 < EXT4_DIR_REC_LEN(1))
1522 break;
1523 i += ext4_rec_len_from_disk(de->rec_len,
1524 extra_size);
1525 }
1526 offset = i;
1527 ctx->pos = offset;
1528 file->f_version = inode->i_version;
1529 }
1530
1531 while (ctx->pos < extra_size) {
1532 if (ctx->pos == 0) {
1533 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1534 goto out;
1535 ctx->pos = dotdot_offset;
1536 continue;
1537 }
1538
1539 if (ctx->pos == dotdot_offset) {
1540 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1541 goto out;
1542 ctx->pos = dotdot_size;
1543 continue;
1544 }
1545
1546 de = (struct ext4_dir_entry_2 *)
1547 (dir_buf + ctx->pos - extra_offset);
1548 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1549 extra_size, ctx->pos))
1550 goto out;
1551 if (le32_to_cpu(de->inode)) {
1552 if (!dir_emit(ctx, de->name, de->name_len,
1553 le32_to_cpu(de->inode),
1554 get_dtype(sb, de->file_type)))
1555 goto out;
1556 }
1557 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1558 }
1559 out:
1560 kfree(dir_buf);
1561 brelse(iloc.bh);
1562 return ret;
1563 }
1564
1565 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1566 struct ext4_dir_entry_2 **parent_de,
1567 int *retval)
1568 {
1569 struct ext4_iloc iloc;
1570
1571 *retval = ext4_get_inode_loc(inode, &iloc);
1572 if (*retval)
1573 return NULL;
1574
1575 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1576
1577 return iloc.bh;
1578 }
1579
1580 /*
1581 * Try to create the inline data for the new dir.
1582 * If it succeeds, return 0, otherwise return the error.
1583 * In case of ENOSPC, the caller should create the normal disk layout dir.
1584 */
1585 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1586 struct inode *inode)
1587 {
1588 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1589 struct ext4_iloc iloc;
1590 struct ext4_dir_entry_2 *de;
1591
1592 ret = ext4_get_inode_loc(inode, &iloc);
1593 if (ret)
1594 return ret;
1595
1596 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1597 if (ret)
1598 goto out;
1599
1600 /*
1601 * For inline dir, we only save the inode information for the ".."
1602 * and create a fake dentry to cover the left space.
1603 */
1604 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1605 de->inode = cpu_to_le32(parent->i_ino);
1606 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1607 de->inode = 0;
1608 de->rec_len = ext4_rec_len_to_disk(
1609 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1610 inline_size);
1611 set_nlink(inode, 2);
1612 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1613 out:
1614 brelse(iloc.bh);
1615 return ret;
1616 }
1617
1618 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1619 struct ext4_filename *fname,
1620 const struct qstr *d_name,
1621 struct ext4_dir_entry_2 **res_dir,
1622 int *has_inline_data)
1623 {
1624 int ret;
1625 struct ext4_iloc iloc;
1626 void *inline_start;
1627 int inline_size;
1628
1629 if (ext4_get_inode_loc(dir, &iloc))
1630 return NULL;
1631
1632 down_read(&EXT4_I(dir)->xattr_sem);
1633 if (!ext4_has_inline_data(dir)) {
1634 *has_inline_data = 0;
1635 goto out;
1636 }
1637
1638 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1639 EXT4_INLINE_DOTDOT_SIZE;
1640 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1641 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1642 dir, fname, d_name, 0, res_dir);
1643 if (ret == 1)
1644 goto out_find;
1645 if (ret < 0)
1646 goto out;
1647
1648 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1649 goto out;
1650
1651 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1652 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1653
1654 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1655 dir, fname, d_name, 0, res_dir);
1656 if (ret == 1)
1657 goto out_find;
1658
1659 out:
1660 brelse(iloc.bh);
1661 iloc.bh = NULL;
1662 out_find:
1663 up_read(&EXT4_I(dir)->xattr_sem);
1664 return iloc.bh;
1665 }
1666
1667 int ext4_delete_inline_entry(handle_t *handle,
1668 struct inode *dir,
1669 struct ext4_dir_entry_2 *de_del,
1670 struct buffer_head *bh,
1671 int *has_inline_data)
1672 {
1673 int err, inline_size;
1674 struct ext4_iloc iloc;
1675 void *inline_start;
1676
1677 err = ext4_get_inode_loc(dir, &iloc);
1678 if (err)
1679 return err;
1680
1681 down_write(&EXT4_I(dir)->xattr_sem);
1682 if (!ext4_has_inline_data(dir)) {
1683 *has_inline_data = 0;
1684 goto out;
1685 }
1686
1687 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1688 EXT4_MIN_INLINE_DATA_SIZE) {
1689 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1690 EXT4_INLINE_DOTDOT_SIZE;
1691 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1692 EXT4_INLINE_DOTDOT_SIZE;
1693 } else {
1694 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1695 inline_size = ext4_get_inline_size(dir) -
1696 EXT4_MIN_INLINE_DATA_SIZE;
1697 }
1698
1699 BUFFER_TRACE(bh, "get_write_access");
1700 err = ext4_journal_get_write_access(handle, bh);
1701 if (err)
1702 goto out;
1703
1704 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1705 inline_start, inline_size, 0);
1706 if (err)
1707 goto out;
1708
1709 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1710 err = ext4_mark_inode_dirty(handle, dir);
1711 if (unlikely(err))
1712 goto out;
1713
1714 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1715 out:
1716 up_write(&EXT4_I(dir)->xattr_sem);
1717 brelse(iloc.bh);
1718 if (err != -ENOENT)
1719 ext4_std_error(dir->i_sb, err);
1720 return err;
1721 }
1722
1723 /*
1724 * Get the inline dentry at offset.
1725 */
1726 static inline struct ext4_dir_entry_2 *
1727 ext4_get_inline_entry(struct inode *inode,
1728 struct ext4_iloc *iloc,
1729 unsigned int offset,
1730 void **inline_start,
1731 int *inline_size)
1732 {
1733 void *inline_pos;
1734
1735 BUG_ON(offset > ext4_get_inline_size(inode));
1736
1737 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1738 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1739 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1740 } else {
1741 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1742 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1743 *inline_size = ext4_get_inline_size(inode) -
1744 EXT4_MIN_INLINE_DATA_SIZE;
1745 }
1746
1747 if (inline_start)
1748 *inline_start = inline_pos;
1749 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1750 }
1751
1752 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1753 {
1754 int err, inline_size;
1755 struct ext4_iloc iloc;
1756 void *inline_pos;
1757 unsigned int offset;
1758 struct ext4_dir_entry_2 *de;
1759 int ret = 1;
1760
1761 err = ext4_get_inode_loc(dir, &iloc);
1762 if (err) {
1763 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1764 err, dir->i_ino);
1765 return 1;
1766 }
1767
1768 down_read(&EXT4_I(dir)->xattr_sem);
1769 if (!ext4_has_inline_data(dir)) {
1770 *has_inline_data = 0;
1771 goto out;
1772 }
1773
1774 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1775 if (!le32_to_cpu(de->inode)) {
1776 ext4_warning(dir->i_sb,
1777 "bad inline directory (dir #%lu) - no `..'",
1778 dir->i_ino);
1779 ret = 1;
1780 goto out;
1781 }
1782
1783 offset = EXT4_INLINE_DOTDOT_SIZE;
1784 while (offset < dir->i_size) {
1785 de = ext4_get_inline_entry(dir, &iloc, offset,
1786 &inline_pos, &inline_size);
1787 if (ext4_check_dir_entry(dir, NULL, de,
1788 iloc.bh, inline_pos,
1789 inline_size, offset)) {
1790 ext4_warning(dir->i_sb,
1791 "bad inline directory (dir #%lu) - "
1792 "inode %u, rec_len %u, name_len %d"
1793 "inline size %d\n",
1794 dir->i_ino, le32_to_cpu(de->inode),
1795 le16_to_cpu(de->rec_len), de->name_len,
1796 inline_size);
1797 ret = 1;
1798 goto out;
1799 }
1800 if (le32_to_cpu(de->inode)) {
1801 ret = 0;
1802 goto out;
1803 }
1804 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1805 }
1806
1807 out:
1808 up_read(&EXT4_I(dir)->xattr_sem);
1809 brelse(iloc.bh);
1810 return ret;
1811 }
1812
1813 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1814 {
1815 int ret;
1816
1817 down_write(&EXT4_I(inode)->xattr_sem);
1818 ret = ext4_destroy_inline_data_nolock(handle, inode);
1819 up_write(&EXT4_I(inode)->xattr_sem);
1820
1821 return ret;
1822 }
1823
1824 int ext4_inline_data_fiemap(struct inode *inode,
1825 struct fiemap_extent_info *fieinfo,
1826 int *has_inline, __u64 start, __u64 len)
1827 {
1828 __u64 physical = 0;
1829 __u64 inline_len;
1830 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1831 FIEMAP_EXTENT_LAST;
1832 int error = 0;
1833 struct ext4_iloc iloc;
1834
1835 down_read(&EXT4_I(inode)->xattr_sem);
1836 if (!ext4_has_inline_data(inode)) {
1837 *has_inline = 0;
1838 goto out;
1839 }
1840 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1841 i_size_read(inode));
1842 if (start >= inline_len)
1843 goto out;
1844 if (start + len < inline_len)
1845 inline_len = start + len;
1846 inline_len -= start;
1847
1848 error = ext4_get_inode_loc(inode, &iloc);
1849 if (error)
1850 goto out;
1851
1852 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1853 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1854 physical += offsetof(struct ext4_inode, i_block);
1855
1856 if (physical)
1857 error = fiemap_fill_next_extent(fieinfo, start, physical,
1858 inline_len, flags);
1859 brelse(iloc.bh);
1860 out:
1861 up_read(&EXT4_I(inode)->xattr_sem);
1862 return (error < 0 ? error : 0);
1863 }
1864
1865 /*
1866 * Called during xattr set, and if we can sparse space 'needed',
1867 * just create the extent tree evict the data to the outer block.
1868 *
1869 * We use jbd2 instead of page cache to move data to the 1st block
1870 * so that the whole transaction can be committed as a whole and
1871 * the data isn't lost because of the delayed page cache write.
1872 */
1873 int ext4_try_to_evict_inline_data(handle_t *handle,
1874 struct inode *inode,
1875 int needed)
1876 {
1877 int error;
1878 struct ext4_xattr_entry *entry;
1879 struct ext4_inode *raw_inode;
1880 struct ext4_iloc iloc;
1881
1882 error = ext4_get_inode_loc(inode, &iloc);
1883 if (error)
1884 return error;
1885
1886 raw_inode = ext4_raw_inode(&iloc);
1887 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1888 EXT4_I(inode)->i_inline_off);
1889 if (EXT4_XATTR_LEN(entry->e_name_len) +
1890 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1891 error = -ENOSPC;
1892 goto out;
1893 }
1894
1895 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1896 out:
1897 brelse(iloc.bh);
1898 return error;
1899 }
1900
1901 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1902 {
1903 handle_t *handle;
1904 int inline_size, value_len, needed_blocks;
1905 size_t i_size;
1906 void *value = NULL;
1907 struct ext4_xattr_ibody_find is = {
1908 .s = { .not_found = -ENODATA, },
1909 };
1910 struct ext4_xattr_info i = {
1911 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1912 .name = EXT4_XATTR_SYSTEM_DATA,
1913 };
1914
1915
1916 needed_blocks = ext4_writepage_trans_blocks(inode);
1917 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1918 if (IS_ERR(handle))
1919 return;
1920
1921 down_write(&EXT4_I(inode)->xattr_sem);
1922 if (!ext4_has_inline_data(inode)) {
1923 *has_inline = 0;
1924 ext4_journal_stop(handle);
1925 return;
1926 }
1927
1928 if (ext4_orphan_add(handle, inode))
1929 goto out;
1930
1931 if (ext4_get_inode_loc(inode, &is.iloc))
1932 goto out;
1933
1934 down_write(&EXT4_I(inode)->i_data_sem);
1935 i_size = inode->i_size;
1936 inline_size = ext4_get_inline_size(inode);
1937 EXT4_I(inode)->i_disksize = i_size;
1938
1939 if (i_size < inline_size) {
1940 /* Clear the content in the xattr space. */
1941 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1942 if (ext4_xattr_ibody_find(inode, &i, &is))
1943 goto out_error;
1944
1945 BUG_ON(is.s.not_found);
1946
1947 value_len = le32_to_cpu(is.s.here->e_value_size);
1948 value = kmalloc(value_len, GFP_NOFS);
1949 if (!value)
1950 goto out_error;
1951
1952 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1953 value, value_len))
1954 goto out_error;
1955
1956 i.value = value;
1957 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1958 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1959 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1960 goto out_error;
1961 }
1962
1963 /* Clear the content within i_blocks. */
1964 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1965 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1966 memset(p + i_size, 0,
1967 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1968 }
1969
1970 EXT4_I(inode)->i_inline_size = i_size <
1971 EXT4_MIN_INLINE_DATA_SIZE ?
1972 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1973 }
1974
1975 out_error:
1976 up_write(&EXT4_I(inode)->i_data_sem);
1977 out:
1978 brelse(is.iloc.bh);
1979 up_write(&EXT4_I(inode)->xattr_sem);
1980 kfree(value);
1981 if (inode->i_nlink)
1982 ext4_orphan_del(handle, inode);
1983
1984 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1985 ext4_mark_inode_dirty(handle, inode);
1986 if (IS_SYNC(inode))
1987 ext4_handle_sync(handle);
1988
1989 ext4_journal_stop(handle);
1990 return;
1991 }
1992
1993 int ext4_convert_inline_data(struct inode *inode)
1994 {
1995 int error, needed_blocks;
1996 handle_t *handle;
1997 struct ext4_iloc iloc;
1998
1999 if (!ext4_has_inline_data(inode)) {
2000 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2001 return 0;
2002 }
2003
2004 needed_blocks = ext4_writepage_trans_blocks(inode);
2005
2006 iloc.bh = NULL;
2007 error = ext4_get_inode_loc(inode, &iloc);
2008 if (error)
2009 return error;
2010
2011 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2012 if (IS_ERR(handle)) {
2013 error = PTR_ERR(handle);
2014 goto out_free;
2015 }
2016
2017 down_write(&EXT4_I(inode)->xattr_sem);
2018 if (!ext4_has_inline_data(inode)) {
2019 up_write(&EXT4_I(inode)->xattr_sem);
2020 goto out;
2021 }
2022
2023 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2024 up_write(&EXT4_I(inode)->xattr_sem);
2025 out:
2026 ext4_journal_stop(handle);
2027 out_free:
2028 brelse(iloc.bh);
2029 return error;
2030 }
Linux with added FPC-III support