]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/ext4/inline.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branches 'for-4.11/upstream-fixes', 'for-4.12/accutouch', 'for-4.12/cp2112...
[mirror_ubuntu-artful-kernel.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 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
219 return;
220
221 BUG_ON(!EXT4_I(inode)->i_inline_off);
222 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
223
224 raw_inode = ext4_raw_inode(iloc);
225 buffer += pos;
226
227 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
228 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
229 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
230 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
231
232 len -= cp_len;
233 buffer += cp_len;
234 pos += cp_len;
235 }
236
237 if (!len)
238 return;
239
240 pos -= EXT4_MIN_INLINE_DATA_SIZE;
241 header = IHDR(inode, raw_inode);
242 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
243 EXT4_I(inode)->i_inline_off);
244
245 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
246 buffer, len);
247 }
248
249 static int ext4_create_inline_data(handle_t *handle,
250 struct inode *inode, unsigned len)
251 {
252 int error;
253 void *value = NULL;
254 struct ext4_xattr_ibody_find is = {
255 .s = { .not_found = -ENODATA, },
256 };
257 struct ext4_xattr_info i = {
258 .name_index = EXT4_XATTR_INDEX_SYSTEM,
259 .name = EXT4_XATTR_SYSTEM_DATA,
260 };
261
262 error = ext4_get_inode_loc(inode, &is.iloc);
263 if (error)
264 return error;
265
266 BUFFER_TRACE(is.iloc.bh, "get_write_access");
267 error = ext4_journal_get_write_access(handle, is.iloc.bh);
268 if (error)
269 goto out;
270
271 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272 value = EXT4_ZERO_XATTR_VALUE;
273 len -= EXT4_MIN_INLINE_DATA_SIZE;
274 } else {
275 value = "";
276 len = 0;
277 }
278
279 /* Insert the the xttr entry. */
280 i.value = value;
281 i.value_len = len;
282
283 error = ext4_xattr_ibody_find(inode, &i, &is);
284 if (error)
285 goto out;
286
287 BUG_ON(!is.s.not_found);
288
289 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
290 if (error) {
291 if (error == -ENOSPC)
292 ext4_clear_inode_state(inode,
293 EXT4_STATE_MAY_INLINE_DATA);
294 goto out;
295 }
296
297 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298 0, EXT4_MIN_INLINE_DATA_SIZE);
299
300 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301 (void *)ext4_raw_inode(&is.iloc));
302 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
305 /*
306 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
307 * get cleared
308 */
309 ext4_set_inode_flags(inode);
310 get_bh(is.iloc.bh);
311 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
312
313 out:
314 brelse(is.iloc.bh);
315 return error;
316 }
317
318 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
319 unsigned int len)
320 {
321 int error;
322 void *value = NULL;
323 struct ext4_xattr_ibody_find is = {
324 .s = { .not_found = -ENODATA, },
325 };
326 struct ext4_xattr_info i = {
327 .name_index = EXT4_XATTR_INDEX_SYSTEM,
328 .name = EXT4_XATTR_SYSTEM_DATA,
329 };
330
331 /* If the old space is ok, write the data directly. */
332 if (len <= EXT4_I(inode)->i_inline_size)
333 return 0;
334
335 error = ext4_get_inode_loc(inode, &is.iloc);
336 if (error)
337 return error;
338
339 error = ext4_xattr_ibody_find(inode, &i, &is);
340 if (error)
341 goto out;
342
343 BUG_ON(is.s.not_found);
344
345 len -= EXT4_MIN_INLINE_DATA_SIZE;
346 value = kzalloc(len, GFP_NOFS);
347 if (!value) {
348 error = -ENOMEM;
349 goto out;
350 }
351
352 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
353 value, len);
354 if (error == -ENODATA)
355 goto out;
356
357 BUFFER_TRACE(is.iloc.bh, "get_write_access");
358 error = ext4_journal_get_write_access(handle, is.iloc.bh);
359 if (error)
360 goto out;
361
362 /* Update the xttr entry. */
363 i.value = value;
364 i.value_len = len;
365
366 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
367 if (error)
368 goto out;
369
370 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
371 (void *)ext4_raw_inode(&is.iloc));
372 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
373 le32_to_cpu(is.s.here->e_value_size);
374 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
375 get_bh(is.iloc.bh);
376 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
377
378 out:
379 kfree(value);
380 brelse(is.iloc.bh);
381 return error;
382 }
383
384 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
385 unsigned int len)
386 {
387 int ret, size, no_expand;
388 struct ext4_inode_info *ei = EXT4_I(inode);
389
390 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
391 return -ENOSPC;
392
393 size = ext4_get_max_inline_size(inode);
394 if (size < len)
395 return -ENOSPC;
396
397 ext4_write_lock_xattr(inode, &no_expand);
398
399 if (ei->i_inline_off)
400 ret = ext4_update_inline_data(handle, inode, len);
401 else
402 ret = ext4_create_inline_data(handle, inode, len);
403
404 ext4_write_unlock_xattr(inode, &no_expand);
405 return ret;
406 }
407
408 static int ext4_destroy_inline_data_nolock(handle_t *handle,
409 struct inode *inode)
410 {
411 struct ext4_inode_info *ei = EXT4_I(inode);
412 struct ext4_xattr_ibody_find is = {
413 .s = { .not_found = 0, },
414 };
415 struct ext4_xattr_info i = {
416 .name_index = EXT4_XATTR_INDEX_SYSTEM,
417 .name = EXT4_XATTR_SYSTEM_DATA,
418 .value = NULL,
419 .value_len = 0,
420 };
421 int error;
422
423 if (!ei->i_inline_off)
424 return 0;
425
426 error = ext4_get_inode_loc(inode, &is.iloc);
427 if (error)
428 return error;
429
430 error = ext4_xattr_ibody_find(inode, &i, &is);
431 if (error)
432 goto out;
433
434 BUFFER_TRACE(is.iloc.bh, "get_write_access");
435 error = ext4_journal_get_write_access(handle, is.iloc.bh);
436 if (error)
437 goto out;
438
439 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
440 if (error)
441 goto out;
442
443 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
444 0, EXT4_MIN_INLINE_DATA_SIZE);
445
446 if (ext4_has_feature_extents(inode->i_sb)) {
447 if (S_ISDIR(inode->i_mode) ||
448 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
449 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
450 ext4_ext_tree_init(handle, inode);
451 }
452 }
453 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
454 /*
455 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
456 * get set.
457 */
458 ext4_set_inode_flags(inode);
459
460 get_bh(is.iloc.bh);
461 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
462
463 EXT4_I(inode)->i_inline_off = 0;
464 EXT4_I(inode)->i_inline_size = 0;
465 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
466 out:
467 brelse(is.iloc.bh);
468 if (error == -ENODATA)
469 error = 0;
470 return error;
471 }
472
473 static int ext4_read_inline_page(struct inode *inode, struct page *page)
474 {
475 void *kaddr;
476 int ret = 0;
477 size_t len;
478 struct ext4_iloc iloc;
479
480 BUG_ON(!PageLocked(page));
481 BUG_ON(!ext4_has_inline_data(inode));
482 BUG_ON(page->index);
483
484 if (!EXT4_I(inode)->i_inline_off) {
485 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
486 inode->i_ino);
487 goto out;
488 }
489
490 ret = ext4_get_inode_loc(inode, &iloc);
491 if (ret)
492 goto out;
493
494 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
495 kaddr = kmap_atomic(page);
496 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
497 flush_dcache_page(page);
498 kunmap_atomic(kaddr);
499 zero_user_segment(page, len, PAGE_SIZE);
500 SetPageUptodate(page);
501 brelse(iloc.bh);
502
503 out:
504 return ret;
505 }
506
507 int ext4_readpage_inline(struct inode *inode, struct page *page)
508 {
509 int ret = 0;
510
511 down_read(&EXT4_I(inode)->xattr_sem);
512 if (!ext4_has_inline_data(inode)) {
513 up_read(&EXT4_I(inode)->xattr_sem);
514 return -EAGAIN;
515 }
516
517 /*
518 * Current inline data can only exist in the 1st page,
519 * So for all the other pages, just set them uptodate.
520 */
521 if (!page->index)
522 ret = ext4_read_inline_page(inode, page);
523 else if (!PageUptodate(page)) {
524 zero_user_segment(page, 0, PAGE_SIZE);
525 SetPageUptodate(page);
526 }
527
528 up_read(&EXT4_I(inode)->xattr_sem);
529
530 unlock_page(page);
531 return ret >= 0 ? 0 : ret;
532 }
533
534 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
535 struct inode *inode,
536 unsigned flags)
537 {
538 int ret, needed_blocks, no_expand;
539 handle_t *handle = NULL;
540 int retries = 0, sem_held = 0;
541 struct page *page = NULL;
542 unsigned from, to;
543 struct ext4_iloc iloc;
544
545 if (!ext4_has_inline_data(inode)) {
546 /*
547 * clear the flag so that no new write
548 * will trap here again.
549 */
550 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
551 return 0;
552 }
553
554 needed_blocks = ext4_writepage_trans_blocks(inode);
555
556 ret = ext4_get_inode_loc(inode, &iloc);
557 if (ret)
558 return ret;
559
560 retry:
561 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
562 if (IS_ERR(handle)) {
563 ret = PTR_ERR(handle);
564 handle = NULL;
565 goto out;
566 }
567
568 /* We cannot recurse into the filesystem as the transaction is already
569 * started */
570 flags |= AOP_FLAG_NOFS;
571
572 page = grab_cache_page_write_begin(mapping, 0, flags);
573 if (!page) {
574 ret = -ENOMEM;
575 goto out;
576 }
577
578 ext4_write_lock_xattr(inode, &no_expand);
579 sem_held = 1;
580 /* If some one has already done this for us, just exit. */
581 if (!ext4_has_inline_data(inode)) {
582 ret = 0;
583 goto out;
584 }
585
586 from = 0;
587 to = ext4_get_inline_size(inode);
588 if (!PageUptodate(page)) {
589 ret = ext4_read_inline_page(inode, page);
590 if (ret < 0)
591 goto out;
592 }
593
594 ret = ext4_destroy_inline_data_nolock(handle, inode);
595 if (ret)
596 goto out;
597
598 if (ext4_should_dioread_nolock(inode)) {
599 ret = __block_write_begin(page, from, to,
600 ext4_get_block_unwritten);
601 } else
602 ret = __block_write_begin(page, from, to, ext4_get_block);
603
604 if (!ret && ext4_should_journal_data(inode)) {
605 ret = ext4_walk_page_buffers(handle, page_buffers(page),
606 from, to, NULL,
607 do_journal_get_write_access);
608 }
609
610 if (ret) {
611 unlock_page(page);
612 put_page(page);
613 page = NULL;
614 ext4_orphan_add(handle, inode);
615 ext4_write_unlock_xattr(inode, &no_expand);
616 sem_held = 0;
617 ext4_journal_stop(handle);
618 handle = NULL;
619 ext4_truncate_failed_write(inode);
620 /*
621 * If truncate failed early the inode might
622 * still be on the orphan list; we need to
623 * make sure the inode is removed from the
624 * orphan list in that case.
625 */
626 if (inode->i_nlink)
627 ext4_orphan_del(NULL, inode);
628 }
629
630 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
631 goto retry;
632
633 if (page)
634 block_commit_write(page, from, to);
635 out:
636 if (page) {
637 unlock_page(page);
638 put_page(page);
639 }
640 if (sem_held)
641 ext4_write_unlock_xattr(inode, &no_expand);
642 if (handle)
643 ext4_journal_stop(handle);
644 brelse(iloc.bh);
645 return ret;
646 }
647
648 /*
649 * Try to write data in the inode.
650 * If the inode has inline data, check whether the new write can be
651 * in the inode also. If not, create the page the handle, move the data
652 * to the page make it update and let the later codes create extent for it.
653 */
654 int ext4_try_to_write_inline_data(struct address_space *mapping,
655 struct inode *inode,
656 loff_t pos, unsigned len,
657 unsigned flags,
658 struct page **pagep)
659 {
660 int ret;
661 handle_t *handle;
662 struct page *page;
663 struct ext4_iloc iloc;
664
665 if (pos + len > ext4_get_max_inline_size(inode))
666 goto convert;
667
668 ret = ext4_get_inode_loc(inode, &iloc);
669 if (ret)
670 return ret;
671
672 /*
673 * The possible write could happen in the inode,
674 * so try to reserve the space in inode first.
675 */
676 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
677 if (IS_ERR(handle)) {
678 ret = PTR_ERR(handle);
679 handle = NULL;
680 goto out;
681 }
682
683 ret = ext4_prepare_inline_data(handle, inode, pos + len);
684 if (ret && ret != -ENOSPC)
685 goto out;
686
687 /* We don't have space in inline inode, so convert it to extent. */
688 if (ret == -ENOSPC) {
689 ext4_journal_stop(handle);
690 brelse(iloc.bh);
691 goto convert;
692 }
693
694 flags |= AOP_FLAG_NOFS;
695
696 page = grab_cache_page_write_begin(mapping, 0, flags);
697 if (!page) {
698 ret = -ENOMEM;
699 goto out;
700 }
701
702 *pagep = page;
703 down_read(&EXT4_I(inode)->xattr_sem);
704 if (!ext4_has_inline_data(inode)) {
705 ret = 0;
706 unlock_page(page);
707 put_page(page);
708 goto out_up_read;
709 }
710
711 if (!PageUptodate(page)) {
712 ret = ext4_read_inline_page(inode, page);
713 if (ret < 0)
714 goto out_up_read;
715 }
716
717 ret = 1;
718 handle = NULL;
719 out_up_read:
720 up_read(&EXT4_I(inode)->xattr_sem);
721 out:
722 if (handle)
723 ext4_journal_stop(handle);
724 brelse(iloc.bh);
725 return ret;
726 convert:
727 return ext4_convert_inline_data_to_extent(mapping,
728 inode, flags);
729 }
730
731 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
732 unsigned copied, struct page *page)
733 {
734 int ret, no_expand;
735 void *kaddr;
736 struct ext4_iloc iloc;
737
738 if (unlikely(copied < len)) {
739 if (!PageUptodate(page)) {
740 copied = 0;
741 goto out;
742 }
743 }
744
745 ret = ext4_get_inode_loc(inode, &iloc);
746 if (ret) {
747 ext4_std_error(inode->i_sb, ret);
748 copied = 0;
749 goto out;
750 }
751
752 ext4_write_lock_xattr(inode, &no_expand);
753 BUG_ON(!ext4_has_inline_data(inode));
754
755 kaddr = kmap_atomic(page);
756 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
757 kunmap_atomic(kaddr);
758 SetPageUptodate(page);
759 /* clear page dirty so that writepages wouldn't work for us. */
760 ClearPageDirty(page);
761
762 ext4_write_unlock_xattr(inode, &no_expand);
763 brelse(iloc.bh);
764 out:
765 return copied;
766 }
767
768 struct buffer_head *
769 ext4_journalled_write_inline_data(struct inode *inode,
770 unsigned len,
771 struct page *page)
772 {
773 int ret, no_expand;
774 void *kaddr;
775 struct ext4_iloc iloc;
776
777 ret = ext4_get_inode_loc(inode, &iloc);
778 if (ret) {
779 ext4_std_error(inode->i_sb, ret);
780 return NULL;
781 }
782
783 ext4_write_lock_xattr(inode, &no_expand);
784 kaddr = kmap_atomic(page);
785 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
786 kunmap_atomic(kaddr);
787 ext4_write_unlock_xattr(inode, &no_expand);
788
789 return iloc.bh;
790 }
791
792 /*
793 * Try to make the page cache and handle ready for the inline data case.
794 * We can call this function in 2 cases:
795 * 1. The inode is created and the first write exceeds inline size. We can
796 * clear the inode state safely.
797 * 2. The inode has inline data, then we need to read the data, make it
798 * update and dirty so that ext4_da_writepages can handle it. We don't
799 * need to start the journal since the file's metatdata isn't changed now.
800 */
801 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
802 struct inode *inode,
803 unsigned flags,
804 void **fsdata)
805 {
806 int ret = 0, inline_size;
807 struct page *page;
808
809 page = grab_cache_page_write_begin(mapping, 0, flags);
810 if (!page)
811 return -ENOMEM;
812
813 down_read(&EXT4_I(inode)->xattr_sem);
814 if (!ext4_has_inline_data(inode)) {
815 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
816 goto out;
817 }
818
819 inline_size = ext4_get_inline_size(inode);
820
821 if (!PageUptodate(page)) {
822 ret = ext4_read_inline_page(inode, page);
823 if (ret < 0)
824 goto out;
825 }
826
827 ret = __block_write_begin(page, 0, inline_size,
828 ext4_da_get_block_prep);
829 if (ret) {
830 up_read(&EXT4_I(inode)->xattr_sem);
831 unlock_page(page);
832 put_page(page);
833 ext4_truncate_failed_write(inode);
834 return ret;
835 }
836
837 SetPageDirty(page);
838 SetPageUptodate(page);
839 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
840 *fsdata = (void *)CONVERT_INLINE_DATA;
841
842 out:
843 up_read(&EXT4_I(inode)->xattr_sem);
844 if (page) {
845 unlock_page(page);
846 put_page(page);
847 }
848 return ret;
849 }
850
851 /*
852 * Prepare the write for the inline data.
853 * If the the data can be written into the inode, we just read
854 * the page and make it uptodate, and start the journal.
855 * Otherwise read the page, makes it dirty so that it can be
856 * handle in writepages(the i_disksize update is left to the
857 * normal ext4_da_write_end).
858 */
859 int ext4_da_write_inline_data_begin(struct address_space *mapping,
860 struct inode *inode,
861 loff_t pos, unsigned len,
862 unsigned flags,
863 struct page **pagep,
864 void **fsdata)
865 {
866 int ret, inline_size;
867 handle_t *handle;
868 struct page *page;
869 struct ext4_iloc iloc;
870 int retries;
871
872 ret = ext4_get_inode_loc(inode, &iloc);
873 if (ret)
874 return ret;
875
876 retry_journal:
877 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
878 if (IS_ERR(handle)) {
879 ret = PTR_ERR(handle);
880 goto out;
881 }
882
883 inline_size = ext4_get_max_inline_size(inode);
884
885 ret = -ENOSPC;
886 if (inline_size >= pos + len) {
887 ret = ext4_prepare_inline_data(handle, inode, pos + len);
888 if (ret && ret != -ENOSPC)
889 goto out_journal;
890 }
891
892 /*
893 * We cannot recurse into the filesystem as the transaction
894 * is already started.
895 */
896 flags |= AOP_FLAG_NOFS;
897
898 if (ret == -ENOSPC) {
899 ret = ext4_da_convert_inline_data_to_extent(mapping,
900 inode,
901 flags,
902 fsdata);
903 ext4_journal_stop(handle);
904 if (ret == -ENOSPC &&
905 ext4_should_retry_alloc(inode->i_sb, &retries))
906 goto retry_journal;
907 goto out;
908 }
909
910
911 page = grab_cache_page_write_begin(mapping, 0, flags);
912 if (!page) {
913 ret = -ENOMEM;
914 goto out_journal;
915 }
916
917 down_read(&EXT4_I(inode)->xattr_sem);
918 if (!ext4_has_inline_data(inode)) {
919 ret = 0;
920 goto out_release_page;
921 }
922
923 if (!PageUptodate(page)) {
924 ret = ext4_read_inline_page(inode, page);
925 if (ret < 0)
926 goto out_release_page;
927 }
928
929 up_read(&EXT4_I(inode)->xattr_sem);
930 *pagep = page;
931 brelse(iloc.bh);
932 return 1;
933 out_release_page:
934 up_read(&EXT4_I(inode)->xattr_sem);
935 unlock_page(page);
936 put_page(page);
937 out_journal:
938 ext4_journal_stop(handle);
939 out:
940 brelse(iloc.bh);
941 return ret;
942 }
943
944 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
945 unsigned len, unsigned copied,
946 struct page *page)
947 {
948 int i_size_changed = 0;
949 int ret;
950
951 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
952 if (ret < 0) {
953 unlock_page(page);
954 put_page(page);
955 return ret;
956 }
957 copied = ret;
958
959 /*
960 * No need to use i_size_read() here, the i_size
961 * cannot change under us because we hold i_mutex.
962 *
963 * But it's important to update i_size while still holding page lock:
964 * page writeout could otherwise come in and zero beyond i_size.
965 */
966 if (pos+copied > inode->i_size) {
967 i_size_write(inode, pos+copied);
968 i_size_changed = 1;
969 }
970 unlock_page(page);
971 put_page(page);
972
973 /*
974 * Don't mark the inode dirty under page lock. First, it unnecessarily
975 * makes the holding time of page lock longer. Second, it forces lock
976 * ordering of page lock and transaction start for journaling
977 * filesystems.
978 */
979 if (i_size_changed)
980 mark_inode_dirty(inode);
981
982 return copied;
983 }
984
985 #ifdef INLINE_DIR_DEBUG
986 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
987 void *inline_start, int inline_size)
988 {
989 int offset;
990 unsigned short de_len;
991 struct ext4_dir_entry_2 *de = inline_start;
992 void *dlimit = inline_start + inline_size;
993
994 trace_printk("inode %lu\n", dir->i_ino);
995 offset = 0;
996 while ((void *)de < dlimit) {
997 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
998 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
999 offset, de_len, de->name_len, de->name,
1000 de->name_len, le32_to_cpu(de->inode));
1001 if (ext4_check_dir_entry(dir, NULL, de, bh,
1002 inline_start, inline_size, offset))
1003 BUG();
1004
1005 offset += de_len;
1006 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1007 }
1008 }
1009 #else
1010 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1011 #endif
1012
1013 /*
1014 * Add a new entry into a inline dir.
1015 * It will return -ENOSPC if no space is available, and -EIO
1016 * and -EEXIST if directory entry already exists.
1017 */
1018 static int ext4_add_dirent_to_inline(handle_t *handle,
1019 struct ext4_filename *fname,
1020 struct inode *dir,
1021 struct inode *inode,
1022 struct ext4_iloc *iloc,
1023 void *inline_start, int inline_size)
1024 {
1025 int err;
1026 struct ext4_dir_entry_2 *de;
1027
1028 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1029 inline_size, fname, &de);
1030 if (err)
1031 return err;
1032
1033 BUFFER_TRACE(iloc->bh, "get_write_access");
1034 err = ext4_journal_get_write_access(handle, iloc->bh);
1035 if (err)
1036 return err;
1037 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1038
1039 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1040
1041 /*
1042 * XXX shouldn't update any times until successful
1043 * completion of syscall, but too many callers depend
1044 * on this.
1045 *
1046 * XXX similarly, too many callers depend on
1047 * ext4_new_inode() setting the times, but error
1048 * recovery deletes the inode, so the worst that can
1049 * happen is that the times are slightly out of date
1050 * and/or different from the directory change time.
1051 */
1052 dir->i_mtime = dir->i_ctime = current_time(dir);
1053 ext4_update_dx_flag(dir);
1054 dir->i_version++;
1055 return 1;
1056 }
1057
1058 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1059 struct ext4_iloc *iloc)
1060 {
1061 struct ext4_xattr_entry *entry;
1062 struct ext4_xattr_ibody_header *header;
1063
1064 BUG_ON(!EXT4_I(inode)->i_inline_off);
1065
1066 header = IHDR(inode, ext4_raw_inode(iloc));
1067 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1068 EXT4_I(inode)->i_inline_off);
1069
1070 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1071 }
1072
1073 /* Set the final de to cover the whole block. */
1074 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1075 {
1076 struct ext4_dir_entry_2 *de, *prev_de;
1077 void *limit;
1078 int de_len;
1079
1080 de = (struct ext4_dir_entry_2 *)de_buf;
1081 if (old_size) {
1082 limit = de_buf + old_size;
1083 do {
1084 prev_de = de;
1085 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1086 de_buf += de_len;
1087 de = (struct ext4_dir_entry_2 *)de_buf;
1088 } while (de_buf < limit);
1089
1090 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1091 old_size, new_size);
1092 } else {
1093 /* this is just created, so create an empty entry. */
1094 de->inode = 0;
1095 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1096 }
1097 }
1098
1099 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1100 struct ext4_iloc *iloc)
1101 {
1102 int ret;
1103 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1104 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1105
1106 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1107 return -ENOSPC;
1108
1109 ret = ext4_update_inline_data(handle, dir,
1110 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1111 if (ret)
1112 return ret;
1113
1114 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1115 EXT4_I(dir)->i_inline_size -
1116 EXT4_MIN_INLINE_DATA_SIZE);
1117 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1118 return 0;
1119 }
1120
1121 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1122 struct ext4_iloc *iloc,
1123 void *buf, int inline_size)
1124 {
1125 ext4_create_inline_data(handle, inode, inline_size);
1126 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1127 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1128 }
1129
1130 static int ext4_finish_convert_inline_dir(handle_t *handle,
1131 struct inode *inode,
1132 struct buffer_head *dir_block,
1133 void *buf,
1134 int inline_size)
1135 {
1136 int err, csum_size = 0, header_size = 0;
1137 struct ext4_dir_entry_2 *de;
1138 struct ext4_dir_entry_tail *t;
1139 void *target = dir_block->b_data;
1140
1141 /*
1142 * First create "." and ".." and then copy the dir information
1143 * back to the block.
1144 */
1145 de = (struct ext4_dir_entry_2 *)target;
1146 de = ext4_init_dot_dotdot(inode, de,
1147 inode->i_sb->s_blocksize, csum_size,
1148 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1149 header_size = (void *)de - target;
1150
1151 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1152 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1153
1154 if (ext4_has_metadata_csum(inode->i_sb))
1155 csum_size = sizeof(struct ext4_dir_entry_tail);
1156
1157 inode->i_size = inode->i_sb->s_blocksize;
1158 i_size_write(inode, inode->i_sb->s_blocksize);
1159 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1160 ext4_update_final_de(dir_block->b_data,
1161 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1162 inode->i_sb->s_blocksize - csum_size);
1163
1164 if (csum_size) {
1165 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1166 inode->i_sb->s_blocksize);
1167 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1168 }
1169 set_buffer_uptodate(dir_block);
1170 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1171 if (err)
1172 goto out;
1173 set_buffer_verified(dir_block);
1174 out:
1175 return err;
1176 }
1177
1178 static int ext4_convert_inline_data_nolock(handle_t *handle,
1179 struct inode *inode,
1180 struct ext4_iloc *iloc)
1181 {
1182 int error;
1183 void *buf = NULL;
1184 struct buffer_head *data_bh = NULL;
1185 struct ext4_map_blocks map;
1186 int inline_size;
1187
1188 inline_size = ext4_get_inline_size(inode);
1189 buf = kmalloc(inline_size, GFP_NOFS);
1190 if (!buf) {
1191 error = -ENOMEM;
1192 goto out;
1193 }
1194
1195 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1196 if (error < 0)
1197 goto out;
1198
1199 /*
1200 * Make sure the inline directory entries pass checks before we try to
1201 * convert them, so that we avoid touching stuff that needs fsck.
1202 */
1203 if (S_ISDIR(inode->i_mode)) {
1204 error = ext4_check_all_de(inode, iloc->bh,
1205 buf + EXT4_INLINE_DOTDOT_SIZE,
1206 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1207 if (error)
1208 goto out;
1209 }
1210
1211 error = ext4_destroy_inline_data_nolock(handle, inode);
1212 if (error)
1213 goto out;
1214
1215 map.m_lblk = 0;
1216 map.m_len = 1;
1217 map.m_flags = 0;
1218 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1219 if (error < 0)
1220 goto out_restore;
1221 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1222 error = -EIO;
1223 goto out_restore;
1224 }
1225
1226 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1227 if (!data_bh) {
1228 error = -ENOMEM;
1229 goto out_restore;
1230 }
1231
1232 lock_buffer(data_bh);
1233 error = ext4_journal_get_create_access(handle, data_bh);
1234 if (error) {
1235 unlock_buffer(data_bh);
1236 error = -EIO;
1237 goto out_restore;
1238 }
1239 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1240
1241 if (!S_ISDIR(inode->i_mode)) {
1242 memcpy(data_bh->b_data, buf, inline_size);
1243 set_buffer_uptodate(data_bh);
1244 error = ext4_handle_dirty_metadata(handle,
1245 inode, data_bh);
1246 } else {
1247 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1248 buf, inline_size);
1249 }
1250
1251 unlock_buffer(data_bh);
1252 out_restore:
1253 if (error)
1254 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1255
1256 out:
1257 brelse(data_bh);
1258 kfree(buf);
1259 return error;
1260 }
1261
1262 /*
1263 * Try to add the new entry to the inline data.
1264 * If succeeds, return 0. If not, extended the inline dir and copied data to
1265 * the new created block.
1266 */
1267 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1268 struct inode *dir, struct inode *inode)
1269 {
1270 int ret, inline_size, no_expand;
1271 void *inline_start;
1272 struct ext4_iloc iloc;
1273
1274 ret = ext4_get_inode_loc(dir, &iloc);
1275 if (ret)
1276 return ret;
1277
1278 ext4_write_lock_xattr(dir, &no_expand);
1279 if (!ext4_has_inline_data(dir))
1280 goto out;
1281
1282 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1283 EXT4_INLINE_DOTDOT_SIZE;
1284 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1285
1286 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1287 inline_start, inline_size);
1288 if (ret != -ENOSPC)
1289 goto out;
1290
1291 /* check whether it can be inserted to inline xattr space. */
1292 inline_size = EXT4_I(dir)->i_inline_size -
1293 EXT4_MIN_INLINE_DATA_SIZE;
1294 if (!inline_size) {
1295 /* Try to use the xattr space.*/
1296 ret = ext4_update_inline_dir(handle, dir, &iloc);
1297 if (ret && ret != -ENOSPC)
1298 goto out;
1299
1300 inline_size = EXT4_I(dir)->i_inline_size -
1301 EXT4_MIN_INLINE_DATA_SIZE;
1302 }
1303
1304 if (inline_size) {
1305 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1306
1307 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1308 inode, &iloc, inline_start,
1309 inline_size);
1310
1311 if (ret != -ENOSPC)
1312 goto out;
1313 }
1314
1315 /*
1316 * The inline space is filled up, so create a new block for it.
1317 * As the extent tree will be created, we have to save the inline
1318 * dir first.
1319 */
1320 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1321
1322 out:
1323 ext4_write_unlock_xattr(dir, &no_expand);
1324 ext4_mark_inode_dirty(handle, dir);
1325 brelse(iloc.bh);
1326 return ret;
1327 }
1328
1329 /*
1330 * This function fills a red-black tree with information from an
1331 * inlined dir. It returns the number directory entries loaded
1332 * into the tree. If there is an error it is returned in err.
1333 */
1334 int htree_inlinedir_to_tree(struct file *dir_file,
1335 struct inode *dir, ext4_lblk_t block,
1336 struct dx_hash_info *hinfo,
1337 __u32 start_hash, __u32 start_minor_hash,
1338 int *has_inline_data)
1339 {
1340 int err = 0, count = 0;
1341 unsigned int parent_ino;
1342 int pos;
1343 struct ext4_dir_entry_2 *de;
1344 struct inode *inode = file_inode(dir_file);
1345 int ret, inline_size = 0;
1346 struct ext4_iloc iloc;
1347 void *dir_buf = NULL;
1348 struct ext4_dir_entry_2 fake;
1349 struct fscrypt_str tmp_str;
1350
1351 ret = ext4_get_inode_loc(inode, &iloc);
1352 if (ret)
1353 return ret;
1354
1355 down_read(&EXT4_I(inode)->xattr_sem);
1356 if (!ext4_has_inline_data(inode)) {
1357 up_read(&EXT4_I(inode)->xattr_sem);
1358 *has_inline_data = 0;
1359 goto out;
1360 }
1361
1362 inline_size = ext4_get_inline_size(inode);
1363 dir_buf = kmalloc(inline_size, GFP_NOFS);
1364 if (!dir_buf) {
1365 ret = -ENOMEM;
1366 up_read(&EXT4_I(inode)->xattr_sem);
1367 goto out;
1368 }
1369
1370 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1371 up_read(&EXT4_I(inode)->xattr_sem);
1372 if (ret < 0)
1373 goto out;
1374
1375 pos = 0;
1376 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1377 while (pos < inline_size) {
1378 /*
1379 * As inlined dir doesn't store any information about '.' and
1380 * only the inode number of '..' is stored, we have to handle
1381 * them differently.
1382 */
1383 if (pos == 0) {
1384 fake.inode = cpu_to_le32(inode->i_ino);
1385 fake.name_len = 1;
1386 strcpy(fake.name, ".");
1387 fake.rec_len = ext4_rec_len_to_disk(
1388 EXT4_DIR_REC_LEN(fake.name_len),
1389 inline_size);
1390 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1391 de = &fake;
1392 pos = EXT4_INLINE_DOTDOT_OFFSET;
1393 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1394 fake.inode = cpu_to_le32(parent_ino);
1395 fake.name_len = 2;
1396 strcpy(fake.name, "..");
1397 fake.rec_len = ext4_rec_len_to_disk(
1398 EXT4_DIR_REC_LEN(fake.name_len),
1399 inline_size);
1400 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1401 de = &fake;
1402 pos = EXT4_INLINE_DOTDOT_SIZE;
1403 } else {
1404 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1405 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1406 if (ext4_check_dir_entry(inode, dir_file, de,
1407 iloc.bh, dir_buf,
1408 inline_size, pos)) {
1409 ret = count;
1410 goto out;
1411 }
1412 }
1413
1414 ext4fs_dirhash(de->name, de->name_len, hinfo);
1415 if ((hinfo->hash < start_hash) ||
1416 ((hinfo->hash == start_hash) &&
1417 (hinfo->minor_hash < start_minor_hash)))
1418 continue;
1419 if (de->inode == 0)
1420 continue;
1421 tmp_str.name = de->name;
1422 tmp_str.len = de->name_len;
1423 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1424 hinfo->minor_hash, de, &tmp_str);
1425 if (err) {
1426 count = err;
1427 goto out;
1428 }
1429 count++;
1430 }
1431 ret = count;
1432 out:
1433 kfree(dir_buf);
1434 brelse(iloc.bh);
1435 return ret;
1436 }
1437
1438 /*
1439 * So this function is called when the volume is mkfsed with
1440 * dir_index disabled. In order to keep f_pos persistent
1441 * after we convert from an inlined dir to a blocked based,
1442 * we just pretend that we are a normal dir and return the
1443 * offset as if '.' and '..' really take place.
1444 *
1445 */
1446 int ext4_read_inline_dir(struct file *file,
1447 struct dir_context *ctx,
1448 int *has_inline_data)
1449 {
1450 unsigned int offset, parent_ino;
1451 int i;
1452 struct ext4_dir_entry_2 *de;
1453 struct super_block *sb;
1454 struct inode *inode = file_inode(file);
1455 int ret, inline_size = 0;
1456 struct ext4_iloc iloc;
1457 void *dir_buf = NULL;
1458 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1459
1460 ret = ext4_get_inode_loc(inode, &iloc);
1461 if (ret)
1462 return ret;
1463
1464 down_read(&EXT4_I(inode)->xattr_sem);
1465 if (!ext4_has_inline_data(inode)) {
1466 up_read(&EXT4_I(inode)->xattr_sem);
1467 *has_inline_data = 0;
1468 goto out;
1469 }
1470
1471 inline_size = ext4_get_inline_size(inode);
1472 dir_buf = kmalloc(inline_size, GFP_NOFS);
1473 if (!dir_buf) {
1474 ret = -ENOMEM;
1475 up_read(&EXT4_I(inode)->xattr_sem);
1476 goto out;
1477 }
1478
1479 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1480 up_read(&EXT4_I(inode)->xattr_sem);
1481 if (ret < 0)
1482 goto out;
1483
1484 ret = 0;
1485 sb = inode->i_sb;
1486 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1487 offset = ctx->pos;
1488
1489 /*
1490 * dotdot_offset and dotdot_size is the real offset and
1491 * size for ".." and "." if the dir is block based while
1492 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1493 * So we will use extra_offset and extra_size to indicate them
1494 * during the inline dir iteration.
1495 */
1496 dotdot_offset = EXT4_DIR_REC_LEN(1);
1497 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1498 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1499 extra_size = extra_offset + inline_size;
1500
1501 /*
1502 * If the version has changed since the last call to
1503 * readdir(2), then we might be pointing to an invalid
1504 * dirent right now. Scan from the start of the inline
1505 * dir to make sure.
1506 */
1507 if (file->f_version != inode->i_version) {
1508 for (i = 0; i < extra_size && i < offset;) {
1509 /*
1510 * "." is with offset 0 and
1511 * ".." is dotdot_offset.
1512 */
1513 if (!i) {
1514 i = dotdot_offset;
1515 continue;
1516 } else if (i == dotdot_offset) {
1517 i = dotdot_size;
1518 continue;
1519 }
1520 /* for other entry, the real offset in
1521 * the buf has to be tuned accordingly.
1522 */
1523 de = (struct ext4_dir_entry_2 *)
1524 (dir_buf + i - extra_offset);
1525 /* It's too expensive to do a full
1526 * dirent test each time round this
1527 * loop, but we do have to test at
1528 * least that it is non-zero. A
1529 * failure will be detected in the
1530 * dirent test below. */
1531 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1532 < EXT4_DIR_REC_LEN(1))
1533 break;
1534 i += ext4_rec_len_from_disk(de->rec_len,
1535 extra_size);
1536 }
1537 offset = i;
1538 ctx->pos = offset;
1539 file->f_version = inode->i_version;
1540 }
1541
1542 while (ctx->pos < extra_size) {
1543 if (ctx->pos == 0) {
1544 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1545 goto out;
1546 ctx->pos = dotdot_offset;
1547 continue;
1548 }
1549
1550 if (ctx->pos == dotdot_offset) {
1551 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1552 goto out;
1553 ctx->pos = dotdot_size;
1554 continue;
1555 }
1556
1557 de = (struct ext4_dir_entry_2 *)
1558 (dir_buf + ctx->pos - extra_offset);
1559 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1560 extra_size, ctx->pos))
1561 goto out;
1562 if (le32_to_cpu(de->inode)) {
1563 if (!dir_emit(ctx, de->name, de->name_len,
1564 le32_to_cpu(de->inode),
1565 get_dtype(sb, de->file_type)))
1566 goto out;
1567 }
1568 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1569 }
1570 out:
1571 kfree(dir_buf);
1572 brelse(iloc.bh);
1573 return ret;
1574 }
1575
1576 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1577 struct ext4_dir_entry_2 **parent_de,
1578 int *retval)
1579 {
1580 struct ext4_iloc iloc;
1581
1582 *retval = ext4_get_inode_loc(inode, &iloc);
1583 if (*retval)
1584 return NULL;
1585
1586 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1587
1588 return iloc.bh;
1589 }
1590
1591 /*
1592 * Try to create the inline data for the new dir.
1593 * If it succeeds, return 0, otherwise return the error.
1594 * In case of ENOSPC, the caller should create the normal disk layout dir.
1595 */
1596 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1597 struct inode *inode)
1598 {
1599 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1600 struct ext4_iloc iloc;
1601 struct ext4_dir_entry_2 *de;
1602
1603 ret = ext4_get_inode_loc(inode, &iloc);
1604 if (ret)
1605 return ret;
1606
1607 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1608 if (ret)
1609 goto out;
1610
1611 /*
1612 * For inline dir, we only save the inode information for the ".."
1613 * and create a fake dentry to cover the left space.
1614 */
1615 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1616 de->inode = cpu_to_le32(parent->i_ino);
1617 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1618 de->inode = 0;
1619 de->rec_len = ext4_rec_len_to_disk(
1620 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1621 inline_size);
1622 set_nlink(inode, 2);
1623 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1624 out:
1625 brelse(iloc.bh);
1626 return ret;
1627 }
1628
1629 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1630 struct ext4_filename *fname,
1631 const struct qstr *d_name,
1632 struct ext4_dir_entry_2 **res_dir,
1633 int *has_inline_data)
1634 {
1635 int ret;
1636 struct ext4_iloc iloc;
1637 void *inline_start;
1638 int inline_size;
1639
1640 if (ext4_get_inode_loc(dir, &iloc))
1641 return NULL;
1642
1643 down_read(&EXT4_I(dir)->xattr_sem);
1644 if (!ext4_has_inline_data(dir)) {
1645 *has_inline_data = 0;
1646 goto out;
1647 }
1648
1649 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1650 EXT4_INLINE_DOTDOT_SIZE;
1651 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1652 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1653 dir, fname, d_name, 0, res_dir);
1654 if (ret == 1)
1655 goto out_find;
1656 if (ret < 0)
1657 goto out;
1658
1659 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1660 goto out;
1661
1662 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1663 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1664
1665 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1666 dir, fname, d_name, 0, res_dir);
1667 if (ret == 1)
1668 goto out_find;
1669
1670 out:
1671 brelse(iloc.bh);
1672 iloc.bh = NULL;
1673 out_find:
1674 up_read(&EXT4_I(dir)->xattr_sem);
1675 return iloc.bh;
1676 }
1677
1678 int ext4_delete_inline_entry(handle_t *handle,
1679 struct inode *dir,
1680 struct ext4_dir_entry_2 *de_del,
1681 struct buffer_head *bh,
1682 int *has_inline_data)
1683 {
1684 int err, inline_size, no_expand;
1685 struct ext4_iloc iloc;
1686 void *inline_start;
1687
1688 err = ext4_get_inode_loc(dir, &iloc);
1689 if (err)
1690 return err;
1691
1692 ext4_write_lock_xattr(dir, &no_expand);
1693 if (!ext4_has_inline_data(dir)) {
1694 *has_inline_data = 0;
1695 goto out;
1696 }
1697
1698 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1699 EXT4_MIN_INLINE_DATA_SIZE) {
1700 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1701 EXT4_INLINE_DOTDOT_SIZE;
1702 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1703 EXT4_INLINE_DOTDOT_SIZE;
1704 } else {
1705 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1706 inline_size = ext4_get_inline_size(dir) -
1707 EXT4_MIN_INLINE_DATA_SIZE;
1708 }
1709
1710 BUFFER_TRACE(bh, "get_write_access");
1711 err = ext4_journal_get_write_access(handle, bh);
1712 if (err)
1713 goto out;
1714
1715 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1716 inline_start, inline_size, 0);
1717 if (err)
1718 goto out;
1719
1720 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1721 out:
1722 ext4_write_unlock_xattr(dir, &no_expand);
1723 if (likely(err == 0))
1724 err = ext4_mark_inode_dirty(handle, dir);
1725 brelse(iloc.bh);
1726 if (err != -ENOENT)
1727 ext4_std_error(dir->i_sb, err);
1728 return err;
1729 }
1730
1731 /*
1732 * Get the inline dentry at offset.
1733 */
1734 static inline struct ext4_dir_entry_2 *
1735 ext4_get_inline_entry(struct inode *inode,
1736 struct ext4_iloc *iloc,
1737 unsigned int offset,
1738 void **inline_start,
1739 int *inline_size)
1740 {
1741 void *inline_pos;
1742
1743 BUG_ON(offset > ext4_get_inline_size(inode));
1744
1745 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1746 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1747 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1748 } else {
1749 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1750 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1751 *inline_size = ext4_get_inline_size(inode) -
1752 EXT4_MIN_INLINE_DATA_SIZE;
1753 }
1754
1755 if (inline_start)
1756 *inline_start = inline_pos;
1757 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1758 }
1759
1760 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1761 {
1762 int err, inline_size;
1763 struct ext4_iloc iloc;
1764 void *inline_pos;
1765 unsigned int offset;
1766 struct ext4_dir_entry_2 *de;
1767 bool ret = true;
1768
1769 err = ext4_get_inode_loc(dir, &iloc);
1770 if (err) {
1771 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1772 err, dir->i_ino);
1773 return true;
1774 }
1775
1776 down_read(&EXT4_I(dir)->xattr_sem);
1777 if (!ext4_has_inline_data(dir)) {
1778 *has_inline_data = 0;
1779 goto out;
1780 }
1781
1782 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1783 if (!le32_to_cpu(de->inode)) {
1784 ext4_warning(dir->i_sb,
1785 "bad inline directory (dir #%lu) - no `..'",
1786 dir->i_ino);
1787 ret = true;
1788 goto out;
1789 }
1790
1791 offset = EXT4_INLINE_DOTDOT_SIZE;
1792 while (offset < dir->i_size) {
1793 de = ext4_get_inline_entry(dir, &iloc, offset,
1794 &inline_pos, &inline_size);
1795 if (ext4_check_dir_entry(dir, NULL, de,
1796 iloc.bh, inline_pos,
1797 inline_size, offset)) {
1798 ext4_warning(dir->i_sb,
1799 "bad inline directory (dir #%lu) - "
1800 "inode %u, rec_len %u, name_len %d"
1801 "inline size %d",
1802 dir->i_ino, le32_to_cpu(de->inode),
1803 le16_to_cpu(de->rec_len), de->name_len,
1804 inline_size);
1805 ret = true;
1806 goto out;
1807 }
1808 if (le32_to_cpu(de->inode)) {
1809 ret = false;
1810 goto out;
1811 }
1812 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1813 }
1814
1815 out:
1816 up_read(&EXT4_I(dir)->xattr_sem);
1817 brelse(iloc.bh);
1818 return ret;
1819 }
1820
1821 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1822 {
1823 int ret, no_expand;
1824
1825 ext4_write_lock_xattr(inode, &no_expand);
1826 ret = ext4_destroy_inline_data_nolock(handle, inode);
1827 ext4_write_unlock_xattr(inode, &no_expand);
1828
1829 return ret;
1830 }
1831
1832 int ext4_inline_data_fiemap(struct inode *inode,
1833 struct fiemap_extent_info *fieinfo,
1834 int *has_inline, __u64 start, __u64 len)
1835 {
1836 __u64 physical = 0;
1837 __u64 inline_len;
1838 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1839 FIEMAP_EXTENT_LAST;
1840 int error = 0;
1841 struct ext4_iloc iloc;
1842
1843 down_read(&EXT4_I(inode)->xattr_sem);
1844 if (!ext4_has_inline_data(inode)) {
1845 *has_inline = 0;
1846 goto out;
1847 }
1848 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1849 i_size_read(inode));
1850 if (start >= inline_len)
1851 goto out;
1852 if (start + len < inline_len)
1853 inline_len = start + len;
1854 inline_len -= start;
1855
1856 error = ext4_get_inode_loc(inode, &iloc);
1857 if (error)
1858 goto out;
1859
1860 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1861 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1862 physical += offsetof(struct ext4_inode, i_block);
1863
1864 if (physical)
1865 error = fiemap_fill_next_extent(fieinfo, start, physical,
1866 inline_len, flags);
1867 brelse(iloc.bh);
1868 out:
1869 up_read(&EXT4_I(inode)->xattr_sem);
1870 return (error < 0 ? error : 0);
1871 }
1872
1873 /*
1874 * Called during xattr set, and if we can sparse space 'needed',
1875 * just create the extent tree evict the data to the outer block.
1876 *
1877 * We use jbd2 instead of page cache to move data to the 1st block
1878 * so that the whole transaction can be committed as a whole and
1879 * the data isn't lost because of the delayed page cache write.
1880 */
1881 int ext4_try_to_evict_inline_data(handle_t *handle,
1882 struct inode *inode,
1883 int needed)
1884 {
1885 int error;
1886 struct ext4_xattr_entry *entry;
1887 struct ext4_inode *raw_inode;
1888 struct ext4_iloc iloc;
1889
1890 error = ext4_get_inode_loc(inode, &iloc);
1891 if (error)
1892 return error;
1893
1894 raw_inode = ext4_raw_inode(&iloc);
1895 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1896 EXT4_I(inode)->i_inline_off);
1897 if (EXT4_XATTR_LEN(entry->e_name_len) +
1898 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1899 error = -ENOSPC;
1900 goto out;
1901 }
1902
1903 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1904 out:
1905 brelse(iloc.bh);
1906 return error;
1907 }
1908
1909 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1910 {
1911 handle_t *handle;
1912 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1913 size_t i_size;
1914 void *value = NULL;
1915 struct ext4_xattr_ibody_find is = {
1916 .s = { .not_found = -ENODATA, },
1917 };
1918 struct ext4_xattr_info i = {
1919 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1920 .name = EXT4_XATTR_SYSTEM_DATA,
1921 };
1922
1923
1924 needed_blocks = ext4_writepage_trans_blocks(inode);
1925 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1926 if (IS_ERR(handle))
1927 return PTR_ERR(handle);
1928
1929 ext4_write_lock_xattr(inode, &no_expand);
1930 if (!ext4_has_inline_data(inode)) {
1931 *has_inline = 0;
1932 ext4_journal_stop(handle);
1933 return 0;
1934 }
1935
1936 if ((err = ext4_orphan_add(handle, inode)) != 0)
1937 goto out;
1938
1939 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1940 goto out;
1941
1942 down_write(&EXT4_I(inode)->i_data_sem);
1943 i_size = inode->i_size;
1944 inline_size = ext4_get_inline_size(inode);
1945 EXT4_I(inode)->i_disksize = i_size;
1946
1947 if (i_size < inline_size) {
1948 /* Clear the content in the xattr space. */
1949 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1950 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1951 goto out_error;
1952
1953 BUG_ON(is.s.not_found);
1954
1955 value_len = le32_to_cpu(is.s.here->e_value_size);
1956 value = kmalloc(value_len, GFP_NOFS);
1957 if (!value) {
1958 err = -ENOMEM;
1959 goto out_error;
1960 }
1961
1962 err = ext4_xattr_ibody_get(inode, i.name_index,
1963 i.name, value, value_len);
1964 if (err <= 0)
1965 goto out_error;
1966
1967 i.value = value;
1968 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1969 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1970 err = ext4_xattr_ibody_inline_set(handle, inode,
1971 &i, &is);
1972 if (err)
1973 goto out_error;
1974 }
1975
1976 /* Clear the content within i_blocks. */
1977 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1978 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1979 memset(p + i_size, 0,
1980 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1981 }
1982
1983 EXT4_I(inode)->i_inline_size = i_size <
1984 EXT4_MIN_INLINE_DATA_SIZE ?
1985 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1986 }
1987
1988 out_error:
1989 up_write(&EXT4_I(inode)->i_data_sem);
1990 out:
1991 brelse(is.iloc.bh);
1992 ext4_write_unlock_xattr(inode, &no_expand);
1993 kfree(value);
1994 if (inode->i_nlink)
1995 ext4_orphan_del(handle, inode);
1996
1997 if (err == 0) {
1998 inode->i_mtime = inode->i_ctime = current_time(inode);
1999 err = ext4_mark_inode_dirty(handle, inode);
2000 if (IS_SYNC(inode))
2001 ext4_handle_sync(handle);
2002 }
2003 ext4_journal_stop(handle);
2004 return err;
2005 }
2006
2007 int ext4_convert_inline_data(struct inode *inode)
2008 {
2009 int error, needed_blocks, no_expand;
2010 handle_t *handle;
2011 struct ext4_iloc iloc;
2012
2013 if (!ext4_has_inline_data(inode)) {
2014 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2015 return 0;
2016 }
2017
2018 needed_blocks = ext4_writepage_trans_blocks(inode);
2019
2020 iloc.bh = NULL;
2021 error = ext4_get_inode_loc(inode, &iloc);
2022 if (error)
2023 return error;
2024
2025 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2026 if (IS_ERR(handle)) {
2027 error = PTR_ERR(handle);
2028 goto out_free;
2029 }
2030
2031 ext4_write_lock_xattr(inode, &no_expand);
2032 if (ext4_has_inline_data(inode))
2033 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2034 ext4_write_unlock_xattr(inode, &no_expand);
2035 ext4_journal_stop(handle);
2036 out_free:
2037 brelse(iloc.bh);
2038 return error;
2039 }
Ubuntu Artful kernel mirror