]> git.ipfire.org Git - people/ms/linux.git/blob - fs/ext4/inline.c
projects / people / ms / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
of/fdt: Add mutex protection for calls to __unflatten_device_tree()
[people/ms/linux.git] / fs / ext4 / inline.c
1 /*
2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 */
14
15 #include <linux/fiemap.h>
16
17 #include "ext4_jbd2.h"
18 #include "ext4.h"
19 #include "xattr.h"
20 #include "truncate.h"
21
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 goto out;
341
342 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
343 value, len);
344 if (error == -ENODATA)
345 goto out;
346
347 BUFFER_TRACE(is.iloc.bh, "get_write_access");
348 error = ext4_journal_get_write_access(handle, is.iloc.bh);
349 if (error)
350 goto out;
351
352 /* Update the xttr entry. */
353 i.value = value;
354 i.value_len = len;
355
356 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
357 if (error)
358 goto out;
359
360 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
361 (void *)ext4_raw_inode(&is.iloc));
362 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
363 le32_to_cpu(is.s.here->e_value_size);
364 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
365 get_bh(is.iloc.bh);
366 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
367
368 out:
369 kfree(value);
370 brelse(is.iloc.bh);
371 return error;
372 }
373
374 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
375 unsigned int len)
376 {
377 int ret, size;
378 struct ext4_inode_info *ei = EXT4_I(inode);
379
380 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
381 return -ENOSPC;
382
383 size = ext4_get_max_inline_size(inode);
384 if (size < len)
385 return -ENOSPC;
386
387 down_write(&EXT4_I(inode)->xattr_sem);
388
389 if (ei->i_inline_off)
390 ret = ext4_update_inline_data(handle, inode, len);
391 else
392 ret = ext4_create_inline_data(handle, inode, len);
393
394 up_write(&EXT4_I(inode)->xattr_sem);
395
396 return ret;
397 }
398
399 static int ext4_destroy_inline_data_nolock(handle_t *handle,
400 struct inode *inode)
401 {
402 struct ext4_inode_info *ei = EXT4_I(inode);
403 struct ext4_xattr_ibody_find is = {
404 .s = { .not_found = 0, },
405 };
406 struct ext4_xattr_info i = {
407 .name_index = EXT4_XATTR_INDEX_SYSTEM,
408 .name = EXT4_XATTR_SYSTEM_DATA,
409 .value = NULL,
410 .value_len = 0,
411 };
412 int error;
413
414 if (!ei->i_inline_off)
415 return 0;
416
417 error = ext4_get_inode_loc(inode, &is.iloc);
418 if (error)
419 return error;
420
421 error = ext4_xattr_ibody_find(inode, &i, &is);
422 if (error)
423 goto out;
424
425 BUFFER_TRACE(is.iloc.bh, "get_write_access");
426 error = ext4_journal_get_write_access(handle, is.iloc.bh);
427 if (error)
428 goto out;
429
430 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
431 if (error)
432 goto out;
433
434 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
435 0, EXT4_MIN_INLINE_DATA_SIZE);
436
437 if (ext4_has_feature_extents(inode->i_sb)) {
438 if (S_ISDIR(inode->i_mode) ||
439 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
440 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
441 ext4_ext_tree_init(handle, inode);
442 }
443 }
444 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
445
446 get_bh(is.iloc.bh);
447 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
448
449 EXT4_I(inode)->i_inline_off = 0;
450 EXT4_I(inode)->i_inline_size = 0;
451 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
452 out:
453 brelse(is.iloc.bh);
454 if (error == -ENODATA)
455 error = 0;
456 return error;
457 }
458
459 static int ext4_read_inline_page(struct inode *inode, struct page *page)
460 {
461 void *kaddr;
462 int ret = 0;
463 size_t len;
464 struct ext4_iloc iloc;
465
466 BUG_ON(!PageLocked(page));
467 BUG_ON(!ext4_has_inline_data(inode));
468 BUG_ON(page->index);
469
470 if (!EXT4_I(inode)->i_inline_off) {
471 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
472 inode->i_ino);
473 goto out;
474 }
475
476 ret = ext4_get_inode_loc(inode, &iloc);
477 if (ret)
478 goto out;
479
480 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
481 kaddr = kmap_atomic(page);
482 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
483 flush_dcache_page(page);
484 kunmap_atomic(kaddr);
485 zero_user_segment(page, len, PAGE_CACHE_SIZE);
486 SetPageUptodate(page);
487 brelse(iloc.bh);
488
489 out:
490 return ret;
491 }
492
493 int ext4_readpage_inline(struct inode *inode, struct page *page)
494 {
495 int ret = 0;
496
497 down_read(&EXT4_I(inode)->xattr_sem);
498 if (!ext4_has_inline_data(inode)) {
499 up_read(&EXT4_I(inode)->xattr_sem);
500 return -EAGAIN;
501 }
502
503 /*
504 * Current inline data can only exist in the 1st page,
505 * So for all the other pages, just set them uptodate.
506 */
507 if (!page->index)
508 ret = ext4_read_inline_page(inode, page);
509 else if (!PageUptodate(page)) {
510 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
511 SetPageUptodate(page);
512 }
513
514 up_read(&EXT4_I(inode)->xattr_sem);
515
516 unlock_page(page);
517 return ret >= 0 ? 0 : ret;
518 }
519
520 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
521 struct inode *inode,
522 unsigned flags)
523 {
524 int ret, needed_blocks;
525 handle_t *handle = NULL;
526 int retries = 0, sem_held = 0;
527 struct page *page = NULL;
528 unsigned from, to;
529 struct ext4_iloc iloc;
530
531 if (!ext4_has_inline_data(inode)) {
532 /*
533 * clear the flag so that no new write
534 * will trap here again.
535 */
536 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
537 return 0;
538 }
539
540 needed_blocks = ext4_writepage_trans_blocks(inode);
541
542 ret = ext4_get_inode_loc(inode, &iloc);
543 if (ret)
544 return ret;
545
546 retry:
547 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
548 if (IS_ERR(handle)) {
549 ret = PTR_ERR(handle);
550 handle = NULL;
551 goto out;
552 }
553
554 /* We cannot recurse into the filesystem as the transaction is already
555 * started */
556 flags |= AOP_FLAG_NOFS;
557
558 page = grab_cache_page_write_begin(mapping, 0, flags);
559 if (!page) {
560 ret = -ENOMEM;
561 goto out;
562 }
563
564 down_write(&EXT4_I(inode)->xattr_sem);
565 sem_held = 1;
566 /* If some one has already done this for us, just exit. */
567 if (!ext4_has_inline_data(inode)) {
568 ret = 0;
569 goto out;
570 }
571
572 from = 0;
573 to = ext4_get_inline_size(inode);
574 if (!PageUptodate(page)) {
575 ret = ext4_read_inline_page(inode, page);
576 if (ret < 0)
577 goto out;
578 }
579
580 ret = ext4_destroy_inline_data_nolock(handle, inode);
581 if (ret)
582 goto out;
583
584 if (ext4_should_dioread_nolock(inode))
585 ret = __block_write_begin(page, from, to, ext4_get_block_write);
586 else
587 ret = __block_write_begin(page, from, to, ext4_get_block);
588
589 if (!ret && ext4_should_journal_data(inode)) {
590 ret = ext4_walk_page_buffers(handle, page_buffers(page),
591 from, to, NULL,
592 do_journal_get_write_access);
593 }
594
595 if (ret) {
596 unlock_page(page);
597 page_cache_release(page);
598 page = NULL;
599 ext4_orphan_add(handle, inode);
600 up_write(&EXT4_I(inode)->xattr_sem);
601 sem_held = 0;
602 ext4_journal_stop(handle);
603 handle = NULL;
604 ext4_truncate_failed_write(inode);
605 /*
606 * If truncate failed early the inode might
607 * still be on the orphan list; we need to
608 * make sure the inode is removed from the
609 * orphan list in that case.
610 */
611 if (inode->i_nlink)
612 ext4_orphan_del(NULL, inode);
613 }
614
615 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
616 goto retry;
617
618 if (page)
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 up_read(&EXT4_I(inode)->xattr_sem);
816 unlock_page(page);
817 page_cache_release(page);
818 ext4_truncate_failed_write(inode);
819 return ret;
820 }
821
822 SetPageDirty(page);
823 SetPageUptodate(page);
824 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
825 *fsdata = (void *)CONVERT_INLINE_DATA;
826
827 out:
828 up_read(&EXT4_I(inode)->xattr_sem);
829 if (page) {
830 unlock_page(page);
831 page_cache_release(page);
832 }
833 return ret;
834 }
835
836 /*
837 * Prepare the write for the inline data.
838 * If the the data can be written into the inode, we just read
839 * the page and make it uptodate, and start the journal.
840 * Otherwise read the page, makes it dirty so that it can be
841 * handle in writepages(the i_disksize update is left to the
842 * normal ext4_da_write_end).
843 */
844 int ext4_da_write_inline_data_begin(struct address_space *mapping,
845 struct inode *inode,
846 loff_t pos, unsigned len,
847 unsigned flags,
848 struct page **pagep,
849 void **fsdata)
850 {
851 int ret, inline_size;
852 handle_t *handle;
853 struct page *page;
854 struct ext4_iloc iloc;
855 int retries;
856
857 ret = ext4_get_inode_loc(inode, &iloc);
858 if (ret)
859 return ret;
860
861 retry_journal:
862 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
863 if (IS_ERR(handle)) {
864 ret = PTR_ERR(handle);
865 goto out;
866 }
867
868 inline_size = ext4_get_max_inline_size(inode);
869
870 ret = -ENOSPC;
871 if (inline_size >= pos + len) {
872 ret = ext4_prepare_inline_data(handle, inode, pos + len);
873 if (ret && ret != -ENOSPC)
874 goto out_journal;
875 }
876
877 /*
878 * We cannot recurse into the filesystem as the transaction
879 * is already started.
880 */
881 flags |= AOP_FLAG_NOFS;
882
883 if (ret == -ENOSPC) {
884 ret = ext4_da_convert_inline_data_to_extent(mapping,
885 inode,
886 flags,
887 fsdata);
888 ext4_journal_stop(handle);
889 if (ret == -ENOSPC &&
890 ext4_should_retry_alloc(inode->i_sb, &retries))
891 goto retry_journal;
892 goto out;
893 }
894
895
896 page = grab_cache_page_write_begin(mapping, 0, flags);
897 if (!page) {
898 ret = -ENOMEM;
899 goto out_journal;
900 }
901
902 down_read(&EXT4_I(inode)->xattr_sem);
903 if (!ext4_has_inline_data(inode)) {
904 ret = 0;
905 goto out_release_page;
906 }
907
908 if (!PageUptodate(page)) {
909 ret = ext4_read_inline_page(inode, page);
910 if (ret < 0)
911 goto out_release_page;
912 }
913
914 up_read(&EXT4_I(inode)->xattr_sem);
915 *pagep = page;
916 brelse(iloc.bh);
917 return 1;
918 out_release_page:
919 up_read(&EXT4_I(inode)->xattr_sem);
920 unlock_page(page);
921 page_cache_release(page);
922 out_journal:
923 ext4_journal_stop(handle);
924 out:
925 brelse(iloc.bh);
926 return ret;
927 }
928
929 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
930 unsigned len, unsigned copied,
931 struct page *page)
932 {
933 int i_size_changed = 0;
934
935 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
936
937 /*
938 * No need to use i_size_read() here, the i_size
939 * cannot change under us because we hold i_mutex.
940 *
941 * But it's important to update i_size while still holding page lock:
942 * page writeout could otherwise come in and zero beyond i_size.
943 */
944 if (pos+copied > inode->i_size) {
945 i_size_write(inode, pos+copied);
946 i_size_changed = 1;
947 }
948 unlock_page(page);
949 page_cache_release(page);
950
951 /*
952 * Don't mark the inode dirty under page lock. First, it unnecessarily
953 * makes the holding time of page lock longer. Second, it forces lock
954 * ordering of page lock and transaction start for journaling
955 * filesystems.
956 */
957 if (i_size_changed)
958 mark_inode_dirty(inode);
959
960 return copied;
961 }
962
963 #ifdef INLINE_DIR_DEBUG
964 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
965 void *inline_start, int inline_size)
966 {
967 int offset;
968 unsigned short de_len;
969 struct ext4_dir_entry_2 *de = inline_start;
970 void *dlimit = inline_start + inline_size;
971
972 trace_printk("inode %lu\n", dir->i_ino);
973 offset = 0;
974 while ((void *)de < dlimit) {
975 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
976 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
977 offset, de_len, de->name_len, de->name,
978 de->name_len, le32_to_cpu(de->inode));
979 if (ext4_check_dir_entry(dir, NULL, de, bh,
980 inline_start, inline_size, offset))
981 BUG();
982
983 offset += de_len;
984 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
985 }
986 }
987 #else
988 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
989 #endif
990
991 /*
992 * Add a new entry into a inline dir.
993 * It will return -ENOSPC if no space is available, and -EIO
994 * and -EEXIST if directory entry already exists.
995 */
996 static int ext4_add_dirent_to_inline(handle_t *handle,
997 struct ext4_filename *fname,
998 struct dentry *dentry,
999 struct inode *inode,
1000 struct ext4_iloc *iloc,
1001 void *inline_start, int inline_size)
1002 {
1003 struct inode *dir = d_inode(dentry->d_parent);
1004 int err;
1005 struct ext4_dir_entry_2 *de;
1006
1007 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1008 inline_size, fname, &de);
1009 if (err)
1010 return err;
1011
1012 BUFFER_TRACE(iloc->bh, "get_write_access");
1013 err = ext4_journal_get_write_access(handle, iloc->bh);
1014 if (err)
1015 return err;
1016 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1017
1018 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1019
1020 /*
1021 * XXX shouldn't update any times until successful
1022 * completion of syscall, but too many callers depend
1023 * on this.
1024 *
1025 * XXX similarly, too many callers depend on
1026 * ext4_new_inode() setting the times, but error
1027 * recovery deletes the inode, so the worst that can
1028 * happen is that the times are slightly out of date
1029 * and/or different from the directory change time.
1030 */
1031 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1032 ext4_update_dx_flag(dir);
1033 dir->i_version++;
1034 ext4_mark_inode_dirty(handle, dir);
1035 return 1;
1036 }
1037
1038 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1039 struct ext4_iloc *iloc)
1040 {
1041 struct ext4_xattr_entry *entry;
1042 struct ext4_xattr_ibody_header *header;
1043
1044 BUG_ON(!EXT4_I(inode)->i_inline_off);
1045
1046 header = IHDR(inode, ext4_raw_inode(iloc));
1047 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1048 EXT4_I(inode)->i_inline_off);
1049
1050 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1051 }
1052
1053 /* Set the final de to cover the whole block. */
1054 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1055 {
1056 struct ext4_dir_entry_2 *de, *prev_de;
1057 void *limit;
1058 int de_len;
1059
1060 de = (struct ext4_dir_entry_2 *)de_buf;
1061 if (old_size) {
1062 limit = de_buf + old_size;
1063 do {
1064 prev_de = de;
1065 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1066 de_buf += de_len;
1067 de = (struct ext4_dir_entry_2 *)de_buf;
1068 } while (de_buf < limit);
1069
1070 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1071 old_size, new_size);
1072 } else {
1073 /* this is just created, so create an empty entry. */
1074 de->inode = 0;
1075 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1076 }
1077 }
1078
1079 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1080 struct ext4_iloc *iloc)
1081 {
1082 int ret;
1083 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1084 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1085
1086 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1087 return -ENOSPC;
1088
1089 ret = ext4_update_inline_data(handle, dir,
1090 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1091 if (ret)
1092 return ret;
1093
1094 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1095 EXT4_I(dir)->i_inline_size -
1096 EXT4_MIN_INLINE_DATA_SIZE);
1097 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1098 return 0;
1099 }
1100
1101 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1102 struct ext4_iloc *iloc,
1103 void *buf, int inline_size)
1104 {
1105 ext4_create_inline_data(handle, inode, inline_size);
1106 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1107 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1108 }
1109
1110 static int ext4_finish_convert_inline_dir(handle_t *handle,
1111 struct inode *inode,
1112 struct buffer_head *dir_block,
1113 void *buf,
1114 int inline_size)
1115 {
1116 int err, csum_size = 0, header_size = 0;
1117 struct ext4_dir_entry_2 *de;
1118 struct ext4_dir_entry_tail *t;
1119 void *target = dir_block->b_data;
1120
1121 /*
1122 * First create "." and ".." and then copy the dir information
1123 * back to the block.
1124 */
1125 de = (struct ext4_dir_entry_2 *)target;
1126 de = ext4_init_dot_dotdot(inode, de,
1127 inode->i_sb->s_blocksize, csum_size,
1128 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1129 header_size = (void *)de - target;
1130
1131 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1132 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1133
1134 if (ext4_has_metadata_csum(inode->i_sb))
1135 csum_size = sizeof(struct ext4_dir_entry_tail);
1136
1137 inode->i_size = inode->i_sb->s_blocksize;
1138 i_size_write(inode, inode->i_sb->s_blocksize);
1139 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1140 ext4_update_final_de(dir_block->b_data,
1141 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1142 inode->i_sb->s_blocksize - csum_size);
1143
1144 if (csum_size) {
1145 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1146 inode->i_sb->s_blocksize);
1147 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1148 }
1149 set_buffer_uptodate(dir_block);
1150 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1151 if (err)
1152 goto out;
1153 set_buffer_verified(dir_block);
1154 out:
1155 return err;
1156 }
1157
1158 static int ext4_convert_inline_data_nolock(handle_t *handle,
1159 struct inode *inode,
1160 struct ext4_iloc *iloc)
1161 {
1162 int error;
1163 void *buf = NULL;
1164 struct buffer_head *data_bh = NULL;
1165 struct ext4_map_blocks map;
1166 int inline_size;
1167
1168 inline_size = ext4_get_inline_size(inode);
1169 buf = kmalloc(inline_size, GFP_NOFS);
1170 if (!buf) {
1171 error = -ENOMEM;
1172 goto out;
1173 }
1174
1175 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1176 if (error < 0)
1177 goto out;
1178
1179 /*
1180 * Make sure the inline directory entries pass checks before we try to
1181 * convert them, so that we avoid touching stuff that needs fsck.
1182 */
1183 if (S_ISDIR(inode->i_mode)) {
1184 error = ext4_check_all_de(inode, iloc->bh,
1185 buf + EXT4_INLINE_DOTDOT_SIZE,
1186 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1187 if (error)
1188 goto out;
1189 }
1190
1191 error = ext4_destroy_inline_data_nolock(handle, inode);
1192 if (error)
1193 goto out;
1194
1195 map.m_lblk = 0;
1196 map.m_len = 1;
1197 map.m_flags = 0;
1198 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1199 if (error < 0)
1200 goto out_restore;
1201 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1202 error = -EIO;
1203 goto out_restore;
1204 }
1205
1206 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1207 if (!data_bh) {
1208 error = -ENOMEM;
1209 goto out_restore;
1210 }
1211
1212 lock_buffer(data_bh);
1213 error = ext4_journal_get_create_access(handle, data_bh);
1214 if (error) {
1215 unlock_buffer(data_bh);
1216 error = -EIO;
1217 goto out_restore;
1218 }
1219 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1220
1221 if (!S_ISDIR(inode->i_mode)) {
1222 memcpy(data_bh->b_data, buf, inline_size);
1223 set_buffer_uptodate(data_bh);
1224 error = ext4_handle_dirty_metadata(handle,
1225 inode, data_bh);
1226 } else {
1227 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1228 buf, inline_size);
1229 }
1230
1231 unlock_buffer(data_bh);
1232 out_restore:
1233 if (error)
1234 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1235
1236 out:
1237 brelse(data_bh);
1238 kfree(buf);
1239 return error;
1240 }
1241
1242 /*
1243 * Try to add the new entry to the inline data.
1244 * If succeeds, return 0. If not, extended the inline dir and copied data to
1245 * the new created block.
1246 */
1247 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1248 struct dentry *dentry, struct inode *inode)
1249 {
1250 int ret, inline_size;
1251 void *inline_start;
1252 struct ext4_iloc iloc;
1253 struct inode *dir = d_inode(dentry->d_parent);
1254
1255 ret = ext4_get_inode_loc(dir, &iloc);
1256 if (ret)
1257 return ret;
1258
1259 down_write(&EXT4_I(dir)->xattr_sem);
1260 if (!ext4_has_inline_data(dir))
1261 goto out;
1262
1263 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1264 EXT4_INLINE_DOTDOT_SIZE;
1265 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1266
1267 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1268 inline_start, inline_size);
1269 if (ret != -ENOSPC)
1270 goto out;
1271
1272 /* check whether it can be inserted to inline xattr space. */
1273 inline_size = EXT4_I(dir)->i_inline_size -
1274 EXT4_MIN_INLINE_DATA_SIZE;
1275 if (!inline_size) {
1276 /* Try to use the xattr space.*/
1277 ret = ext4_update_inline_dir(handle, dir, &iloc);
1278 if (ret && ret != -ENOSPC)
1279 goto out;
1280
1281 inline_size = EXT4_I(dir)->i_inline_size -
1282 EXT4_MIN_INLINE_DATA_SIZE;
1283 }
1284
1285 if (inline_size) {
1286 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1287
1288 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1289 inode, &iloc, inline_start,
1290 inline_size);
1291
1292 if (ret != -ENOSPC)
1293 goto out;
1294 }
1295
1296 /*
1297 * The inline space is filled up, so create a new block for it.
1298 * As the extent tree will be created, we have to save the inline
1299 * dir first.
1300 */
1301 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1302
1303 out:
1304 ext4_mark_inode_dirty(handle, dir);
1305 up_write(&EXT4_I(dir)->xattr_sem);
1306 brelse(iloc.bh);
1307 return ret;
1308 }
1309
1310 /*
1311 * This function fills a red-black tree with information from an
1312 * inlined dir. It returns the number directory entries loaded
1313 * into the tree. If there is an error it is returned in err.
1314 */
1315 int htree_inlinedir_to_tree(struct file *dir_file,
1316 struct inode *dir, ext4_lblk_t block,
1317 struct dx_hash_info *hinfo,
1318 __u32 start_hash, __u32 start_minor_hash,
1319 int *has_inline_data)
1320 {
1321 int err = 0, count = 0;
1322 unsigned int parent_ino;
1323 int pos;
1324 struct ext4_dir_entry_2 *de;
1325 struct inode *inode = file_inode(dir_file);
1326 int ret, inline_size = 0;
1327 struct ext4_iloc iloc;
1328 void *dir_buf = NULL;
1329 struct ext4_dir_entry_2 fake;
1330 struct ext4_str tmp_str;
1331
1332 ret = ext4_get_inode_loc(inode, &iloc);
1333 if (ret)
1334 return ret;
1335
1336 down_read(&EXT4_I(inode)->xattr_sem);
1337 if (!ext4_has_inline_data(inode)) {
1338 up_read(&EXT4_I(inode)->xattr_sem);
1339 *has_inline_data = 0;
1340 goto out;
1341 }
1342
1343 inline_size = ext4_get_inline_size(inode);
1344 dir_buf = kmalloc(inline_size, GFP_NOFS);
1345 if (!dir_buf) {
1346 ret = -ENOMEM;
1347 up_read(&EXT4_I(inode)->xattr_sem);
1348 goto out;
1349 }
1350
1351 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1352 up_read(&EXT4_I(inode)->xattr_sem);
1353 if (ret < 0)
1354 goto out;
1355
1356 pos = 0;
1357 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1358 while (pos < inline_size) {
1359 /*
1360 * As inlined dir doesn't store any information about '.' and
1361 * only the inode number of '..' is stored, we have to handle
1362 * them differently.
1363 */
1364 if (pos == 0) {
1365 fake.inode = cpu_to_le32(inode->i_ino);
1366 fake.name_len = 1;
1367 strcpy(fake.name, ".");
1368 fake.rec_len = ext4_rec_len_to_disk(
1369 EXT4_DIR_REC_LEN(fake.name_len),
1370 inline_size);
1371 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1372 de = &fake;
1373 pos = EXT4_INLINE_DOTDOT_OFFSET;
1374 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1375 fake.inode = cpu_to_le32(parent_ino);
1376 fake.name_len = 2;
1377 strcpy(fake.name, "..");
1378 fake.rec_len = ext4_rec_len_to_disk(
1379 EXT4_DIR_REC_LEN(fake.name_len),
1380 inline_size);
1381 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1382 de = &fake;
1383 pos = EXT4_INLINE_DOTDOT_SIZE;
1384 } else {
1385 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1386 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1387 if (ext4_check_dir_entry(inode, dir_file, de,
1388 iloc.bh, dir_buf,
1389 inline_size, pos)) {
1390 ret = count;
1391 goto out;
1392 }
1393 }
1394
1395 ext4fs_dirhash(de->name, de->name_len, hinfo);
1396 if ((hinfo->hash < start_hash) ||
1397 ((hinfo->hash == start_hash) &&
1398 (hinfo->minor_hash < start_minor_hash)))
1399 continue;
1400 if (de->inode == 0)
1401 continue;
1402 tmp_str.name = de->name;
1403 tmp_str.len = de->name_len;
1404 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1405 hinfo->minor_hash, de, &tmp_str);
1406 if (err) {
1407 count = err;
1408 goto out;
1409 }
1410 count++;
1411 }
1412 ret = count;
1413 out:
1414 kfree(dir_buf);
1415 brelse(iloc.bh);
1416 return ret;
1417 }
1418
1419 /*
1420 * So this function is called when the volume is mkfsed with
1421 * dir_index disabled. In order to keep f_pos persistent
1422 * after we convert from an inlined dir to a blocked based,
1423 * we just pretend that we are a normal dir and return the
1424 * offset as if '.' and '..' really take place.
1425 *
1426 */
1427 int ext4_read_inline_dir(struct file *file,
1428 struct dir_context *ctx,
1429 int *has_inline_data)
1430 {
1431 unsigned int offset, parent_ino;
1432 int i;
1433 struct ext4_dir_entry_2 *de;
1434 struct super_block *sb;
1435 struct inode *inode = file_inode(file);
1436 int ret, inline_size = 0;
1437 struct ext4_iloc iloc;
1438 void *dir_buf = NULL;
1439 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1440
1441 ret = ext4_get_inode_loc(inode, &iloc);
1442 if (ret)
1443 return ret;
1444
1445 down_read(&EXT4_I(inode)->xattr_sem);
1446 if (!ext4_has_inline_data(inode)) {
1447 up_read(&EXT4_I(inode)->xattr_sem);
1448 *has_inline_data = 0;
1449 goto out;
1450 }
1451
1452 inline_size = ext4_get_inline_size(inode);
1453 dir_buf = kmalloc(inline_size, GFP_NOFS);
1454 if (!dir_buf) {
1455 ret = -ENOMEM;
1456 up_read(&EXT4_I(inode)->xattr_sem);
1457 goto out;
1458 }
1459
1460 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1461 up_read(&EXT4_I(inode)->xattr_sem);
1462 if (ret < 0)
1463 goto out;
1464
1465 ret = 0;
1466 sb = inode->i_sb;
1467 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1468 offset = ctx->pos;
1469
1470 /*
1471 * dotdot_offset and dotdot_size is the real offset and
1472 * size for ".." and "." if the dir is block based while
1473 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1474 * So we will use extra_offset and extra_size to indicate them
1475 * during the inline dir iteration.
1476 */
1477 dotdot_offset = EXT4_DIR_REC_LEN(1);
1478 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1479 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1480 extra_size = extra_offset + inline_size;
1481
1482 /*
1483 * If the version has changed since the last call to
1484 * readdir(2), then we might be pointing to an invalid
1485 * dirent right now. Scan from the start of the inline
1486 * dir to make sure.
1487 */
1488 if (file->f_version != inode->i_version) {
1489 for (i = 0; i < extra_size && i < offset;) {
1490 /*
1491 * "." is with offset 0 and
1492 * ".." is dotdot_offset.
1493 */
1494 if (!i) {
1495 i = dotdot_offset;
1496 continue;
1497 } else if (i == dotdot_offset) {
1498 i = dotdot_size;
1499 continue;
1500 }
1501 /* for other entry, the real offset in
1502 * the buf has to be tuned accordingly.
1503 */
1504 de = (struct ext4_dir_entry_2 *)
1505 (dir_buf + i - extra_offset);
1506 /* It's too expensive to do a full
1507 * dirent test each time round this
1508 * loop, but we do have to test at
1509 * least that it is non-zero. A
1510 * failure will be detected in the
1511 * dirent test below. */
1512 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1513 < EXT4_DIR_REC_LEN(1))
1514 break;
1515 i += ext4_rec_len_from_disk(de->rec_len,
1516 extra_size);
1517 }
1518 offset = i;
1519 ctx->pos = offset;
1520 file->f_version = inode->i_version;
1521 }
1522
1523 while (ctx->pos < extra_size) {
1524 if (ctx->pos == 0) {
1525 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1526 goto out;
1527 ctx->pos = dotdot_offset;
1528 continue;
1529 }
1530
1531 if (ctx->pos == dotdot_offset) {
1532 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1533 goto out;
1534 ctx->pos = dotdot_size;
1535 continue;
1536 }
1537
1538 de = (struct ext4_dir_entry_2 *)
1539 (dir_buf + ctx->pos - extra_offset);
1540 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1541 extra_size, ctx->pos))
1542 goto out;
1543 if (le32_to_cpu(de->inode)) {
1544 if (!dir_emit(ctx, de->name, de->name_len,
1545 le32_to_cpu(de->inode),
1546 get_dtype(sb, de->file_type)))
1547 goto out;
1548 }
1549 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1550 }
1551 out:
1552 kfree(dir_buf);
1553 brelse(iloc.bh);
1554 return ret;
1555 }
1556
1557 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1558 struct ext4_dir_entry_2 **parent_de,
1559 int *retval)
1560 {
1561 struct ext4_iloc iloc;
1562
1563 *retval = ext4_get_inode_loc(inode, &iloc);
1564 if (*retval)
1565 return NULL;
1566
1567 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1568
1569 return iloc.bh;
1570 }
1571
1572 /*
1573 * Try to create the inline data for the new dir.
1574 * If it succeeds, return 0, otherwise return the error.
1575 * In case of ENOSPC, the caller should create the normal disk layout dir.
1576 */
1577 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1578 struct inode *inode)
1579 {
1580 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1581 struct ext4_iloc iloc;
1582 struct ext4_dir_entry_2 *de;
1583
1584 ret = ext4_get_inode_loc(inode, &iloc);
1585 if (ret)
1586 return ret;
1587
1588 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1589 if (ret)
1590 goto out;
1591
1592 /*
1593 * For inline dir, we only save the inode information for the ".."
1594 * and create a fake dentry to cover the left space.
1595 */
1596 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1597 de->inode = cpu_to_le32(parent->i_ino);
1598 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1599 de->inode = 0;
1600 de->rec_len = ext4_rec_len_to_disk(
1601 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1602 inline_size);
1603 set_nlink(inode, 2);
1604 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1605 out:
1606 brelse(iloc.bh);
1607 return ret;
1608 }
1609
1610 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1611 struct ext4_filename *fname,
1612 const struct qstr *d_name,
1613 struct ext4_dir_entry_2 **res_dir,
1614 int *has_inline_data)
1615 {
1616 int ret;
1617 struct ext4_iloc iloc;
1618 void *inline_start;
1619 int inline_size;
1620
1621 if (ext4_get_inode_loc(dir, &iloc))
1622 return NULL;
1623
1624 down_read(&EXT4_I(dir)->xattr_sem);
1625 if (!ext4_has_inline_data(dir)) {
1626 *has_inline_data = 0;
1627 goto out;
1628 }
1629
1630 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1631 EXT4_INLINE_DOTDOT_SIZE;
1632 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1633 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1634 dir, fname, d_name, 0, res_dir);
1635 if (ret == 1)
1636 goto out_find;
1637 if (ret < 0)
1638 goto out;
1639
1640 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1641 goto out;
1642
1643 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1644 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1645
1646 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1647 dir, fname, d_name, 0, res_dir);
1648 if (ret == 1)
1649 goto out_find;
1650
1651 out:
1652 brelse(iloc.bh);
1653 iloc.bh = NULL;
1654 out_find:
1655 up_read(&EXT4_I(dir)->xattr_sem);
1656 return iloc.bh;
1657 }
1658
1659 int ext4_delete_inline_entry(handle_t *handle,
1660 struct inode *dir,
1661 struct ext4_dir_entry_2 *de_del,
1662 struct buffer_head *bh,
1663 int *has_inline_data)
1664 {
1665 int err, inline_size;
1666 struct ext4_iloc iloc;
1667 void *inline_start;
1668
1669 err = ext4_get_inode_loc(dir, &iloc);
1670 if (err)
1671 return err;
1672
1673 down_write(&EXT4_I(dir)->xattr_sem);
1674 if (!ext4_has_inline_data(dir)) {
1675 *has_inline_data = 0;
1676 goto out;
1677 }
1678
1679 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1680 EXT4_MIN_INLINE_DATA_SIZE) {
1681 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1682 EXT4_INLINE_DOTDOT_SIZE;
1683 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1684 EXT4_INLINE_DOTDOT_SIZE;
1685 } else {
1686 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1687 inline_size = ext4_get_inline_size(dir) -
1688 EXT4_MIN_INLINE_DATA_SIZE;
1689 }
1690
1691 BUFFER_TRACE(bh, "get_write_access");
1692 err = ext4_journal_get_write_access(handle, bh);
1693 if (err)
1694 goto out;
1695
1696 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1697 inline_start, inline_size, 0);
1698 if (err)
1699 goto out;
1700
1701 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1702 err = ext4_mark_inode_dirty(handle, dir);
1703 if (unlikely(err))
1704 goto out;
1705
1706 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1707 out:
1708 up_write(&EXT4_I(dir)->xattr_sem);
1709 brelse(iloc.bh);
1710 if (err != -ENOENT)
1711 ext4_std_error(dir->i_sb, err);
1712 return err;
1713 }
1714
1715 /*
1716 * Get the inline dentry at offset.
1717 */
1718 static inline struct ext4_dir_entry_2 *
1719 ext4_get_inline_entry(struct inode *inode,
1720 struct ext4_iloc *iloc,
1721 unsigned int offset,
1722 void **inline_start,
1723 int *inline_size)
1724 {
1725 void *inline_pos;
1726
1727 BUG_ON(offset > ext4_get_inline_size(inode));
1728
1729 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1730 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1731 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1732 } else {
1733 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1734 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1735 *inline_size = ext4_get_inline_size(inode) -
1736 EXT4_MIN_INLINE_DATA_SIZE;
1737 }
1738
1739 if (inline_start)
1740 *inline_start = inline_pos;
1741 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1742 }
1743
1744 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1745 {
1746 int err, inline_size;
1747 struct ext4_iloc iloc;
1748 void *inline_pos;
1749 unsigned int offset;
1750 struct ext4_dir_entry_2 *de;
1751 int ret = 1;
1752
1753 err = ext4_get_inode_loc(dir, &iloc);
1754 if (err) {
1755 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1756 err, dir->i_ino);
1757 return 1;
1758 }
1759
1760 down_read(&EXT4_I(dir)->xattr_sem);
1761 if (!ext4_has_inline_data(dir)) {
1762 *has_inline_data = 0;
1763 goto out;
1764 }
1765
1766 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1767 if (!le32_to_cpu(de->inode)) {
1768 ext4_warning(dir->i_sb,
1769 "bad inline directory (dir #%lu) - no `..'",
1770 dir->i_ino);
1771 ret = 1;
1772 goto out;
1773 }
1774
1775 offset = EXT4_INLINE_DOTDOT_SIZE;
1776 while (offset < dir->i_size) {
1777 de = ext4_get_inline_entry(dir, &iloc, offset,
1778 &inline_pos, &inline_size);
1779 if (ext4_check_dir_entry(dir, NULL, de,
1780 iloc.bh, inline_pos,
1781 inline_size, offset)) {
1782 ext4_warning(dir->i_sb,
1783 "bad inline directory (dir #%lu) - "
1784 "inode %u, rec_len %u, name_len %d"
1785 "inline size %d\n",
1786 dir->i_ino, le32_to_cpu(de->inode),
1787 le16_to_cpu(de->rec_len), de->name_len,
1788 inline_size);
1789 ret = 1;
1790 goto out;
1791 }
1792 if (le32_to_cpu(de->inode)) {
1793 ret = 0;
1794 goto out;
1795 }
1796 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1797 }
1798
1799 out:
1800 up_read(&EXT4_I(dir)->xattr_sem);
1801 brelse(iloc.bh);
1802 return ret;
1803 }
1804
1805 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1806 {
1807 int ret;
1808
1809 down_write(&EXT4_I(inode)->xattr_sem);
1810 ret = ext4_destroy_inline_data_nolock(handle, inode);
1811 up_write(&EXT4_I(inode)->xattr_sem);
1812
1813 return ret;
1814 }
1815
1816 int ext4_inline_data_fiemap(struct inode *inode,
1817 struct fiemap_extent_info *fieinfo,
1818 int *has_inline, __u64 start, __u64 len)
1819 {
1820 __u64 physical = 0;
1821 __u64 inline_len;
1822 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1823 FIEMAP_EXTENT_LAST;
1824 int error = 0;
1825 struct ext4_iloc iloc;
1826
1827 down_read(&EXT4_I(inode)->xattr_sem);
1828 if (!ext4_has_inline_data(inode)) {
1829 *has_inline = 0;
1830 goto out;
1831 }
1832 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1833 i_size_read(inode));
1834 if (start >= inline_len)
1835 goto out;
1836 if (start + len < inline_len)
1837 inline_len = start + len;
1838 inline_len -= start;
1839
1840 error = ext4_get_inode_loc(inode, &iloc);
1841 if (error)
1842 goto out;
1843
1844 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1845 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1846 physical += offsetof(struct ext4_inode, i_block);
1847
1848 if (physical)
1849 error = fiemap_fill_next_extent(fieinfo, start, physical,
1850 inline_len, flags);
1851 brelse(iloc.bh);
1852 out:
1853 up_read(&EXT4_I(inode)->xattr_sem);
1854 return (error < 0 ? error : 0);
1855 }
1856
1857 /*
1858 * Called during xattr set, and if we can sparse space 'needed',
1859 * just create the extent tree evict the data to the outer block.
1860 *
1861 * We use jbd2 instead of page cache to move data to the 1st block
1862 * so that the whole transaction can be committed as a whole and
1863 * the data isn't lost because of the delayed page cache write.
1864 */
1865 int ext4_try_to_evict_inline_data(handle_t *handle,
1866 struct inode *inode,
1867 int needed)
1868 {
1869 int error;
1870 struct ext4_xattr_entry *entry;
1871 struct ext4_inode *raw_inode;
1872 struct ext4_iloc iloc;
1873
1874 error = ext4_get_inode_loc(inode, &iloc);
1875 if (error)
1876 return error;
1877
1878 raw_inode = ext4_raw_inode(&iloc);
1879 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1880 EXT4_I(inode)->i_inline_off);
1881 if (EXT4_XATTR_LEN(entry->e_name_len) +
1882 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1883 error = -ENOSPC;
1884 goto out;
1885 }
1886
1887 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1888 out:
1889 brelse(iloc.bh);
1890 return error;
1891 }
1892
1893 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1894 {
1895 handle_t *handle;
1896 int inline_size, value_len, needed_blocks;
1897 size_t i_size;
1898 void *value = NULL;
1899 struct ext4_xattr_ibody_find is = {
1900 .s = { .not_found = -ENODATA, },
1901 };
1902 struct ext4_xattr_info i = {
1903 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1904 .name = EXT4_XATTR_SYSTEM_DATA,
1905 };
1906
1907
1908 needed_blocks = ext4_writepage_trans_blocks(inode);
1909 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1910 if (IS_ERR(handle))
1911 return;
1912
1913 down_write(&EXT4_I(inode)->xattr_sem);
1914 if (!ext4_has_inline_data(inode)) {
1915 *has_inline = 0;
1916 ext4_journal_stop(handle);
1917 return;
1918 }
1919
1920 if (ext4_orphan_add(handle, inode))
1921 goto out;
1922
1923 if (ext4_get_inode_loc(inode, &is.iloc))
1924 goto out;
1925
1926 down_write(&EXT4_I(inode)->i_data_sem);
1927 i_size = inode->i_size;
1928 inline_size = ext4_get_inline_size(inode);
1929 EXT4_I(inode)->i_disksize = i_size;
1930
1931 if (i_size < inline_size) {
1932 /* Clear the content in the xattr space. */
1933 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1934 if (ext4_xattr_ibody_find(inode, &i, &is))
1935 goto out_error;
1936
1937 BUG_ON(is.s.not_found);
1938
1939 value_len = le32_to_cpu(is.s.here->e_value_size);
1940 value = kmalloc(value_len, GFP_NOFS);
1941 if (!value)
1942 goto out_error;
1943
1944 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1945 value, value_len))
1946 goto out_error;
1947
1948 i.value = value;
1949 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1950 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1951 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1952 goto out_error;
1953 }
1954
1955 /* Clear the content within i_blocks. */
1956 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1957 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1958 memset(p + i_size, 0,
1959 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1960 }
1961
1962 EXT4_I(inode)->i_inline_size = i_size <
1963 EXT4_MIN_INLINE_DATA_SIZE ?
1964 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1965 }
1966
1967 out_error:
1968 up_write(&EXT4_I(inode)->i_data_sem);
1969 out:
1970 brelse(is.iloc.bh);
1971 up_write(&EXT4_I(inode)->xattr_sem);
1972 kfree(value);
1973 if (inode->i_nlink)
1974 ext4_orphan_del(handle, inode);
1975
1976 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1977 ext4_mark_inode_dirty(handle, inode);
1978 if (IS_SYNC(inode))
1979 ext4_handle_sync(handle);
1980
1981 ext4_journal_stop(handle);
1982 return;
1983 }
1984
1985 int ext4_convert_inline_data(struct inode *inode)
1986 {
1987 int error, needed_blocks;
1988 handle_t *handle;
1989 struct ext4_iloc iloc;
1990
1991 if (!ext4_has_inline_data(inode)) {
1992 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1993 return 0;
1994 }
1995
1996 needed_blocks = ext4_writepage_trans_blocks(inode);
1997
1998 iloc.bh = NULL;
1999 error = ext4_get_inode_loc(inode, &iloc);
2000 if (error)
2001 return error;
2002
2003 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2004 if (IS_ERR(handle)) {
2005 error = PTR_ERR(handle);
2006 goto out_free;
2007 }
2008
2009 down_write(&EXT4_I(inode)->xattr_sem);
2010 if (!ext4_has_inline_data(inode)) {
2011 up_write(&EXT4_I(inode)->xattr_sem);
2012 goto out;
2013 }
2014
2015 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2016 up_write(&EXT4_I(inode)->xattr_sem);
2017 out:
2018 ext4_journal_stop(handle);
2019 out_free:
2020 brelse(iloc.bh);
2021 return error;
2022 }
Michael's Linux tree.