1 // SPDX-License-Identifier: GPL-2.0-only 1
2 /*
3 * fs/libfs.c
4 * Library for filesystems writers.
5 */
6
7 #include <linux/blkdev.h>
8 #include <linux/export.h>
9 #include <linux/pagemap.h>
10 #include <linux/slab.h>
11 #include <linux/cred.h>
12 #include <linux/mount.h>
13 #include <linux/vfs.h>
14 #include <linux/quotaops.h>
15 #include <linux/mutex.h>
16 #include <linux/namei.h>
17 #include <linux/exportfs.h>
18 #include <linux/iversion.h>
19 #include <linux/writeback.h>
20 #include <linux/buffer_head.h> /* sync_mapping
21 #include <linux/fs_context.h>
22 #include <linux/pseudo_fs.h>
23 #include <linux/fsnotify.h>
24 #include <linux/unicode.h>
25 #include <linux/fscrypt.h>
26 #include <linux/pidfs.h>
27
28 #include <linux/uaccess.h>
29
30 #include "internal.h"
31
32 int simple_getattr(struct mnt_idmap *idmap, co
33 struct kstat *stat, u32 req
34 unsigned int query_flags)
35 {
36 struct inode *inode = d_inode(path->de
37 generic_fillattr(&nop_mnt_idmap, reque
38 stat->blocks = inode->i_mapping->nrpag
39 return 0;
40 }
41 EXPORT_SYMBOL(simple_getattr);
42
43 int simple_statfs(struct dentry *dentry, struc
44 {
45 u64 id = huge_encode_dev(dentry->d_sb-
46
47 buf->f_fsid = u64_to_fsid(id);
48 buf->f_type = dentry->d_sb->s_magic;
49 buf->f_bsize = PAGE_SIZE;
50 buf->f_namelen = NAME_MAX;
51 return 0;
52 }
53 EXPORT_SYMBOL(simple_statfs);
54
55 /*
56 * Retaining negative dentries for an in-memor
57 * memory and lookup time: arrange for them to
58 */
59 int always_delete_dentry(const struct dentry *
60 {
61 return 1;
62 }
63 EXPORT_SYMBOL(always_delete_dentry);
64
65 const struct dentry_operations simple_dentry_o
66 .d_delete = always_delete_dentry,
67 };
68 EXPORT_SYMBOL(simple_dentry_operations);
69
70 /*
71 * Lookup the data. This is trivial - if the d
72 * exist, we know it is negative. Set d_op to
73 */
74 struct dentry *simple_lookup(struct inode *dir
75 {
76 if (dentry->d_name.len > NAME_MAX)
77 return ERR_PTR(-ENAMETOOLONG);
78 if (!dentry->d_sb->s_d_op)
79 d_set_d_op(dentry, &simple_den
80 d_add(dentry, NULL);
81 return NULL;
82 }
83 EXPORT_SYMBOL(simple_lookup);
84
85 int dcache_dir_open(struct inode *inode, struc
86 {
87 file->private_data = d_alloc_cursor(fi
88
89 return file->private_data ? 0 : -ENOME
90 }
91 EXPORT_SYMBOL(dcache_dir_open);
92
93 int dcache_dir_close(struct inode *inode, stru
94 {
95 dput(file->private_data);
96 return 0;
97 }
98 EXPORT_SYMBOL(dcache_dir_close);
99
100 /* parent is locked at least shared */
101 /*
102 * Returns an element of siblings' list.
103 * We are looking for <count>th positive after
104 * found, dentry is grabbed and returned to ca
105 * If no such element exists, NULL is returned
106 */
107 static struct dentry *scan_positives(struct de
108 struct
109 loff_t
110 struct
111 {
112 struct dentry *dentry = cursor->d_pare
113
114 spin_lock(&dentry->d_lock);
115 while (*p) {
116 struct dentry *d = hlist_entry
117 p = &d->d_sib.next;
118 // we must at least skip curso
119 if (d->d_flags & DCACHE_DENTRY
120 continue;
121 if (simple_positive(d) && !--c
122 spin_lock_nested(&d->d
123 if (simple_positive(d)
124 found = dget_d
125 spin_unlock(&d->d_lock
126 if (likely(found))
127 break;
128 count = 1;
129 }
130 if (need_resched()) {
131 if (!hlist_unhashed(&c
132 __hlist_del(&c
133 hlist_add_behind(&curs
134 p = &cursor->d_sib.nex
135 spin_unlock(&dentry->d
136 cond_resched();
137 spin_lock(&dentry->d_l
138 }
139 }
140 spin_unlock(&dentry->d_lock);
141 dput(last);
142 return found;
143 }
144
145 loff_t dcache_dir_lseek(struct file *file, lof
146 {
147 struct dentry *dentry = file->f_path.d
148 switch (whence) {
149 case 1:
150 offset += file->f_pos;
151 fallthrough;
152 case 0:
153 if (offset >= 0)
154 break;
155 fallthrough;
156 default:
157 return -EINVAL;
158 }
159 if (offset != file->f_pos) {
160 struct dentry *cursor = file->
161 struct dentry *to = NULL;
162
163 inode_lock_shared(dentry->d_in
164
165 if (offset > 2)
166 to = scan_positives(cu
167 of
168 spin_lock(&dentry->d_lock);
169 hlist_del_init(&cursor->d_sib)
170 if (to)
171 hlist_add_behind(&curs
172 spin_unlock(&dentry->d_lock);
173 dput(to);
174
175 file->f_pos = offset;
176
177 inode_unlock_shared(dentry->d_
178 }
179 return offset;
180 }
181 EXPORT_SYMBOL(dcache_dir_lseek);
182
183 /*
184 * Directory is locked and all positive dentri
185 * for ramfs-type trees they can't go away wit
186 * both impossible due to the lock on director
187 */
188
189 int dcache_readdir(struct file *file, struct d
190 {
191 struct dentry *dentry = file->f_path.d
192 struct dentry *cursor = file->private_
193 struct dentry *next = NULL;
194 struct hlist_node **p;
195
196 if (!dir_emit_dots(file, ctx))
197 return 0;
198
199 if (ctx->pos == 2)
200 p = &dentry->d_children.first;
201 else
202 p = &cursor->d_sib.next;
203
204 while ((next = scan_positives(cursor,
205 if (!dir_emit(ctx, next->d_nam
206 d_inode(next)->i
207 fs_umode_to_dtyp
208 break;
209 ctx->pos++;
210 p = &next->d_sib.next;
211 }
212 spin_lock(&dentry->d_lock);
213 hlist_del_init(&cursor->d_sib);
214 if (next)
215 hlist_add_before(&cursor->d_si
216 spin_unlock(&dentry->d_lock);
217 dput(next);
218
219 return 0;
220 }
221 EXPORT_SYMBOL(dcache_readdir);
222
223 ssize_t generic_read_dir(struct file *filp, ch
224 {
225 return -EISDIR;
226 }
227 EXPORT_SYMBOL(generic_read_dir);
228
229 const struct file_operations simple_dir_operat
230 .open = dcache_dir_open,
231 .release = dcache_dir_close,
232 .llseek = dcache_dir_lseek,
233 .read = generic_read_dir,
234 .iterate_shared = dcache_readdir,
235 .fsync = noop_fsync,
236 };
237 EXPORT_SYMBOL(simple_dir_operations);
238
239 const struct inode_operations simple_dir_inode
240 .lookup = simple_lookup,
241 };
242 EXPORT_SYMBOL(simple_dir_inode_operations);
243
244 /* 0 is '.', 1 is '..', so always start with o
245 enum {
246 DIR_OFFSET_MIN = 2,
247 };
248
249 static void offset_set(struct dentry *dentry,
250 {
251 dentry->d_fsdata = (void *)offset;
252 }
253
254 static long dentry2offset(struct dentry *dentr
255 {
256 return (long)dentry->d_fsdata;
257 }
258
259 static struct lock_class_key simple_offset_loc
260
261 /**
262 * simple_offset_init - initialize an offset_c
263 * @octx: directory offset map to be initializ
264 *
265 */
266 void simple_offset_init(struct offset_ctx *oct
267 {
268 mt_init_flags(&octx->mt, MT_FLAGS_ALLO
269 lockdep_set_class(&octx->mt.ma_lock, &
270 octx->next_offset = DIR_OFFSET_MIN;
271 }
272
273 /**
274 * simple_offset_add - Add an entry to a direc
275 * @octx: directory offset ctx to be updated
276 * @dentry: new dentry being added
277 *
278 * Returns zero on success. @octx and the dent
279 * Otherwise, a negative errno value is return
280 */
281 int simple_offset_add(struct offset_ctx *octx,
282 {
283 unsigned long offset;
284 int ret;
285
286 if (dentry2offset(dentry) != 0)
287 return -EBUSY;
288
289 ret = mtree_alloc_cyclic(&octx->mt, &o
290 LONG_MAX, &oc
291 if (ret < 0)
292 return ret;
293
294 offset_set(dentry, offset);
295 return 0;
296 }
297
298 static int simple_offset_replace(struct offset
299 long offset)
300 {
301 int ret;
302
303 ret = mtree_store(&octx->mt, offset, d
304 if (ret)
305 return ret;
306 offset_set(dentry, offset);
307 return 0;
308 }
309
310 /**
311 * simple_offset_remove - Remove an entry to a
312 * @octx: directory offset ctx to be updated
313 * @dentry: dentry being removed
314 *
315 */
316 void simple_offset_remove(struct offset_ctx *o
317 {
318 long offset;
319
320 offset = dentry2offset(dentry);
321 if (offset == 0)
322 return;
323
324 mtree_erase(&octx->mt, offset);
325 offset_set(dentry, 0);
326 }
327
328 /**
329 * simple_offset_empty - Check if a dentry can
330 * @dentry: dentry to be tested
331 *
332 * Returns 0 if @dentry is a non-empty directo
333 */
334 int simple_offset_empty(struct dentry *dentry)
335 {
336 struct inode *inode = d_inode(dentry);
337 struct offset_ctx *octx;
338 struct dentry *child;
339 unsigned long index;
340 int ret = 1;
341
342 if (!inode || !S_ISDIR(inode->i_mode))
343 return ret;
344
345 index = DIR_OFFSET_MIN;
346 octx = inode->i_op->get_offset_ctx(ino
347 mt_for_each(&octx->mt, child, index, L
348 spin_lock(&child->d_lock);
349 if (simple_positive(child)) {
350 spin_unlock(&child->d_
351 ret = 0;
352 break;
353 }
354 spin_unlock(&child->d_lock);
355 }
356
357 return ret;
358 }
359
360 /**
361 * simple_offset_rename - handle directory off
362 * @old_dir: parent directory of source entry
363 * @old_dentry: dentry of source entry
364 * @new_dir: parent_directory of destination e
365 * @new_dentry: dentry of destination
366 *
367 * Caller provides appropriate serialization.
368 *
369 * User space expects the directory offset val
370 * (new) directory entry to be unchanged after
371 *
372 * Returns zero on success, a negative errno v
373 */
374 int simple_offset_rename(struct inode *old_dir
375 struct inode *new_dir
376 {
377 struct offset_ctx *old_ctx = old_dir->
378 struct offset_ctx *new_ctx = new_dir->
379 long new_offset = dentry2offset(new_de
380
381 simple_offset_remove(old_ctx, old_dent
382
383 if (new_offset) {
384 offset_set(new_dentry, 0);
385 return simple_offset_replace(n
386 }
387 return simple_offset_add(new_ctx, old_
388 }
389
390 /**
391 * simple_offset_rename_exchange - exchange re
392 * @old_dir: parent of dentry being moved
393 * @old_dentry: dentry being moved
394 * @new_dir: destination parent
395 * @new_dentry: destination dentry
396 *
397 * This API preserves the directory offset val
398 * appropriate serialization.
399 *
400 * Returns zero on success. Otherwise a negati
401 * rename is rolled back.
402 */
403 int simple_offset_rename_exchange(struct inode
404 struct dentr
405 struct inode
406 struct dentr
407 {
408 struct offset_ctx *old_ctx = old_dir->
409 struct offset_ctx *new_ctx = new_dir->
410 long old_index = dentry2offset(old_den
411 long new_index = dentry2offset(new_den
412 int ret;
413
414 simple_offset_remove(old_ctx, old_dent
415 simple_offset_remove(new_ctx, new_dent
416
417 ret = simple_offset_replace(new_ctx, o
418 if (ret)
419 goto out_restore;
420
421 ret = simple_offset_replace(old_ctx, n
422 if (ret) {
423 simple_offset_remove(new_ctx,
424 goto out_restore;
425 }
426
427 ret = simple_rename_exchange(old_dir,
428 if (ret) {
429 simple_offset_remove(new_ctx,
430 simple_offset_remove(old_ctx,
431 goto out_restore;
432 }
433 return 0;
434
435 out_restore:
436 (void)simple_offset_replace(old_ctx, o
437 (void)simple_offset_replace(new_ctx, n
438 return ret;
439 }
440
441 /**
442 * simple_offset_destroy - Release offset map
443 * @octx: directory offset ctx that is about t
444 *
445 * During fs teardown (eg. umount), a director
446 * contain entries. xa_destroy() cleans out an
447 */
448 void simple_offset_destroy(struct offset_ctx *
449 {
450 mtree_destroy(&octx->mt);
451 }
452
453 static int offset_dir_open(struct inode *inode
454 {
455 struct offset_ctx *ctx = inode->i_op->
456
457 file->private_data = (void *)ctx->next
458 return 0;
459 }
460
461 /**
462 * offset_dir_llseek - Advance the read positi
463 * @file: an open directory whose position is
464 * @offset: a byte offset
465 * @whence: enumerator describing the starting
466 *
467 * SEEK_END, SEEK_DATA, and SEEK_HOLE are not
468 *
469 * Returns the updated read position if succes
470 * negative errno is returned and the read pos
471 */
472 static loff_t offset_dir_llseek(struct file *f
473 {
474 struct inode *inode = file->f_inode;
475 struct offset_ctx *ctx = inode->i_op->
476
477 switch (whence) {
478 case SEEK_CUR:
479 offset += file->f_pos;
480 fallthrough;
481 case SEEK_SET:
482 if (offset >= 0)
483 break;
484 fallthrough;
485 default:
486 return -EINVAL;
487 }
488
489 /* In this case, ->private_data is pro
490 if (!offset)
491 file->private_data = (void *)c
492 return vfs_setpos(file, offset, LONG_M
493 }
494
495 static struct dentry *offset_find_next(struct
496 {
497 MA_STATE(mas, &octx->mt, offset, offse
498 struct dentry *child, *found = NULL;
499
500 rcu_read_lock();
501 child = mas_find(&mas, LONG_MAX);
502 if (!child)
503 goto out;
504 spin_lock(&child->d_lock);
505 if (simple_positive(child))
506 found = dget_dlock(child);
507 spin_unlock(&child->d_lock);
508 out:
509 rcu_read_unlock();
510 return found;
511 }
512
513 static bool offset_dir_emit(struct dir_context
514 {
515 struct inode *inode = d_inode(dentry);
516 long offset = dentry2offset(dentry);
517
518 return ctx->actor(ctx, dentry->d_name.
519 inode->i_ino, fs_umo
520 }
521
522 static void offset_iterate_dir(struct inode *i
523 {
524 struct offset_ctx *octx = inode->i_op-
525 struct dentry *dentry;
526
527 while (true) {
528 dentry = offset_find_next(octx
529 if (!dentry)
530 return;
531
532 if (dentry2offset(dentry) >= l
533 dput(dentry);
534 return;
535 }
536
537 if (!offset_dir_emit(ctx, dent
538 dput(dentry);
539 return;
540 }
541
542 ctx->pos = dentry2offset(dentr
543 dput(dentry);
544 }
545 }
546
547 /**
548 * offset_readdir - Emit entries starting at o
549 * @file: an open directory to iterate over
550 * @ctx: directory iteration context
551 *
552 * Caller must hold @file's i_rwsem to prevent
553 * entries during this call.
554 *
555 * On entry, @ctx->pos contains an offset that
556 * to be read from the directory.
557 *
558 * The operation continues until there are no
559 * until the ctx->actor indicates there is no
560 * output buffer.
561 *
562 * On return, @ctx->pos contains an offset tha
563 * in this directory when offset_readdir() is
564 *
565 * Return values:
566 * %0 - Complete
567 */
568 static int offset_readdir(struct file *file, s
569 {
570 struct dentry *dir = file->f_path.dent
571 long last_index = (long)file->private_
572
573 lockdep_assert_held(&d_inode(dir)->i_r
574
575 if (!dir_emit_dots(file, ctx))
576 return 0;
577
578 offset_iterate_dir(d_inode(dir), ctx,
579 return 0;
580 }
581
582 const struct file_operations simple_offset_dir
583 .open = offset_dir_open,
584 .llseek = offset_dir_llseek,
585 .iterate_shared = offset_readdir,
586 .read = generic_read_dir,
587 .fsync = noop_fsync,
588 };
589
590 static struct dentry *find_next_child(struct d
591 {
592 struct dentry *child = NULL, *d;
593
594 spin_lock(&parent->d_lock);
595 d = prev ? d_next_sibling(prev) : d_fi
596 hlist_for_each_entry_from(d, d_sib) {
597 if (simple_positive(d)) {
598 spin_lock_nested(&d->d
599 if (simple_positive(d)
600 child = dget_d
601 spin_unlock(&d->d_lock
602 if (likely(child))
603 break;
604 }
605 }
606 spin_unlock(&parent->d_lock);
607 dput(prev);
608 return child;
609 }
610
611 void simple_recursive_removal(struct dentry *d
612 void (*callback)
613 {
614 struct dentry *this = dget(dentry);
615 while (true) {
616 struct dentry *victim = NULL,
617 struct inode *inode = this->d_
618
619 inode_lock(inode);
620 if (d_is_dir(this))
621 inode->i_flags |= S_DE
622 while ((child = find_next_chil
623 // kill and ascend
624 // update metadata whi
625 inode_set_ctime_curren
626 clear_nlink(inode);
627 inode_unlock(inode);
628 victim = this;
629 this = this->d_parent;
630 inode = this->d_inode;
631 inode_lock(inode);
632 if (simple_positive(vi
633 d_invalidate(v
634 if (d_is_dir(v
635 fsnoti
636 else
637 fsnoti
638 if (callback)
639 callba
640 dput(victim);
641 }
642 if (victim == dentry)
643 inode_set_mtim
644
645 if (d_is_dir(d
646 drop_n
647 inode_unlock(i
648 dput(dentry);
649 return;
650 }
651 }
652 inode_unlock(inode);
653 this = child;
654 }
655 }
656 EXPORT_SYMBOL(simple_recursive_removal);
657
658 static const struct super_operations simple_su
659 .statfs = simple_statfs,
660 };
661
662 static int pseudo_fs_fill_super(struct super_b
663 {
664 struct pseudo_fs_context *ctx = fc->fs
665 struct inode *root;
666
667 s->s_maxbytes = MAX_LFS_FILESIZE;
668 s->s_blocksize = PAGE_SIZE;
669 s->s_blocksize_bits = PAGE_SHIFT;
670 s->s_magic = ctx->magic;
671 s->s_op = ctx->ops ?: &simple_super_op
672 s->s_xattr = ctx->xattr;
673 s->s_time_gran = 1;
674 root = new_inode(s);
675 if (!root)
676 return -ENOMEM;
677
678 /*
679 * since this is the first inode, make
680 * after this must take care not to co
681 * max_reserved of 1 to iunique).
682 */
683 root->i_ino = 1;
684 root->i_mode = S_IFDIR | S_IRUSR | S_I
685 simple_inode_init_ts(root);
686 s->s_root = d_make_root(root);
687 if (!s->s_root)
688 return -ENOMEM;
689 s->s_d_op = ctx->dops;
690 return 0;
691 }
692
693 static int pseudo_fs_get_tree(struct fs_contex
694 {
695 return get_tree_nodev(fc, pseudo_fs_fi
696 }
697
698 static void pseudo_fs_free(struct fs_context *
699 {
700 kfree(fc->fs_private);
701 }
702
703 static const struct fs_context_operations pseu
704 .free = pseudo_fs_free,
705 .get_tree = pseudo_fs_get_tree,
706 };
707
708 /*
709 * Common helper for pseudo-filesystems (sockf
710 * will never be mountable)
711 */
712 struct pseudo_fs_context *init_pseudo(struct f
713 unsign
714 {
715 struct pseudo_fs_context *ctx;
716
717 ctx = kzalloc(sizeof(struct pseudo_fs_
718 if (likely(ctx)) {
719 ctx->magic = magic;
720 fc->fs_private = ctx;
721 fc->ops = &pseudo_fs_context_o
722 fc->sb_flags |= SB_NOUSER;
723 fc->global = true;
724 }
725 return ctx;
726 }
727 EXPORT_SYMBOL(init_pseudo);
728
729 int simple_open(struct inode *inode, struct fi
730 {
731 if (inode->i_private)
732 file->private_data = inode->i_
733 return 0;
734 }
735 EXPORT_SYMBOL(simple_open);
736
737 int simple_link(struct dentry *old_dentry, str
738 {
739 struct inode *inode = d_inode(old_dent
740
741 inode_set_mtime_to_ts(dir,
742 inode_set_ctime_
743 inc_nlink(inode);
744 ihold(inode);
745 dget(dentry);
746 d_instantiate(dentry, inode);
747 return 0;
748 }
749 EXPORT_SYMBOL(simple_link);
750
751 int simple_empty(struct dentry *dentry)
752 {
753 struct dentry *child;
754 int ret = 0;
755
756 spin_lock(&dentry->d_lock);
757 hlist_for_each_entry(child, &dentry->d
758 spin_lock_nested(&child->d_loc
759 if (simple_positive(child)) {
760 spin_unlock(&child->d_
761 goto out;
762 }
763 spin_unlock(&child->d_lock);
764 }
765 ret = 1;
766 out:
767 spin_unlock(&dentry->d_lock);
768 return ret;
769 }
770 EXPORT_SYMBOL(simple_empty);
771
772 int simple_unlink(struct inode *dir, struct de
773 {
774 struct inode *inode = d_inode(dentry);
775
776 inode_set_mtime_to_ts(dir,
777 inode_set_ctime_
778 drop_nlink(inode);
779 dput(dentry);
780 return 0;
781 }
782 EXPORT_SYMBOL(simple_unlink);
783
784 int simple_rmdir(struct inode *dir, struct den
785 {
786 if (!simple_empty(dentry))
787 return -ENOTEMPTY;
788
789 drop_nlink(d_inode(dentry));
790 simple_unlink(dir, dentry);
791 drop_nlink(dir);
792 return 0;
793 }
794 EXPORT_SYMBOL(simple_rmdir);
795
796 /**
797 * simple_rename_timestamp - update the variou
798 * @old_dir: old parent directory
799 * @old_dentry: dentry that is being renamed
800 * @new_dir: new parent directory
801 * @new_dentry: target for rename
802 *
803 * POSIX mandates that the old and new parent
804 * mtime updated, and that inodes of @old_dent
805 * their ctime updated.
806 */
807 void simple_rename_timestamp(struct inode *old
808 struct inode *new
809 {
810 struct inode *newino = d_inode(new_den
811
812 inode_set_mtime_to_ts(old_dir, inode_s
813 if (new_dir != old_dir)
814 inode_set_mtime_to_ts(new_dir,
815 inode_se
816 inode_set_ctime_current(d_inode(old_de
817 if (newino)
818 inode_set_ctime_current(newino
819 }
820 EXPORT_SYMBOL_GPL(simple_rename_timestamp);
821
822 int simple_rename_exchange(struct inode *old_d
823 struct inode *new_d
824 {
825 bool old_is_dir = d_is_dir(old_dentry)
826 bool new_is_dir = d_is_dir(new_dentry)
827
828 if (old_dir != new_dir && old_is_dir !
829 if (old_is_dir) {
830 drop_nlink(old_dir);
831 inc_nlink(new_dir);
832 } else {
833 drop_nlink(new_dir);
834 inc_nlink(old_dir);
835 }
836 }
837 simple_rename_timestamp(old_dir, old_d
838 return 0;
839 }
840 EXPORT_SYMBOL_GPL(simple_rename_exchange);
841
842 int simple_rename(struct mnt_idmap *idmap, str
843 struct dentry *old_dentry, s
844 struct dentry *new_dentry, u
845 {
846 int they_are_dirs = d_is_dir(old_dentr
847
848 if (flags & ~(RENAME_NOREPLACE | RENAM
849 return -EINVAL;
850
851 if (flags & RENAME_EXCHANGE)
852 return simple_rename_exchange(
853
854 if (!simple_empty(new_dentry))
855 return -ENOTEMPTY;
856
857 if (d_really_is_positive(new_dentry))
858 simple_unlink(new_dir, new_den
859 if (they_are_dirs) {
860 drop_nlink(d_inode(new
861 drop_nlink(old_dir);
862 }
863 } else if (they_are_dirs) {
864 drop_nlink(old_dir);
865 inc_nlink(new_dir);
866 }
867
868 simple_rename_timestamp(old_dir, old_d
869 return 0;
870 }
871 EXPORT_SYMBOL(simple_rename);
872
873 /**
874 * simple_setattr - setattr for simple filesys
875 * @idmap: idmap of the target mount
876 * @dentry: dentry
877 * @iattr: iattr structure
878 *
879 * Returns 0 on success, -error on failure.
880 *
881 * simple_setattr is a simple ->setattr implem
882 * implementation of size changes.
883 *
884 * It can either be used for in-memory filesys
885 * on simple regular filesystems. Anything th
886 * or wire state on size changes needs its own
887 */
888 int simple_setattr(struct mnt_idmap *idmap, st
889 struct iattr *iattr)
890 {
891 struct inode *inode = d_inode(dentry);
892 int error;
893
894 error = setattr_prepare(idmap, dentry,
895 if (error)
896 return error;
897
898 if (iattr->ia_valid & ATTR_SIZE)
899 truncate_setsize(inode, iattr-
900 setattr_copy(idmap, inode, iattr);
901 mark_inode_dirty(inode);
902 return 0;
903 }
904 EXPORT_SYMBOL(simple_setattr);
905
906 static int simple_read_folio(struct file *file
907 {
908 folio_zero_range(folio, 0, folio_size(
909 flush_dcache_folio(folio);
910 folio_mark_uptodate(folio);
911 folio_unlock(folio);
912 return 0;
913 }
914
915 int simple_write_begin(struct file *file, stru
916 loff_t pos, unsigned l
917 struct folio **foliop,
918 {
919 struct folio *folio;
920
921 folio = __filemap_get_folio(mapping, p
922 mapping_gfp_mask(mappi
923 if (IS_ERR(folio))
924 return PTR_ERR(folio);
925
926 *foliop = folio;
927
928 if (!folio_test_uptodate(folio) && (le
929 size_t from = offset_in_folio(
930
931 folio_zero_segments(folio, 0,
932 from + len, fo
933 }
934 return 0;
935 }
936 EXPORT_SYMBOL(simple_write_begin);
937
938 /**
939 * simple_write_end - .write_end helper for no
940 * @file: See .write_end of address_space_oper
941 * @mapping: "
942 * @pos: "
943 * @len: "
944 * @copied: "
945 * @folio: "
946 * @fsdata: "
947 *
948 * simple_write_end does the minimum needed fo
949 * writing is done. It has the same API signat
950 * address_space_operations vector. So it can
951 * FSes that don't need any other processing.
952 * Block based filesystems should use generic_
953 * NOTE: Even though i_size might get updated
954 * is not called, so a filesystem that actuall
955 * should extend on what's done here with a ca
956 * case that i_size has changed.
957 *
958 * Use *ONLY* with simple_read_folio()
959 */
960 static int simple_write_end(struct file *file,
961 loff_t pos, unsigned l
962 struct folio *folio, v
963 {
964 struct inode *inode = folio->mapping->
965 loff_t last_pos = pos + copied;
966
967 /* zero the stale part of the folio if
968 if (!folio_test_uptodate(folio)) {
969 if (copied < len) {
970 size_t from = offset_i
971
972 folio_zero_range(folio
973 }
974 folio_mark_uptodate(folio);
975 }
976 /*
977 * No need to use i_size_read() here,
978 * cannot change under us because we h
979 */
980 if (last_pos > inode->i_size)
981 i_size_write(inode, last_pos);
982
983 folio_mark_dirty(folio);
984 folio_unlock(folio);
985 folio_put(folio);
986
987 return copied;
988 }
989
990 /*
991 * Provides ramfs-style behavior: data in the
992 */
993 const struct address_space_operations ram_aops
994 .read_folio = simple_read_folio,
995 .write_begin = simple_write_begin,
996 .write_end = simple_write_end,
997 .dirty_folio = noop_dirty_folio,
998 };
999 EXPORT_SYMBOL(ram_aops);
1000
1001 /*
1002 * the inodes created here are not hashed. If
1003 * unique inode values later for this filesys
1004 * to pass it an appropriate max_reserved val
1005 */
1006 int simple_fill_super(struct super_block *s,
1007 const struct tree_descr
1008 {
1009 struct inode *inode;
1010 struct dentry *dentry;
1011 int i;
1012
1013 s->s_blocksize = PAGE_SIZE;
1014 s->s_blocksize_bits = PAGE_SHIFT;
1015 s->s_magic = magic;
1016 s->s_op = &simple_super_operations;
1017 s->s_time_gran = 1;
1018
1019 inode = new_inode(s);
1020 if (!inode)
1021 return -ENOMEM;
1022 /*
1023 * because the root inode is 1, the f
1024 * entry at index 1
1025 */
1026 inode->i_ino = 1;
1027 inode->i_mode = S_IFDIR | 0755;
1028 simple_inode_init_ts(inode);
1029 inode->i_op = &simple_dir_inode_opera
1030 inode->i_fop = &simple_dir_operations
1031 set_nlink(inode, 2);
1032 s->s_root = d_make_root(inode);
1033 if (!s->s_root)
1034 return -ENOMEM;
1035 for (i = 0; !files->name || files->na
1036 if (!files->name)
1037 continue;
1038
1039 /* warn if it tries to confli
1040 if (unlikely(i == 1))
1041 printk(KERN_WARNING "
1042 "with an inde
1043 s->s_type->na
1044
1045 dentry = d_alloc_name(s->s_ro
1046 if (!dentry)
1047 return -ENOMEM;
1048 inode = new_inode(s);
1049 if (!inode) {
1050 dput(dentry);
1051 return -ENOMEM;
1052 }
1053 inode->i_mode = S_IFREG | fil
1054 simple_inode_init_ts(inode);
1055 inode->i_fop = files->ops;
1056 inode->i_ino = i;
1057 d_add(dentry, inode);
1058 }
1059 return 0;
1060 }
1061 EXPORT_SYMBOL(simple_fill_super);
1062
1063 static DEFINE_SPINLOCK(pin_fs_lock);
1064
1065 int simple_pin_fs(struct file_system_type *ty
1066 {
1067 struct vfsmount *mnt = NULL;
1068 spin_lock(&pin_fs_lock);
1069 if (unlikely(!*mount)) {
1070 spin_unlock(&pin_fs_lock);
1071 mnt = vfs_kern_mount(type, SB
1072 if (IS_ERR(mnt))
1073 return PTR_ERR(mnt);
1074 spin_lock(&pin_fs_lock);
1075 if (!*mount)
1076 *mount = mnt;
1077 }
1078 mntget(*mount);
1079 ++*count;
1080 spin_unlock(&pin_fs_lock);
1081 mntput(mnt);
1082 return 0;
1083 }
1084 EXPORT_SYMBOL(simple_pin_fs);
1085
1086 void simple_release_fs(struct vfsmount **moun
1087 {
1088 struct vfsmount *mnt;
1089 spin_lock(&pin_fs_lock);
1090 mnt = *mount;
1091 if (!--*count)
1092 *mount = NULL;
1093 spin_unlock(&pin_fs_lock);
1094 mntput(mnt);
1095 }
1096 EXPORT_SYMBOL(simple_release_fs);
1097
1098 /**
1099 * simple_read_from_buffer - copy data from t
1100 * @to: the user space buffer to read to
1101 * @count: the maximum number of bytes to rea
1102 * @ppos: the current position in the buffer
1103 * @from: the buffer to read from
1104 * @available: the size of the buffer
1105 *
1106 * The simple_read_from_buffer() function rea
1107 * buffer @from at offset @ppos into the user
1108 *
1109 * On success, the number of bytes read is re
1110 * advanced by this number, or negative value
1111 **/
1112 ssize_t simple_read_from_buffer(void __user *
1113 const void *f
1114 {
1115 loff_t pos = *ppos;
1116 size_t ret;
1117
1118 if (pos < 0)
1119 return -EINVAL;
1120 if (pos >= available || !count)
1121 return 0;
1122 if (count > available - pos)
1123 count = available - pos;
1124 ret = copy_to_user(to, from + pos, co
1125 if (ret == count)
1126 return -EFAULT;
1127 count -= ret;
1128 *ppos = pos + count;
1129 return count;
1130 }
1131 EXPORT_SYMBOL(simple_read_from_buffer);
1132
1133 /**
1134 * simple_write_to_buffer - copy data from us
1135 * @to: the buffer to write to
1136 * @available: the size of the buffer
1137 * @ppos: the current position in the buffer
1138 * @from: the user space buffer to read from
1139 * @count: the maximum number of bytes to rea
1140 *
1141 * The simple_write_to_buffer() function read
1142 * space address starting at @from into the b
1143 *
1144 * On success, the number of bytes written is
1145 * advanced by this number, or negative value
1146 **/
1147 ssize_t simple_write_to_buffer(void *to, size
1148 const void __user *from, size
1149 {
1150 loff_t pos = *ppos;
1151 size_t res;
1152
1153 if (pos < 0)
1154 return -EINVAL;
1155 if (pos >= available || !count)
1156 return 0;
1157 if (count > available - pos)
1158 count = available - pos;
1159 res = copy_from_user(to + pos, from,
1160 if (res == count)
1161 return -EFAULT;
1162 count -= res;
1163 *ppos = pos + count;
1164 return count;
1165 }
1166 EXPORT_SYMBOL(simple_write_to_buffer);
1167
1168 /**
1169 * memory_read_from_buffer - copy data from t
1170 * @to: the kernel space buffer to read to
1171 * @count: the maximum number of bytes to rea
1172 * @ppos: the current position in the buffer
1173 * @from: the buffer to read from
1174 * @available: the size of the buffer
1175 *
1176 * The memory_read_from_buffer() function rea
1177 * buffer @from at offset @ppos into the kern
1178 *
1179 * On success, the number of bytes read is re
1180 * advanced by this number, or negative value
1181 **/
1182 ssize_t memory_read_from_buffer(void *to, siz
1183 const void *f
1184 {
1185 loff_t pos = *ppos;
1186
1187 if (pos < 0)
1188 return -EINVAL;
1189 if (pos >= available)
1190 return 0;
1191 if (count > available - pos)
1192 count = available - pos;
1193 memcpy(to, from + pos, count);
1194 *ppos = pos + count;
1195
1196 return count;
1197 }
1198 EXPORT_SYMBOL(memory_read_from_buffer);
1199
1200 /*
1201 * Transaction based IO.
1202 * The file expects a single write which trig
1203 * possibly a read which collects the result
1204 * file-local buffer.
1205 */
1206
1207 void simple_transaction_set(struct file *file
1208 {
1209 struct simple_transaction_argresp *ar
1210
1211 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1212
1213 /*
1214 * The barrier ensures that ar->size
1215 * ar->data is ready for reading.
1216 */
1217 smp_mb();
1218 ar->size = n;
1219 }
1220 EXPORT_SYMBOL(simple_transaction_set);
1221
1222 char *simple_transaction_get(struct file *fil
1223 {
1224 struct simple_transaction_argresp *ar
1225 static DEFINE_SPINLOCK(simple_transac
1226
1227 if (size > SIMPLE_TRANSACTION_LIMIT -
1228 return ERR_PTR(-EFBIG);
1229
1230 ar = (struct simple_transaction_argre
1231 if (!ar)
1232 return ERR_PTR(-ENOMEM);
1233
1234 spin_lock(&simple_transaction_lock);
1235
1236 /* only one write allowed per open */
1237 if (file->private_data) {
1238 spin_unlock(&simple_transacti
1239 free_page((unsigned long)ar);
1240 return ERR_PTR(-EBUSY);
1241 }
1242
1243 file->private_data = ar;
1244
1245 spin_unlock(&simple_transaction_lock)
1246
1247 if (copy_from_user(ar->data, buf, siz
1248 return ERR_PTR(-EFAULT);
1249
1250 return ar->data;
1251 }
1252 EXPORT_SYMBOL(simple_transaction_get);
1253
1254 ssize_t simple_transaction_read(struct file *
1255 {
1256 struct simple_transaction_argresp *ar
1257
1258 if (!ar)
1259 return 0;
1260 return simple_read_from_buffer(buf, s
1261 }
1262 EXPORT_SYMBOL(simple_transaction_read);
1263
1264 int simple_transaction_release(struct inode *
1265 {
1266 free_page((unsigned long)file->privat
1267 return 0;
1268 }
1269 EXPORT_SYMBOL(simple_transaction_release);
1270
1271 /* Simple attribute files */
1272
1273 struct simple_attr {
1274 int (*get)(void *, u64 *);
1275 int (*set)(void *, u64);
1276 char get_buf[24]; /* enough to
1277 char set_buf[24];
1278 void *data;
1279 const char *fmt; /* format for
1280 struct mutex mutex; /* protects a
1281 };
1282
1283 /* simple_attr_open is called by an actual at
1284 * to set the attribute specific access opera
1285 int simple_attr_open(struct inode *inode, str
1286 int (*get)(void *, u64 *
1287 const char *fmt)
1288 {
1289 struct simple_attr *attr;
1290
1291 attr = kzalloc(sizeof(*attr), GFP_KER
1292 if (!attr)
1293 return -ENOMEM;
1294
1295 attr->get = get;
1296 attr->set = set;
1297 attr->data = inode->i_private;
1298 attr->fmt = fmt;
1299 mutex_init(&attr->mutex);
1300
1301 file->private_data = attr;
1302
1303 return nonseekable_open(inode, file);
1304 }
1305 EXPORT_SYMBOL_GPL(simple_attr_open);
1306
1307 int simple_attr_release(struct inode *inode,
1308 {
1309 kfree(file->private_data);
1310 return 0;
1311 }
1312 EXPORT_SYMBOL_GPL(simple_attr_release); /* GP
1313
1314 /* read from the buffer that is filled with t
1315 ssize_t simple_attr_read(struct file *file, c
1316 size_t len, loff_t *
1317 {
1318 struct simple_attr *attr;
1319 size_t size;
1320 ssize_t ret;
1321
1322 attr = file->private_data;
1323
1324 if (!attr->get)
1325 return -EACCES;
1326
1327 ret = mutex_lock_interruptible(&attr-
1328 if (ret)
1329 return ret;
1330
1331 if (*ppos && attr->get_buf[0]) {
1332 /* continued read */
1333 size = strlen(attr->get_buf);
1334 } else {
1335 /* first read */
1336 u64 val;
1337 ret = attr->get(attr->data, &
1338 if (ret)
1339 goto out;
1340
1341 size = scnprintf(attr->get_bu
1342 attr->fmt, (
1343 }
1344
1345 ret = simple_read_from_buffer(buf, le
1346 out:
1347 mutex_unlock(&attr->mutex);
1348 return ret;
1349 }
1350 EXPORT_SYMBOL_GPL(simple_attr_read);
1351
1352 /* interpret the buffer as a number to call t
1353 static ssize_t simple_attr_write_xsigned(stru
1354 size_t len, loff_t
1355 {
1356 struct simple_attr *attr;
1357 unsigned long long val;
1358 size_t size;
1359 ssize_t ret;
1360
1361 attr = file->private_data;
1362 if (!attr->set)
1363 return -EACCES;
1364
1365 ret = mutex_lock_interruptible(&attr-
1366 if (ret)
1367 return ret;
1368
1369 ret = -EFAULT;
1370 size = min(sizeof(attr->set_buf) - 1,
1371 if (copy_from_user(attr->set_buf, buf
1372 goto out;
1373
1374 attr->set_buf[size] = '\0';
1375 if (is_signed)
1376 ret = kstrtoll(attr->set_buf,
1377 else
1378 ret = kstrtoull(attr->set_buf
1379 if (ret)
1380 goto out;
1381 ret = attr->set(attr->data, val);
1382 if (ret == 0)
1383 ret = len; /* on success, cla
1384 out:
1385 mutex_unlock(&attr->mutex);
1386 return ret;
1387 }
1388
1389 ssize_t simple_attr_write(struct file *file,
1390 size_t len, loff_t
1391 {
1392 return simple_attr_write_xsigned(file
1393 }
1394 EXPORT_SYMBOL_GPL(simple_attr_write);
1395
1396 ssize_t simple_attr_write_signed(struct file
1397 size_t len, loff_t
1398 {
1399 return simple_attr_write_xsigned(file
1400 }
1401 EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1402
1403 /**
1404 * generic_encode_ino32_fh - generic export_o
1405 * @inode: the object to encode
1406 * @fh: where to store the file handle f
1407 * @max_len: maximum length to store there (i
1408 * @parent: parent directory inode, if wante
1409 *
1410 * This generic encode_fh function assumes th
1411 * is suitable for locating an inode, and tha
1412 * can be used to check that it is still vali
1413 * filehandle fragment where export_decode_fh
1414 */
1415 int generic_encode_ino32_fh(struct inode *ino
1416 struct inode *par
1417 {
1418 struct fid *fid = (void *)fh;
1419 int len = *max_len;
1420 int type = FILEID_INO32_GEN;
1421
1422 if (parent && (len < 4)) {
1423 *max_len = 4;
1424 return FILEID_INVALID;
1425 } else if (len < 2) {
1426 *max_len = 2;
1427 return FILEID_INVALID;
1428 }
1429
1430 len = 2;
1431 fid->i32.ino = inode->i_ino;
1432 fid->i32.gen = inode->i_generation;
1433 if (parent) {
1434 fid->i32.parent_ino = parent-
1435 fid->i32.parent_gen = parent-
1436 len = 4;
1437 type = FILEID_INO32_GEN_PAREN
1438 }
1439 *max_len = len;
1440 return type;
1441 }
1442 EXPORT_SYMBOL_GPL(generic_encode_ino32_fh);
1443
1444 /**
1445 * generic_fh_to_dentry - generic helper for
1446 * @sb: filesystem to do the file han
1447 * @fid: file handle to convert
1448 * @fh_len: length of the file handle in
1449 * @fh_type: type of file handle
1450 * @get_inode: filesystem callback to retrie
1451 *
1452 * This function decodes @fid as long as it h
1453 * Linux filehandle types and calls @get_inod
1454 * inode for the object specified in the file
1455 */
1456 struct dentry *generic_fh_to_dentry(struct su
1457 int fh_len, int fh_type, stru
1458 (struct super_block *
1459 {
1460 struct inode *inode = NULL;
1461
1462 if (fh_len < 2)
1463 return NULL;
1464
1465 switch (fh_type) {
1466 case FILEID_INO32_GEN:
1467 case FILEID_INO32_GEN_PARENT:
1468 inode = get_inode(sb, fid->i3
1469 break;
1470 }
1471
1472 return d_obtain_alias(inode);
1473 }
1474 EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1475
1476 /**
1477 * generic_fh_to_parent - generic helper for
1478 * @sb: filesystem to do the file han
1479 * @fid: file handle to convert
1480 * @fh_len: length of the file handle in
1481 * @fh_type: type of file handle
1482 * @get_inode: filesystem callback to retrie
1483 *
1484 * This function decodes @fid as long as it h
1485 * Linux filehandle types and calls @get_inod
1486 * inode for the _parent_ object specified in
1487 * is specified in the file handle, or NULL o
1488 */
1489 struct dentry *generic_fh_to_parent(struct su
1490 int fh_len, int fh_type, stru
1491 (struct super_block *
1492 {
1493 struct inode *inode = NULL;
1494
1495 if (fh_len <= 2)
1496 return NULL;
1497
1498 switch (fh_type) {
1499 case FILEID_INO32_GEN_PARENT:
1500 inode = get_inode(sb, fid->i3
1501 (fh_len > 3
1502 break;
1503 }
1504
1505 return d_obtain_alias(inode);
1506 }
1507 EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1508
1509 /**
1510 * __generic_file_fsync - generic fsync imple
1511 *
1512 * @file: file to synchronize
1513 * @start: start offset in bytes
1514 * @end: end offset in bytes (inclusiv
1515 * @datasync: only synchronize essential me
1516 *
1517 * This is a generic implementation of the fs
1518 * filesystems which track all non-inode meta
1519 * hanging off the address_space structure.
1520 */
1521 int __generic_file_fsync(struct file *file, l
1522 int datasync
1523 {
1524 struct inode *inode = file->f_mapping
1525 int err;
1526 int ret;
1527
1528 err = file_write_and_wait_range(file,
1529 if (err)
1530 return err;
1531
1532 inode_lock(inode);
1533 ret = sync_mapping_buffers(inode->i_m
1534 if (!(inode->i_state & I_DIRTY_ALL))
1535 goto out;
1536 if (datasync && !(inode->i_state & I_
1537 goto out;
1538
1539 err = sync_inode_metadata(inode, 1);
1540 if (ret == 0)
1541 ret = err;
1542
1543 out:
1544 inode_unlock(inode);
1545 /* check and advance again to catch e
1546 err = file_check_and_advance_wb_err(f
1547 if (ret == 0)
1548 ret = err;
1549 return ret;
1550 }
1551 EXPORT_SYMBOL(__generic_file_fsync);
1552
1553 /**
1554 * generic_file_fsync - generic fsync impleme
1555 * with flush
1556 * @file: file to synchronize
1557 * @start: start offset in bytes
1558 * @end: end offset in bytes (inclusiv
1559 * @datasync: only synchronize essential me
1560 *
1561 */
1562
1563 int generic_file_fsync(struct file *file, lof
1564 int datasync)
1565 {
1566 struct inode *inode = file->f_mapping
1567 int err;
1568
1569 err = __generic_file_fsync(file, star
1570 if (err)
1571 return err;
1572 return blkdev_issue_flush(inode->i_sb
1573 }
1574 EXPORT_SYMBOL(generic_file_fsync);
1575
1576 /**
1577 * generic_check_addressable - Check addressa
1578 * @blocksize_bits: log of file system bl
1579 * @num_blocks: number of blocks in f
1580 *
1581 * Determine whether a file system with @num_
1582 * block size of 2**@blocksize_bits) is addre
1583 * and page cache of the system. Return 0 if
1584 */
1585 int generic_check_addressable(unsigned blocks
1586 {
1587 u64 last_fs_block = num_blocks - 1;
1588 u64 last_fs_page =
1589 last_fs_block >> (PAGE_SHIFT
1590
1591 if (unlikely(num_blocks == 0))
1592 return 0;
1593
1594 if ((blocksize_bits < 9) || (blocksiz
1595 return -EINVAL;
1596
1597 if ((last_fs_block > (sector_t)(~0ULL
1598 (last_fs_page > (pgoff_t)(~0ULL))
1599 return -EFBIG;
1600 }
1601 return 0;
1602 }
1603 EXPORT_SYMBOL(generic_check_addressable);
1604
1605 /*
1606 * No-op implementation of ->fsync for in-mem
1607 */
1608 int noop_fsync(struct file *file, loff_t star
1609 {
1610 return 0;
1611 }
1612 EXPORT_SYMBOL(noop_fsync);
1613
1614 ssize_t noop_direct_IO(struct kiocb *iocb, st
1615 {
1616 /*
1617 * iomap based filesystems support di
1618 * this callback. However, it still n
1619 * inode->a_ops so that open/fcntl kn
1620 * generally supported.
1621 */
1622 return -EINVAL;
1623 }
1624 EXPORT_SYMBOL_GPL(noop_direct_IO);
1625
1626 /* Because kfree isn't assignment-compatible
1627 void kfree_link(void *p)
1628 {
1629 kfree(p);
1630 }
1631 EXPORT_SYMBOL(kfree_link);
1632
1633 struct inode *alloc_anon_inode(struct super_b
1634 {
1635 static const struct address_space_ope
1636 .dirty_folio = noop_dirty_
1637 };
1638 struct inode *inode = new_inode_pseud
1639
1640 if (!inode)
1641 return ERR_PTR(-ENOMEM);
1642
1643 inode->i_ino = get_next_ino();
1644 inode->i_mapping->a_ops = &anon_aops;
1645
1646 /*
1647 * Mark the inode dirty from the very
1648 * that way it will never be moved to
1649 * list because mark_inode_dirty() wi
1650 * that it already _is_ on the dirty
1651 */
1652 inode->i_state = I_DIRTY;
1653 inode->i_mode = S_IRUSR | S_IWUSR;
1654 inode->i_uid = current_fsuid();
1655 inode->i_gid = current_fsgid();
1656 inode->i_flags |= S_PRIVATE;
1657 simple_inode_init_ts(inode);
1658 return inode;
1659 }
1660 EXPORT_SYMBOL(alloc_anon_inode);
1661
1662 /**
1663 * simple_nosetlease - generic helper for pro
1664 * @filp: file pointer
1665 * @arg: type of lease to obtain
1666 * @flp: new lease supplied for insertion
1667 * @priv: private data for lm_setup operation
1668 *
1669 * Generic helper for filesystems that do not
1670 * All arguments are ignored and it just retu
1671 */
1672 int
1673 simple_nosetlease(struct file *filp, int arg,
1674 void **priv)
1675 {
1676 return -EINVAL;
1677 }
1678 EXPORT_SYMBOL(simple_nosetlease);
1679
1680 /**
1681 * simple_get_link - generic helper to get th
1682 * @dentry: not used here
1683 * @inode: the symlink inode
1684 * @done: not used here
1685 *
1686 * Generic helper for filesystems to use for
1687 * the symlink target is stored in ->i_link.
1688 * since as an optimization the path lookup c
1689 * directly, without calling ->get_link(). B
1690 * to mark the inode_operations as being for
1691 *
1692 * Return: the symlink target
1693 */
1694 const char *simple_get_link(struct dentry *de
1695 struct delayed_ca
1696 {
1697 return inode->i_link;
1698 }
1699 EXPORT_SYMBOL(simple_get_link);
1700
1701 const struct inode_operations simple_symlink_
1702 .get_link = simple_get_link,
1703 };
1704 EXPORT_SYMBOL(simple_symlink_inode_operations
1705
1706 /*
1707 * Operations for a permanently empty directo
1708 */
1709 static struct dentry *empty_dir_lookup(struct
1710 {
1711 return ERR_PTR(-ENOENT);
1712 }
1713
1714 static int empty_dir_getattr(struct mnt_idmap
1715 const struct pat
1716 u32 request_mask
1717 {
1718 struct inode *inode = d_inode(path->d
1719 generic_fillattr(&nop_mnt_idmap, requ
1720 return 0;
1721 }
1722
1723 static int empty_dir_setattr(struct mnt_idmap
1724 struct dentry *d
1725 {
1726 return -EPERM;
1727 }
1728
1729 static ssize_t empty_dir_listxattr(struct den
1730 {
1731 return -EOPNOTSUPP;
1732 }
1733
1734 static const struct inode_operations empty_di
1735 .lookup = empty_dir_lookup,
1736 .permission = generic_permission,
1737 .setattr = empty_dir_setattr,
1738 .getattr = empty_dir_getattr,
1739 .listxattr = empty_dir_listxattr
1740 };
1741
1742 static loff_t empty_dir_llseek(struct file *f
1743 {
1744 /* An empty directory has two entries
1745 return generic_file_llseek_size(file,
1746 }
1747
1748 static int empty_dir_readdir(struct file *fil
1749 {
1750 dir_emit_dots(file, ctx);
1751 return 0;
1752 }
1753
1754 static const struct file_operations empty_dir
1755 .llseek = empty_dir_llseek,
1756 .read = generic_read_dir,
1757 .iterate_shared = empty_dir_readdir,
1758 .fsync = noop_fsync,
1759 };
1760
1761
1762 void make_empty_dir_inode(struct inode *inode
1763 {
1764 set_nlink(inode, 2);
1765 inode->i_mode = S_IFDIR | S_IRUGO | S
1766 inode->i_uid = GLOBAL_ROOT_UID;
1767 inode->i_gid = GLOBAL_ROOT_GID;
1768 inode->i_rdev = 0;
1769 inode->i_size = 0;
1770 inode->i_blkbits = PAGE_SHIFT;
1771 inode->i_blocks = 0;
1772
1773 inode->i_op = &empty_dir_inode_operat
1774 inode->i_opflags &= ~IOP_XATTR;
1775 inode->i_fop = &empty_dir_operations;
1776 }
1777
1778 bool is_empty_dir_inode(struct inode *inode)
1779 {
1780 return (inode->i_fop == &empty_dir_op
1781 (inode->i_op == &empty_dir_in
1782 }
1783
1784 #if IS_ENABLED(CONFIG_UNICODE)
1785 /**
1786 * generic_ci_d_compare - generic d_compare i
1787 * @dentry: dentry whose name we are chec
1788 * @len: len of name of dentry
1789 * @str: str pointer to name of dentry
1790 * @name: Name to compare against
1791 *
1792 * Return: 0 if names match, 1 if mismatch, o
1793 */
1794 static int generic_ci_d_compare(const struct
1795 const char *s
1796 {
1797 const struct dentry *parent;
1798 const struct inode *dir;
1799 char strbuf[DNAME_INLINE_LEN];
1800 struct qstr qstr;
1801
1802 /*
1803 * Attempt a case-sensitive match fir
1804 * should cover most lookups, includi
1805 * applications that expect a case-se
1806 *
1807 * This comparison is safe under RCU
1808 * guarantees the consistency between
1809 * __d_lookup_rcu_op_compare() for de
1810 */
1811 if (len == name->len && !memcmp(str,
1812 return 0;
1813
1814 parent = READ_ONCE(dentry->d_parent);
1815 dir = READ_ONCE(parent->d_inode);
1816 if (!dir || !IS_CASEFOLDED(dir))
1817 return 1;
1818
1819 /*
1820 * If the dentry name is stored in-li
1821 * modified by a rename. If this hap
1822 * the lookup, so it doesn't matter w
1823 * However, it's unsafe to call utf8_
1824 * string. Therefore, we have to cop
1825 */
1826 if (len <= DNAME_INLINE_LEN - 1) {
1827 memcpy(strbuf, str, len);
1828 strbuf[len] = 0;
1829 str = strbuf;
1830 /* prevent compiler from opti
1831 barrier();
1832 }
1833 qstr.len = len;
1834 qstr.name = str;
1835
1836 return utf8_strncasecmp(dentry->d_sb-
1837 }
1838
1839 /**
1840 * generic_ci_d_hash - generic d_hash impleme
1841 * @dentry: dentry of the parent director
1842 * @str: qstr of name whose hash we sh
1843 *
1844 * Return: 0 if hash was successful or unchan
1845 */
1846 static int generic_ci_d_hash(const struct den
1847 {
1848 const struct inode *dir = READ_ONCE(d
1849 struct super_block *sb = dentry->d_sb
1850 const struct unicode_map *um = sb->s_
1851 int ret;
1852
1853 if (!dir || !IS_CASEFOLDED(dir))
1854 return 0;
1855
1856 ret = utf8_casefold_hash(um, dentry,
1857 if (ret < 0 && sb_has_strict_encoding
1858 return -EINVAL;
1859 return 0;
1860 }
1861
1862 static const struct dentry_operations generic
1863 .d_hash = generic_ci_d_hash,
1864 .d_compare = generic_ci_d_compare,
1865 #ifdef CONFIG_FS_ENCRYPTION
1866 .d_revalidate = fscrypt_d_revalidate,
1867 #endif
1868 };
1869
1870 /**
1871 * generic_ci_match() - Match a name (case-in
1872 * This is a filesystem helper for comparison
1873 * generic_ci_d_compare should be used in VFS
1874 *
1875 * @parent: Inode of the parent of the dirent
1876 * @name: name under lookup.
1877 * @folded_name: Optional pre-folded name und
1878 * @de_name: Dirent name.
1879 * @de_name_len: dirent name length.
1880 *
1881 * Test whether a case-insensitive directory
1882 * being searched. If @folded_name is provid
1883 * recalculating the casefold of @name.
1884 *
1885 * Return: > 0 if the directory entry matches
1886 * < 0 on error.
1887 */
1888 int generic_ci_match(const struct inode *pare
1889 const struct qstr *name,
1890 const struct qstr *folde
1891 const u8 *de_name, u32 d
1892 {
1893 const struct super_block *sb = parent
1894 const struct unicode_map *um = sb->s_
1895 struct fscrypt_str decrypted_name = F
1896 struct qstr dirent = QSTR_INIT(de_nam
1897 int res = 0;
1898
1899 if (IS_ENCRYPTED(parent)) {
1900 const struct fscrypt_str encr
1901 FSTR_INIT((u8 *) de_n
1902
1903 if (WARN_ON_ONCE(!fscrypt_has
1904 return -EINVAL;
1905
1906 decrypted_name.name = kmalloc
1907 if (!decrypted_name.name)
1908 return -ENOMEM;
1909 res = fscrypt_fname_disk_to_u
1910
1911 if (res < 0) {
1912 kfree(decrypted_name.
1913 return res;
1914 }
1915 dirent.name = decrypted_name.
1916 dirent.len = decrypted_name.l
1917 }
1918
1919 /*
1920 * Attempt a case-sensitive match fir
1921 * should cover most lookups, includi
1922 * applications that expect a case-se
1923 */
1924
1925 if (dirent.len == name->len &&
1926 !memcmp(name->name, dirent.name,
1927 goto out;
1928
1929 if (folded_name->name)
1930 res = utf8_strncasecmp_folded
1931 else
1932 res = utf8_strncasecmp(um, na
1933
1934 out:
1935 kfree(decrypted_name.name);
1936 if (res < 0 && sb_has_strict_encoding
1937 pr_err_ratelimited("Directory
1938 return 0;
1939 }
1940 return !res;
1941 }
1942 EXPORT_SYMBOL(generic_ci_match);
1943 #endif
1944
1945 #ifdef CONFIG_FS_ENCRYPTION
1946 static const struct dentry_operations generic
1947 .d_revalidate = fscrypt_d_revalidate,
1948 };
1949 #endif
1950
1951 /**
1952 * generic_set_sb_d_ops - helper for choosing
1953 * filesystem-wide dentry operations for the
1954 * @sb: superblock to be configured
1955 *
1956 * Filesystems supporting casefolding and/or
1957 * helper at mount-time to configure sb->s_d_
1958 * operations required for the enabled featur
1959 * called after these have been configured, b
1960 * is created.
1961 */
1962 void generic_set_sb_d_ops(struct super_block
1963 {
1964 #if IS_ENABLED(CONFIG_UNICODE)
1965 if (sb->s_encoding) {
1966 sb->s_d_op = &generic_ci_dent
1967 return;
1968 }
1969 #endif
1970 #ifdef CONFIG_FS_ENCRYPTION
1971 if (sb->s_cop) {
1972 sb->s_d_op = &generic_encrypt
1973 return;
1974 }
1975 #endif
1976 }
1977 EXPORT_SYMBOL(generic_set_sb_d_ops);
1978
1979 /**
1980 * inode_maybe_inc_iversion - increments i_ve
1981 * @inode: inode with the i_version that shou
1982 * @force: increment the counter even if it's
1983 *
1984 * Every time the inode is modified, the i_ve
1985 * changed by any observer.
1986 *
1987 * If "force" is set or the QUERIED flag is s
1988 * the value, and clear the queried flag.
1989 *
1990 * In the common case where neither is set, t
1991 * updating i_version.
1992 *
1993 * If this function returns false, and no oth
1994 * can avoid logging the metadata.
1995 */
1996 bool inode_maybe_inc_iversion(struct inode *i
1997 {
1998 u64 cur, new;
1999
2000 /*
2001 * The i_version field is not strictl
2002 * information, but the legacy inode_
2003 * to serialize increments.
2004 *
2005 * We add a full memory barrier to en
2006 * with other state is preserved (eit
2007 * or explicitly from smp_mb if we do
2008 * the former).
2009 *
2010 * These barriers pair with inode_que
2011 */
2012 cur = inode_peek_iversion_raw(inode);
2013 if (!force && !(cur & I_VERSION_QUERI
2014 smp_mb();
2015 cur = inode_peek_iversion_raw
2016 }
2017
2018 do {
2019 /* If flag is clear then we n
2020 if (!force && !(cur & I_VERSI
2021 return false;
2022
2023 /* Since lowest bit is flag,
2024 new = (cur & ~I_VERSION_QUERI
2025 } while (!atomic64_try_cmpxchg(&inode
2026 return true;
2027 }
2028 EXPORT_SYMBOL(inode_maybe_inc_iversion);
2029
2030 /**
2031 * inode_query_iversion - read i_version for
2032 * @inode: inode from which i_version should
2033 *
2034 * Read the inode i_version counter. This sho
2035 * to store the returned i_version for later
2036 * that a later query of the i_version will r
2037 * anything has changed.
2038 *
2039 * In this implementation, we fetch the curre
2040 * then try to swap it into place with a cmpx
2041 * that fails, we try again with the newly fe
2042 */
2043 u64 inode_query_iversion(struct inode *inode)
2044 {
2045 u64 cur, new;
2046 bool fenced = false;
2047
2048 /*
2049 * Memory barriers (implicit in cmpxc
2050 * inode_maybe_inc_iversion(), see th
2051 */
2052 cur = inode_peek_iversion_raw(inode);
2053 do {
2054 /* If flag is already set, th
2055 if (cur & I_VERSION_QUERIED)
2056 if (!fenced)
2057 smp_mb();
2058 break;
2059 }
2060
2061 fenced = true;
2062 new = cur | I_VERSION_QUERIED
2063 } while (!atomic64_try_cmpxchg(&inode
2064 return cur >> I_VERSION_QUERIED_SHIFT
2065 }
2066 EXPORT_SYMBOL(inode_query_iversion);
2067
2068 ssize_t direct_write_fallback(struct kiocb *i
2069 ssize_t direct_written, ssize
2070 {
2071 struct address_space *mapping = iocb-
2072 loff_t pos = iocb->ki_pos - buffered_
2073 loff_t end = iocb->ki_pos - 1;
2074 int err;
2075
2076 /*
2077 * If the buffered write fallback ret
2078 * the number of bytes which were wri
2079 * code if that was zero.
2080 *
2081 * Note that this differs from normal
2082 * return -EFOO even if some bytes we
2083 */
2084 if (unlikely(buffered_written < 0)) {
2085 if (direct_written)
2086 return direct_written
2087 return buffered_written;
2088 }
2089
2090 /*
2091 * We need to ensure that the page ca
2092 * invalidated to preserve the expect
2093 */
2094 err = filemap_write_and_wait_range(ma
2095 if (err < 0) {
2096 /*
2097 * We don't know how much we
2098 * bytes which were direct-wr
2099 */
2100 iocb->ki_pos -= buffered_writ
2101 if (direct_written)
2102 return direct_written
2103 return err;
2104 }
2105 invalidate_mapping_pages(mapping, pos
2106 return direct_written + buffered_writ
2107 }
2108 EXPORT_SYMBOL_GPL(direct_write_fallback);
2109
2110 /**
2111 * simple_inode_init_ts - initialize the time
2112 * @inode: inode to be initialized
2113 *
2114 * When a new inode is created, most filesyst
2115 * current time. Add a helper to do this.
2116 */
2117 struct timespec64 simple_inode_init_ts(struct
2118 {
2119 struct timespec64 ts = inode_set_ctim
2120
2121 inode_set_atime_to_ts(inode, ts);
2122 inode_set_mtime_to_ts(inode, ts);
2123 return ts;
2124 }
2125 EXPORT_SYMBOL(simple_inode_init_ts);
2126
2127 static inline struct dentry *get_stashed_dent
2128 {
2129 struct dentry *dentry;
2130
2131 guard(rcu)();
2132 dentry = rcu_dereference(*stashed);
2133 if (!dentry)
2134 return NULL;
2135 if (!lockref_get_not_dead(&dentry->d_
2136 return NULL;
2137 return dentry;
2138 }
2139
2140 static struct dentry *prepare_anon_dentry(str
2141 str
2142 voi
2143 {
2144 struct dentry *dentry;
2145 struct inode *inode;
2146 const struct stashed_operations *sops
2147 int ret;
2148
2149 inode = new_inode_pseudo(sb);
2150 if (!inode) {
2151 sops->put_data(data);
2152 return ERR_PTR(-ENOMEM);
2153 }
2154
2155 inode->i_flags |= S_IMMUTABLE;
2156 inode->i_mode = S_IFREG;
2157 simple_inode_init_ts(inode);
2158
2159 ret = sops->init_inode(inode, data);
2160 if (ret < 0) {
2161 iput(inode);
2162 return ERR_PTR(ret);
2163 }
2164
2165 /* Notice when this is changed. */
2166 WARN_ON_ONCE(!S_ISREG(inode->i_mode))
2167 WARN_ON_ONCE(!IS_IMMUTABLE(inode));
2168
2169 dentry = d_alloc_anon(sb);
2170 if (!dentry) {
2171 iput(inode);
2172 return ERR_PTR(-ENOMEM);
2173 }
2174
2175 /* Store address of location where de
2176 dentry->d_fsdata = stashed;
2177
2178 /* @data is now owned by the fs */
2179 d_instantiate(dentry, inode);
2180 return dentry;
2181 }
2182
2183 static struct dentry *stash_dentry(struct den
2184 struct den
2185 {
2186 guard(rcu)();
2187 for (;;) {
2188 struct dentry *old;
2189
2190 /* Assume any old dentry was
2191 old = cmpxchg(stashed, NULL,
2192 if (likely(!old))
2193 return dentry;
2194
2195 /* Check if somebody else ins
2196 if (lockref_get_not_dead(&old
2197 return old;
2198
2199 /* There's an old dead dentry
2200 if (likely(try_cmpxchg(stashe
2201 return dentry;
2202 }
2203 }
2204
2205 /**
2206 * path_from_stashed - create path from stash
2207 * @stashed: where to retrieve or stash de
2208 * @mnt: mnt of the filesystems to use
2209 * @data: data to store in inode->i_pri
2210 * @path: path to create
2211 *
2212 * The function tries to retrieve a stashed d
2213 * is still valid then it will be reused. If
2214 * will allocate a new dentry and inode. It w
2215 * can reuse an existing dentry in case one h
2216 * update @stashed with the newly added dentr
2217 *
2218 * Special-purpose helper for nsfs and pidfs.
2219 *
2220 * Return: On success zero and on failure a n
2221 */
2222 int path_from_stashed(struct dentry **stashed
2223 struct path *path)
2224 {
2225 struct dentry *dentry;
2226 const struct stashed_operations *sops
2227
2228 /* See if dentry can be reused. */
2229 path->dentry = get_stashed_dentry(sta
2230 if (path->dentry) {
2231 sops->put_data(data);
2232 goto out_path;
2233 }
2234
2235 /* Allocate a new dentry. */
2236 dentry = prepare_anon_dentry(stashed,
2237 if (IS_ERR(dentry))
2238 return PTR_ERR(dentry);
2239
2240 /* Added a new dentry. @data is now o
2241 path->dentry = stash_dentry(stashed,
2242 if (path->dentry != dentry)
2243 dput(dentry);
2244
2245 out_path:
2246 WARN_ON_ONCE(path->dentry->d_fsdata !
2247 WARN_ON_ONCE(d_inode(path->dentry)->i
2248 path->mnt = mntget(mnt);
2249 return 0;
2250 }
2251
2252 void stashed_dentry_prune(struct dentry *dent
2253 {
2254 struct dentry **stashed = dentry->d_f
2255 struct inode *inode = d_inode(dentry)
2256
2257 if (WARN_ON_ONCE(!stashed))
2258 return;
2259
2260 if (!inode)
2261 return;
2262
2263 /*
2264 * Only replace our own @dentry as so
2265 * already cleared out @dentry and st
2266 * dentry in there.
2267 */
2268 cmpxchg(stashed, dentry, NULL);
2269 }
2270
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.