1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/dir.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <asm/unaligned.h>
9 #include <linux/fs.h>
10 #include <linux/f2fs_fs.h>
11 #include <linux/sched/signal.h>
12 #include <linux/unicode.h>
13 #include "f2fs.h"
14 #include "node.h"
15 #include "acl.h"
16 #include "xattr.h"
17 #include <trace/events/f2fs.h>
18
19 #if IS_ENABLED(CONFIG_UNICODE)
20 extern struct kmem_cache *f2fs_cf_name_slab;
21 #endif
22
23 static unsigned long dir_blocks(struct inode *inode)
24 {
25 return ((unsigned long long) (i_size_read(inode) + PAGE_SIZE - 1))
26 >> PAGE_SHIFT;
27 }
28
29 static unsigned int dir_buckets(unsigned int level, int dir_level)
30 {
31 if (level + dir_level < MAX_DIR_HASH_DEPTH / 2)
32 return BIT(level + dir_level);
33 else
34 return MAX_DIR_BUCKETS;
35 }
36
37 static unsigned int bucket_blocks(unsigned int level)
38 {
39 if (level < MAX_DIR_HASH_DEPTH / 2)
40 return 2;
41 else
42 return 4;
43 }
44
45 #if IS_ENABLED(CONFIG_UNICODE)
46 /* If @dir is casefolded, initialize @fname->cf_name from @fname->usr_fname. */
47 int f2fs_init_casefolded_name(const struct inode *dir,
48 struct f2fs_filename *fname)
49 {
50 struct super_block *sb = dir->i_sb;
51 unsigned char *buf;
52 int len;
53
54 if (IS_CASEFOLDED(dir) &&
55 !is_dot_dotdot(fname->usr_fname->name, fname->usr_fname->len)) {
56 buf = f2fs_kmem_cache_alloc(f2fs_cf_name_slab,
57 GFP_NOFS, false, F2FS_SB(sb));
58 if (!buf)
59 return -ENOMEM;
60
61 len = utf8_casefold(sb->s_encoding, fname->usr_fname,
62 buf, F2FS_NAME_LEN);
63 if (len <= 0) {
64 kmem_cache_free(f2fs_cf_name_slab, buf);
65 if (sb_has_strict_encoding(sb))
66 return -EINVAL;
67 /* fall back to treating name as opaque byte sequence */
68 return 0;
69 }
70 fname->cf_name.name = buf;
71 fname->cf_name.len = len;
72 }
73
74 return 0;
75 }
76
77 void f2fs_free_casefolded_name(struct f2fs_filename *fname)
78 {
79 unsigned char *buf = (unsigned char *)fname->cf_name.name;
80
81 if (buf) {
82 kmem_cache_free(f2fs_cf_name_slab, buf);
83 fname->cf_name.name = NULL;
84 }
85 }
86 #endif /* CONFIG_UNICODE */
87
88 static int __f2fs_setup_filename(const struct inode *dir,
89 const struct fscrypt_name *crypt_name,
90 struct f2fs_filename *fname)
91 {
92 int err;
93
94 memset(fname, 0, sizeof(*fname));
95
96 fname->usr_fname = crypt_name->usr_fname;
97 fname->disk_name = crypt_name->disk_name;
98 #ifdef CONFIG_FS_ENCRYPTION
99 fname->crypto_buf = crypt_name->crypto_buf;
100 #endif
101 if (crypt_name->is_nokey_name) {
102 /* hash was decoded from the no-key name */
103 fname->hash = cpu_to_le32(crypt_name->hash);
104 } else {
105 err = f2fs_init_casefolded_name(dir, fname);
106 if (err) {
107 f2fs_free_filename(fname);
108 return err;
109 }
110 f2fs_hash_filename(dir, fname);
111 }
112 return 0;
113 }
114
115 /*
116 * Prepare to search for @iname in @dir. This is similar to
117 * fscrypt_setup_filename(), but this also handles computing the casefolded name
118 * and the f2fs dirhash if needed, then packing all the information about this
119 * filename up into a 'struct f2fs_filename'.
120 */
121 int f2fs_setup_filename(struct inode *dir, const struct qstr *iname,
122 int lookup, struct f2fs_filename *fname)
123 {
124 struct fscrypt_name crypt_name;
125 int err;
126
127 err = fscrypt_setup_filename(dir, iname, lookup, &crypt_name);
128 if (err)
129 return err;
130
131 return __f2fs_setup_filename(dir, &crypt_name, fname);
132 }
133
134 /*
135 * Prepare to look up @dentry in @dir. This is similar to
136 * fscrypt_prepare_lookup(), but this also handles computing the casefolded name
137 * and the f2fs dirhash if needed, then packing all the information about this
138 * filename up into a 'struct f2fs_filename'.
139 */
140 int f2fs_prepare_lookup(struct inode *dir, struct dentry *dentry,
141 struct f2fs_filename *fname)
142 {
143 struct fscrypt_name crypt_name;
144 int err;
145
146 err = fscrypt_prepare_lookup(dir, dentry, &crypt_name);
147 if (err)
148 return err;
149
150 return __f2fs_setup_filename(dir, &crypt_name, fname);
151 }
152
153 void f2fs_free_filename(struct f2fs_filename *fname)
154 {
155 #ifdef CONFIG_FS_ENCRYPTION
156 kfree(fname->crypto_buf.name);
157 fname->crypto_buf.name = NULL;
158 #endif
159 f2fs_free_casefolded_name(fname);
160 }
161
162 static unsigned long dir_block_index(unsigned int level,
163 int dir_level, unsigned int idx)
164 {
165 unsigned long i;
166 unsigned long bidx = 0;
167
168 for (i = 0; i < level; i++)
169 bidx += mul_u32_u32(dir_buckets(i, dir_level),
170 bucket_blocks(i));
171 bidx += idx * bucket_blocks(level);
172 return bidx;
173 }
174
175 static struct f2fs_dir_entry *find_in_block(struct inode *dir,
176 struct page *dentry_page,
177 const struct f2fs_filename *fname,
178 int *max_slots)
179 {
180 struct f2fs_dentry_block *dentry_blk;
181 struct f2fs_dentry_ptr d;
182
183 dentry_blk = (struct f2fs_dentry_block *)page_address(dentry_page);
184
185 make_dentry_ptr_block(dir, &d, dentry_blk);
186 return f2fs_find_target_dentry(&d, fname, max_slots);
187 }
188
189 static inline int f2fs_match_name(const struct inode *dir,
190 const struct f2fs_filename *fname,
191 const u8 *de_name, u32 de_name_len)
192 {
193 struct fscrypt_name f;
194
195 #if IS_ENABLED(CONFIG_UNICODE)
196 if (fname->cf_name.name)
197 return generic_ci_match(dir, fname->usr_fname,
198 &fname->cf_name,
199 de_name, de_name_len);
200
201 #endif
202 f.usr_fname = fname->usr_fname;
203 f.disk_name = fname->disk_name;
204 #ifdef CONFIG_FS_ENCRYPTION
205 f.crypto_buf = fname->crypto_buf;
206 #endif
207 return fscrypt_match_name(&f, de_name, de_name_len);
208 }
209
210 struct f2fs_dir_entry *f2fs_find_target_dentry(const struct f2fs_dentry_ptr *d,
211 const struct f2fs_filename *fname, int *max_slots)
212 {
213 struct f2fs_dir_entry *de;
214 unsigned long bit_pos = 0;
215 int max_len = 0;
216 int res = 0;
217
218 if (max_slots)
219 *max_slots = 0;
220 while (bit_pos < d->max) {
221 if (!test_bit_le(bit_pos, d->bitmap)) {
222 bit_pos++;
223 max_len++;
224 continue;
225 }
226
227 de = &d->dentry[bit_pos];
228
229 if (unlikely(!de->name_len)) {
230 bit_pos++;
231 continue;
232 }
233
234 if (de->hash_code == fname->hash) {
235 res = f2fs_match_name(d->inode, fname,
236 d->filename[bit_pos],
237 le16_to_cpu(de->name_len));
238 if (res < 0)
239 return ERR_PTR(res);
240 if (res)
241 goto found;
242 }
243
244 if (max_slots && max_len > *max_slots)
245 *max_slots = max_len;
246 max_len = 0;
247
248 bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
249 }
250
251 de = NULL;
252 found:
253 if (max_slots && max_len > *max_slots)
254 *max_slots = max_len;
255 return de;
256 }
257
258 static struct f2fs_dir_entry *find_in_level(struct inode *dir,
259 unsigned int level,
260 const struct f2fs_filename *fname,
261 struct page **res_page)
262 {
263 int s = GET_DENTRY_SLOTS(fname->disk_name.len);
264 unsigned int nbucket, nblock;
265 unsigned int bidx, end_block;
266 struct page *dentry_page;
267 struct f2fs_dir_entry *de = NULL;
268 pgoff_t next_pgofs;
269 bool room = false;
270 int max_slots;
271
272 nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
273 nblock = bucket_blocks(level);
274
275 bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
276 le32_to_cpu(fname->hash) % nbucket);
277 end_block = bidx + nblock;
278
279 while (bidx < end_block) {
280 /* no need to allocate new dentry pages to all the indices */
281 dentry_page = f2fs_find_data_page(dir, bidx, &next_pgofs);
282 if (IS_ERR(dentry_page)) {
283 if (PTR_ERR(dentry_page) == -ENOENT) {
284 room = true;
285 bidx = next_pgofs;
286 continue;
287 } else {
288 *res_page = dentry_page;
289 break;
290 }
291 }
292
293 de = find_in_block(dir, dentry_page, fname, &max_slots);
294 if (IS_ERR(de)) {
295 *res_page = ERR_CAST(de);
296 de = NULL;
297 break;
298 } else if (de) {
299 *res_page = dentry_page;
300 break;
301 }
302
303 if (max_slots >= s)
304 room = true;
305 f2fs_put_page(dentry_page, 0);
306
307 bidx++;
308 }
309
310 if (!de && room && F2FS_I(dir)->chash != fname->hash) {
311 F2FS_I(dir)->chash = fname->hash;
312 F2FS_I(dir)->clevel = level;
313 }
314
315 return de;
316 }
317
318 struct f2fs_dir_entry *__f2fs_find_entry(struct inode *dir,
319 const struct f2fs_filename *fname,
320 struct page **res_page)
321 {
322 unsigned long npages = dir_blocks(dir);
323 struct f2fs_dir_entry *de = NULL;
324 unsigned int max_depth;
325 unsigned int level;
326
327 *res_page = NULL;
328
329 if (f2fs_has_inline_dentry(dir)) {
330 de = f2fs_find_in_inline_dir(dir, fname, res_page);
331 goto out;
332 }
333
334 if (npages == 0)
335 goto out;
336
337 max_depth = F2FS_I(dir)->i_current_depth;
338 if (unlikely(max_depth > MAX_DIR_HASH_DEPTH)) {
339 f2fs_warn(F2FS_I_SB(dir), "Corrupted max_depth of %lu: %u",
340 dir->i_ino, max_depth);
341 max_depth = MAX_DIR_HASH_DEPTH;
342 f2fs_i_depth_write(dir, max_depth);
343 }
344
345 for (level = 0; level < max_depth; level++) {
346 de = find_in_level(dir, level, fname, res_page);
347 if (de || IS_ERR(*res_page))
348 break;
349 }
350 out:
351 /* This is to increase the speed of f2fs_create */
352 if (!de)
353 F2FS_I(dir)->task = current;
354 return de;
355 }
356
357 /*
358 * Find an entry in the specified directory with the wanted name.
359 * It returns the page where the entry was found (as a parameter - res_page),
360 * and the entry itself. Page is returned mapped and unlocked.
361 * Entry is guaranteed to be valid.
362 */
363 struct f2fs_dir_entry *f2fs_find_entry(struct inode *dir,
364 const struct qstr *child, struct page **res_page)
365 {
366 struct f2fs_dir_entry *de = NULL;
367 struct f2fs_filename fname;
368 int err;
369
370 err = f2fs_setup_filename(dir, child, 1, &fname);
371 if (err) {
372 if (err == -ENOENT)
373 *res_page = NULL;
374 else
375 *res_page = ERR_PTR(err);
376 return NULL;
377 }
378
379 de = __f2fs_find_entry(dir, &fname, res_page);
380
381 f2fs_free_filename(&fname);
382 return de;
383 }
384
385 struct f2fs_dir_entry *f2fs_parent_dir(struct inode *dir, struct page **p)
386 {
387 return f2fs_find_entry(dir, &dotdot_name, p);
388 }
389
390 ino_t f2fs_inode_by_name(struct inode *dir, const struct qstr *qstr,
391 struct page **page)
392 {
393 ino_t res = 0;
394 struct f2fs_dir_entry *de;
395
396 de = f2fs_find_entry(dir, qstr, page);
397 if (de) {
398 res = le32_to_cpu(de->ino);
399 f2fs_put_page(*page, 0);
400 }
401
402 return res;
403 }
404
405 void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de,
406 struct page *page, struct inode *inode)
407 {
408 enum page_type type = f2fs_has_inline_dentry(dir) ? NODE : DATA;
409
410 lock_page(page);
411 f2fs_wait_on_page_writeback(page, type, true, true);
412 de->ino = cpu_to_le32(inode->i_ino);
413 de->file_type = fs_umode_to_ftype(inode->i_mode);
414 set_page_dirty(page);
415
416 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
417 f2fs_mark_inode_dirty_sync(dir, false);
418 f2fs_put_page(page, 1);
419 }
420
421 static void init_dent_inode(struct inode *dir, struct inode *inode,
422 const struct f2fs_filename *fname,
423 struct page *ipage)
424 {
425 struct f2fs_inode *ri;
426
427 if (!fname) /* tmpfile case? */
428 return;
429
430 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
431
432 /* copy name info. to this inode page */
433 ri = F2FS_INODE(ipage);
434 ri->i_namelen = cpu_to_le32(fname->disk_name.len);
435 memcpy(ri->i_name, fname->disk_name.name, fname->disk_name.len);
436 if (IS_ENCRYPTED(dir)) {
437 file_set_enc_name(inode);
438 /*
439 * Roll-forward recovery doesn't have encryption keys available,
440 * so it can't compute the dirhash for encrypted+casefolded
441 * filenames. Append it to i_name if possible. Else, disable
442 * roll-forward recovery of the dentry (i.e., make fsync'ing the
443 * file force a checkpoint) by setting LOST_PINO.
444 */
445 if (IS_CASEFOLDED(dir)) {
446 if (fname->disk_name.len + sizeof(f2fs_hash_t) <=
447 F2FS_NAME_LEN)
448 put_unaligned(fname->hash, (f2fs_hash_t *)
449 &ri->i_name[fname->disk_name.len]);
450 else
451 file_lost_pino(inode);
452 }
453 }
454 set_page_dirty(ipage);
455 }
456
457 void f2fs_do_make_empty_dir(struct inode *inode, struct inode *parent,
458 struct f2fs_dentry_ptr *d)
459 {
460 struct fscrypt_str dot = FSTR_INIT(".", 1);
461 struct fscrypt_str dotdot = FSTR_INIT("..", 2);
462
463 /* update dirent of "." */
464 f2fs_update_dentry(inode->i_ino, inode->i_mode, d, &dot, 0, 0);
465
466 /* update dirent of ".." */
467 f2fs_update_dentry(parent->i_ino, parent->i_mode, d, &dotdot, 0, 1);
468 }
469
470 static int make_empty_dir(struct inode *inode,
471 struct inode *parent, struct page *page)
472 {
473 struct page *dentry_page;
474 struct f2fs_dentry_block *dentry_blk;
475 struct f2fs_dentry_ptr d;
476
477 if (f2fs_has_inline_dentry(inode))
478 return f2fs_make_empty_inline_dir(inode, parent, page);
479
480 dentry_page = f2fs_get_new_data_page(inode, page, 0, true);
481 if (IS_ERR(dentry_page))
482 return PTR_ERR(dentry_page);
483
484 dentry_blk = page_address(dentry_page);
485
486 make_dentry_ptr_block(NULL, &d, dentry_blk);
487 f2fs_do_make_empty_dir(inode, parent, &d);
488
489 set_page_dirty(dentry_page);
490 f2fs_put_page(dentry_page, 1);
491 return 0;
492 }
493
494 struct page *f2fs_init_inode_metadata(struct inode *inode, struct inode *dir,
495 const struct f2fs_filename *fname, struct page *dpage)
496 {
497 struct page *page;
498 int err;
499
500 if (is_inode_flag_set(inode, FI_NEW_INODE)) {
501 page = f2fs_new_inode_page(inode);
502 if (IS_ERR(page))
503 return page;
504
505 if (S_ISDIR(inode->i_mode)) {
506 /* in order to handle error case */
507 get_page(page);
508 err = make_empty_dir(inode, dir, page);
509 if (err) {
510 lock_page(page);
511 goto put_error;
512 }
513 put_page(page);
514 }
515
516 err = f2fs_init_acl(inode, dir, page, dpage);
517 if (err)
518 goto put_error;
519
520 err = f2fs_init_security(inode, dir,
521 fname ? fname->usr_fname : NULL, page);
522 if (err)
523 goto put_error;
524
525 if (IS_ENCRYPTED(inode)) {
526 err = fscrypt_set_context(inode, page);
527 if (err)
528 goto put_error;
529 }
530 } else {
531 page = f2fs_get_node_page(F2FS_I_SB(dir), inode->i_ino);
532 if (IS_ERR(page))
533 return page;
534 }
535
536 init_dent_inode(dir, inode, fname, page);
537
538 /*
539 * This file should be checkpointed during fsync.
540 * We lost i_pino from now on.
541 */
542 if (is_inode_flag_set(inode, FI_INC_LINK)) {
543 if (!S_ISDIR(inode->i_mode))
544 file_lost_pino(inode);
545 /*
546 * If link the tmpfile to alias through linkat path,
547 * we should remove this inode from orphan list.
548 */
549 if (inode->i_nlink == 0)
550 f2fs_remove_orphan_inode(F2FS_I_SB(dir), inode->i_ino);
551 f2fs_i_links_write(inode, true);
552 }
553 return page;
554
555 put_error:
556 clear_nlink(inode);
557 f2fs_update_inode(inode, page);
558 f2fs_put_page(page, 1);
559 return ERR_PTR(err);
560 }
561
562 void f2fs_update_parent_metadata(struct inode *dir, struct inode *inode,
563 unsigned int current_depth)
564 {
565 if (inode && is_inode_flag_set(inode, FI_NEW_INODE)) {
566 if (S_ISDIR(inode->i_mode))
567 f2fs_i_links_write(dir, true);
568 clear_inode_flag(inode, FI_NEW_INODE);
569 }
570 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
571 f2fs_mark_inode_dirty_sync(dir, false);
572
573 if (F2FS_I(dir)->i_current_depth != current_depth)
574 f2fs_i_depth_write(dir, current_depth);
575
576 if (inode && is_inode_flag_set(inode, FI_INC_LINK))
577 clear_inode_flag(inode, FI_INC_LINK);
578 }
579
580 int f2fs_room_for_filename(const void *bitmap, int slots, int max_slots)
581 {
582 int bit_start = 0;
583 int zero_start, zero_end;
584 next:
585 zero_start = find_next_zero_bit_le(bitmap, max_slots, bit_start);
586 if (zero_start >= max_slots)
587 return max_slots;
588
589 zero_end = find_next_bit_le(bitmap, max_slots, zero_start);
590 if (zero_end - zero_start >= slots)
591 return zero_start;
592
593 bit_start = zero_end + 1;
594
595 if (zero_end + 1 >= max_slots)
596 return max_slots;
597 goto next;
598 }
599
600 bool f2fs_has_enough_room(struct inode *dir, struct page *ipage,
601 const struct f2fs_filename *fname)
602 {
603 struct f2fs_dentry_ptr d;
604 unsigned int bit_pos;
605 int slots = GET_DENTRY_SLOTS(fname->disk_name.len);
606
607 make_dentry_ptr_inline(dir, &d, inline_data_addr(dir, ipage));
608
609 bit_pos = f2fs_room_for_filename(d.bitmap, slots, d.max);
610
611 return bit_pos < d.max;
612 }
613
614 void f2fs_update_dentry(nid_t ino, umode_t mode, struct f2fs_dentry_ptr *d,
615 const struct fscrypt_str *name, f2fs_hash_t name_hash,
616 unsigned int bit_pos)
617 {
618 struct f2fs_dir_entry *de;
619 int slots = GET_DENTRY_SLOTS(name->len);
620 int i;
621
622 de = &d->dentry[bit_pos];
623 de->hash_code = name_hash;
624 de->name_len = cpu_to_le16(name->len);
625 memcpy(d->filename[bit_pos], name->name, name->len);
626 de->ino = cpu_to_le32(ino);
627 de->file_type = fs_umode_to_ftype(mode);
628 for (i = 0; i < slots; i++) {
629 __set_bit_le(bit_pos + i, (void *)d->bitmap);
630 /* avoid wrong garbage data for readdir */
631 if (i)
632 (de + i)->name_len = 0;
633 }
634 }
635
636 int f2fs_add_regular_entry(struct inode *dir, const struct f2fs_filename *fname,
637 struct inode *inode, nid_t ino, umode_t mode)
638 {
639 unsigned int bit_pos;
640 unsigned int level;
641 unsigned int current_depth;
642 unsigned long bidx, block;
643 unsigned int nbucket, nblock;
644 struct page *dentry_page = NULL;
645 struct f2fs_dentry_block *dentry_blk = NULL;
646 struct f2fs_dentry_ptr d;
647 struct page *page = NULL;
648 int slots, err = 0;
649
650 level = 0;
651 slots = GET_DENTRY_SLOTS(fname->disk_name.len);
652
653 current_depth = F2FS_I(dir)->i_current_depth;
654 if (F2FS_I(dir)->chash == fname->hash) {
655 level = F2FS_I(dir)->clevel;
656 F2FS_I(dir)->chash = 0;
657 }
658
659 start:
660 if (time_to_inject(F2FS_I_SB(dir), FAULT_DIR_DEPTH))
661 return -ENOSPC;
662
663 if (unlikely(current_depth == MAX_DIR_HASH_DEPTH))
664 return -ENOSPC;
665
666 /* Increase the depth, if required */
667 if (level == current_depth)
668 ++current_depth;
669
670 nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
671 nblock = bucket_blocks(level);
672
673 bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
674 (le32_to_cpu(fname->hash) % nbucket));
675
676 for (block = bidx; block <= (bidx + nblock - 1); block++) {
677 dentry_page = f2fs_get_new_data_page(dir, NULL, block, true);
678 if (IS_ERR(dentry_page))
679 return PTR_ERR(dentry_page);
680
681 dentry_blk = page_address(dentry_page);
682 bit_pos = f2fs_room_for_filename(&dentry_blk->dentry_bitmap,
683 slots, NR_DENTRY_IN_BLOCK);
684 if (bit_pos < NR_DENTRY_IN_BLOCK)
685 goto add_dentry;
686
687 f2fs_put_page(dentry_page, 1);
688 }
689
690 /* Move to next level to find the empty slot for new dentry */
691 ++level;
692 goto start;
693 add_dentry:
694 f2fs_wait_on_page_writeback(dentry_page, DATA, true, true);
695
696 if (inode) {
697 f2fs_down_write(&F2FS_I(inode)->i_sem);
698 page = f2fs_init_inode_metadata(inode, dir, fname, NULL);
699 if (IS_ERR(page)) {
700 err = PTR_ERR(page);
701 goto fail;
702 }
703 }
704
705 make_dentry_ptr_block(NULL, &d, dentry_blk);
706 f2fs_update_dentry(ino, mode, &d, &fname->disk_name, fname->hash,
707 bit_pos);
708
709 set_page_dirty(dentry_page);
710
711 if (inode) {
712 f2fs_i_pino_write(inode, dir->i_ino);
713
714 /* synchronize inode page's data from inode cache */
715 if (is_inode_flag_set(inode, FI_NEW_INODE))
716 f2fs_update_inode(inode, page);
717
718 f2fs_put_page(page, 1);
719 }
720
721 f2fs_update_parent_metadata(dir, inode, current_depth);
722 fail:
723 if (inode)
724 f2fs_up_write(&F2FS_I(inode)->i_sem);
725
726 f2fs_put_page(dentry_page, 1);
727
728 return err;
729 }
730
731 int f2fs_add_dentry(struct inode *dir, const struct f2fs_filename *fname,
732 struct inode *inode, nid_t ino, umode_t mode)
733 {
734 int err = -EAGAIN;
735
736 if (f2fs_has_inline_dentry(dir)) {
737 /*
738 * Should get i_xattr_sem to keep the lock order:
739 * i_xattr_sem -> inode_page lock used by f2fs_setxattr.
740 */
741 f2fs_down_read(&F2FS_I(dir)->i_xattr_sem);
742 err = f2fs_add_inline_entry(dir, fname, inode, ino, mode);
743 f2fs_up_read(&F2FS_I(dir)->i_xattr_sem);
744 }
745 if (err == -EAGAIN)
746 err = f2fs_add_regular_entry(dir, fname, inode, ino, mode);
747
748 f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
749 return err;
750 }
751
752 /*
753 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
754 * f2fs_unlock_op().
755 */
756 int f2fs_do_add_link(struct inode *dir, const struct qstr *name,
757 struct inode *inode, nid_t ino, umode_t mode)
758 {
759 struct f2fs_filename fname;
760 struct page *page = NULL;
761 struct f2fs_dir_entry *de = NULL;
762 int err;
763
764 err = f2fs_setup_filename(dir, name, 0, &fname);
765 if (err)
766 return err;
767
768 /*
769 * An immature stackable filesystem shows a race condition between lookup
770 * and create. If we have same task when doing lookup and create, it's
771 * definitely fine as expected by VFS normally. Otherwise, let's just
772 * verify on-disk dentry one more time, which guarantees filesystem
773 * consistency more.
774 */
775 if (current != F2FS_I(dir)->task) {
776 de = __f2fs_find_entry(dir, &fname, &page);
777 F2FS_I(dir)->task = NULL;
778 }
779 if (de) {
780 f2fs_put_page(page, 0);
781 err = -EEXIST;
782 } else if (IS_ERR(page)) {
783 err = PTR_ERR(page);
784 } else {
785 err = f2fs_add_dentry(dir, &fname, inode, ino, mode);
786 }
787 f2fs_free_filename(&fname);
788 return err;
789 }
790
791 int f2fs_do_tmpfile(struct inode *inode, struct inode *dir,
792 struct f2fs_filename *fname)
793 {
794 struct page *page;
795 int err = 0;
796
797 f2fs_down_write(&F2FS_I(inode)->i_sem);
798 page = f2fs_init_inode_metadata(inode, dir, fname, NULL);
799 if (IS_ERR(page)) {
800 err = PTR_ERR(page);
801 goto fail;
802 }
803 f2fs_put_page(page, 1);
804
805 clear_inode_flag(inode, FI_NEW_INODE);
806 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
807 fail:
808 f2fs_up_write(&F2FS_I(inode)->i_sem);
809 return err;
810 }
811
812 void f2fs_drop_nlink(struct inode *dir, struct inode *inode)
813 {
814 struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
815
816 f2fs_down_write(&F2FS_I(inode)->i_sem);
817
818 if (S_ISDIR(inode->i_mode))
819 f2fs_i_links_write(dir, false);
820 inode_set_ctime_current(inode);
821
822 f2fs_i_links_write(inode, false);
823 if (S_ISDIR(inode->i_mode)) {
824 f2fs_i_links_write(inode, false);
825 f2fs_i_size_write(inode, 0);
826 }
827 f2fs_up_write(&F2FS_I(inode)->i_sem);
828
829 if (inode->i_nlink == 0)
830 f2fs_add_orphan_inode(inode);
831 else
832 f2fs_release_orphan_inode(sbi);
833 }
834
835 /*
836 * It only removes the dentry from the dentry page, corresponding name
837 * entry in name page does not need to be touched during deletion.
838 */
839 void f2fs_delete_entry(struct f2fs_dir_entry *dentry, struct page *page,
840 struct inode *dir, struct inode *inode)
841 {
842 struct f2fs_dentry_block *dentry_blk;
843 unsigned int bit_pos;
844 int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len));
845 int i;
846
847 f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
848
849 if (F2FS_OPTION(F2FS_I_SB(dir)).fsync_mode == FSYNC_MODE_STRICT)
850 f2fs_add_ino_entry(F2FS_I_SB(dir), dir->i_ino, TRANS_DIR_INO);
851
852 if (f2fs_has_inline_dentry(dir))
853 return f2fs_delete_inline_entry(dentry, page, dir, inode);
854
855 lock_page(page);
856 f2fs_wait_on_page_writeback(page, DATA, true, true);
857
858 dentry_blk = page_address(page);
859 bit_pos = dentry - dentry_blk->dentry;
860 for (i = 0; i < slots; i++)
861 __clear_bit_le(bit_pos + i, &dentry_blk->dentry_bitmap);
862
863 /* Let's check and deallocate this dentry page */
864 bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
865 NR_DENTRY_IN_BLOCK,
866 0);
867 set_page_dirty(page);
868
869 if (bit_pos == NR_DENTRY_IN_BLOCK &&
870 !f2fs_truncate_hole(dir, page->index, page->index + 1)) {
871 f2fs_clear_page_cache_dirty_tag(page);
872 clear_page_dirty_for_io(page);
873 ClearPageUptodate(page);
874 clear_page_private_all(page);
875
876 inode_dec_dirty_pages(dir);
877 f2fs_remove_dirty_inode(dir);
878 }
879 f2fs_put_page(page, 1);
880
881 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
882 f2fs_mark_inode_dirty_sync(dir, false);
883
884 if (inode)
885 f2fs_drop_nlink(dir, inode);
886 }
887
888 bool f2fs_empty_dir(struct inode *dir)
889 {
890 unsigned long bidx = 0;
891 struct page *dentry_page;
892 unsigned int bit_pos;
893 struct f2fs_dentry_block *dentry_blk;
894 unsigned long nblock = dir_blocks(dir);
895
896 if (f2fs_has_inline_dentry(dir))
897 return f2fs_empty_inline_dir(dir);
898
899 while (bidx < nblock) {
900 pgoff_t next_pgofs;
901
902 dentry_page = f2fs_find_data_page(dir, bidx, &next_pgofs);
903 if (IS_ERR(dentry_page)) {
904 if (PTR_ERR(dentry_page) == -ENOENT) {
905 bidx = next_pgofs;
906 continue;
907 } else {
908 return false;
909 }
910 }
911
912 dentry_blk = page_address(dentry_page);
913 if (bidx == 0)
914 bit_pos = 2;
915 else
916 bit_pos = 0;
917 bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
918 NR_DENTRY_IN_BLOCK,
919 bit_pos);
920
921 f2fs_put_page(dentry_page, 0);
922
923 if (bit_pos < NR_DENTRY_IN_BLOCK)
924 return false;
925
926 bidx++;
927 }
928 return true;
929 }
930
931 int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
932 unsigned int start_pos, struct fscrypt_str *fstr)
933 {
934 unsigned char d_type = DT_UNKNOWN;
935 unsigned int bit_pos;
936 struct f2fs_dir_entry *de = NULL;
937 struct fscrypt_str de_name = FSTR_INIT(NULL, 0);
938 struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
939 struct blk_plug plug;
940 bool readdir_ra = sbi->readdir_ra;
941 bool found_valid_dirent = false;
942 int err = 0;
943
944 bit_pos = ((unsigned long)ctx->pos % d->max);
945
946 if (readdir_ra)
947 blk_start_plug(&plug);
948
949 while (bit_pos < d->max) {
950 bit_pos = find_next_bit_le(d->bitmap, d->max, bit_pos);
951 if (bit_pos >= d->max)
952 break;
953
954 de = &d->dentry[bit_pos];
955 if (de->name_len == 0) {
956 if (found_valid_dirent || !bit_pos) {
957 f2fs_warn_ratelimited(sbi,
958 "invalid namelen(0), ino:%u, run fsck to fix.",
959 le32_to_cpu(de->ino));
960 set_sbi_flag(sbi, SBI_NEED_FSCK);
961 }
962 bit_pos++;
963 ctx->pos = start_pos + bit_pos;
964 continue;
965 }
966
967 d_type = fs_ftype_to_dtype(de->file_type);
968
969 de_name.name = d->filename[bit_pos];
970 de_name.len = le16_to_cpu(de->name_len);
971
972 /* check memory boundary before moving forward */
973 bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
974 if (unlikely(bit_pos > d->max ||
975 le16_to_cpu(de->name_len) > F2FS_NAME_LEN)) {
976 f2fs_warn(sbi, "%s: corrupted namelen=%d, run fsck to fix.",
977 __func__, le16_to_cpu(de->name_len));
978 set_sbi_flag(sbi, SBI_NEED_FSCK);
979 err = -EFSCORRUPTED;
980 f2fs_handle_error(sbi, ERROR_CORRUPTED_DIRENT);
981 goto out;
982 }
983
984 if (IS_ENCRYPTED(d->inode)) {
985 int save_len = fstr->len;
986
987 err = fscrypt_fname_disk_to_usr(d->inode,
988 (u32)le32_to_cpu(de->hash_code),
989 0, &de_name, fstr);
990 if (err)
991 goto out;
992
993 de_name = *fstr;
994 fstr->len = save_len;
995 }
996
997 if (!dir_emit(ctx, de_name.name, de_name.len,
998 le32_to_cpu(de->ino), d_type)) {
999 err = 1;
1000 goto out;
1001 }
1002
1003 if (readdir_ra)
1004 f2fs_ra_node_page(sbi, le32_to_cpu(de->ino));
1005
1006 ctx->pos = start_pos + bit_pos;
1007 found_valid_dirent = true;
1008 }
1009 out:
1010 if (readdir_ra)
1011 blk_finish_plug(&plug);
1012 return err;
1013 }
1014
1015 static int f2fs_readdir(struct file *file, struct dir_context *ctx)
1016 {
1017 struct inode *inode = file_inode(file);
1018 unsigned long npages = dir_blocks(inode);
1019 struct f2fs_dentry_block *dentry_blk = NULL;
1020 struct page *dentry_page = NULL;
1021 struct file_ra_state *ra = &file->f_ra;
1022 loff_t start_pos = ctx->pos;
1023 unsigned int n = ((unsigned long)ctx->pos / NR_DENTRY_IN_BLOCK);
1024 struct f2fs_dentry_ptr d;
1025 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
1026 int err = 0;
1027
1028 if (IS_ENCRYPTED(inode)) {
1029 err = fscrypt_prepare_readdir(inode);
1030 if (err)
1031 goto out;
1032
1033 err = fscrypt_fname_alloc_buffer(F2FS_NAME_LEN, &fstr);
1034 if (err < 0)
1035 goto out;
1036 }
1037
1038 if (f2fs_has_inline_dentry(inode)) {
1039 err = f2fs_read_inline_dir(file, ctx, &fstr);
1040 goto out_free;
1041 }
1042
1043 for (; n < npages; ctx->pos = n * NR_DENTRY_IN_BLOCK) {
1044 pgoff_t next_pgofs;
1045
1046 /* allow readdir() to be interrupted */
1047 if (fatal_signal_pending(current)) {
1048 err = -ERESTARTSYS;
1049 goto out_free;
1050 }
1051 cond_resched();
1052
1053 /* readahead for multi pages of dir */
1054 if (npages - n > 1 && !ra_has_index(ra, n))
1055 page_cache_sync_readahead(inode->i_mapping, ra, file, n,
1056 min(npages - n, (pgoff_t)MAX_DIR_RA_PAGES));
1057
1058 dentry_page = f2fs_find_data_page(inode, n, &next_pgofs);
1059 if (IS_ERR(dentry_page)) {
1060 err = PTR_ERR(dentry_page);
1061 if (err == -ENOENT) {
1062 err = 0;
1063 n = next_pgofs;
1064 continue;
1065 } else {
1066 goto out_free;
1067 }
1068 }
1069
1070 dentry_blk = page_address(dentry_page);
1071
1072 make_dentry_ptr_block(inode, &d, dentry_blk);
1073
1074 err = f2fs_fill_dentries(ctx, &d,
1075 n * NR_DENTRY_IN_BLOCK, &fstr);
1076 if (err) {
1077 f2fs_put_page(dentry_page, 0);
1078 break;
1079 }
1080
1081 f2fs_put_page(dentry_page, 0);
1082
1083 n++;
1084 }
1085 out_free:
1086 fscrypt_fname_free_buffer(&fstr);
1087 out:
1088 trace_f2fs_readdir(inode, start_pos, ctx->pos, err);
1089 return err < 0 ? err : 0;
1090 }
1091
1092 const struct file_operations f2fs_dir_operations = {
1093 .llseek = generic_file_llseek,
1094 .read = generic_read_dir,
1095 .iterate_shared = f2fs_readdir,
1096 .fsync = f2fs_sync_file,
1097 .unlocked_ioctl = f2fs_ioctl,
1098 #ifdef CONFIG_COMPAT
1099 .compat_ioctl = f2fs_compat_ioctl,
1100 #endif
1101 };
1102
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.