?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
selftests/powerpc: Add test for VPHN
[linux/fpc-iii.git] / fs / ext4 / inline.c
blob4b143febf21f98d782892279b11731a50af5f359
1 /*
2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 */
14 #include "ext4_jbd2.h"
15 #include "ext4.h"
16 #include "xattr.h"
17 #include "truncate.h"
18 #include <linux/fiemap.h>
19
20 #define EXT4_XATTR_SYSTEM_DATA "data"
21 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
22 #define EXT4_INLINE_DOTDOT_OFFSET 2
23 #define EXT4_INLINE_DOTDOT_SIZE 4
24
25 static int ext4_get_inline_size(struct inode *inode)
26 {
27 if (EXT4_I(inode)->i_inline_off)
28 return EXT4_I(inode)->i_inline_size;
29
30 return 0;
31 }
32
33 static int get_max_inline_xattr_value_size(struct inode *inode,
34 struct ext4_iloc *iloc)
35 {
36 struct ext4_xattr_ibody_header *header;
37 struct ext4_xattr_entry *entry;
38 struct ext4_inode *raw_inode;
39 int free, min_offs;
40
41 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42 EXT4_GOOD_OLD_INODE_SIZE -
43 EXT4_I(inode)->i_extra_isize -
44 sizeof(struct ext4_xattr_ibody_header);
45
46 /*
47 * We need to subtract another sizeof(__u32) since an in-inode xattr
48 * needs an empty 4 bytes to indicate the gap between the xattr entry
49 * and the name/value pair.
50 */
51 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52 return EXT4_XATTR_SIZE(min_offs -
53 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54 EXT4_XATTR_ROUND - sizeof(__u32));
55
56 raw_inode = ext4_raw_inode(iloc);
57 header = IHDR(inode, raw_inode);
58 entry = IFIRST(header);
59
60 /* Compute min_offs. */
61 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
62 if (!entry->e_value_block && entry->e_value_size) {
63 size_t offs = le16_to_cpu(entry->e_value_offs);
64 if (offs < min_offs)
65 min_offs = offs;
66 }
67 }
68 free = min_offs -
69 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
70
71 if (EXT4_I(inode)->i_inline_off) {
72 entry = (struct ext4_xattr_entry *)
73 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
74
75 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
76 goto out;
77 }
78
79 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
80
81 if (free > EXT4_XATTR_ROUND)
82 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
83 else
84 free = 0;
85
86 out:
87 return free;
88 }
89
90 /*
91 * Get the maximum size we now can store in an inode.
92 * If we can't find the space for a xattr entry, don't use the space
93 * of the extents since we have no space to indicate the inline data.
94 */
95 int ext4_get_max_inline_size(struct inode *inode)
96 {
97 int error, max_inline_size;
98 struct ext4_iloc iloc;
99
100 if (EXT4_I(inode)->i_extra_isize == 0)
101 return 0;
102
103 error = ext4_get_inode_loc(inode, &iloc);
104 if (error) {
105 ext4_error_inode(inode, __func__, __LINE__, 0,
106 "can't get inode location %lu",
107 inode->i_ino);
108 return 0;
109 }
110
111 down_read(&EXT4_I(inode)->xattr_sem);
112 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
113 up_read(&EXT4_I(inode)->xattr_sem);
114
115 brelse(iloc.bh);
116
117 if (!max_inline_size)
118 return 0;
119
120 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
121 }
122
123 /*
124 * this function does not take xattr_sem, which is OK because it is
125 * currently only used in a code path coming form ext4_iget, before
126 * the new inode has been unlocked
127 */
128 int ext4_find_inline_data_nolock(struct inode *inode)
129 {
130 struct ext4_xattr_ibody_find is = {
131 .s = { .not_found = -ENODATA, },
132 };
133 struct ext4_xattr_info i = {
134 .name_index = EXT4_XATTR_INDEX_SYSTEM,
135 .name = EXT4_XATTR_SYSTEM_DATA,
136 };
137 int error;
138
139 if (EXT4_I(inode)->i_extra_isize == 0)
140 return 0;
141
142 error = ext4_get_inode_loc(inode, &is.iloc);
143 if (error)
144 return error;
145
146 error = ext4_xattr_ibody_find(inode, &i, &is);
147 if (error)
148 goto out;
149
150 if (!is.s.not_found) {
151 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
152 (void *)ext4_raw_inode(&is.iloc));
153 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
154 le32_to_cpu(is.s.here->e_value_size);
155 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
156 }
157 out:
158 brelse(is.iloc.bh);
159 return error;
160 }
161
162 static int ext4_read_inline_data(struct inode *inode, void *buffer,
163 unsigned int len,
164 struct ext4_iloc *iloc)
165 {
166 struct ext4_xattr_entry *entry;
167 struct ext4_xattr_ibody_header *header;
168 int cp_len = 0;
169 struct ext4_inode *raw_inode;
170
171 if (!len)
172 return 0;
173
174 BUG_ON(len > EXT4_I(inode)->i_inline_size);
175
176 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
177 len : EXT4_MIN_INLINE_DATA_SIZE;
178
179 raw_inode = ext4_raw_inode(iloc);
180 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
181
182 len -= cp_len;
183 buffer += cp_len;
184
185 if (!len)
186 goto out;
187
188 header = IHDR(inode, raw_inode);
189 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
190 EXT4_I(inode)->i_inline_off);
191 len = min_t(unsigned int, len,
192 (unsigned int)le32_to_cpu(entry->e_value_size));
193
194 memcpy(buffer,
195 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
196 cp_len += len;
197
198 out:
199 return cp_len;
200 }
201
202 /*
203 * write the buffer to the inline inode.
204 * If 'create' is set, we don't need to do the extra copy in the xattr
205 * value since it is already handled by ext4_xattr_ibody_inline_set.
206 * That saves us one memcpy.
207 */
208 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
209 void *buffer, loff_t pos, unsigned int len)
210 {
211 struct ext4_xattr_entry *entry;
212 struct ext4_xattr_ibody_header *header;
213 struct ext4_inode *raw_inode;
214 int cp_len = 0;
215
216 BUG_ON(!EXT4_I(inode)->i_inline_off);
217 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
218
219 raw_inode = ext4_raw_inode(iloc);
220 buffer += pos;
221
222 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
223 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
224 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
225 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
226
227 len -= cp_len;
228 buffer += cp_len;
229 pos += cp_len;
230 }
231
232 if (!len)
233 return;
234
235 pos -= EXT4_MIN_INLINE_DATA_SIZE;
236 header = IHDR(inode, raw_inode);
237 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
238 EXT4_I(inode)->i_inline_off);
239
240 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
241 buffer, len);
242 }
243
244 static int ext4_create_inline_data(handle_t *handle,
245 struct inode *inode, unsigned len)
246 {
247 int error;
248 void *value = NULL;
249 struct ext4_xattr_ibody_find is = {
250 .s = { .not_found = -ENODATA, },
251 };
252 struct ext4_xattr_info i = {
253 .name_index = EXT4_XATTR_INDEX_SYSTEM,
254 .name = EXT4_XATTR_SYSTEM_DATA,
255 };
256
257 error = ext4_get_inode_loc(inode, &is.iloc);
258 if (error)
259 return error;
260
261 BUFFER_TRACE(is.iloc.bh, "get_write_access");
262 error = ext4_journal_get_write_access(handle, is.iloc.bh);
263 if (error)
264 goto out;
265
266 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
267 value = EXT4_ZERO_XATTR_VALUE;
268 len -= EXT4_MIN_INLINE_DATA_SIZE;
269 } else {
270 value = "";
271 len = 0;
272 }
273
274 /* Insert the the xttr entry. */
275 i.value = value;
276 i.value_len = len;
277
278 error = ext4_xattr_ibody_find(inode, &i, &is);
279 if (error)
280 goto out;
281
282 BUG_ON(!is.s.not_found);
283
284 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
285 if (error) {
286 if (error == -ENOSPC)
287 ext4_clear_inode_state(inode,
288 EXT4_STATE_MAY_INLINE_DATA);
289 goto out;
290 }
291
292 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
293 0, EXT4_MIN_INLINE_DATA_SIZE);
294
295 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
296 (void *)ext4_raw_inode(&is.iloc));
297 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
298 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
299 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
300 get_bh(is.iloc.bh);
301 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
302
303 out:
304 brelse(is.iloc.bh);
305 return error;
306 }
307
308 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
309 unsigned int len)
310 {
311 int error;
312 void *value = NULL;
313 struct ext4_xattr_ibody_find is = {
314 .s = { .not_found = -ENODATA, },
315 };
316 struct ext4_xattr_info i = {
317 .name_index = EXT4_XATTR_INDEX_SYSTEM,
318 .name = EXT4_XATTR_SYSTEM_DATA,
319 };
320
321 /* If the old space is ok, write the data directly. */
322 if (len <= EXT4_I(inode)->i_inline_size)
323 return 0;
324
325 error = ext4_get_inode_loc(inode, &is.iloc);
326 if (error)
327 return error;
328
329 error = ext4_xattr_ibody_find(inode, &i, &is);
330 if (error)
331 goto out;
332
333 BUG_ON(is.s.not_found);
334
335 len -= EXT4_MIN_INLINE_DATA_SIZE;
336 value = kzalloc(len, GFP_NOFS);
337 if (!value)
338 goto out;
339
340 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
341 value, len);
342 if (error == -ENODATA)
343 goto out;
344
345 BUFFER_TRACE(is.iloc.bh, "get_write_access");
346 error = ext4_journal_get_write_access(handle, is.iloc.bh);
347 if (error)
348 goto out;
349
350 /* Update the xttr entry. */
351 i.value = value;
352 i.value_len = len;
353
354 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
355 if (error)
356 goto out;
357
358 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
359 (void *)ext4_raw_inode(&is.iloc));
360 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
361 le32_to_cpu(is.s.here->e_value_size);
362 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
363 get_bh(is.iloc.bh);
364 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
365
366 out:
367 kfree(value);
368 brelse(is.iloc.bh);
369 return error;
370 }
371
372 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
373 unsigned int len)
374 {
375 int ret, size;
376 struct ext4_inode_info *ei = EXT4_I(inode);
377
378 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
379 return -ENOSPC;
380
381 size = ext4_get_max_inline_size(inode);
382 if (size < len)
383 return -ENOSPC;
384
385 down_write(&EXT4_I(inode)->xattr_sem);
386
387 if (ei->i_inline_off)
388 ret = ext4_update_inline_data(handle, inode, len);
389 else
390 ret = ext4_create_inline_data(handle, inode, len);
391
392 up_write(&EXT4_I(inode)->xattr_sem);
393
394 return ret;
395 }
396
397 static int ext4_destroy_inline_data_nolock(handle_t *handle,
398 struct inode *inode)
399 {
400 struct ext4_inode_info *ei = EXT4_I(inode);
401 struct ext4_xattr_ibody_find is = {
402 .s = { .not_found = 0, },
403 };
404 struct ext4_xattr_info i = {
405 .name_index = EXT4_XATTR_INDEX_SYSTEM,
406 .name = EXT4_XATTR_SYSTEM_DATA,
407 .value = NULL,
408 .value_len = 0,
409 };
410 int error;
411
412 if (!ei->i_inline_off)
413 return 0;
414
415 error = ext4_get_inode_loc(inode, &is.iloc);
416 if (error)
417 return error;
418
419 error = ext4_xattr_ibody_find(inode, &i, &is);
420 if (error)
421 goto out;
422
423 BUFFER_TRACE(is.iloc.bh, "get_write_access");
424 error = ext4_journal_get_write_access(handle, is.iloc.bh);
425 if (error)
426 goto out;
427
428 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
429 if (error)
430 goto out;
431
432 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
433 0, EXT4_MIN_INLINE_DATA_SIZE);
434
435 if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
436 EXT4_FEATURE_INCOMPAT_EXTENTS)) {
437 if (S_ISDIR(inode->i_mode) ||
438 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
439 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
440 ext4_ext_tree_init(handle, inode);
441 }
442 }
443 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
444
445 get_bh(is.iloc.bh);
446 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
447
448 EXT4_I(inode)->i_inline_off = 0;
449 EXT4_I(inode)->i_inline_size = 0;
450 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
451 out:
452 brelse(is.iloc.bh);
453 if (error == -ENODATA)
454 error = 0;
455 return error;
456 }
457
458 static int ext4_read_inline_page(struct inode *inode, struct page *page)
459 {
460 void *kaddr;
461 int ret = 0;
462 size_t len;
463 struct ext4_iloc iloc;
464
465 BUG_ON(!PageLocked(page));
466 BUG_ON(!ext4_has_inline_data(inode));
467 BUG_ON(page->index);
468
469 if (!EXT4_I(inode)->i_inline_off) {
470 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
471 inode->i_ino);
472 goto out;
473 }
474
475 ret = ext4_get_inode_loc(inode, &iloc);
476 if (ret)
477 goto out;
478
479 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
480 kaddr = kmap_atomic(page);
481 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
482 flush_dcache_page(page);
483 kunmap_atomic(kaddr);
484 zero_user_segment(page, len, PAGE_CACHE_SIZE);
485 SetPageUptodate(page);
486 brelse(iloc.bh);
487
488 out:
489 return ret;
490 }
491
492 int ext4_readpage_inline(struct inode *inode, struct page *page)
493 {
494 int ret = 0;
495
496 down_read(&EXT4_I(inode)->xattr_sem);
497 if (!ext4_has_inline_data(inode)) {
498 up_read(&EXT4_I(inode)->xattr_sem);
499 return -EAGAIN;
500 }
501
502 /*
503 * Current inline data can only exist in the 1st page,
504 * So for all the other pages, just set them uptodate.
505 */
506 if (!page->index)
507 ret = ext4_read_inline_page(inode, page);
508 else if (!PageUptodate(page)) {
509 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
510 SetPageUptodate(page);
511 }
512
513 up_read(&EXT4_I(inode)->xattr_sem);
514
515 unlock_page(page);
516 return ret >= 0 ? 0 : ret;
517 }
518
519 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
520 struct inode *inode,
521 unsigned flags)
522 {
523 int ret, needed_blocks;
524 handle_t *handle = NULL;
525 int retries = 0, sem_held = 0;
526 struct page *page = NULL;
527 unsigned from, to;
528 struct ext4_iloc iloc;
529
530 if (!ext4_has_inline_data(inode)) {
531 /*
532 * clear the flag so that no new write
533 * will trap here again.
534 */
535 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
536 return 0;
537 }
538
539 needed_blocks = ext4_writepage_trans_blocks(inode);
540
541 ret = ext4_get_inode_loc(inode, &iloc);
542 if (ret)
543 return ret;
544
545 retry:
546 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
547 if (IS_ERR(handle)) {
548 ret = PTR_ERR(handle);
549 handle = NULL;
550 goto out;
551 }
552
553 /* We cannot recurse into the filesystem as the transaction is already
554 * started */
555 flags |= AOP_FLAG_NOFS;
556
557 page = grab_cache_page_write_begin(mapping, 0, flags);
558 if (!page) {
559 ret = -ENOMEM;
560 goto out;
561 }
562
563 down_write(&EXT4_I(inode)->xattr_sem);
564 sem_held = 1;
565 /* If some one has already done this for us, just exit. */
566 if (!ext4_has_inline_data(inode)) {
567 ret = 0;
568 goto out;
569 }
570
571 from = 0;
572 to = ext4_get_inline_size(inode);
573 if (!PageUptodate(page)) {
574 ret = ext4_read_inline_page(inode, page);
575 if (ret < 0)
576 goto out;
577 }
578
579 ret = ext4_destroy_inline_data_nolock(handle, inode);
580 if (ret)
581 goto out;
582
583 if (ext4_should_dioread_nolock(inode))
584 ret = __block_write_begin(page, from, to, ext4_get_block_write);
585 else
586 ret = __block_write_begin(page, from, to, ext4_get_block);
587
588 if (!ret && ext4_should_journal_data(inode)) {
589 ret = ext4_walk_page_buffers(handle, page_buffers(page),
590 from, to, NULL,
591 do_journal_get_write_access);
592 }
593
594 if (ret) {
595 unlock_page(page);
596 page_cache_release(page);
597 page = NULL;
598 ext4_orphan_add(handle, inode);
599 up_write(&EXT4_I(inode)->xattr_sem);
600 sem_held = 0;
601 ext4_journal_stop(handle);
602 handle = NULL;
603 ext4_truncate_failed_write(inode);
604 /*
605 * If truncate failed early the inode might
606 * still be on the orphan list; we need to
607 * make sure the inode is removed from the
608 * orphan list in that case.
609 */
610 if (inode->i_nlink)
611 ext4_orphan_del(NULL, inode);
612 }
613
614 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
615 goto retry;
616
617 if (page)
618 block_commit_write(page, from, to);
619 out:
620 if (page) {
621 unlock_page(page);
622 page_cache_release(page);
623 }
624 if (sem_held)
625 up_write(&EXT4_I(inode)->xattr_sem);
626 if (handle)
627 ext4_journal_stop(handle);
628 brelse(iloc.bh);
629 return ret;
630 }
631
632 /*
633 * Try to write data in the inode.
634 * If the inode has inline data, check whether the new write can be
635 * in the inode also. If not, create the page the handle, move the data
636 * to the page make it update and let the later codes create extent for it.
637 */
638 int ext4_try_to_write_inline_data(struct address_space *mapping,
639 struct inode *inode,
640 loff_t pos, unsigned len,
641 unsigned flags,
642 struct page **pagep)
643 {
644 int ret;
645 handle_t *handle;
646 struct page *page;
647 struct ext4_iloc iloc;
648
649 if (pos + len > ext4_get_max_inline_size(inode))
650 goto convert;
651
652 ret = ext4_get_inode_loc(inode, &iloc);
653 if (ret)
654 return ret;
655
656 /*
657 * The possible write could happen in the inode,
658 * so try to reserve the space in inode first.
659 */
660 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
661 if (IS_ERR(handle)) {
662 ret = PTR_ERR(handle);
663 handle = NULL;
664 goto out;
665 }
666
667 ret = ext4_prepare_inline_data(handle, inode, pos + len);
668 if (ret && ret != -ENOSPC)
669 goto out;
670
671 /* We don't have space in inline inode, so convert it to extent. */
672 if (ret == -ENOSPC) {
673 ext4_journal_stop(handle);
674 brelse(iloc.bh);
675 goto convert;
676 }
677
678 flags |= AOP_FLAG_NOFS;
679
680 page = grab_cache_page_write_begin(mapping, 0, flags);
681 if (!page) {
682 ret = -ENOMEM;
683 goto out;
684 }
685
686 *pagep = page;
687 down_read(&EXT4_I(inode)->xattr_sem);
688 if (!ext4_has_inline_data(inode)) {
689 ret = 0;
690 unlock_page(page);
691 page_cache_release(page);
692 goto out_up_read;
693 }
694
695 if (!PageUptodate(page)) {
696 ret = ext4_read_inline_page(inode, page);
697 if (ret < 0)
698 goto out_up_read;
699 }
700
701 ret = 1;
702 handle = NULL;
703 out_up_read:
704 up_read(&EXT4_I(inode)->xattr_sem);
705 out:
706 if (handle)
707 ext4_journal_stop(handle);
708 brelse(iloc.bh);
709 return ret;
710 convert:
711 return ext4_convert_inline_data_to_extent(mapping,
712 inode, flags);
713 }
714
715 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
716 unsigned copied, struct page *page)
717 {
718 int ret;
719 void *kaddr;
720 struct ext4_iloc iloc;
721
722 if (unlikely(copied < len)) {
723 if (!PageUptodate(page)) {
724 copied = 0;
725 goto out;
726 }
727 }
728
729 ret = ext4_get_inode_loc(inode, &iloc);
730 if (ret) {
731 ext4_std_error(inode->i_sb, ret);
732 copied = 0;
733 goto out;
734 }
735
736 down_write(&EXT4_I(inode)->xattr_sem);
737 BUG_ON(!ext4_has_inline_data(inode));
738
739 kaddr = kmap_atomic(page);
740 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
741 kunmap_atomic(kaddr);
742 SetPageUptodate(page);
743 /* clear page dirty so that writepages wouldn't work for us. */
744 ClearPageDirty(page);
745
746 up_write(&EXT4_I(inode)->xattr_sem);
747 brelse(iloc.bh);
748 out:
749 return copied;
750 }
751
752 struct buffer_head *
753 ext4_journalled_write_inline_data(struct inode *inode,
754 unsigned len,
755 struct page *page)
756 {
757 int ret;
758 void *kaddr;
759 struct ext4_iloc iloc;
760
761 ret = ext4_get_inode_loc(inode, &iloc);
762 if (ret) {
763 ext4_std_error(inode->i_sb, ret);
764 return NULL;
765 }
766
767 down_write(&EXT4_I(inode)->xattr_sem);
768 kaddr = kmap_atomic(page);
769 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
770 kunmap_atomic(kaddr);
771 up_write(&EXT4_I(inode)->xattr_sem);
772
773 return iloc.bh;
774 }
775
776 /*
777 * Try to make the page cache and handle ready for the inline data case.
778 * We can call this function in 2 cases:
779 * 1. The inode is created and the first write exceeds inline size. We can
780 * clear the inode state safely.
781 * 2. The inode has inline data, then we need to read the data, make it
782 * update and dirty so that ext4_da_writepages can handle it. We don't
783 * need to start the journal since the file's metatdata isn't changed now.
784 */
785 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
786 struct inode *inode,
787 unsigned flags,
788 void **fsdata)
789 {
790 int ret = 0, inline_size;
791 struct page *page;
792
793 page = grab_cache_page_write_begin(mapping, 0, flags);
794 if (!page)
795 return -ENOMEM;
796
797 down_read(&EXT4_I(inode)->xattr_sem);
798 if (!ext4_has_inline_data(inode)) {
799 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
800 goto out;
801 }
802
803 inline_size = ext4_get_inline_size(inode);
804
805 if (!PageUptodate(page)) {
806 ret = ext4_read_inline_page(inode, page);
807 if (ret < 0)
808 goto out;
809 }
810
811 ret = __block_write_begin(page, 0, inline_size,
812 ext4_da_get_block_prep);
813 if (ret) {
814 up_read(&EXT4_I(inode)->xattr_sem);
815 unlock_page(page);
816 page_cache_release(page);
817 ext4_truncate_failed_write(inode);
818 return ret;
819 }
820
821 SetPageDirty(page);
822 SetPageUptodate(page);
823 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
824 *fsdata = (void *)CONVERT_INLINE_DATA;
825
826 out:
827 up_read(&EXT4_I(inode)->xattr_sem);
828 if (page) {
829 unlock_page(page);
830 page_cache_release(page);
831 }
832 return ret;
833 }
834
835 /*
836 * Prepare the write for the inline data.
837 * If the the data can be written into the inode, we just read
838 * the page and make it uptodate, and start the journal.
839 * Otherwise read the page, makes it dirty so that it can be
840 * handle in writepages(the i_disksize update is left to the
841 * normal ext4_da_write_end).
842 */
843 int ext4_da_write_inline_data_begin(struct address_space *mapping,
844 struct inode *inode,
845 loff_t pos, unsigned len,
846 unsigned flags,
847 struct page **pagep,
848 void **fsdata)
849 {
850 int ret, inline_size;
851 handle_t *handle;
852 struct page *page;
853 struct ext4_iloc iloc;
854 int retries;
855
856 ret = ext4_get_inode_loc(inode, &iloc);
857 if (ret)
858 return ret;
859
860 retry_journal:
861 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
862 if (IS_ERR(handle)) {
863 ret = PTR_ERR(handle);
864 goto out;
865 }
866
867 inline_size = ext4_get_max_inline_size(inode);
868
869 ret = -ENOSPC;
870 if (inline_size >= pos + len) {
871 ret = ext4_prepare_inline_data(handle, inode, pos + len);
872 if (ret && ret != -ENOSPC)
873 goto out_journal;
874 }
875
876 /*
877 * We cannot recurse into the filesystem as the transaction
878 * is already started.
879 */
880 flags |= AOP_FLAG_NOFS;
881
882 if (ret == -ENOSPC) {
883 ret = ext4_da_convert_inline_data_to_extent(mapping,
884 inode,
885 flags,
886 fsdata);
887 ext4_journal_stop(handle);
888 if (ret == -ENOSPC &&
889 ext4_should_retry_alloc(inode->i_sb, &retries))
890 goto retry_journal;
891 goto out;
892 }
893
894
895 page = grab_cache_page_write_begin(mapping, 0, flags);
896 if (!page) {
897 ret = -ENOMEM;
898 goto out_journal;
899 }
900
901 down_read(&EXT4_I(inode)->xattr_sem);
902 if (!ext4_has_inline_data(inode)) {
903 ret = 0;
904 goto out_release_page;
905 }
906
907 if (!PageUptodate(page)) {
908 ret = ext4_read_inline_page(inode, page);
909 if (ret < 0)
910 goto out_release_page;
911 }
912
913 up_read(&EXT4_I(inode)->xattr_sem);
914 *pagep = page;
915 brelse(iloc.bh);
916 return 1;
917 out_release_page:
918 up_read(&EXT4_I(inode)->xattr_sem);
919 unlock_page(page);
920 page_cache_release(page);
921 out_journal:
922 ext4_journal_stop(handle);
923 out:
924 brelse(iloc.bh);
925 return ret;
926 }
927
928 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
929 unsigned len, unsigned copied,
930 struct page *page)
931 {
932 int i_size_changed = 0;
933
934 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
935
936 /*
937 * No need to use i_size_read() here, the i_size
938 * cannot change under us because we hold i_mutex.
939 *
940 * But it's important to update i_size while still holding page lock:
941 * page writeout could otherwise come in and zero beyond i_size.
942 */
943 if (pos+copied > inode->i_size) {
944 i_size_write(inode, pos+copied);
945 i_size_changed = 1;
946 }
947 unlock_page(page);
948 page_cache_release(page);
949
950 /*
951 * Don't mark the inode dirty under page lock. First, it unnecessarily
952 * makes the holding time of page lock longer. Second, it forces lock
953 * ordering of page lock and transaction start for journaling
954 * filesystems.
955 */
956 if (i_size_changed)
957 mark_inode_dirty(inode);
958
959 return copied;
960 }
961
962 #ifdef INLINE_DIR_DEBUG
963 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
964 void *inline_start, int inline_size)
965 {
966 int offset;
967 unsigned short de_len;
968 struct ext4_dir_entry_2 *de = inline_start;
969 void *dlimit = inline_start + inline_size;
970
971 trace_printk("inode %lu\n", dir->i_ino);
972 offset = 0;
973 while ((void *)de < dlimit) {
974 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
975 trace_printk("de: off %u rlen %u name %*.s nlen %u ino %u\n",
976 offset, de_len, de->name_len, de->name,
977 de->name_len, le32_to_cpu(de->inode));
978 if (ext4_check_dir_entry(dir, NULL, de, bh,
979 inline_start, inline_size, offset))
980 BUG();
981
982 offset += de_len;
983 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
984 }
985 }
986 #else
987 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
988 #endif
989
990 /*
991 * Add a new entry into a inline dir.
992 * It will return -ENOSPC if no space is available, and -EIO
993 * and -EEXIST if directory entry already exists.
994 */
995 static int ext4_add_dirent_to_inline(handle_t *handle,
996 struct dentry *dentry,
997 struct inode *inode,
998 struct ext4_iloc *iloc,
999 void *inline_start, int inline_size)
1000 {
1001 struct inode *dir = dentry->d_parent->d_inode;
1002 const char *name = dentry->d_name.name;
1003 int namelen = dentry->d_name.len;
1004 int err;
1005 struct ext4_dir_entry_2 *de;
1006
1007 err = ext4_find_dest_de(dir, inode, iloc->bh,
1008 inline_start, inline_size,
1009 name, namelen, &de);
1010 if (err)
1011 return err;
1012
1013 BUFFER_TRACE(iloc->bh, "get_write_access");
1014 err = ext4_journal_get_write_access(handle, iloc->bh);
1015 if (err)
1016 return err;
1017 ext4_insert_dentry(inode, de, inline_size, name, namelen);
1018
1019 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1020
1021 /*
1022 * XXX shouldn't update any times until successful
1023 * completion of syscall, but too many callers depend
1024 * on this.
1025 *
1026 * XXX similarly, too many callers depend on
1027 * ext4_new_inode() setting the times, but error
1028 * recovery deletes the inode, so the worst that can
1029 * happen is that the times are slightly out of date
1030 * and/or different from the directory change time.
1031 */
1032 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1033 ext4_update_dx_flag(dir);
1034 dir->i_version++;
1035 ext4_mark_inode_dirty(handle, dir);
1036 return 1;
1037 }
1038
1039 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1040 struct ext4_iloc *iloc)
1041 {
1042 struct ext4_xattr_entry *entry;
1043 struct ext4_xattr_ibody_header *header;
1044
1045 BUG_ON(!EXT4_I(inode)->i_inline_off);
1046
1047 header = IHDR(inode, ext4_raw_inode(iloc));
1048 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1049 EXT4_I(inode)->i_inline_off);
1050
1051 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1052 }
1053
1054 /* Set the final de to cover the whole block. */
1055 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1056 {
1057 struct ext4_dir_entry_2 *de, *prev_de;
1058 void *limit;
1059 int de_len;
1060
1061 de = (struct ext4_dir_entry_2 *)de_buf;
1062 if (old_size) {
1063 limit = de_buf + old_size;
1064 do {
1065 prev_de = de;
1066 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1067 de_buf += de_len;
1068 de = (struct ext4_dir_entry_2 *)de_buf;
1069 } while (de_buf < limit);
1070
1071 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1072 old_size, new_size);
1073 } else {
1074 /* this is just created, so create an empty entry. */
1075 de->inode = 0;
1076 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1077 }
1078 }
1079
1080 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1081 struct ext4_iloc *iloc)
1082 {
1083 int ret;
1084 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1085 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1086
1087 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1088 return -ENOSPC;
1089
1090 ret = ext4_update_inline_data(handle, dir,
1091 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1092 if (ret)
1093 return ret;
1094
1095 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1096 EXT4_I(dir)->i_inline_size -
1097 EXT4_MIN_INLINE_DATA_SIZE);
1098 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1099 return 0;
1100 }
1101
1102 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1103 struct ext4_iloc *iloc,
1104 void *buf, int inline_size)
1105 {
1106 ext4_create_inline_data(handle, inode, inline_size);
1107 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1108 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1109 }
1110
1111 static int ext4_finish_convert_inline_dir(handle_t *handle,
1112 struct inode *inode,
1113 struct buffer_head *dir_block,
1114 void *buf,
1115 int inline_size)
1116 {
1117 int err, csum_size = 0, header_size = 0;
1118 struct ext4_dir_entry_2 *de;
1119 struct ext4_dir_entry_tail *t;
1120 void *target = dir_block->b_data;
1121
1122 /*
1123 * First create "." and ".." and then copy the dir information
1124 * back to the block.
1125 */
1126 de = (struct ext4_dir_entry_2 *)target;
1127 de = ext4_init_dot_dotdot(inode, de,
1128 inode->i_sb->s_blocksize, csum_size,
1129 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1130 header_size = (void *)de - target;
1131
1132 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1133 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1134
1135 if (ext4_has_metadata_csum(inode->i_sb))
1136 csum_size = sizeof(struct ext4_dir_entry_tail);
1137
1138 inode->i_size = inode->i_sb->s_blocksize;
1139 i_size_write(inode, inode->i_sb->s_blocksize);
1140 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1141 ext4_update_final_de(dir_block->b_data,
1142 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1143 inode->i_sb->s_blocksize - csum_size);
1144
1145 if (csum_size) {
1146 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1147 inode->i_sb->s_blocksize);
1148 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1149 }
1150 set_buffer_uptodate(dir_block);
1151 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1152 if (err)
1153 goto out;
1154 set_buffer_verified(dir_block);
1155 out:
1156 return err;
1157 }
1158
1159 static int ext4_convert_inline_data_nolock(handle_t *handle,
1160 struct inode *inode,
1161 struct ext4_iloc *iloc)
1162 {
1163 int error;
1164 void *buf = NULL;
1165 struct buffer_head *data_bh = NULL;
1166 struct ext4_map_blocks map;
1167 int inline_size;
1168
1169 inline_size = ext4_get_inline_size(inode);
1170 buf = kmalloc(inline_size, GFP_NOFS);
1171 if (!buf) {
1172 error = -ENOMEM;
1173 goto out;
1174 }
1175
1176 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1177 if (error < 0)
1178 goto out;
1179
1180 /*
1181 * Make sure the inline directory entries pass checks before we try to
1182 * convert them, so that we avoid touching stuff that needs fsck.
1183 */
1184 if (S_ISDIR(inode->i_mode)) {
1185 error = ext4_check_all_de(inode, iloc->bh,
1186 buf + EXT4_INLINE_DOTDOT_SIZE,
1187 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1188 if (error)
1189 goto out;
1190 }
1191
1192 error = ext4_destroy_inline_data_nolock(handle, inode);
1193 if (error)
1194 goto out;
1195
1196 map.m_lblk = 0;
1197 map.m_len = 1;
1198 map.m_flags = 0;
1199 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1200 if (error < 0)
1201 goto out_restore;
1202 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1203 error = -EIO;
1204 goto out_restore;
1205 }
1206
1207 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1208 if (!data_bh) {
1209 error = -ENOMEM;
1210 goto out_restore;
1211 }
1212
1213 lock_buffer(data_bh);
1214 error = ext4_journal_get_create_access(handle, data_bh);
1215 if (error) {
1216 unlock_buffer(data_bh);
1217 error = -EIO;
1218 goto out_restore;
1219 }
1220 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1221
1222 if (!S_ISDIR(inode->i_mode)) {
1223 memcpy(data_bh->b_data, buf, inline_size);
1224 set_buffer_uptodate(data_bh);
1225 error = ext4_handle_dirty_metadata(handle,
1226 inode, data_bh);
1227 } else {
1228 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1229 buf, inline_size);
1230 }
1231
1232 unlock_buffer(data_bh);
1233 out_restore:
1234 if (error)
1235 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1236
1237 out:
1238 brelse(data_bh);
1239 kfree(buf);
1240 return error;
1241 }
1242
1243 /*
1244 * Try to add the new entry to the inline data.
1245 * If succeeds, return 0. If not, extended the inline dir and copied data to
1246 * the new created block.
1247 */
1248 int ext4_try_add_inline_entry(handle_t *handle, struct dentry *dentry,
1249 struct inode *inode)
1250 {
1251 int ret, inline_size;
1252 void *inline_start;
1253 struct ext4_iloc iloc;
1254 struct inode *dir = dentry->d_parent->d_inode;
1255
1256 ret = ext4_get_inode_loc(dir, &iloc);
1257 if (ret)
1258 return ret;
1259
1260 down_write(&EXT4_I(dir)->xattr_sem);
1261 if (!ext4_has_inline_data(dir))
1262 goto out;
1263
1264 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1265 EXT4_INLINE_DOTDOT_SIZE;
1266 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1267
1268 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1269 inline_start, inline_size);
1270 if (ret != -ENOSPC)
1271 goto out;
1272
1273 /* check whether it can be inserted to inline xattr space. */
1274 inline_size = EXT4_I(dir)->i_inline_size -
1275 EXT4_MIN_INLINE_DATA_SIZE;
1276 if (!inline_size) {
1277 /* Try to use the xattr space.*/
1278 ret = ext4_update_inline_dir(handle, dir, &iloc);
1279 if (ret && ret != -ENOSPC)
1280 goto out;
1281
1282 inline_size = EXT4_I(dir)->i_inline_size -
1283 EXT4_MIN_INLINE_DATA_SIZE;
1284 }
1285
1286 if (inline_size) {
1287 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1288
1289 ret = ext4_add_dirent_to_inline(handle, dentry, inode, &iloc,
1290 inline_start, 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
1331 ret = ext4_get_inode_loc(inode, &iloc);
1332 if (ret)
1333 return ret;
1334
1335 down_read(&EXT4_I(inode)->xattr_sem);
1336 if (!ext4_has_inline_data(inode)) {
1337 up_read(&EXT4_I(inode)->xattr_sem);
1338 *has_inline_data = 0;
1339 goto out;
1340 }
1341
1342 inline_size = ext4_get_inline_size(inode);
1343 dir_buf = kmalloc(inline_size, GFP_NOFS);
1344 if (!dir_buf) {
1345 ret = -ENOMEM;
1346 up_read(&EXT4_I(inode)->xattr_sem);
1347 goto out;
1348 }
1349
1350 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1351 up_read(&EXT4_I(inode)->xattr_sem);
1352 if (ret < 0)
1353 goto out;
1354
1355 pos = 0;
1356 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1357 while (pos < inline_size) {
1358 /*
1359 * As inlined dir doesn't store any information about '.' and
1360 * only the inode number of '..' is stored, we have to handle
1361 * them differently.
1362 */
1363 if (pos == 0) {
1364 fake.inode = cpu_to_le32(inode->i_ino);
1365 fake.name_len = 1;
1366 strcpy(fake.name, ".");
1367 fake.rec_len = ext4_rec_len_to_disk(
1368 EXT4_DIR_REC_LEN(fake.name_len),
1369 inline_size);
1370 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1371 de = &fake;
1372 pos = EXT4_INLINE_DOTDOT_OFFSET;
1373 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1374 fake.inode = cpu_to_le32(parent_ino);
1375 fake.name_len = 2;
1376 strcpy(fake.name, "..");
1377 fake.rec_len = ext4_rec_len_to_disk(
1378 EXT4_DIR_REC_LEN(fake.name_len),
1379 inline_size);
1380 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1381 de = &fake;
1382 pos = EXT4_INLINE_DOTDOT_SIZE;
1383 } else {
1384 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1385 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1386 if (ext4_check_dir_entry(inode, dir_file, de,
1387 iloc.bh, dir_buf,
1388 inline_size, pos)) {
1389 ret = count;
1390 goto out;
1391 }
1392 }
1393
1394 ext4fs_dirhash(de->name, de->name_len, hinfo);
1395 if ((hinfo->hash < start_hash) ||
1396 ((hinfo->hash == start_hash) &&
1397 (hinfo->minor_hash < start_minor_hash)))
1398 continue;
1399 if (de->inode == 0)
1400 continue;
1401 err = ext4_htree_store_dirent(dir_file,
1402 hinfo->hash, hinfo->minor_hash, de);
1403 if (err) {
1404 count = err;
1405 goto out;
1406 }
1407 count++;
1408 }
1409 ret = count;
1410 out:
1411 kfree(dir_buf);
1412 brelse(iloc.bh);
1413 return ret;
1414 }
1415
1416 /*
1417 * So this function is called when the volume is mkfsed with
1418 * dir_index disabled. In order to keep f_pos persistent
1419 * after we convert from an inlined dir to a blocked based,
1420 * we just pretend that we are a normal dir and return the
1421 * offset as if '.' and '..' really take place.
1422 *
1423 */
1424 int ext4_read_inline_dir(struct file *file,
1425 struct dir_context *ctx,
1426 int *has_inline_data)
1427 {
1428 unsigned int offset, parent_ino;
1429 int i;
1430 struct ext4_dir_entry_2 *de;
1431 struct super_block *sb;
1432 struct inode *inode = file_inode(file);
1433 int ret, inline_size = 0;
1434 struct ext4_iloc iloc;
1435 void *dir_buf = NULL;
1436 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1437
1438 ret = ext4_get_inode_loc(inode, &iloc);
1439 if (ret)
1440 return ret;
1441
1442 down_read(&EXT4_I(inode)->xattr_sem);
1443 if (!ext4_has_inline_data(inode)) {
1444 up_read(&EXT4_I(inode)->xattr_sem);
1445 *has_inline_data = 0;
1446 goto out;
1447 }
1448
1449 inline_size = ext4_get_inline_size(inode);
1450 dir_buf = kmalloc(inline_size, GFP_NOFS);
1451 if (!dir_buf) {
1452 ret = -ENOMEM;
1453 up_read(&EXT4_I(inode)->xattr_sem);
1454 goto out;
1455 }
1456
1457 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1458 up_read(&EXT4_I(inode)->xattr_sem);
1459 if (ret < 0)
1460 goto out;
1461
1462 ret = 0;
1463 sb = inode->i_sb;
1464 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1465 offset = ctx->pos;
1466
1467 /*
1468 * dotdot_offset and dotdot_size is the real offset and
1469 * size for ".." and "." if the dir is block based while
1470 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1471 * So we will use extra_offset and extra_size to indicate them
1472 * during the inline dir iteration.
1473 */
1474 dotdot_offset = EXT4_DIR_REC_LEN(1);
1475 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1476 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1477 extra_size = extra_offset + inline_size;
1478
1479 /*
1480 * If the version has changed since the last call to
1481 * readdir(2), then we might be pointing to an invalid
1482 * dirent right now. Scan from the start of the inline
1483 * dir to make sure.
1484 */
1485 if (file->f_version != inode->i_version) {
1486 for (i = 0; i < extra_size && i < offset;) {
1487 /*
1488 * "." is with offset 0 and
1489 * ".." is dotdot_offset.
1490 */
1491 if (!i) {
1492 i = dotdot_offset;
1493 continue;
1494 } else if (i == dotdot_offset) {
1495 i = dotdot_size;
1496 continue;
1497 }
1498 /* for other entry, the real offset in
1499 * the buf has to be tuned accordingly.
1500 */
1501 de = (struct ext4_dir_entry_2 *)
1502 (dir_buf + i - extra_offset);
1503 /* It's too expensive to do a full
1504 * dirent test each time round this
1505 * loop, but we do have to test at
1506 * least that it is non-zero. A
1507 * failure will be detected in the
1508 * dirent test below. */
1509 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1510 < EXT4_DIR_REC_LEN(1))
1511 break;
1512 i += ext4_rec_len_from_disk(de->rec_len,
1513 extra_size);
1514 }
1515 offset = i;
1516 ctx->pos = offset;
1517 file->f_version = inode->i_version;
1518 }
1519
1520 while (ctx->pos < extra_size) {
1521 if (ctx->pos == 0) {
1522 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1523 goto out;
1524 ctx->pos = dotdot_offset;
1525 continue;
1526 }
1527
1528 if (ctx->pos == dotdot_offset) {
1529 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1530 goto out;
1531 ctx->pos = dotdot_size;
1532 continue;
1533 }
1534
1535 de = (struct ext4_dir_entry_2 *)
1536 (dir_buf + ctx->pos - extra_offset);
1537 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1538 extra_size, ctx->pos))
1539 goto out;
1540 if (le32_to_cpu(de->inode)) {
1541 if (!dir_emit(ctx, de->name, de->name_len,
1542 le32_to_cpu(de->inode),
1543 get_dtype(sb, de->file_type)))
1544 goto out;
1545 }
1546 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1547 }
1548 out:
1549 kfree(dir_buf);
1550 brelse(iloc.bh);
1551 return ret;
1552 }
1553
1554 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1555 struct ext4_dir_entry_2 **parent_de,
1556 int *retval)
1557 {
1558 struct ext4_iloc iloc;
1559
1560 *retval = ext4_get_inode_loc(inode, &iloc);
1561 if (*retval)
1562 return NULL;
1563
1564 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1565
1566 return iloc.bh;
1567 }
1568
1569 /*
1570 * Try to create the inline data for the new dir.
1571 * If it succeeds, return 0, otherwise return the error.
1572 * In case of ENOSPC, the caller should create the normal disk layout dir.
1573 */
1574 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1575 struct inode *inode)
1576 {
1577 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1578 struct ext4_iloc iloc;
1579 struct ext4_dir_entry_2 *de;
1580
1581 ret = ext4_get_inode_loc(inode, &iloc);
1582 if (ret)
1583 return ret;
1584
1585 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1586 if (ret)
1587 goto out;
1588
1589 /*
1590 * For inline dir, we only save the inode information for the ".."
1591 * and create a fake dentry to cover the left space.
1592 */
1593 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1594 de->inode = cpu_to_le32(parent->i_ino);
1595 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1596 de->inode = 0;
1597 de->rec_len = ext4_rec_len_to_disk(
1598 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1599 inline_size);
1600 set_nlink(inode, 2);
1601 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1602 out:
1603 brelse(iloc.bh);
1604 return ret;
1605 }
1606
1607 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1608 const struct qstr *d_name,
1609 struct ext4_dir_entry_2 **res_dir,
1610 int *has_inline_data)
1611 {
1612 int ret;
1613 struct ext4_iloc iloc;
1614 void *inline_start;
1615 int inline_size;
1616
1617 if (ext4_get_inode_loc(dir, &iloc))
1618 return NULL;
1619
1620 down_read(&EXT4_I(dir)->xattr_sem);
1621 if (!ext4_has_inline_data(dir)) {
1622 *has_inline_data = 0;
1623 goto out;
1624 }
1625
1626 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1627 EXT4_INLINE_DOTDOT_SIZE;
1628 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1629 ret = search_dir(iloc.bh, inline_start, inline_size,
1630 dir, d_name, 0, res_dir);
1631 if (ret == 1)
1632 goto out_find;
1633 if (ret < 0)
1634 goto out;
1635
1636 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1637 goto out;
1638
1639 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1640 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1641
1642 ret = search_dir(iloc.bh, inline_start, inline_size,
1643 dir, d_name, 0, res_dir);
1644 if (ret == 1)
1645 goto out_find;
1646
1647 out:
1648 brelse(iloc.bh);
1649 iloc.bh = NULL;
1650 out_find:
1651 up_read(&EXT4_I(dir)->xattr_sem);
1652 return iloc.bh;
1653 }
1654
1655 int ext4_delete_inline_entry(handle_t *handle,
1656 struct inode *dir,
1657 struct ext4_dir_entry_2 *de_del,
1658 struct buffer_head *bh,
1659 int *has_inline_data)
1660 {
1661 int err, inline_size;
1662 struct ext4_iloc iloc;
1663 void *inline_start;
1664
1665 err = ext4_get_inode_loc(dir, &iloc);
1666 if (err)
1667 return err;
1668
1669 down_write(&EXT4_I(dir)->xattr_sem);
1670 if (!ext4_has_inline_data(dir)) {
1671 *has_inline_data = 0;
1672 goto out;
1673 }
1674
1675 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1676 EXT4_MIN_INLINE_DATA_SIZE) {
1677 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1678 EXT4_INLINE_DOTDOT_SIZE;
1679 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1680 EXT4_INLINE_DOTDOT_SIZE;
1681 } else {
1682 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1683 inline_size = ext4_get_inline_size(dir) -
1684 EXT4_MIN_INLINE_DATA_SIZE;
1685 }
1686
1687 BUFFER_TRACE(bh, "get_write_access");
1688 err = ext4_journal_get_write_access(handle, bh);
1689 if (err)
1690 goto out;
1691
1692 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1693 inline_start, inline_size, 0);
1694 if (err)
1695 goto out;
1696
1697 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1698 err = ext4_mark_inode_dirty(handle, dir);
1699 if (unlikely(err))
1700 goto out;
1701
1702 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1703 out:
1704 up_write(&EXT4_I(dir)->xattr_sem);
1705 brelse(iloc.bh);
1706 if (err != -ENOENT)
1707 ext4_std_error(dir->i_sb, err);
1708 return err;
1709 }
1710
1711 /*
1712 * Get the inline dentry at offset.
1713 */
1714 static inline struct ext4_dir_entry_2 *
1715 ext4_get_inline_entry(struct inode *inode,
1716 struct ext4_iloc *iloc,
1717 unsigned int offset,
1718 void **inline_start,
1719 int *inline_size)
1720 {
1721 void *inline_pos;
1722
1723 BUG_ON(offset > ext4_get_inline_size(inode));
1724
1725 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1726 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1727 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1728 } else {
1729 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1730 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1731 *inline_size = ext4_get_inline_size(inode) -
1732 EXT4_MIN_INLINE_DATA_SIZE;
1733 }
1734
1735 if (inline_start)
1736 *inline_start = inline_pos;
1737 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1738 }
1739
1740 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1741 {
1742 int err, inline_size;
1743 struct ext4_iloc iloc;
1744 void *inline_pos;
1745 unsigned int offset;
1746 struct ext4_dir_entry_2 *de;
1747 int ret = 1;
1748
1749 err = ext4_get_inode_loc(dir, &iloc);
1750 if (err) {
1751 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1752 err, dir->i_ino);
1753 return 1;
1754 }
1755
1756 down_read(&EXT4_I(dir)->xattr_sem);
1757 if (!ext4_has_inline_data(dir)) {
1758 *has_inline_data = 0;
1759 goto out;
1760 }
1761
1762 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1763 if (!le32_to_cpu(de->inode)) {
1764 ext4_warning(dir->i_sb,
1765 "bad inline directory (dir #%lu) - no `..'",
1766 dir->i_ino);
1767 ret = 1;
1768 goto out;
1769 }
1770
1771 offset = EXT4_INLINE_DOTDOT_SIZE;
1772 while (offset < dir->i_size) {
1773 de = ext4_get_inline_entry(dir, &iloc, offset,
1774 &inline_pos, &inline_size);
1775 if (ext4_check_dir_entry(dir, NULL, de,
1776 iloc.bh, inline_pos,
1777 inline_size, offset)) {
1778 ext4_warning(dir->i_sb,
1779 "bad inline directory (dir #%lu) - "
1780 "inode %u, rec_len %u, name_len %d"
1781 "inline size %d\n",
1782 dir->i_ino, le32_to_cpu(de->inode),
1783 le16_to_cpu(de->rec_len), de->name_len,
1784 inline_size);
1785 ret = 1;
1786 goto out;
1787 }
1788 if (le32_to_cpu(de->inode)) {
1789 ret = 0;
1790 goto out;
1791 }
1792 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1793 }
1794
1795 out:
1796 up_read(&EXT4_I(dir)->xattr_sem);
1797 brelse(iloc.bh);
1798 return ret;
1799 }
1800
1801 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1802 {
1803 int ret;
1804
1805 down_write(&EXT4_I(inode)->xattr_sem);
1806 ret = ext4_destroy_inline_data_nolock(handle, inode);
1807 up_write(&EXT4_I(inode)->xattr_sem);
1808
1809 return ret;
1810 }
1811
1812 int ext4_inline_data_fiemap(struct inode *inode,
1813 struct fiemap_extent_info *fieinfo,
1814 int *has_inline, __u64 start, __u64 len)
1815 {
1816 __u64 physical = 0;
1817 __u64 inline_len;
1818 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1819 FIEMAP_EXTENT_LAST;
1820 int error = 0;
1821 struct ext4_iloc iloc;
1822
1823 down_read(&EXT4_I(inode)->xattr_sem);
1824 if (!ext4_has_inline_data(inode)) {
1825 *has_inline = 0;
1826 goto out;
1827 }
1828 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1829 i_size_read(inode));
1830 if (start >= inline_len)
1831 goto out;
1832 if (start + len < inline_len)
1833 inline_len = start + len;
1834 inline_len -= start;
1835
1836 error = ext4_get_inode_loc(inode, &iloc);
1837 if (error)
1838 goto out;
1839
1840 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1841 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1842 physical += offsetof(struct ext4_inode, i_block);
1843
1844 if (physical)
1845 error = fiemap_fill_next_extent(fieinfo, start, physical,
1846 inline_len, flags);
1847 brelse(iloc.bh);
1848 out:
1849 up_read(&EXT4_I(inode)->xattr_sem);
1850 return (error < 0 ? error : 0);
1851 }
1852
1853 /*
1854 * Called during xattr set, and if we can sparse space 'needed',
1855 * just create the extent tree evict the data to the outer block.
1856 *
1857 * We use jbd2 instead of page cache to move data to the 1st block
1858 * so that the whole transaction can be committed as a whole and
1859 * the data isn't lost because of the delayed page cache write.
1860 */
1861 int ext4_try_to_evict_inline_data(handle_t *handle,
1862 struct inode *inode,
1863 int needed)
1864 {
1865 int error;
1866 struct ext4_xattr_entry *entry;
1867 struct ext4_inode *raw_inode;
1868 struct ext4_iloc iloc;
1869
1870 error = ext4_get_inode_loc(inode, &iloc);
1871 if (error)
1872 return error;
1873
1874 raw_inode = ext4_raw_inode(&iloc);
1875 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1876 EXT4_I(inode)->i_inline_off);
1877 if (EXT4_XATTR_LEN(entry->e_name_len) +
1878 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1879 error = -ENOSPC;
1880 goto out;
1881 }
1882
1883 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1884 out:
1885 brelse(iloc.bh);
1886 return error;
1887 }
1888
1889 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1890 {
1891 handle_t *handle;
1892 int inline_size, value_len, needed_blocks;
1893 size_t i_size;
1894 void *value = NULL;
1895 struct ext4_xattr_ibody_find is = {
1896 .s = { .not_found = -ENODATA, },
1897 };
1898 struct ext4_xattr_info i = {
1899 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1900 .name = EXT4_XATTR_SYSTEM_DATA,
1901 };
1902
1903
1904 needed_blocks = ext4_writepage_trans_blocks(inode);
1905 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1906 if (IS_ERR(handle))
1907 return;
1908
1909 down_write(&EXT4_I(inode)->xattr_sem);
1910 if (!ext4_has_inline_data(inode)) {
1911 *has_inline = 0;
1912 ext4_journal_stop(handle);
1913 return;
1914 }
1915
1916 if (ext4_orphan_add(handle, inode))
1917 goto out;
1918
1919 if (ext4_get_inode_loc(inode, &is.iloc))
1920 goto out;
1921
1922 down_write(&EXT4_I(inode)->i_data_sem);
1923 i_size = inode->i_size;
1924 inline_size = ext4_get_inline_size(inode);
1925 EXT4_I(inode)->i_disksize = i_size;
1926
1927 if (i_size < inline_size) {
1928 /* Clear the content in the xattr space. */
1929 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1930 if (ext4_xattr_ibody_find(inode, &i, &is))
1931 goto out_error;
1932
1933 BUG_ON(is.s.not_found);
1934
1935 value_len = le32_to_cpu(is.s.here->e_value_size);
1936 value = kmalloc(value_len, GFP_NOFS);
1937 if (!value)
1938 goto out_error;
1939
1940 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1941 value, value_len))
1942 goto out_error;
1943
1944 i.value = value;
1945 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1946 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1947 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1948 goto out_error;
1949 }
1950
1951 /* Clear the content within i_blocks. */
1952 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1953 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1954 memset(p + i_size, 0,
1955 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1956 }
1957
1958 EXT4_I(inode)->i_inline_size = i_size <
1959 EXT4_MIN_INLINE_DATA_SIZE ?
1960 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1961 }
1962
1963 out_error:
1964 up_write(&EXT4_I(inode)->i_data_sem);
1965 out:
1966 brelse(is.iloc.bh);
1967 up_write(&EXT4_I(inode)->xattr_sem);
1968 kfree(value);
1969 if (inode->i_nlink)
1970 ext4_orphan_del(handle, inode);
1971
1972 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1973 ext4_mark_inode_dirty(handle, inode);
1974 if (IS_SYNC(inode))
1975 ext4_handle_sync(handle);
1976
1977 ext4_journal_stop(handle);
1978 return;
1979 }
1980
1981 int ext4_convert_inline_data(struct inode *inode)
1982 {
1983 int error, needed_blocks;
1984 handle_t *handle;
1985 struct ext4_iloc iloc;
1986
1987 if (!ext4_has_inline_data(inode)) {
1988 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1989 return 0;
1990 }
1991
1992 needed_blocks = ext4_writepage_trans_blocks(inode);
1993
1994 iloc.bh = NULL;
1995 error = ext4_get_inode_loc(inode, &iloc);
1996 if (error)
1997 return error;
1998
1999 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2000 if (IS_ERR(handle)) {
2001 error = PTR_ERR(handle);
2002 goto out_free;
2003 }
2004
2005 down_write(&EXT4_I(inode)->xattr_sem);
2006 if (!ext4_has_inline_data(inode)) {
2007 up_write(&EXT4_I(inode)->xattr_sem);
2008 goto out;
2009 }
2010
2011 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2012 up_write(&EXT4_I(inode)->xattr_sem);
2013 out:
2014 ext4_journal_stop(handle);
2015 out_free:
2016 brelse(iloc.bh);
2017 return error;
2018 }
Linux with added FPC-III support