Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 <linux/unaligned.h>
9#include <linux/fs.h>
10#include <linux/f2fs_fs.h>
11#include <linux/filelock.h>
12#include <linux/sched/signal.h>
13#include <linux/unicode.h>
14#include "f2fs.h"
15#include "node.h"
16#include "acl.h"
17#include "xattr.h"
18#include <trace/events/f2fs.h>
19
20static inline bool f2fs_should_fallback_to_linear(struct inode *dir)
21{
22 struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
23
24 switch (F2FS_OPTION(sbi).lookup_mode) {
25 case LOOKUP_PERF:
26 return false;
27 case LOOKUP_COMPAT:
28 return true;
29 case LOOKUP_AUTO:
30 return !sb_no_casefold_compat_fallback(sbi->sb);
31 }
32 return false;
33}
34
35#if IS_ENABLED(CONFIG_UNICODE)
36extern struct kmem_cache *f2fs_cf_name_slab;
37#endif
38
39static unsigned long dir_blocks(struct inode *inode)
40{
41 return ((unsigned long long) (i_size_read(inode) + PAGE_SIZE - 1))
42 >> PAGE_SHIFT;
43}
44
45static unsigned int dir_buckets(unsigned int level, int dir_level)
46{
47 if (level + dir_level < MAX_DIR_HASH_DEPTH / 2)
48 return BIT(level + dir_level);
49 else
50 return MAX_DIR_BUCKETS;
51}
52
53static unsigned int bucket_blocks(unsigned int level)
54{
55 if (level < MAX_DIR_HASH_DEPTH / 2)
56 return 2;
57 else
58 return 4;
59}
60
61#if IS_ENABLED(CONFIG_UNICODE)
62/* If @dir is casefolded, initialize @fname->cf_name from @fname->usr_fname. */
63int f2fs_init_casefolded_name(const struct inode *dir,
64 struct f2fs_filename *fname)
65{
66 struct super_block *sb = dir->i_sb;
67 unsigned char *buf;
68 int len;
69
70 if (IS_CASEFOLDED(dir) &&
71 !name_is_dot_dotdot(fname->usr_fname->name, fname->usr_fname->len)) {
72 buf = f2fs_kmem_cache_alloc(f2fs_cf_name_slab,
73 GFP_NOFS, false, F2FS_SB(sb));
74 if (!buf)
75 return -ENOMEM;
76
77 len = utf8_casefold(sb->s_encoding, fname->usr_fname,
78 buf, F2FS_NAME_LEN);
79 if (len <= 0) {
80 kmem_cache_free(f2fs_cf_name_slab, buf);
81 if (sb_has_strict_encoding(sb))
82 return -EINVAL;
83 /* fall back to treating name as opaque byte sequence */
84 return 0;
85 }
86 fname->cf_name.name = buf;
87 fname->cf_name.len = len;
88 }
89
90 return 0;
91}
92
93void f2fs_free_casefolded_name(struct f2fs_filename *fname)
94{
95 unsigned char *buf = (unsigned char *)fname->cf_name.name;
96
97 if (buf) {
98 kmem_cache_free(f2fs_cf_name_slab, buf);
99 fname->cf_name.name = NULL;
100 }
101}
102#endif /* CONFIG_UNICODE */
103
104static int __f2fs_setup_filename(const struct inode *dir,
105 const struct fscrypt_name *crypt_name,
106 struct f2fs_filename *fname)
107{
108 int err;
109
110 memset(fname, 0, sizeof(*fname));
111
112 fname->usr_fname = crypt_name->usr_fname;
113 fname->disk_name = crypt_name->disk_name;
114#ifdef CONFIG_FS_ENCRYPTION
115 fname->crypto_buf = crypt_name->crypto_buf;
116#endif
117 if (crypt_name->is_nokey_name) {
118 /* hash was decoded from the no-key name */
119 fname->hash = cpu_to_le32(crypt_name->hash);
120 } else {
121 err = f2fs_init_casefolded_name(dir, fname);
122 if (err) {
123 f2fs_free_filename(fname);
124 return err;
125 }
126 f2fs_hash_filename(dir, fname);
127 }
128 return 0;
129}
130
131/*
132 * Prepare to search for @iname in @dir. This is similar to
133 * fscrypt_setup_filename(), but this also handles computing the casefolded name
134 * and the f2fs dirhash if needed, then packing all the information about this
135 * filename up into a 'struct f2fs_filename'.
136 */
137int f2fs_setup_filename(struct inode *dir, const struct qstr *iname,
138 int lookup, struct f2fs_filename *fname)
139{
140 struct fscrypt_name crypt_name;
141 int err;
142
143 err = fscrypt_setup_filename(dir, iname, lookup, &crypt_name);
144 if (err)
145 return err;
146
147 return __f2fs_setup_filename(dir, &crypt_name, fname);
148}
149
150/*
151 * Prepare to look up @dentry in @dir. This is similar to
152 * fscrypt_prepare_lookup(), but this also handles computing the casefolded name
153 * and the f2fs dirhash if needed, then packing all the information about this
154 * filename up into a 'struct f2fs_filename'.
155 */
156int f2fs_prepare_lookup(struct inode *dir, struct dentry *dentry,
157 struct f2fs_filename *fname)
158{
159 struct fscrypt_name crypt_name;
160 int err;
161
162 err = fscrypt_prepare_lookup(dir, dentry, &crypt_name);
163 if (err)
164 return err;
165
166 return __f2fs_setup_filename(dir, &crypt_name, fname);
167}
168
169void f2fs_free_filename(struct f2fs_filename *fname)
170{
171#ifdef CONFIG_FS_ENCRYPTION
172 kfree(fname->crypto_buf.name);
173 fname->crypto_buf.name = NULL;
174#endif
175 f2fs_free_casefolded_name(fname);
176}
177
178static unsigned long dir_block_index(unsigned int level,
179 int dir_level, unsigned int idx)
180{
181 unsigned long i;
182 unsigned long bidx = 0;
183
184 for (i = 0; i < level; i++)
185 bidx += mul_u32_u32(dir_buckets(i, dir_level),
186 bucket_blocks(i));
187 bidx += idx * bucket_blocks(level);
188 return bidx;
189}
190
191static struct f2fs_dir_entry *find_in_block(struct inode *dir,
192 struct folio *dentry_folio,
193 const struct f2fs_filename *fname,
194 int *max_slots,
195 bool use_hash)
196{
197 struct f2fs_dentry_block *dentry_blk;
198 struct f2fs_dentry_ptr d;
199
200 dentry_blk = folio_address(dentry_folio);
201
202 make_dentry_ptr_block(dir, &d, dentry_blk);
203 return f2fs_find_target_dentry(&d, fname, max_slots, use_hash);
204}
205
206static inline int f2fs_match_name(const struct inode *dir,
207 const struct f2fs_filename *fname,
208 const u8 *de_name, u32 de_name_len)
209{
210 struct fscrypt_name f;
211
212#if IS_ENABLED(CONFIG_UNICODE)
213 if (fname->cf_name.name)
214 return generic_ci_match(dir, fname->usr_fname,
215 &fname->cf_name,
216 de_name, de_name_len);
217
218#endif
219 f.usr_fname = fname->usr_fname;
220 f.disk_name = fname->disk_name;
221#ifdef CONFIG_FS_ENCRYPTION
222 f.crypto_buf = fname->crypto_buf;
223#endif
224 return fscrypt_match_name(&f, de_name, de_name_len);
225}
226
227struct f2fs_dir_entry *f2fs_find_target_dentry(const struct f2fs_dentry_ptr *d,
228 const struct f2fs_filename *fname, int *max_slots,
229 bool use_hash)
230{
231 struct f2fs_dir_entry *de;
232 unsigned long bit_pos = 0;
233 int max_len = 0;
234 int res = 0;
235
236 if (max_slots)
237 *max_slots = 0;
238 while (bit_pos < d->max) {
239 if (!test_bit_le(bit_pos, d->bitmap)) {
240 bit_pos++;
241 max_len++;
242 continue;
243 }
244
245 de = &d->dentry[bit_pos];
246
247 if (unlikely(!de->name_len)) {
248 bit_pos++;
249 continue;
250 }
251
252 if (!use_hash || de->hash_code == fname->hash) {
253 res = f2fs_match_name(d->inode, fname,
254 d->filename[bit_pos],
255 le16_to_cpu(de->name_len));
256 if (res < 0)
257 return ERR_PTR(res);
258 if (res)
259 goto found;
260 }
261
262 if (max_slots && max_len > *max_slots)
263 *max_slots = max_len;
264 max_len = 0;
265
266 bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
267 }
268
269 de = NULL;
270found:
271 if (max_slots && max_len > *max_slots)
272 *max_slots = max_len;
273 return de;
274}
275
276static struct f2fs_dir_entry *find_in_level(struct inode *dir,
277 unsigned int level,
278 const struct f2fs_filename *fname,
279 struct folio **res_folio,
280 bool use_hash)
281{
282 int s = GET_DENTRY_SLOTS(fname->disk_name.len);
283 unsigned int nbucket, nblock;
284 unsigned int bidx, end_block, bucket_no;
285 struct f2fs_dir_entry *de = NULL;
286 pgoff_t next_pgofs;
287 bool room = false;
288 int max_slots;
289
290 nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
291 nblock = bucket_blocks(level);
292
293 bucket_no = use_hash ? le32_to_cpu(fname->hash) % nbucket : 0;
294
295start_find_bucket:
296 bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
297 bucket_no);
298 end_block = bidx + nblock;
299
300 while (bidx < end_block) {
301 /* no need to allocate new dentry pages to all the indices */
302 struct folio *dentry_folio;
303 dentry_folio = f2fs_find_data_folio(dir, bidx, &next_pgofs);
304 if (IS_ERR(dentry_folio)) {
305 if (PTR_ERR(dentry_folio) == -ENOENT) {
306 room = true;
307 bidx = next_pgofs;
308 continue;
309 } else {
310 *res_folio = dentry_folio;
311 break;
312 }
313 }
314
315 de = find_in_block(dir, dentry_folio, fname, &max_slots, use_hash);
316 if (IS_ERR(de)) {
317 *res_folio = ERR_CAST(de);
318 de = NULL;
319 break;
320 } else if (de) {
321 *res_folio = dentry_folio;
322 break;
323 }
324
325 if (max_slots >= s)
326 room = true;
327 f2fs_folio_put(dentry_folio, false);
328
329 bidx++;
330 }
331
332 if (de)
333 return de;
334
335 if (likely(use_hash)) {
336 if (room && F2FS_I(dir)->chash != fname->hash) {
337 F2FS_I(dir)->chash = fname->hash;
338 F2FS_I(dir)->clevel = level;
339 }
340 } else if (++bucket_no < nbucket) {
341 goto start_find_bucket;
342 }
343 return NULL;
344}
345
346struct f2fs_dir_entry *__f2fs_find_entry(struct inode *dir,
347 const struct f2fs_filename *fname,
348 struct folio **res_folio)
349{
350 unsigned long npages = dir_blocks(dir);
351 struct f2fs_dir_entry *de = NULL;
352 unsigned int max_depth;
353 unsigned int level;
354 bool use_hash = true;
355
356 *res_folio = NULL;
357
358#if IS_ENABLED(CONFIG_UNICODE)
359start_find_entry:
360#endif
361 if (f2fs_has_inline_dentry(dir)) {
362 de = f2fs_find_in_inline_dir(dir, fname, res_folio, use_hash);
363 goto out;
364 }
365
366 if (npages == 0)
367 goto out;
368
369 max_depth = F2FS_I(dir)->i_current_depth;
370 if (unlikely(max_depth > MAX_DIR_HASH_DEPTH)) {
371 f2fs_warn(F2FS_I_SB(dir), "Corrupted max_depth of %llu: %u",
372 dir->i_ino, max_depth);
373 max_depth = MAX_DIR_HASH_DEPTH;
374 f2fs_i_depth_write(dir, max_depth);
375 }
376
377 for (level = 0; level < max_depth; level++) {
378 de = find_in_level(dir, level, fname, res_folio, use_hash);
379 if (de || IS_ERR(*res_folio))
380 break;
381 }
382
383out:
384#if IS_ENABLED(CONFIG_UNICODE)
385 if (f2fs_should_fallback_to_linear(dir) &&
386 IS_CASEFOLDED(dir) && !de && use_hash) {
387 use_hash = false;
388 goto start_find_entry;
389 }
390#endif
391 /* This is to increase the speed of f2fs_create */
392 if (!de)
393 F2FS_I(dir)->task = current;
394 return de;
395}
396
397/*
398 * Find an entry in the specified directory with the wanted name.
399 * It returns the page where the entry was found (as a parameter - res_page),
400 * and the entry itself. Page is returned mapped and unlocked.
401 * Entry is guaranteed to be valid.
402 */
403struct f2fs_dir_entry *f2fs_find_entry(struct inode *dir,
404 const struct qstr *child, struct folio **res_folio)
405{
406 struct f2fs_dir_entry *de = NULL;
407 struct f2fs_filename fname;
408 int err;
409
410 err = f2fs_setup_filename(dir, child, 1, &fname);
411 if (err) {
412 if (err == -ENOENT)
413 *res_folio = NULL;
414 else
415 *res_folio = ERR_PTR(err);
416 return NULL;
417 }
418
419 de = __f2fs_find_entry(dir, &fname, res_folio);
420
421 f2fs_free_filename(&fname);
422 return de;
423}
424
425struct f2fs_dir_entry *f2fs_parent_dir(struct inode *dir, struct folio **f)
426{
427 return f2fs_find_entry(dir, &dotdot_name, f);
428}
429
430ino_t f2fs_inode_by_name(struct inode *dir, const struct qstr *qstr,
431 struct folio **folio)
432{
433 ino_t res = 0;
434 struct f2fs_dir_entry *de;
435
436 de = f2fs_find_entry(dir, qstr, folio);
437 if (de) {
438 res = le32_to_cpu(de->ino);
439 f2fs_folio_put(*folio, false);
440 }
441
442 return res;
443}
444
445void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de,
446 struct folio *folio, struct inode *inode)
447{
448 enum page_type type = f2fs_has_inline_dentry(dir) ? NODE : DATA;
449
450 folio_lock(folio);
451 f2fs_folio_wait_writeback(folio, type, true, true);
452 de->ino = cpu_to_le32(inode->i_ino);
453 de->file_type = fs_umode_to_ftype(inode->i_mode);
454 folio_mark_dirty(folio);
455
456 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
457 f2fs_mark_inode_dirty_sync(dir, false);
458 f2fs_folio_put(folio, true);
459}
460
461static void init_dent_inode(struct inode *dir, struct inode *inode,
462 const struct f2fs_filename *fname,
463 struct folio *ifolio)
464{
465 struct f2fs_inode *ri;
466
467 if (!fname) /* tmpfile case? */
468 return;
469
470 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
471
472 /* copy name info. to this inode folio */
473 ri = F2FS_INODE(ifolio);
474 ri->i_namelen = cpu_to_le32(fname->disk_name.len);
475 memcpy(ri->i_name, fname->disk_name.name, fname->disk_name.len);
476 if (IS_ENCRYPTED(dir)) {
477 file_set_enc_name(inode);
478 /*
479 * Roll-forward recovery doesn't have encryption keys available,
480 * so it can't compute the dirhash for encrypted+casefolded
481 * filenames. Append it to i_name if possible. Else, disable
482 * roll-forward recovery of the dentry (i.e., make fsync'ing the
483 * file force a checkpoint) by setting LOST_PINO.
484 */
485 if (IS_CASEFOLDED(dir)) {
486 if (fname->disk_name.len + sizeof(f2fs_hash_t) <=
487 F2FS_NAME_LEN)
488 put_unaligned(fname->hash, (f2fs_hash_t *)
489 &ri->i_name[fname->disk_name.len]);
490 else
491 file_lost_pino(inode);
492 }
493 }
494 folio_mark_dirty(ifolio);
495}
496
497void f2fs_do_make_empty_dir(struct inode *inode, struct inode *parent,
498 struct f2fs_dentry_ptr *d)
499{
500 struct fscrypt_str dot = FSTR_INIT(".", 1);
501 struct fscrypt_str dotdot = FSTR_INIT("..", 2);
502
503 /* update dirent of "." */
504 f2fs_update_dentry(inode->i_ino, inode->i_mode, d, &dot, 0, 0);
505
506 /* update dirent of ".." */
507 f2fs_update_dentry(parent->i_ino, parent->i_mode, d, &dotdot, 0, 1);
508}
509
510static int make_empty_dir(struct inode *inode,
511 struct inode *parent, struct folio *folio)
512{
513 struct folio *dentry_folio;
514 struct f2fs_dentry_block *dentry_blk;
515 struct f2fs_dentry_ptr d;
516
517 if (f2fs_has_inline_dentry(inode))
518 return f2fs_make_empty_inline_dir(inode, parent, folio);
519
520 dentry_folio = f2fs_get_new_data_folio(inode, folio, 0, true);
521 if (IS_ERR(dentry_folio))
522 return PTR_ERR(dentry_folio);
523
524 dentry_blk = folio_address(dentry_folio);
525
526 make_dentry_ptr_block(NULL, &d, dentry_blk);
527 f2fs_do_make_empty_dir(inode, parent, &d);
528
529 folio_mark_dirty(dentry_folio);
530 f2fs_folio_put(dentry_folio, true);
531 return 0;
532}
533
534struct folio *f2fs_init_inode_metadata(struct inode *inode, struct inode *dir,
535 const struct f2fs_filename *fname, struct folio *dfolio)
536{
537 struct folio *folio;
538 int err;
539
540 if (is_inode_flag_set(inode, FI_NEW_INODE)) {
541 folio = f2fs_new_inode_folio(inode);
542 if (IS_ERR(folio))
543 return folio;
544
545 if (S_ISDIR(inode->i_mode)) {
546 /* in order to handle error case */
547 folio_get(folio);
548 err = make_empty_dir(inode, dir, folio);
549 if (err) {
550 folio_lock(folio);
551 goto put_error;
552 }
553 folio_put(folio);
554 }
555
556 err = f2fs_init_acl(inode, dir, folio, dfolio);
557 if (err)
558 goto put_error;
559
560 err = f2fs_init_security(inode, dir,
561 fname ? fname->usr_fname : NULL,
562 folio);
563 if (err)
564 goto put_error;
565
566 if (IS_ENCRYPTED(inode)) {
567 err = fscrypt_set_context(inode, folio);
568 if (err)
569 goto put_error;
570 }
571 } else {
572 folio = f2fs_get_inode_folio(F2FS_I_SB(dir), inode->i_ino);
573 if (IS_ERR(folio))
574 return folio;
575 }
576
577 init_dent_inode(dir, inode, fname, folio);
578
579 /*
580 * This file should be checkpointed during fsync.
581 * We lost i_pino from now on.
582 */
583 if (is_inode_flag_set(inode, FI_INC_LINK)) {
584 if (!S_ISDIR(inode->i_mode))
585 file_lost_pino(inode);
586 /*
587 * If link the tmpfile to alias through linkat path,
588 * we should remove this inode from orphan list.
589 */
590 if (inode->i_nlink == 0)
591 f2fs_remove_orphan_inode(F2FS_I_SB(dir), inode->i_ino);
592 f2fs_i_links_write(inode, true);
593 }
594 return folio;
595
596put_error:
597 clear_nlink(inode);
598 f2fs_update_inode(inode, folio);
599 f2fs_folio_put(folio, true);
600 return ERR_PTR(err);
601}
602
603void f2fs_update_parent_metadata(struct inode *dir, struct inode *inode,
604 unsigned int current_depth)
605{
606 if (inode && is_inode_flag_set(inode, FI_NEW_INODE)) {
607 if (S_ISDIR(inode->i_mode))
608 f2fs_i_links_write(dir, true);
609 clear_inode_flag(inode, FI_NEW_INODE);
610 }
611 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
612 f2fs_mark_inode_dirty_sync(dir, false);
613
614 if (F2FS_I(dir)->i_current_depth != current_depth)
615 f2fs_i_depth_write(dir, current_depth);
616
617 if (inode && is_inode_flag_set(inode, FI_INC_LINK))
618 clear_inode_flag(inode, FI_INC_LINK);
619}
620
621int f2fs_room_for_filename(const void *bitmap, int slots, int max_slots)
622{
623 int bit_start = 0;
624 int zero_start, zero_end;
625next:
626 zero_start = find_next_zero_bit_le(bitmap, max_slots, bit_start);
627 if (zero_start >= max_slots)
628 return max_slots;
629
630 zero_end = find_next_bit_le(bitmap, max_slots, zero_start);
631 if (zero_end - zero_start >= slots)
632 return zero_start;
633
634 bit_start = zero_end + 1;
635
636 if (zero_end + 1 >= max_slots)
637 return max_slots;
638 goto next;
639}
640
641bool f2fs_has_enough_room(struct inode *dir, struct folio *ifolio,
642 const struct f2fs_filename *fname)
643{
644 struct f2fs_dentry_ptr d;
645 unsigned int bit_pos;
646 int slots = GET_DENTRY_SLOTS(fname->disk_name.len);
647
648 make_dentry_ptr_inline(dir, &d, inline_data_addr(dir, ifolio));
649
650 bit_pos = f2fs_room_for_filename(d.bitmap, slots, d.max);
651
652 return bit_pos < d.max;
653}
654
655void f2fs_update_dentry(nid_t ino, umode_t mode, struct f2fs_dentry_ptr *d,
656 const struct fscrypt_str *name, f2fs_hash_t name_hash,
657 unsigned int bit_pos)
658{
659 struct f2fs_dir_entry *de;
660 int slots = GET_DENTRY_SLOTS(name->len);
661 int i;
662
663 de = &d->dentry[bit_pos];
664 de->hash_code = name_hash;
665 de->name_len = cpu_to_le16(name->len);
666 memcpy(d->filename[bit_pos], name->name, name->len);
667 de->ino = cpu_to_le32(ino);
668 de->file_type = fs_umode_to_ftype(mode);
669 for (i = 0; i < slots; i++) {
670 __set_bit_le(bit_pos + i, (void *)d->bitmap);
671 /* avoid wrong garbage data for readdir */
672 if (i)
673 (de + i)->name_len = 0;
674 }
675}
676
677int f2fs_add_regular_entry(struct inode *dir, const struct f2fs_filename *fname,
678 struct inode *inode, nid_t ino, umode_t mode)
679{
680 unsigned int bit_pos;
681 unsigned int level;
682 unsigned int current_depth;
683 unsigned long bidx, block;
684 unsigned int nbucket, nblock;
685 struct folio *dentry_folio = NULL;
686 struct f2fs_dentry_block *dentry_blk = NULL;
687 struct f2fs_dentry_ptr d;
688 struct folio *folio = NULL;
689 int slots, err = 0;
690
691 level = 0;
692 slots = GET_DENTRY_SLOTS(fname->disk_name.len);
693
694 current_depth = F2FS_I(dir)->i_current_depth;
695 if (F2FS_I(dir)->chash == fname->hash) {
696 level = F2FS_I(dir)->clevel;
697 F2FS_I(dir)->chash = 0;
698 }
699
700start:
701 if (time_to_inject(F2FS_I_SB(dir), FAULT_DIR_DEPTH))
702 return -ENOSPC;
703
704 if (unlikely(current_depth == MAX_DIR_HASH_DEPTH))
705 return -ENOSPC;
706
707 /* Increase the depth, if required */
708 if (level == current_depth)
709 ++current_depth;
710
711 nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
712 nblock = bucket_blocks(level);
713
714 bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
715 (le32_to_cpu(fname->hash) % nbucket));
716
717 for (block = bidx; block <= (bidx + nblock - 1); block++) {
718 dentry_folio = f2fs_get_new_data_folio(dir, NULL, block, true);
719 if (IS_ERR(dentry_folio))
720 return PTR_ERR(dentry_folio);
721
722 dentry_blk = folio_address(dentry_folio);
723 bit_pos = f2fs_room_for_filename(&dentry_blk->dentry_bitmap,
724 slots, NR_DENTRY_IN_BLOCK);
725 if (bit_pos < NR_DENTRY_IN_BLOCK)
726 goto add_dentry;
727
728 f2fs_folio_put(dentry_folio, true);
729 }
730
731 /* Move to next level to find the empty slot for new dentry */
732 ++level;
733 goto start;
734add_dentry:
735 f2fs_folio_wait_writeback(dentry_folio, DATA, true, true);
736
737 if (inode) {
738 f2fs_down_write(&F2FS_I(inode)->i_sem);
739 folio = f2fs_init_inode_metadata(inode, dir, fname, NULL);
740 if (IS_ERR(folio)) {
741 err = PTR_ERR(folio);
742 goto fail;
743 }
744 }
745
746 make_dentry_ptr_block(NULL, &d, dentry_blk);
747 f2fs_update_dentry(ino, mode, &d, &fname->disk_name, fname->hash,
748 bit_pos);
749
750 folio_mark_dirty(dentry_folio);
751
752 if (inode) {
753 f2fs_i_pino_write(inode, dir->i_ino);
754
755 /* synchronize inode page's data from inode cache */
756 if (is_inode_flag_set(inode, FI_NEW_INODE))
757 f2fs_update_inode(inode, folio);
758
759 f2fs_folio_put(folio, true);
760 }
761
762 f2fs_update_parent_metadata(dir, inode, current_depth);
763fail:
764 if (inode)
765 f2fs_up_write(&F2FS_I(inode)->i_sem);
766
767 f2fs_folio_put(dentry_folio, true);
768
769 return err;
770}
771
772int f2fs_add_dentry(struct inode *dir, const struct f2fs_filename *fname,
773 struct inode *inode, nid_t ino, umode_t mode)
774{
775 int err = -EAGAIN;
776
777 if (f2fs_has_inline_dentry(dir)) {
778 /*
779 * Should get i_xattr_sem to keep the lock order:
780 * i_xattr_sem -> inode_page lock used by f2fs_setxattr.
781 */
782 f2fs_down_read(&F2FS_I(dir)->i_xattr_sem);
783 err = f2fs_add_inline_entry(dir, fname, inode, ino, mode);
784 f2fs_up_read(&F2FS_I(dir)->i_xattr_sem);
785 }
786 if (err == -EAGAIN)
787 err = f2fs_add_regular_entry(dir, fname, inode, ino, mode);
788
789 f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
790 return err;
791}
792
793/*
794 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
795 * f2fs_unlock_op().
796 */
797int f2fs_do_add_link(struct inode *dir, const struct qstr *name,
798 struct inode *inode, nid_t ino, umode_t mode)
799{
800 struct f2fs_filename fname;
801 struct folio *folio = NULL;
802 struct f2fs_dir_entry *de = NULL;
803 int err;
804
805 err = f2fs_setup_filename(dir, name, 0, &fname);
806 if (err)
807 return err;
808
809 /*
810 * An immature stackable filesystem shows a race condition between lookup
811 * and create. If we have same task when doing lookup and create, it's
812 * definitely fine as expected by VFS normally. Otherwise, let's just
813 * verify on-disk dentry one more time, which guarantees filesystem
814 * consistency more.
815 */
816 if (current != F2FS_I(dir)->task) {
817 de = __f2fs_find_entry(dir, &fname, &folio);
818 F2FS_I(dir)->task = NULL;
819 }
820 if (de) {
821 f2fs_folio_put(folio, false);
822 err = -EEXIST;
823 } else if (IS_ERR(folio)) {
824 err = PTR_ERR(folio);
825 } else {
826 err = f2fs_add_dentry(dir, &fname, inode, ino, mode);
827 }
828 f2fs_free_filename(&fname);
829 return err;
830}
831
832int f2fs_do_tmpfile(struct inode *inode, struct inode *dir,
833 struct f2fs_filename *fname)
834{
835 struct folio *folio;
836 int err = 0;
837
838 f2fs_down_write(&F2FS_I(inode)->i_sem);
839 folio = f2fs_init_inode_metadata(inode, dir, fname, NULL);
840 if (IS_ERR(folio)) {
841 err = PTR_ERR(folio);
842 goto fail;
843 }
844 f2fs_folio_put(folio, true);
845
846 clear_inode_flag(inode, FI_NEW_INODE);
847 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
848fail:
849 f2fs_up_write(&F2FS_I(inode)->i_sem);
850 return err;
851}
852
853void f2fs_drop_nlink(struct inode *dir, struct inode *inode)
854{
855 struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
856
857 f2fs_down_write(&F2FS_I(inode)->i_sem);
858
859 if (S_ISDIR(inode->i_mode))
860 f2fs_i_links_write(dir, false);
861 inode_set_ctime_current(inode);
862
863 f2fs_i_links_write(inode, false);
864 if (S_ISDIR(inode->i_mode)) {
865 f2fs_i_links_write(inode, false);
866 f2fs_i_size_write(inode, 0);
867 }
868 f2fs_up_write(&F2FS_I(inode)->i_sem);
869
870 if (inode->i_nlink == 0)
871 f2fs_add_orphan_inode(inode);
872 else
873 f2fs_release_orphan_inode(sbi);
874}
875
876/*
877 * It only removes the dentry from the dentry page, corresponding name
878 * entry in name page does not need to be touched during deletion.
879 */
880void f2fs_delete_entry(struct f2fs_dir_entry *dentry, struct folio *folio,
881 struct inode *dir, struct inode *inode)
882{
883 struct f2fs_dentry_block *dentry_blk;
884 unsigned int bit_pos;
885 int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len));
886 pgoff_t index = folio->index;
887 int i;
888
889 f2fs_update_time(F2FS_I_SB(dir), REQ_TIME);
890
891 if (F2FS_OPTION(F2FS_I_SB(dir)).fsync_mode == FSYNC_MODE_STRICT)
892 f2fs_add_ino_entry(F2FS_I_SB(dir), dir->i_ino, TRANS_DIR_INO);
893
894 if (f2fs_has_inline_dentry(dir))
895 return f2fs_delete_inline_entry(dentry, folio, dir, inode);
896
897 folio_lock(folio);
898 f2fs_folio_wait_writeback(folio, DATA, true, true);
899
900 dentry_blk = folio_address(folio);
901 bit_pos = dentry - dentry_blk->dentry;
902 for (i = 0; i < slots; i++)
903 __clear_bit_le(bit_pos + i, &dentry_blk->dentry_bitmap);
904
905 /* Let's check and deallocate this dentry page */
906 bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
907 NR_DENTRY_IN_BLOCK,
908 0);
909 folio_mark_dirty(folio);
910
911 if (bit_pos == NR_DENTRY_IN_BLOCK &&
912 !f2fs_truncate_hole(dir, index, index + 1)) {
913 f2fs_clear_page_cache_dirty_tag(folio);
914 folio_clear_dirty_for_io(folio);
915 folio_clear_uptodate(folio);
916 folio_detach_private(folio);
917
918 inode_dec_dirty_pages(dir);
919 f2fs_remove_dirty_inode(dir);
920 }
921 f2fs_folio_put(folio, true);
922
923 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
924 f2fs_mark_inode_dirty_sync(dir, false);
925
926 if (inode)
927 f2fs_drop_nlink(dir, inode);
928}
929
930bool f2fs_empty_dir(struct inode *dir)
931{
932 unsigned long bidx = 0;
933 unsigned int bit_pos;
934 struct f2fs_dentry_block *dentry_blk;
935 unsigned long nblock = dir_blocks(dir);
936
937 if (f2fs_has_inline_dentry(dir))
938 return f2fs_empty_inline_dir(dir);
939
940 while (bidx < nblock) {
941 pgoff_t next_pgofs;
942 struct folio *dentry_folio;
943
944 dentry_folio = f2fs_find_data_folio(dir, bidx, &next_pgofs);
945 if (IS_ERR(dentry_folio)) {
946 if (PTR_ERR(dentry_folio) == -ENOENT) {
947 bidx = next_pgofs;
948 continue;
949 } else {
950 return false;
951 }
952 }
953
954 dentry_blk = folio_address(dentry_folio);
955 if (bidx == 0)
956 bit_pos = 2;
957 else
958 bit_pos = 0;
959 bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
960 NR_DENTRY_IN_BLOCK,
961 bit_pos);
962
963 f2fs_folio_put(dentry_folio, false);
964
965 if (bit_pos < NR_DENTRY_IN_BLOCK)
966 return false;
967
968 bidx++;
969 }
970 return true;
971}
972
973int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
974 unsigned int start_pos, struct fscrypt_str *fstr)
975{
976 unsigned char d_type = DT_UNKNOWN;
977 unsigned int bit_pos;
978 struct f2fs_dir_entry *de = NULL;
979 struct fscrypt_str de_name = FSTR_INIT(NULL, 0);
980 struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
981 struct blk_plug plug;
982 bool readdir_ra = sbi->readdir_ra;
983 bool found_valid_dirent = false;
984 int err = 0;
985
986 bit_pos = ((unsigned long)ctx->pos % d->max);
987
988 if (readdir_ra)
989 blk_start_plug(&plug);
990
991 while (bit_pos < d->max) {
992 bit_pos = find_next_bit_le(d->bitmap, d->max, bit_pos);
993 if (bit_pos >= d->max)
994 break;
995
996 de = &d->dentry[bit_pos];
997 if (de->name_len == 0) {
998 if (found_valid_dirent || !bit_pos) {
999 f2fs_warn_ratelimited(sbi,
1000 "invalid namelen(0), ino:%u, run fsck to fix.",
1001 le32_to_cpu(de->ino));
1002 set_sbi_flag(sbi, SBI_NEED_FSCK);
1003 }
1004 bit_pos++;
1005 ctx->pos = start_pos + bit_pos;
1006 continue;
1007 }
1008
1009 d_type = fs_ftype_to_dtype(de->file_type);
1010
1011 de_name.name = d->filename[bit_pos];
1012 de_name.len = le16_to_cpu(de->name_len);
1013
1014 /* check memory boundary before moving forward */
1015 bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
1016 if (unlikely(bit_pos > d->max ||
1017 le16_to_cpu(de->name_len) > F2FS_NAME_LEN)) {
1018 f2fs_warn(sbi, "%s: corrupted namelen=%d, run fsck to fix.",
1019 __func__, le16_to_cpu(de->name_len));
1020 set_sbi_flag(sbi, SBI_NEED_FSCK);
1021 err = -EFSCORRUPTED;
1022 f2fs_handle_error(sbi, ERROR_CORRUPTED_DIRENT);
1023 goto out;
1024 }
1025
1026 if (IS_ENCRYPTED(d->inode)) {
1027 int save_len = fstr->len;
1028
1029 err = fscrypt_fname_disk_to_usr(d->inode,
1030 (u32)le32_to_cpu(de->hash_code),
1031 0, &de_name, fstr);
1032 if (err)
1033 goto out;
1034
1035 de_name = *fstr;
1036 fstr->len = save_len;
1037 }
1038
1039 if (!dir_emit(ctx, de_name.name, de_name.len,
1040 le32_to_cpu(de->ino), d_type)) {
1041 err = 1;
1042 goto out;
1043 }
1044
1045 if (readdir_ra)
1046 f2fs_ra_node_page(sbi, le32_to_cpu(de->ino));
1047
1048 ctx->pos = start_pos + bit_pos;
1049 found_valid_dirent = true;
1050 }
1051out:
1052 if (readdir_ra)
1053 blk_finish_plug(&plug);
1054 return err;
1055}
1056
1057static int f2fs_readdir(struct file *file, struct dir_context *ctx)
1058{
1059 struct inode *inode = file_inode(file);
1060 unsigned long npages = dir_blocks(inode);
1061 struct f2fs_dentry_block *dentry_blk = NULL;
1062 struct file_ra_state *ra = &file->f_ra;
1063 loff_t start_pos = ctx->pos;
1064 unsigned int n = ((unsigned long)ctx->pos / NR_DENTRY_IN_BLOCK);
1065 struct f2fs_dentry_ptr d;
1066 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
1067 int err = 0;
1068
1069 if (IS_ENCRYPTED(inode)) {
1070 err = fscrypt_prepare_readdir(inode);
1071 if (err)
1072 goto out;
1073
1074 err = fscrypt_fname_alloc_buffer(F2FS_NAME_LEN, &fstr);
1075 if (err < 0)
1076 goto out;
1077 }
1078
1079 if (f2fs_has_inline_dentry(inode)) {
1080 err = f2fs_read_inline_dir(file, ctx, &fstr);
1081 goto out_free;
1082 }
1083
1084 for (; n < npages; ctx->pos = n * NR_DENTRY_IN_BLOCK) {
1085 struct folio *dentry_folio;
1086 pgoff_t next_pgofs;
1087
1088 /* allow readdir() to be interrupted */
1089 if (fatal_signal_pending(current)) {
1090 err = -ERESTARTSYS;
1091 goto out_free;
1092 }
1093 cond_resched();
1094
1095 /* readahead for multi pages of dir */
1096 if (npages - n > 1 && !ra_has_index(ra, n))
1097 page_cache_sync_readahead(inode->i_mapping, ra, file, n,
1098 min(npages - n, (pgoff_t)MAX_DIR_RA_PAGES));
1099
1100 dentry_folio = f2fs_find_data_folio(inode, n, &next_pgofs);
1101 if (IS_ERR(dentry_folio)) {
1102 err = PTR_ERR(dentry_folio);
1103 if (err == -ENOENT) {
1104 err = 0;
1105 n = next_pgofs;
1106 continue;
1107 } else {
1108 goto out_free;
1109 }
1110 }
1111
1112 dentry_blk = folio_address(dentry_folio);
1113
1114 make_dentry_ptr_block(inode, &d, dentry_blk);
1115
1116 err = f2fs_fill_dentries(ctx, &d,
1117 n * NR_DENTRY_IN_BLOCK, &fstr);
1118 f2fs_folio_put(dentry_folio, false);
1119 if (err)
1120 break;
1121
1122 n++;
1123 }
1124out_free:
1125 fscrypt_fname_free_buffer(&fstr);
1126out:
1127 trace_f2fs_readdir(inode, start_pos, ctx->pos, err);
1128 return err < 0 ? err : 0;
1129}
1130
1131const struct file_operations f2fs_dir_operations = {
1132 .llseek = generic_file_llseek,
1133 .read = generic_read_dir,
1134 .iterate_shared = f2fs_readdir,
1135 .fsync = f2fs_sync_file,
1136 .unlocked_ioctl = f2fs_ioctl,
1137#ifdef CONFIG_COMPAT
1138 .compat_ioctl = f2fs_compat_ioctl,
1139#endif
1140 .setlease = generic_setlease,
1141};