Loading...
1// SPDX-License-Identifier: GPL-2.0-only
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/writeback.h>
19#include <linux/buffer_head.h> /* sync_mapping_buffers */
20#include <linux/fs_context.h>
21#include <linux/pseudo_fs.h>
22
23#include <linux/uaccess.h>
24
25#include "internal.h"
26
27int simple_getattr(const struct path *path, struct kstat *stat,
28 u32 request_mask, unsigned int query_flags)
29{
30 struct inode *inode = d_inode(path->dentry);
31 generic_fillattr(inode, stat);
32 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
33 return 0;
34}
35EXPORT_SYMBOL(simple_getattr);
36
37int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
38{
39 buf->f_type = dentry->d_sb->s_magic;
40 buf->f_bsize = PAGE_SIZE;
41 buf->f_namelen = NAME_MAX;
42 return 0;
43}
44EXPORT_SYMBOL(simple_statfs);
45
46/*
47 * Retaining negative dentries for an in-memory filesystem just wastes
48 * memory and lookup time: arrange for them to be deleted immediately.
49 */
50int always_delete_dentry(const struct dentry *dentry)
51{
52 return 1;
53}
54EXPORT_SYMBOL(always_delete_dentry);
55
56const struct dentry_operations simple_dentry_operations = {
57 .d_delete = always_delete_dentry,
58};
59EXPORT_SYMBOL(simple_dentry_operations);
60
61/*
62 * Lookup the data. This is trivial - if the dentry didn't already
63 * exist, we know it is negative. Set d_op to delete negative dentries.
64 */
65struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
66{
67 if (dentry->d_name.len > NAME_MAX)
68 return ERR_PTR(-ENAMETOOLONG);
69 if (!dentry->d_sb->s_d_op)
70 d_set_d_op(dentry, &simple_dentry_operations);
71 d_add(dentry, NULL);
72 return NULL;
73}
74EXPORT_SYMBOL(simple_lookup);
75
76int dcache_dir_open(struct inode *inode, struct file *file)
77{
78 file->private_data = d_alloc_cursor(file->f_path.dentry);
79
80 return file->private_data ? 0 : -ENOMEM;
81}
82EXPORT_SYMBOL(dcache_dir_open);
83
84int dcache_dir_close(struct inode *inode, struct file *file)
85{
86 dput(file->private_data);
87 return 0;
88}
89EXPORT_SYMBOL(dcache_dir_close);
90
91/* parent is locked at least shared */
92/*
93 * Returns an element of siblings' list.
94 * We are looking for <count>th positive after <p>; if
95 * found, dentry is grabbed and returned to caller.
96 * If no such element exists, NULL is returned.
97 */
98static struct dentry *scan_positives(struct dentry *cursor,
99 struct list_head *p,
100 loff_t count,
101 struct dentry *last)
102{
103 struct dentry *dentry = cursor->d_parent, *found = NULL;
104
105 spin_lock(&dentry->d_lock);
106 while ((p = p->next) != &dentry->d_subdirs) {
107 struct dentry *d = list_entry(p, struct dentry, d_child);
108 // we must at least skip cursors, to avoid livelocks
109 if (d->d_flags & DCACHE_DENTRY_CURSOR)
110 continue;
111 if (simple_positive(d) && !--count) {
112 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
113 if (simple_positive(d))
114 found = dget_dlock(d);
115 spin_unlock(&d->d_lock);
116 if (likely(found))
117 break;
118 count = 1;
119 }
120 if (need_resched()) {
121 list_move(&cursor->d_child, p);
122 p = &cursor->d_child;
123 spin_unlock(&dentry->d_lock);
124 cond_resched();
125 spin_lock(&dentry->d_lock);
126 }
127 }
128 spin_unlock(&dentry->d_lock);
129 dput(last);
130 return found;
131}
132
133loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
134{
135 struct dentry *dentry = file->f_path.dentry;
136 switch (whence) {
137 case 1:
138 offset += file->f_pos;
139 /* fall through */
140 case 0:
141 if (offset >= 0)
142 break;
143 /* fall through */
144 default:
145 return -EINVAL;
146 }
147 if (offset != file->f_pos) {
148 struct dentry *cursor = file->private_data;
149 struct dentry *to = NULL;
150
151 inode_lock_shared(dentry->d_inode);
152
153 if (offset > 2)
154 to = scan_positives(cursor, &dentry->d_subdirs,
155 offset - 2, NULL);
156 spin_lock(&dentry->d_lock);
157 if (to)
158 list_move(&cursor->d_child, &to->d_child);
159 else
160 list_del_init(&cursor->d_child);
161 spin_unlock(&dentry->d_lock);
162 dput(to);
163
164 file->f_pos = offset;
165
166 inode_unlock_shared(dentry->d_inode);
167 }
168 return offset;
169}
170EXPORT_SYMBOL(dcache_dir_lseek);
171
172/* Relationship between i_mode and the DT_xxx types */
173static inline unsigned char dt_type(struct inode *inode)
174{
175 return (inode->i_mode >> 12) & 15;
176}
177
178/*
179 * Directory is locked and all positive dentries in it are safe, since
180 * for ramfs-type trees they can't go away without unlink() or rmdir(),
181 * both impossible due to the lock on directory.
182 */
183
184int dcache_readdir(struct file *file, struct dir_context *ctx)
185{
186 struct dentry *dentry = file->f_path.dentry;
187 struct dentry *cursor = file->private_data;
188 struct list_head *anchor = &dentry->d_subdirs;
189 struct dentry *next = NULL;
190 struct list_head *p;
191
192 if (!dir_emit_dots(file, ctx))
193 return 0;
194
195 if (ctx->pos == 2)
196 p = anchor;
197 else if (!list_empty(&cursor->d_child))
198 p = &cursor->d_child;
199 else
200 return 0;
201
202 while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
203 if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
204 d_inode(next)->i_ino, dt_type(d_inode(next))))
205 break;
206 ctx->pos++;
207 p = &next->d_child;
208 }
209 spin_lock(&dentry->d_lock);
210 if (next)
211 list_move_tail(&cursor->d_child, &next->d_child);
212 else
213 list_del_init(&cursor->d_child);
214 spin_unlock(&dentry->d_lock);
215 dput(next);
216
217 return 0;
218}
219EXPORT_SYMBOL(dcache_readdir);
220
221ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
222{
223 return -EISDIR;
224}
225EXPORT_SYMBOL(generic_read_dir);
226
227const struct file_operations simple_dir_operations = {
228 .open = dcache_dir_open,
229 .release = dcache_dir_close,
230 .llseek = dcache_dir_lseek,
231 .read = generic_read_dir,
232 .iterate_shared = dcache_readdir,
233 .fsync = noop_fsync,
234};
235EXPORT_SYMBOL(simple_dir_operations);
236
237const struct inode_operations simple_dir_inode_operations = {
238 .lookup = simple_lookup,
239};
240EXPORT_SYMBOL(simple_dir_inode_operations);
241
242static const struct super_operations simple_super_operations = {
243 .statfs = simple_statfs,
244};
245
246static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
247{
248 struct pseudo_fs_context *ctx = fc->fs_private;
249 struct inode *root;
250
251 s->s_maxbytes = MAX_LFS_FILESIZE;
252 s->s_blocksize = PAGE_SIZE;
253 s->s_blocksize_bits = PAGE_SHIFT;
254 s->s_magic = ctx->magic;
255 s->s_op = ctx->ops ?: &simple_super_operations;
256 s->s_xattr = ctx->xattr;
257 s->s_time_gran = 1;
258 root = new_inode(s);
259 if (!root)
260 return -ENOMEM;
261
262 /*
263 * since this is the first inode, make it number 1. New inodes created
264 * after this must take care not to collide with it (by passing
265 * max_reserved of 1 to iunique).
266 */
267 root->i_ino = 1;
268 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
269 root->i_atime = root->i_mtime = root->i_ctime = current_time(root);
270 s->s_root = d_make_root(root);
271 if (!s->s_root)
272 return -ENOMEM;
273 s->s_d_op = ctx->dops;
274 return 0;
275}
276
277static int pseudo_fs_get_tree(struct fs_context *fc)
278{
279 return get_tree_nodev(fc, pseudo_fs_fill_super);
280}
281
282static void pseudo_fs_free(struct fs_context *fc)
283{
284 kfree(fc->fs_private);
285}
286
287static const struct fs_context_operations pseudo_fs_context_ops = {
288 .free = pseudo_fs_free,
289 .get_tree = pseudo_fs_get_tree,
290};
291
292/*
293 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
294 * will never be mountable)
295 */
296struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
297 unsigned long magic)
298{
299 struct pseudo_fs_context *ctx;
300
301 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
302 if (likely(ctx)) {
303 ctx->magic = magic;
304 fc->fs_private = ctx;
305 fc->ops = &pseudo_fs_context_ops;
306 fc->sb_flags |= SB_NOUSER;
307 fc->global = true;
308 }
309 return ctx;
310}
311EXPORT_SYMBOL(init_pseudo);
312
313int simple_open(struct inode *inode, struct file *file)
314{
315 if (inode->i_private)
316 file->private_data = inode->i_private;
317 return 0;
318}
319EXPORT_SYMBOL(simple_open);
320
321int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
322{
323 struct inode *inode = d_inode(old_dentry);
324
325 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
326 inc_nlink(inode);
327 ihold(inode);
328 dget(dentry);
329 d_instantiate(dentry, inode);
330 return 0;
331}
332EXPORT_SYMBOL(simple_link);
333
334int simple_empty(struct dentry *dentry)
335{
336 struct dentry *child;
337 int ret = 0;
338
339 spin_lock(&dentry->d_lock);
340 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
341 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
342 if (simple_positive(child)) {
343 spin_unlock(&child->d_lock);
344 goto out;
345 }
346 spin_unlock(&child->d_lock);
347 }
348 ret = 1;
349out:
350 spin_unlock(&dentry->d_lock);
351 return ret;
352}
353EXPORT_SYMBOL(simple_empty);
354
355int simple_unlink(struct inode *dir, struct dentry *dentry)
356{
357 struct inode *inode = d_inode(dentry);
358
359 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
360 drop_nlink(inode);
361 dput(dentry);
362 return 0;
363}
364EXPORT_SYMBOL(simple_unlink);
365
366int simple_rmdir(struct inode *dir, struct dentry *dentry)
367{
368 if (!simple_empty(dentry))
369 return -ENOTEMPTY;
370
371 drop_nlink(d_inode(dentry));
372 simple_unlink(dir, dentry);
373 drop_nlink(dir);
374 return 0;
375}
376EXPORT_SYMBOL(simple_rmdir);
377
378int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
379 struct inode *new_dir, struct dentry *new_dentry,
380 unsigned int flags)
381{
382 struct inode *inode = d_inode(old_dentry);
383 int they_are_dirs = d_is_dir(old_dentry);
384
385 if (flags & ~RENAME_NOREPLACE)
386 return -EINVAL;
387
388 if (!simple_empty(new_dentry))
389 return -ENOTEMPTY;
390
391 if (d_really_is_positive(new_dentry)) {
392 simple_unlink(new_dir, new_dentry);
393 if (they_are_dirs) {
394 drop_nlink(d_inode(new_dentry));
395 drop_nlink(old_dir);
396 }
397 } else if (they_are_dirs) {
398 drop_nlink(old_dir);
399 inc_nlink(new_dir);
400 }
401
402 old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
403 new_dir->i_mtime = inode->i_ctime = current_time(old_dir);
404
405 return 0;
406}
407EXPORT_SYMBOL(simple_rename);
408
409/**
410 * simple_setattr - setattr for simple filesystem
411 * @dentry: dentry
412 * @iattr: iattr structure
413 *
414 * Returns 0 on success, -error on failure.
415 *
416 * simple_setattr is a simple ->setattr implementation without a proper
417 * implementation of size changes.
418 *
419 * It can either be used for in-memory filesystems or special files
420 * on simple regular filesystems. Anything that needs to change on-disk
421 * or wire state on size changes needs its own setattr method.
422 */
423int simple_setattr(struct dentry *dentry, struct iattr *iattr)
424{
425 struct inode *inode = d_inode(dentry);
426 int error;
427
428 error = setattr_prepare(dentry, iattr);
429 if (error)
430 return error;
431
432 if (iattr->ia_valid & ATTR_SIZE)
433 truncate_setsize(inode, iattr->ia_size);
434 setattr_copy(inode, iattr);
435 mark_inode_dirty(inode);
436 return 0;
437}
438EXPORT_SYMBOL(simple_setattr);
439
440int simple_readpage(struct file *file, struct page *page)
441{
442 clear_highpage(page);
443 flush_dcache_page(page);
444 SetPageUptodate(page);
445 unlock_page(page);
446 return 0;
447}
448EXPORT_SYMBOL(simple_readpage);
449
450int simple_write_begin(struct file *file, struct address_space *mapping,
451 loff_t pos, unsigned len, unsigned flags,
452 struct page **pagep, void **fsdata)
453{
454 struct page *page;
455 pgoff_t index;
456
457 index = pos >> PAGE_SHIFT;
458
459 page = grab_cache_page_write_begin(mapping, index, flags);
460 if (!page)
461 return -ENOMEM;
462
463 *pagep = page;
464
465 if (!PageUptodate(page) && (len != PAGE_SIZE)) {
466 unsigned from = pos & (PAGE_SIZE - 1);
467
468 zero_user_segments(page, 0, from, from + len, PAGE_SIZE);
469 }
470 return 0;
471}
472EXPORT_SYMBOL(simple_write_begin);
473
474/**
475 * simple_write_end - .write_end helper for non-block-device FSes
476 * @file: See .write_end of address_space_operations
477 * @mapping: "
478 * @pos: "
479 * @len: "
480 * @copied: "
481 * @page: "
482 * @fsdata: "
483 *
484 * simple_write_end does the minimum needed for updating a page after writing is
485 * done. It has the same API signature as the .write_end of
486 * address_space_operations vector. So it can just be set onto .write_end for
487 * FSes that don't need any other processing. i_mutex is assumed to be held.
488 * Block based filesystems should use generic_write_end().
489 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
490 * is not called, so a filesystem that actually does store data in .write_inode
491 * should extend on what's done here with a call to mark_inode_dirty() in the
492 * case that i_size has changed.
493 *
494 * Use *ONLY* with simple_readpage()
495 */
496int simple_write_end(struct file *file, struct address_space *mapping,
497 loff_t pos, unsigned len, unsigned copied,
498 struct page *page, void *fsdata)
499{
500 struct inode *inode = page->mapping->host;
501 loff_t last_pos = pos + copied;
502
503 /* zero the stale part of the page if we did a short copy */
504 if (!PageUptodate(page)) {
505 if (copied < len) {
506 unsigned from = pos & (PAGE_SIZE - 1);
507
508 zero_user(page, from + copied, len - copied);
509 }
510 SetPageUptodate(page);
511 }
512 /*
513 * No need to use i_size_read() here, the i_size
514 * cannot change under us because we hold the i_mutex.
515 */
516 if (last_pos > inode->i_size)
517 i_size_write(inode, last_pos);
518
519 set_page_dirty(page);
520 unlock_page(page);
521 put_page(page);
522
523 return copied;
524}
525EXPORT_SYMBOL(simple_write_end);
526
527/*
528 * the inodes created here are not hashed. If you use iunique to generate
529 * unique inode values later for this filesystem, then you must take care
530 * to pass it an appropriate max_reserved value to avoid collisions.
531 */
532int simple_fill_super(struct super_block *s, unsigned long magic,
533 const struct tree_descr *files)
534{
535 struct inode *inode;
536 struct dentry *root;
537 struct dentry *dentry;
538 int i;
539
540 s->s_blocksize = PAGE_SIZE;
541 s->s_blocksize_bits = PAGE_SHIFT;
542 s->s_magic = magic;
543 s->s_op = &simple_super_operations;
544 s->s_time_gran = 1;
545
546 inode = new_inode(s);
547 if (!inode)
548 return -ENOMEM;
549 /*
550 * because the root inode is 1, the files array must not contain an
551 * entry at index 1
552 */
553 inode->i_ino = 1;
554 inode->i_mode = S_IFDIR | 0755;
555 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
556 inode->i_op = &simple_dir_inode_operations;
557 inode->i_fop = &simple_dir_operations;
558 set_nlink(inode, 2);
559 root = d_make_root(inode);
560 if (!root)
561 return -ENOMEM;
562 for (i = 0; !files->name || files->name[0]; i++, files++) {
563 if (!files->name)
564 continue;
565
566 /* warn if it tries to conflict with the root inode */
567 if (unlikely(i == 1))
568 printk(KERN_WARNING "%s: %s passed in a files array"
569 "with an index of 1!\n", __func__,
570 s->s_type->name);
571
572 dentry = d_alloc_name(root, files->name);
573 if (!dentry)
574 goto out;
575 inode = new_inode(s);
576 if (!inode) {
577 dput(dentry);
578 goto out;
579 }
580 inode->i_mode = S_IFREG | files->mode;
581 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
582 inode->i_fop = files->ops;
583 inode->i_ino = i;
584 d_add(dentry, inode);
585 }
586 s->s_root = root;
587 return 0;
588out:
589 d_genocide(root);
590 shrink_dcache_parent(root);
591 dput(root);
592 return -ENOMEM;
593}
594EXPORT_SYMBOL(simple_fill_super);
595
596static DEFINE_SPINLOCK(pin_fs_lock);
597
598int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
599{
600 struct vfsmount *mnt = NULL;
601 spin_lock(&pin_fs_lock);
602 if (unlikely(!*mount)) {
603 spin_unlock(&pin_fs_lock);
604 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
605 if (IS_ERR(mnt))
606 return PTR_ERR(mnt);
607 spin_lock(&pin_fs_lock);
608 if (!*mount)
609 *mount = mnt;
610 }
611 mntget(*mount);
612 ++*count;
613 spin_unlock(&pin_fs_lock);
614 mntput(mnt);
615 return 0;
616}
617EXPORT_SYMBOL(simple_pin_fs);
618
619void simple_release_fs(struct vfsmount **mount, int *count)
620{
621 struct vfsmount *mnt;
622 spin_lock(&pin_fs_lock);
623 mnt = *mount;
624 if (!--*count)
625 *mount = NULL;
626 spin_unlock(&pin_fs_lock);
627 mntput(mnt);
628}
629EXPORT_SYMBOL(simple_release_fs);
630
631/**
632 * simple_read_from_buffer - copy data from the buffer to user space
633 * @to: the user space buffer to read to
634 * @count: the maximum number of bytes to read
635 * @ppos: the current position in the buffer
636 * @from: the buffer to read from
637 * @available: the size of the buffer
638 *
639 * The simple_read_from_buffer() function reads up to @count bytes from the
640 * buffer @from at offset @ppos into the user space address starting at @to.
641 *
642 * On success, the number of bytes read is returned and the offset @ppos is
643 * advanced by this number, or negative value is returned on error.
644 **/
645ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
646 const void *from, size_t available)
647{
648 loff_t pos = *ppos;
649 size_t ret;
650
651 if (pos < 0)
652 return -EINVAL;
653 if (pos >= available || !count)
654 return 0;
655 if (count > available - pos)
656 count = available - pos;
657 ret = copy_to_user(to, from + pos, count);
658 if (ret == count)
659 return -EFAULT;
660 count -= ret;
661 *ppos = pos + count;
662 return count;
663}
664EXPORT_SYMBOL(simple_read_from_buffer);
665
666/**
667 * simple_write_to_buffer - copy data from user space to the buffer
668 * @to: the buffer to write to
669 * @available: the size of the buffer
670 * @ppos: the current position in the buffer
671 * @from: the user space buffer to read from
672 * @count: the maximum number of bytes to read
673 *
674 * The simple_write_to_buffer() function reads up to @count bytes from the user
675 * space address starting at @from into the buffer @to at offset @ppos.
676 *
677 * On success, the number of bytes written is returned and the offset @ppos is
678 * advanced by this number, or negative value is returned on error.
679 **/
680ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
681 const void __user *from, size_t count)
682{
683 loff_t pos = *ppos;
684 size_t res;
685
686 if (pos < 0)
687 return -EINVAL;
688 if (pos >= available || !count)
689 return 0;
690 if (count > available - pos)
691 count = available - pos;
692 res = copy_from_user(to + pos, from, count);
693 if (res == count)
694 return -EFAULT;
695 count -= res;
696 *ppos = pos + count;
697 return count;
698}
699EXPORT_SYMBOL(simple_write_to_buffer);
700
701/**
702 * memory_read_from_buffer - copy data from the buffer
703 * @to: the kernel space buffer to read to
704 * @count: the maximum number of bytes to read
705 * @ppos: the current position in the buffer
706 * @from: the buffer to read from
707 * @available: the size of the buffer
708 *
709 * The memory_read_from_buffer() function reads up to @count bytes from the
710 * buffer @from at offset @ppos into the kernel space address starting at @to.
711 *
712 * On success, the number of bytes read is returned and the offset @ppos is
713 * advanced by this number, or negative value is returned on error.
714 **/
715ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
716 const void *from, size_t available)
717{
718 loff_t pos = *ppos;
719
720 if (pos < 0)
721 return -EINVAL;
722 if (pos >= available)
723 return 0;
724 if (count > available - pos)
725 count = available - pos;
726 memcpy(to, from + pos, count);
727 *ppos = pos + count;
728
729 return count;
730}
731EXPORT_SYMBOL(memory_read_from_buffer);
732
733/*
734 * Transaction based IO.
735 * The file expects a single write which triggers the transaction, and then
736 * possibly a read which collects the result - which is stored in a
737 * file-local buffer.
738 */
739
740void simple_transaction_set(struct file *file, size_t n)
741{
742 struct simple_transaction_argresp *ar = file->private_data;
743
744 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
745
746 /*
747 * The barrier ensures that ar->size will really remain zero until
748 * ar->data is ready for reading.
749 */
750 smp_mb();
751 ar->size = n;
752}
753EXPORT_SYMBOL(simple_transaction_set);
754
755char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
756{
757 struct simple_transaction_argresp *ar;
758 static DEFINE_SPINLOCK(simple_transaction_lock);
759
760 if (size > SIMPLE_TRANSACTION_LIMIT - 1)
761 return ERR_PTR(-EFBIG);
762
763 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
764 if (!ar)
765 return ERR_PTR(-ENOMEM);
766
767 spin_lock(&simple_transaction_lock);
768
769 /* only one write allowed per open */
770 if (file->private_data) {
771 spin_unlock(&simple_transaction_lock);
772 free_page((unsigned long)ar);
773 return ERR_PTR(-EBUSY);
774 }
775
776 file->private_data = ar;
777
778 spin_unlock(&simple_transaction_lock);
779
780 if (copy_from_user(ar->data, buf, size))
781 return ERR_PTR(-EFAULT);
782
783 return ar->data;
784}
785EXPORT_SYMBOL(simple_transaction_get);
786
787ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
788{
789 struct simple_transaction_argresp *ar = file->private_data;
790
791 if (!ar)
792 return 0;
793 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
794}
795EXPORT_SYMBOL(simple_transaction_read);
796
797int simple_transaction_release(struct inode *inode, struct file *file)
798{
799 free_page((unsigned long)file->private_data);
800 return 0;
801}
802EXPORT_SYMBOL(simple_transaction_release);
803
804/* Simple attribute files */
805
806struct simple_attr {
807 int (*get)(void *, u64 *);
808 int (*set)(void *, u64);
809 char get_buf[24]; /* enough to store a u64 and "\n\0" */
810 char set_buf[24];
811 void *data;
812 const char *fmt; /* format for read operation */
813 struct mutex mutex; /* protects access to these buffers */
814};
815
816/* simple_attr_open is called by an actual attribute open file operation
817 * to set the attribute specific access operations. */
818int simple_attr_open(struct inode *inode, struct file *file,
819 int (*get)(void *, u64 *), int (*set)(void *, u64),
820 const char *fmt)
821{
822 struct simple_attr *attr;
823
824 attr = kmalloc(sizeof(*attr), GFP_KERNEL);
825 if (!attr)
826 return -ENOMEM;
827
828 attr->get = get;
829 attr->set = set;
830 attr->data = inode->i_private;
831 attr->fmt = fmt;
832 mutex_init(&attr->mutex);
833
834 file->private_data = attr;
835
836 return nonseekable_open(inode, file);
837}
838EXPORT_SYMBOL_GPL(simple_attr_open);
839
840int simple_attr_release(struct inode *inode, struct file *file)
841{
842 kfree(file->private_data);
843 return 0;
844}
845EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */
846
847/* read from the buffer that is filled with the get function */
848ssize_t simple_attr_read(struct file *file, char __user *buf,
849 size_t len, loff_t *ppos)
850{
851 struct simple_attr *attr;
852 size_t size;
853 ssize_t ret;
854
855 attr = file->private_data;
856
857 if (!attr->get)
858 return -EACCES;
859
860 ret = mutex_lock_interruptible(&attr->mutex);
861 if (ret)
862 return ret;
863
864 if (*ppos) { /* continued read */
865 size = strlen(attr->get_buf);
866 } else { /* first read */
867 u64 val;
868 ret = attr->get(attr->data, &val);
869 if (ret)
870 goto out;
871
872 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
873 attr->fmt, (unsigned long long)val);
874 }
875
876 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
877out:
878 mutex_unlock(&attr->mutex);
879 return ret;
880}
881EXPORT_SYMBOL_GPL(simple_attr_read);
882
883/* interpret the buffer as a number to call the set function with */
884ssize_t simple_attr_write(struct file *file, const char __user *buf,
885 size_t len, loff_t *ppos)
886{
887 struct simple_attr *attr;
888 u64 val;
889 size_t size;
890 ssize_t ret;
891
892 attr = file->private_data;
893 if (!attr->set)
894 return -EACCES;
895
896 ret = mutex_lock_interruptible(&attr->mutex);
897 if (ret)
898 return ret;
899
900 ret = -EFAULT;
901 size = min(sizeof(attr->set_buf) - 1, len);
902 if (copy_from_user(attr->set_buf, buf, size))
903 goto out;
904
905 attr->set_buf[size] = '\0';
906 val = simple_strtoll(attr->set_buf, NULL, 0);
907 ret = attr->set(attr->data, val);
908 if (ret == 0)
909 ret = len; /* on success, claim we got the whole input */
910out:
911 mutex_unlock(&attr->mutex);
912 return ret;
913}
914EXPORT_SYMBOL_GPL(simple_attr_write);
915
916/**
917 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
918 * @sb: filesystem to do the file handle conversion on
919 * @fid: file handle to convert
920 * @fh_len: length of the file handle in bytes
921 * @fh_type: type of file handle
922 * @get_inode: filesystem callback to retrieve inode
923 *
924 * This function decodes @fid as long as it has one of the well-known
925 * Linux filehandle types and calls @get_inode on it to retrieve the
926 * inode for the object specified in the file handle.
927 */
928struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
929 int fh_len, int fh_type, struct inode *(*get_inode)
930 (struct super_block *sb, u64 ino, u32 gen))
931{
932 struct inode *inode = NULL;
933
934 if (fh_len < 2)
935 return NULL;
936
937 switch (fh_type) {
938 case FILEID_INO32_GEN:
939 case FILEID_INO32_GEN_PARENT:
940 inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
941 break;
942 }
943
944 return d_obtain_alias(inode);
945}
946EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
947
948/**
949 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
950 * @sb: filesystem to do the file handle conversion on
951 * @fid: file handle to convert
952 * @fh_len: length of the file handle in bytes
953 * @fh_type: type of file handle
954 * @get_inode: filesystem callback to retrieve inode
955 *
956 * This function decodes @fid as long as it has one of the well-known
957 * Linux filehandle types and calls @get_inode on it to retrieve the
958 * inode for the _parent_ object specified in the file handle if it
959 * is specified in the file handle, or NULL otherwise.
960 */
961struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
962 int fh_len, int fh_type, struct inode *(*get_inode)
963 (struct super_block *sb, u64 ino, u32 gen))
964{
965 struct inode *inode = NULL;
966
967 if (fh_len <= 2)
968 return NULL;
969
970 switch (fh_type) {
971 case FILEID_INO32_GEN_PARENT:
972 inode = get_inode(sb, fid->i32.parent_ino,
973 (fh_len > 3 ? fid->i32.parent_gen : 0));
974 break;
975 }
976
977 return d_obtain_alias(inode);
978}
979EXPORT_SYMBOL_GPL(generic_fh_to_parent);
980
981/**
982 * __generic_file_fsync - generic fsync implementation for simple filesystems
983 *
984 * @file: file to synchronize
985 * @start: start offset in bytes
986 * @end: end offset in bytes (inclusive)
987 * @datasync: only synchronize essential metadata if true
988 *
989 * This is a generic implementation of the fsync method for simple
990 * filesystems which track all non-inode metadata in the buffers list
991 * hanging off the address_space structure.
992 */
993int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
994 int datasync)
995{
996 struct inode *inode = file->f_mapping->host;
997 int err;
998 int ret;
999
1000 err = file_write_and_wait_range(file, start, end);
1001 if (err)
1002 return err;
1003
1004 inode_lock(inode);
1005 ret = sync_mapping_buffers(inode->i_mapping);
1006 if (!(inode->i_state & I_DIRTY_ALL))
1007 goto out;
1008 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1009 goto out;
1010
1011 err = sync_inode_metadata(inode, 1);
1012 if (ret == 0)
1013 ret = err;
1014
1015out:
1016 inode_unlock(inode);
1017 /* check and advance again to catch errors after syncing out buffers */
1018 err = file_check_and_advance_wb_err(file);
1019 if (ret == 0)
1020 ret = err;
1021 return ret;
1022}
1023EXPORT_SYMBOL(__generic_file_fsync);
1024
1025/**
1026 * generic_file_fsync - generic fsync implementation for simple filesystems
1027 * with flush
1028 * @file: file to synchronize
1029 * @start: start offset in bytes
1030 * @end: end offset in bytes (inclusive)
1031 * @datasync: only synchronize essential metadata if true
1032 *
1033 */
1034
1035int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1036 int datasync)
1037{
1038 struct inode *inode = file->f_mapping->host;
1039 int err;
1040
1041 err = __generic_file_fsync(file, start, end, datasync);
1042 if (err)
1043 return err;
1044 return blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
1045}
1046EXPORT_SYMBOL(generic_file_fsync);
1047
1048/**
1049 * generic_check_addressable - Check addressability of file system
1050 * @blocksize_bits: log of file system block size
1051 * @num_blocks: number of blocks in file system
1052 *
1053 * Determine whether a file system with @num_blocks blocks (and a
1054 * block size of 2**@blocksize_bits) is addressable by the sector_t
1055 * and page cache of the system. Return 0 if so and -EFBIG otherwise.
1056 */
1057int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1058{
1059 u64 last_fs_block = num_blocks - 1;
1060 u64 last_fs_page =
1061 last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1062
1063 if (unlikely(num_blocks == 0))
1064 return 0;
1065
1066 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1067 return -EINVAL;
1068
1069 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1070 (last_fs_page > (pgoff_t)(~0ULL))) {
1071 return -EFBIG;
1072 }
1073 return 0;
1074}
1075EXPORT_SYMBOL(generic_check_addressable);
1076
1077/*
1078 * No-op implementation of ->fsync for in-memory filesystems.
1079 */
1080int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1081{
1082 return 0;
1083}
1084EXPORT_SYMBOL(noop_fsync);
1085
1086int noop_set_page_dirty(struct page *page)
1087{
1088 /*
1089 * Unlike __set_page_dirty_no_writeback that handles dirty page
1090 * tracking in the page object, dax does all dirty tracking in
1091 * the inode address_space in response to mkwrite faults. In the
1092 * dax case we only need to worry about potentially dirty CPU
1093 * caches, not dirty page cache pages to write back.
1094 *
1095 * This callback is defined to prevent fallback to
1096 * __set_page_dirty_buffers() in set_page_dirty().
1097 */
1098 return 0;
1099}
1100EXPORT_SYMBOL_GPL(noop_set_page_dirty);
1101
1102void noop_invalidatepage(struct page *page, unsigned int offset,
1103 unsigned int length)
1104{
1105 /*
1106 * There is no page cache to invalidate in the dax case, however
1107 * we need this callback defined to prevent falling back to
1108 * block_invalidatepage() in do_invalidatepage().
1109 */
1110}
1111EXPORT_SYMBOL_GPL(noop_invalidatepage);
1112
1113ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1114{
1115 /*
1116 * iomap based filesystems support direct I/O without need for
1117 * this callback. However, it still needs to be set in
1118 * inode->a_ops so that open/fcntl know that direct I/O is
1119 * generally supported.
1120 */
1121 return -EINVAL;
1122}
1123EXPORT_SYMBOL_GPL(noop_direct_IO);
1124
1125/* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1126void kfree_link(void *p)
1127{
1128 kfree(p);
1129}
1130EXPORT_SYMBOL(kfree_link);
1131
1132/*
1133 * nop .set_page_dirty method so that people can use .page_mkwrite on
1134 * anon inodes.
1135 */
1136static int anon_set_page_dirty(struct page *page)
1137{
1138 return 0;
1139};
1140
1141/*
1142 * A single inode exists for all anon_inode files. Contrary to pipes,
1143 * anon_inode inodes have no associated per-instance data, so we need
1144 * only allocate one of them.
1145 */
1146struct inode *alloc_anon_inode(struct super_block *s)
1147{
1148 static const struct address_space_operations anon_aops = {
1149 .set_page_dirty = anon_set_page_dirty,
1150 };
1151 struct inode *inode = new_inode_pseudo(s);
1152
1153 if (!inode)
1154 return ERR_PTR(-ENOMEM);
1155
1156 inode->i_ino = get_next_ino();
1157 inode->i_mapping->a_ops = &anon_aops;
1158
1159 /*
1160 * Mark the inode dirty from the very beginning,
1161 * that way it will never be moved to the dirty
1162 * list because mark_inode_dirty() will think
1163 * that it already _is_ on the dirty list.
1164 */
1165 inode->i_state = I_DIRTY;
1166 inode->i_mode = S_IRUSR | S_IWUSR;
1167 inode->i_uid = current_fsuid();
1168 inode->i_gid = current_fsgid();
1169 inode->i_flags |= S_PRIVATE;
1170 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
1171 return inode;
1172}
1173EXPORT_SYMBOL(alloc_anon_inode);
1174
1175/**
1176 * simple_nosetlease - generic helper for prohibiting leases
1177 * @filp: file pointer
1178 * @arg: type of lease to obtain
1179 * @flp: new lease supplied for insertion
1180 * @priv: private data for lm_setup operation
1181 *
1182 * Generic helper for filesystems that do not wish to allow leases to be set.
1183 * All arguments are ignored and it just returns -EINVAL.
1184 */
1185int
1186simple_nosetlease(struct file *filp, long arg, struct file_lock **flp,
1187 void **priv)
1188{
1189 return -EINVAL;
1190}
1191EXPORT_SYMBOL(simple_nosetlease);
1192
1193/**
1194 * simple_get_link - generic helper to get the target of "fast" symlinks
1195 * @dentry: not used here
1196 * @inode: the symlink inode
1197 * @done: not used here
1198 *
1199 * Generic helper for filesystems to use for symlink inodes where a pointer to
1200 * the symlink target is stored in ->i_link. NOTE: this isn't normally called,
1201 * since as an optimization the path lookup code uses any non-NULL ->i_link
1202 * directly, without calling ->get_link(). But ->get_link() still must be set,
1203 * to mark the inode_operations as being for a symlink.
1204 *
1205 * Return: the symlink target
1206 */
1207const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1208 struct delayed_call *done)
1209{
1210 return inode->i_link;
1211}
1212EXPORT_SYMBOL(simple_get_link);
1213
1214const struct inode_operations simple_symlink_inode_operations = {
1215 .get_link = simple_get_link,
1216};
1217EXPORT_SYMBOL(simple_symlink_inode_operations);
1218
1219/*
1220 * Operations for a permanently empty directory.
1221 */
1222static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1223{
1224 return ERR_PTR(-ENOENT);
1225}
1226
1227static int empty_dir_getattr(const struct path *path, struct kstat *stat,
1228 u32 request_mask, unsigned int query_flags)
1229{
1230 struct inode *inode = d_inode(path->dentry);
1231 generic_fillattr(inode, stat);
1232 return 0;
1233}
1234
1235static int empty_dir_setattr(struct dentry *dentry, struct iattr *attr)
1236{
1237 return -EPERM;
1238}
1239
1240static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1241{
1242 return -EOPNOTSUPP;
1243}
1244
1245static const struct inode_operations empty_dir_inode_operations = {
1246 .lookup = empty_dir_lookup,
1247 .permission = generic_permission,
1248 .setattr = empty_dir_setattr,
1249 .getattr = empty_dir_getattr,
1250 .listxattr = empty_dir_listxattr,
1251};
1252
1253static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1254{
1255 /* An empty directory has two entries . and .. at offsets 0 and 1 */
1256 return generic_file_llseek_size(file, offset, whence, 2, 2);
1257}
1258
1259static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1260{
1261 dir_emit_dots(file, ctx);
1262 return 0;
1263}
1264
1265static const struct file_operations empty_dir_operations = {
1266 .llseek = empty_dir_llseek,
1267 .read = generic_read_dir,
1268 .iterate_shared = empty_dir_readdir,
1269 .fsync = noop_fsync,
1270};
1271
1272
1273void make_empty_dir_inode(struct inode *inode)
1274{
1275 set_nlink(inode, 2);
1276 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1277 inode->i_uid = GLOBAL_ROOT_UID;
1278 inode->i_gid = GLOBAL_ROOT_GID;
1279 inode->i_rdev = 0;
1280 inode->i_size = 0;
1281 inode->i_blkbits = PAGE_SHIFT;
1282 inode->i_blocks = 0;
1283
1284 inode->i_op = &empty_dir_inode_operations;
1285 inode->i_opflags &= ~IOP_XATTR;
1286 inode->i_fop = &empty_dir_operations;
1287}
1288
1289bool is_empty_dir_inode(struct inode *inode)
1290{
1291 return (inode->i_fop == &empty_dir_operations) &&
1292 (inode->i_op == &empty_dir_inode_operations);
1293}
1// SPDX-License-Identifier: GPL-2.0-only
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_buffers */
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
32int simple_getattr(struct mnt_idmap *idmap, const struct path *path,
33 struct kstat *stat, u32 request_mask,
34 unsigned int query_flags)
35{
36 struct inode *inode = d_inode(path->dentry);
37 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
38 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
39 return 0;
40}
41EXPORT_SYMBOL(simple_getattr);
42
43int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
44{
45 u64 id = huge_encode_dev(dentry->d_sb->s_dev);
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}
53EXPORT_SYMBOL(simple_statfs);
54
55/*
56 * Retaining negative dentries for an in-memory filesystem just wastes
57 * memory and lookup time: arrange for them to be deleted immediately.
58 */
59int always_delete_dentry(const struct dentry *dentry)
60{
61 return 1;
62}
63EXPORT_SYMBOL(always_delete_dentry);
64
65const struct dentry_operations simple_dentry_operations = {
66 .d_delete = always_delete_dentry,
67};
68EXPORT_SYMBOL(simple_dentry_operations);
69
70/*
71 * Lookup the data. This is trivial - if the dentry didn't already
72 * exist, we know it is negative. Set d_op to delete negative dentries.
73 */
74struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
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_dentry_operations);
80 d_add(dentry, NULL);
81 return NULL;
82}
83EXPORT_SYMBOL(simple_lookup);
84
85int dcache_dir_open(struct inode *inode, struct file *file)
86{
87 file->private_data = d_alloc_cursor(file->f_path.dentry);
88
89 return file->private_data ? 0 : -ENOMEM;
90}
91EXPORT_SYMBOL(dcache_dir_open);
92
93int dcache_dir_close(struct inode *inode, struct file *file)
94{
95 dput(file->private_data);
96 return 0;
97}
98EXPORT_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 <p>; if
104 * found, dentry is grabbed and returned to caller.
105 * If no such element exists, NULL is returned.
106 */
107static struct dentry *scan_positives(struct dentry *cursor,
108 struct hlist_node **p,
109 loff_t count,
110 struct dentry *last)
111{
112 struct dentry *dentry = cursor->d_parent, *found = NULL;
113
114 spin_lock(&dentry->d_lock);
115 while (*p) {
116 struct dentry *d = hlist_entry(*p, struct dentry, d_sib);
117 p = &d->d_sib.next;
118 // we must at least skip cursors, to avoid livelocks
119 if (d->d_flags & DCACHE_DENTRY_CURSOR)
120 continue;
121 if (simple_positive(d) && !--count) {
122 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
123 if (simple_positive(d))
124 found = dget_dlock(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(&cursor->d_sib))
132 __hlist_del(&cursor->d_sib);
133 hlist_add_behind(&cursor->d_sib, &d->d_sib);
134 p = &cursor->d_sib.next;
135 spin_unlock(&dentry->d_lock);
136 cond_resched();
137 spin_lock(&dentry->d_lock);
138 }
139 }
140 spin_unlock(&dentry->d_lock);
141 dput(last);
142 return found;
143}
144
145loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
146{
147 struct dentry *dentry = file->f_path.dentry;
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->private_data;
161 struct dentry *to = NULL;
162
163 inode_lock_shared(dentry->d_inode);
164
165 if (offset > 2)
166 to = scan_positives(cursor, &dentry->d_children.first,
167 offset - 2, NULL);
168 spin_lock(&dentry->d_lock);
169 hlist_del_init(&cursor->d_sib);
170 if (to)
171 hlist_add_behind(&cursor->d_sib, &to->d_sib);
172 spin_unlock(&dentry->d_lock);
173 dput(to);
174
175 file->f_pos = offset;
176
177 inode_unlock_shared(dentry->d_inode);
178 }
179 return offset;
180}
181EXPORT_SYMBOL(dcache_dir_lseek);
182
183/*
184 * Directory is locked and all positive dentries in it are safe, since
185 * for ramfs-type trees they can't go away without unlink() or rmdir(),
186 * both impossible due to the lock on directory.
187 */
188
189int dcache_readdir(struct file *file, struct dir_context *ctx)
190{
191 struct dentry *dentry = file->f_path.dentry;
192 struct dentry *cursor = file->private_data;
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, p, 1, next)) != NULL) {
205 if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
206 d_inode(next)->i_ino,
207 fs_umode_to_dtype(d_inode(next)->i_mode)))
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_sib, &next->d_sib);
216 spin_unlock(&dentry->d_lock);
217 dput(next);
218
219 return 0;
220}
221EXPORT_SYMBOL(dcache_readdir);
222
223ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
224{
225 return -EISDIR;
226}
227EXPORT_SYMBOL(generic_read_dir);
228
229const struct file_operations simple_dir_operations = {
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};
237EXPORT_SYMBOL(simple_dir_operations);
238
239const struct inode_operations simple_dir_inode_operations = {
240 .lookup = simple_lookup,
241};
242EXPORT_SYMBOL(simple_dir_inode_operations);
243
244/* 0 is '.', 1 is '..', so always start with offset 2 or more */
245enum {
246 DIR_OFFSET_MIN = 2,
247};
248
249static void offset_set(struct dentry *dentry, long offset)
250{
251 dentry->d_fsdata = (void *)offset;
252}
253
254static long dentry2offset(struct dentry *dentry)
255{
256 return (long)dentry->d_fsdata;
257}
258
259static struct lock_class_key simple_offset_lock_class;
260
261/**
262 * simple_offset_init - initialize an offset_ctx
263 * @octx: directory offset map to be initialized
264 *
265 */
266void simple_offset_init(struct offset_ctx *octx)
267{
268 mt_init_flags(&octx->mt, MT_FLAGS_ALLOC_RANGE);
269 lockdep_set_class(&octx->mt.ma_lock, &simple_offset_lock_class);
270 octx->next_offset = DIR_OFFSET_MIN;
271}
272
273/**
274 * simple_offset_add - Add an entry to a directory's offset map
275 * @octx: directory offset ctx to be updated
276 * @dentry: new dentry being added
277 *
278 * Returns zero on success. @octx and the dentry's offset are updated.
279 * Otherwise, a negative errno value is returned.
280 */
281int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
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, &offset, dentry, DIR_OFFSET_MIN,
290 LONG_MAX, &octx->next_offset, GFP_KERNEL);
291 if (ret < 0)
292 return ret;
293
294 offset_set(dentry, offset);
295 return 0;
296}
297
298static int simple_offset_replace(struct offset_ctx *octx, struct dentry *dentry,
299 long offset)
300{
301 int ret;
302
303 ret = mtree_store(&octx->mt, offset, dentry, GFP_KERNEL);
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 directory's offset map
312 * @octx: directory offset ctx to be updated
313 * @dentry: dentry being removed
314 *
315 */
316void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
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 be unlinked
330 * @dentry: dentry to be tested
331 *
332 * Returns 0 if @dentry is a non-empty directory; otherwise returns 1.
333 */
334int 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(inode);
347 mt_for_each(&octx->mt, child, index, LONG_MAX) {
348 spin_lock(&child->d_lock);
349 if (simple_positive(child)) {
350 spin_unlock(&child->d_lock);
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 offsets for rename
362 * @old_dir: parent directory of source entry
363 * @old_dentry: dentry of source entry
364 * @new_dir: parent_directory of destination entry
365 * @new_dentry: dentry of destination
366 *
367 * Caller provides appropriate serialization.
368 *
369 * User space expects the directory offset value of the replaced
370 * (new) directory entry to be unchanged after a rename.
371 *
372 * Returns zero on success, a negative errno value on failure.
373 */
374int simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
375 struct inode *new_dir, struct dentry *new_dentry)
376{
377 struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
378 struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
379 long new_offset = dentry2offset(new_dentry);
380
381 simple_offset_remove(old_ctx, old_dentry);
382
383 if (new_offset) {
384 offset_set(new_dentry, 0);
385 return simple_offset_replace(new_ctx, old_dentry, new_offset);
386 }
387 return simple_offset_add(new_ctx, old_dentry);
388}
389
390/**
391 * simple_offset_rename_exchange - exchange rename with directory offsets
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 values. Caller provides
398 * appropriate serialization.
399 *
400 * Returns zero on success. Otherwise a negative errno is returned and the
401 * rename is rolled back.
402 */
403int simple_offset_rename_exchange(struct inode *old_dir,
404 struct dentry *old_dentry,
405 struct inode *new_dir,
406 struct dentry *new_dentry)
407{
408 struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
409 struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
410 long old_index = dentry2offset(old_dentry);
411 long new_index = dentry2offset(new_dentry);
412 int ret;
413
414 simple_offset_remove(old_ctx, old_dentry);
415 simple_offset_remove(new_ctx, new_dentry);
416
417 ret = simple_offset_replace(new_ctx, old_dentry, new_index);
418 if (ret)
419 goto out_restore;
420
421 ret = simple_offset_replace(old_ctx, new_dentry, old_index);
422 if (ret) {
423 simple_offset_remove(new_ctx, old_dentry);
424 goto out_restore;
425 }
426
427 ret = simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
428 if (ret) {
429 simple_offset_remove(new_ctx, old_dentry);
430 simple_offset_remove(old_ctx, new_dentry);
431 goto out_restore;
432 }
433 return 0;
434
435out_restore:
436 (void)simple_offset_replace(old_ctx, old_dentry, old_index);
437 (void)simple_offset_replace(new_ctx, new_dentry, new_index);
438 return ret;
439}
440
441/**
442 * simple_offset_destroy - Release offset map
443 * @octx: directory offset ctx that is about to be destroyed
444 *
445 * During fs teardown (eg. umount), a directory's offset map might still
446 * contain entries. xa_destroy() cleans out anything that remains.
447 */
448void simple_offset_destroy(struct offset_ctx *octx)
449{
450 mtree_destroy(&octx->mt);
451}
452
453/**
454 * offset_dir_llseek - Advance the read position of a directory descriptor
455 * @file: an open directory whose position is to be updated
456 * @offset: a byte offset
457 * @whence: enumerator describing the starting position for this update
458 *
459 * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories.
460 *
461 * Returns the updated read position if successful; otherwise a
462 * negative errno is returned and the read position remains unchanged.
463 */
464static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
465{
466 switch (whence) {
467 case SEEK_CUR:
468 offset += file->f_pos;
469 fallthrough;
470 case SEEK_SET:
471 if (offset >= 0)
472 break;
473 fallthrough;
474 default:
475 return -EINVAL;
476 }
477
478 /* In this case, ->private_data is protected by f_pos_lock */
479 file->private_data = NULL;
480 return vfs_setpos(file, offset, LONG_MAX);
481}
482
483static struct dentry *offset_find_next(struct offset_ctx *octx, loff_t offset)
484{
485 MA_STATE(mas, &octx->mt, offset, offset);
486 struct dentry *child, *found = NULL;
487
488 rcu_read_lock();
489 child = mas_find(&mas, LONG_MAX);
490 if (!child)
491 goto out;
492 spin_lock(&child->d_lock);
493 if (simple_positive(child))
494 found = dget_dlock(child);
495 spin_unlock(&child->d_lock);
496out:
497 rcu_read_unlock();
498 return found;
499}
500
501static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
502{
503 struct inode *inode = d_inode(dentry);
504 long offset = dentry2offset(dentry);
505
506 return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset,
507 inode->i_ino, fs_umode_to_dtype(inode->i_mode));
508}
509
510static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
511{
512 struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
513 struct dentry *dentry;
514
515 while (true) {
516 dentry = offset_find_next(octx, ctx->pos);
517 if (!dentry)
518 return ERR_PTR(-ENOENT);
519
520 if (!offset_dir_emit(ctx, dentry)) {
521 dput(dentry);
522 break;
523 }
524
525 ctx->pos = dentry2offset(dentry) + 1;
526 dput(dentry);
527 }
528 return NULL;
529}
530
531/**
532 * offset_readdir - Emit entries starting at offset @ctx->pos
533 * @file: an open directory to iterate over
534 * @ctx: directory iteration context
535 *
536 * Caller must hold @file's i_rwsem to prevent insertion or removal of
537 * entries during this call.
538 *
539 * On entry, @ctx->pos contains an offset that represents the first entry
540 * to be read from the directory.
541 *
542 * The operation continues until there are no more entries to read, or
543 * until the ctx->actor indicates there is no more space in the caller's
544 * output buffer.
545 *
546 * On return, @ctx->pos contains an offset that will read the next entry
547 * in this directory when offset_readdir() is called again with @ctx.
548 *
549 * Return values:
550 * %0 - Complete
551 */
552static int offset_readdir(struct file *file, struct dir_context *ctx)
553{
554 struct dentry *dir = file->f_path.dentry;
555
556 lockdep_assert_held(&d_inode(dir)->i_rwsem);
557
558 if (!dir_emit_dots(file, ctx))
559 return 0;
560
561 /* In this case, ->private_data is protected by f_pos_lock */
562 if (ctx->pos == DIR_OFFSET_MIN)
563 file->private_data = NULL;
564 else if (file->private_data == ERR_PTR(-ENOENT))
565 return 0;
566 file->private_data = offset_iterate_dir(d_inode(dir), ctx);
567 return 0;
568}
569
570const struct file_operations simple_offset_dir_operations = {
571 .llseek = offset_dir_llseek,
572 .iterate_shared = offset_readdir,
573 .read = generic_read_dir,
574 .fsync = noop_fsync,
575};
576
577static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
578{
579 struct dentry *child = NULL, *d;
580
581 spin_lock(&parent->d_lock);
582 d = prev ? d_next_sibling(prev) : d_first_child(parent);
583 hlist_for_each_entry_from(d, d_sib) {
584 if (simple_positive(d)) {
585 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
586 if (simple_positive(d))
587 child = dget_dlock(d);
588 spin_unlock(&d->d_lock);
589 if (likely(child))
590 break;
591 }
592 }
593 spin_unlock(&parent->d_lock);
594 dput(prev);
595 return child;
596}
597
598void simple_recursive_removal(struct dentry *dentry,
599 void (*callback)(struct dentry *))
600{
601 struct dentry *this = dget(dentry);
602 while (true) {
603 struct dentry *victim = NULL, *child;
604 struct inode *inode = this->d_inode;
605
606 inode_lock(inode);
607 if (d_is_dir(this))
608 inode->i_flags |= S_DEAD;
609 while ((child = find_next_child(this, victim)) == NULL) {
610 // kill and ascend
611 // update metadata while it's still locked
612 inode_set_ctime_current(inode);
613 clear_nlink(inode);
614 inode_unlock(inode);
615 victim = this;
616 this = this->d_parent;
617 inode = this->d_inode;
618 inode_lock(inode);
619 if (simple_positive(victim)) {
620 d_invalidate(victim); // avoid lost mounts
621 if (d_is_dir(victim))
622 fsnotify_rmdir(inode, victim);
623 else
624 fsnotify_unlink(inode, victim);
625 if (callback)
626 callback(victim);
627 dput(victim); // unpin it
628 }
629 if (victim == dentry) {
630 inode_set_mtime_to_ts(inode,
631 inode_set_ctime_current(inode));
632 if (d_is_dir(dentry))
633 drop_nlink(inode);
634 inode_unlock(inode);
635 dput(dentry);
636 return;
637 }
638 }
639 inode_unlock(inode);
640 this = child;
641 }
642}
643EXPORT_SYMBOL(simple_recursive_removal);
644
645static const struct super_operations simple_super_operations = {
646 .statfs = simple_statfs,
647};
648
649static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
650{
651 struct pseudo_fs_context *ctx = fc->fs_private;
652 struct inode *root;
653
654 s->s_maxbytes = MAX_LFS_FILESIZE;
655 s->s_blocksize = PAGE_SIZE;
656 s->s_blocksize_bits = PAGE_SHIFT;
657 s->s_magic = ctx->magic;
658 s->s_op = ctx->ops ?: &simple_super_operations;
659 s->s_xattr = ctx->xattr;
660 s->s_time_gran = 1;
661 root = new_inode(s);
662 if (!root)
663 return -ENOMEM;
664
665 /*
666 * since this is the first inode, make it number 1. New inodes created
667 * after this must take care not to collide with it (by passing
668 * max_reserved of 1 to iunique).
669 */
670 root->i_ino = 1;
671 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
672 simple_inode_init_ts(root);
673 s->s_root = d_make_root(root);
674 if (!s->s_root)
675 return -ENOMEM;
676 s->s_d_op = ctx->dops;
677 return 0;
678}
679
680static int pseudo_fs_get_tree(struct fs_context *fc)
681{
682 return get_tree_nodev(fc, pseudo_fs_fill_super);
683}
684
685static void pseudo_fs_free(struct fs_context *fc)
686{
687 kfree(fc->fs_private);
688}
689
690static const struct fs_context_operations pseudo_fs_context_ops = {
691 .free = pseudo_fs_free,
692 .get_tree = pseudo_fs_get_tree,
693};
694
695/*
696 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
697 * will never be mountable)
698 */
699struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
700 unsigned long magic)
701{
702 struct pseudo_fs_context *ctx;
703
704 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
705 if (likely(ctx)) {
706 ctx->magic = magic;
707 fc->fs_private = ctx;
708 fc->ops = &pseudo_fs_context_ops;
709 fc->sb_flags |= SB_NOUSER;
710 fc->global = true;
711 }
712 return ctx;
713}
714EXPORT_SYMBOL(init_pseudo);
715
716int simple_open(struct inode *inode, struct file *file)
717{
718 if (inode->i_private)
719 file->private_data = inode->i_private;
720 return 0;
721}
722EXPORT_SYMBOL(simple_open);
723
724int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
725{
726 struct inode *inode = d_inode(old_dentry);
727
728 inode_set_mtime_to_ts(dir,
729 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
730 inc_nlink(inode);
731 ihold(inode);
732 dget(dentry);
733 d_instantiate(dentry, inode);
734 return 0;
735}
736EXPORT_SYMBOL(simple_link);
737
738int simple_empty(struct dentry *dentry)
739{
740 struct dentry *child;
741 int ret = 0;
742
743 spin_lock(&dentry->d_lock);
744 hlist_for_each_entry(child, &dentry->d_children, d_sib) {
745 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
746 if (simple_positive(child)) {
747 spin_unlock(&child->d_lock);
748 goto out;
749 }
750 spin_unlock(&child->d_lock);
751 }
752 ret = 1;
753out:
754 spin_unlock(&dentry->d_lock);
755 return ret;
756}
757EXPORT_SYMBOL(simple_empty);
758
759int simple_unlink(struct inode *dir, struct dentry *dentry)
760{
761 struct inode *inode = d_inode(dentry);
762
763 inode_set_mtime_to_ts(dir,
764 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
765 drop_nlink(inode);
766 dput(dentry);
767 return 0;
768}
769EXPORT_SYMBOL(simple_unlink);
770
771int simple_rmdir(struct inode *dir, struct dentry *dentry)
772{
773 if (!simple_empty(dentry))
774 return -ENOTEMPTY;
775
776 drop_nlink(d_inode(dentry));
777 simple_unlink(dir, dentry);
778 drop_nlink(dir);
779 return 0;
780}
781EXPORT_SYMBOL(simple_rmdir);
782
783/**
784 * simple_rename_timestamp - update the various inode timestamps for rename
785 * @old_dir: old parent directory
786 * @old_dentry: dentry that is being renamed
787 * @new_dir: new parent directory
788 * @new_dentry: target for rename
789 *
790 * POSIX mandates that the old and new parent directories have their ctime and
791 * mtime updated, and that inodes of @old_dentry and @new_dentry (if any), have
792 * their ctime updated.
793 */
794void simple_rename_timestamp(struct inode *old_dir, struct dentry *old_dentry,
795 struct inode *new_dir, struct dentry *new_dentry)
796{
797 struct inode *newino = d_inode(new_dentry);
798
799 inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
800 if (new_dir != old_dir)
801 inode_set_mtime_to_ts(new_dir,
802 inode_set_ctime_current(new_dir));
803 inode_set_ctime_current(d_inode(old_dentry));
804 if (newino)
805 inode_set_ctime_current(newino);
806}
807EXPORT_SYMBOL_GPL(simple_rename_timestamp);
808
809int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
810 struct inode *new_dir, struct dentry *new_dentry)
811{
812 bool old_is_dir = d_is_dir(old_dentry);
813 bool new_is_dir = d_is_dir(new_dentry);
814
815 if (old_dir != new_dir && old_is_dir != new_is_dir) {
816 if (old_is_dir) {
817 drop_nlink(old_dir);
818 inc_nlink(new_dir);
819 } else {
820 drop_nlink(new_dir);
821 inc_nlink(old_dir);
822 }
823 }
824 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
825 return 0;
826}
827EXPORT_SYMBOL_GPL(simple_rename_exchange);
828
829int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir,
830 struct dentry *old_dentry, struct inode *new_dir,
831 struct dentry *new_dentry, unsigned int flags)
832{
833 int they_are_dirs = d_is_dir(old_dentry);
834
835 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
836 return -EINVAL;
837
838 if (flags & RENAME_EXCHANGE)
839 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
840
841 if (!simple_empty(new_dentry))
842 return -ENOTEMPTY;
843
844 if (d_really_is_positive(new_dentry)) {
845 simple_unlink(new_dir, new_dentry);
846 if (they_are_dirs) {
847 drop_nlink(d_inode(new_dentry));
848 drop_nlink(old_dir);
849 }
850 } else if (they_are_dirs) {
851 drop_nlink(old_dir);
852 inc_nlink(new_dir);
853 }
854
855 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
856 return 0;
857}
858EXPORT_SYMBOL(simple_rename);
859
860/**
861 * simple_setattr - setattr for simple filesystem
862 * @idmap: idmap of the target mount
863 * @dentry: dentry
864 * @iattr: iattr structure
865 *
866 * Returns 0 on success, -error on failure.
867 *
868 * simple_setattr is a simple ->setattr implementation without a proper
869 * implementation of size changes.
870 *
871 * It can either be used for in-memory filesystems or special files
872 * on simple regular filesystems. Anything that needs to change on-disk
873 * or wire state on size changes needs its own setattr method.
874 */
875int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
876 struct iattr *iattr)
877{
878 struct inode *inode = d_inode(dentry);
879 int error;
880
881 error = setattr_prepare(idmap, dentry, iattr);
882 if (error)
883 return error;
884
885 if (iattr->ia_valid & ATTR_SIZE)
886 truncate_setsize(inode, iattr->ia_size);
887 setattr_copy(idmap, inode, iattr);
888 mark_inode_dirty(inode);
889 return 0;
890}
891EXPORT_SYMBOL(simple_setattr);
892
893static int simple_read_folio(struct file *file, struct folio *folio)
894{
895 folio_zero_range(folio, 0, folio_size(folio));
896 flush_dcache_folio(folio);
897 folio_mark_uptodate(folio);
898 folio_unlock(folio);
899 return 0;
900}
901
902int simple_write_begin(struct file *file, struct address_space *mapping,
903 loff_t pos, unsigned len,
904 struct page **pagep, void **fsdata)
905{
906 struct folio *folio;
907
908 folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
909 mapping_gfp_mask(mapping));
910 if (IS_ERR(folio))
911 return PTR_ERR(folio);
912
913 *pagep = &folio->page;
914
915 if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
916 size_t from = offset_in_folio(folio, pos);
917
918 folio_zero_segments(folio, 0, from,
919 from + len, folio_size(folio));
920 }
921 return 0;
922}
923EXPORT_SYMBOL(simple_write_begin);
924
925/**
926 * simple_write_end - .write_end helper for non-block-device FSes
927 * @file: See .write_end of address_space_operations
928 * @mapping: "
929 * @pos: "
930 * @len: "
931 * @copied: "
932 * @page: "
933 * @fsdata: "
934 *
935 * simple_write_end does the minimum needed for updating a page after writing is
936 * done. It has the same API signature as the .write_end of
937 * address_space_operations vector. So it can just be set onto .write_end for
938 * FSes that don't need any other processing. i_mutex is assumed to be held.
939 * Block based filesystems should use generic_write_end().
940 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
941 * is not called, so a filesystem that actually does store data in .write_inode
942 * should extend on what's done here with a call to mark_inode_dirty() in the
943 * case that i_size has changed.
944 *
945 * Use *ONLY* with simple_read_folio()
946 */
947static int simple_write_end(struct file *file, struct address_space *mapping,
948 loff_t pos, unsigned len, unsigned copied,
949 struct page *page, void *fsdata)
950{
951 struct folio *folio = page_folio(page);
952 struct inode *inode = folio->mapping->host;
953 loff_t last_pos = pos + copied;
954
955 /* zero the stale part of the folio if we did a short copy */
956 if (!folio_test_uptodate(folio)) {
957 if (copied < len) {
958 size_t from = offset_in_folio(folio, pos);
959
960 folio_zero_range(folio, from + copied, len - copied);
961 }
962 folio_mark_uptodate(folio);
963 }
964 /*
965 * No need to use i_size_read() here, the i_size
966 * cannot change under us because we hold the i_mutex.
967 */
968 if (last_pos > inode->i_size)
969 i_size_write(inode, last_pos);
970
971 folio_mark_dirty(folio);
972 folio_unlock(folio);
973 folio_put(folio);
974
975 return copied;
976}
977
978/*
979 * Provides ramfs-style behavior: data in the pagecache, but no writeback.
980 */
981const struct address_space_operations ram_aops = {
982 .read_folio = simple_read_folio,
983 .write_begin = simple_write_begin,
984 .write_end = simple_write_end,
985 .dirty_folio = noop_dirty_folio,
986};
987EXPORT_SYMBOL(ram_aops);
988
989/*
990 * the inodes created here are not hashed. If you use iunique to generate
991 * unique inode values later for this filesystem, then you must take care
992 * to pass it an appropriate max_reserved value to avoid collisions.
993 */
994int simple_fill_super(struct super_block *s, unsigned long magic,
995 const struct tree_descr *files)
996{
997 struct inode *inode;
998 struct dentry *dentry;
999 int i;
1000
1001 s->s_blocksize = PAGE_SIZE;
1002 s->s_blocksize_bits = PAGE_SHIFT;
1003 s->s_magic = magic;
1004 s->s_op = &simple_super_operations;
1005 s->s_time_gran = 1;
1006
1007 inode = new_inode(s);
1008 if (!inode)
1009 return -ENOMEM;
1010 /*
1011 * because the root inode is 1, the files array must not contain an
1012 * entry at index 1
1013 */
1014 inode->i_ino = 1;
1015 inode->i_mode = S_IFDIR | 0755;
1016 simple_inode_init_ts(inode);
1017 inode->i_op = &simple_dir_inode_operations;
1018 inode->i_fop = &simple_dir_operations;
1019 set_nlink(inode, 2);
1020 s->s_root = d_make_root(inode);
1021 if (!s->s_root)
1022 return -ENOMEM;
1023 for (i = 0; !files->name || files->name[0]; i++, files++) {
1024 if (!files->name)
1025 continue;
1026
1027 /* warn if it tries to conflict with the root inode */
1028 if (unlikely(i == 1))
1029 printk(KERN_WARNING "%s: %s passed in a files array"
1030 "with an index of 1!\n", __func__,
1031 s->s_type->name);
1032
1033 dentry = d_alloc_name(s->s_root, files->name);
1034 if (!dentry)
1035 return -ENOMEM;
1036 inode = new_inode(s);
1037 if (!inode) {
1038 dput(dentry);
1039 return -ENOMEM;
1040 }
1041 inode->i_mode = S_IFREG | files->mode;
1042 simple_inode_init_ts(inode);
1043 inode->i_fop = files->ops;
1044 inode->i_ino = i;
1045 d_add(dentry, inode);
1046 }
1047 return 0;
1048}
1049EXPORT_SYMBOL(simple_fill_super);
1050
1051static DEFINE_SPINLOCK(pin_fs_lock);
1052
1053int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
1054{
1055 struct vfsmount *mnt = NULL;
1056 spin_lock(&pin_fs_lock);
1057 if (unlikely(!*mount)) {
1058 spin_unlock(&pin_fs_lock);
1059 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
1060 if (IS_ERR(mnt))
1061 return PTR_ERR(mnt);
1062 spin_lock(&pin_fs_lock);
1063 if (!*mount)
1064 *mount = mnt;
1065 }
1066 mntget(*mount);
1067 ++*count;
1068 spin_unlock(&pin_fs_lock);
1069 mntput(mnt);
1070 return 0;
1071}
1072EXPORT_SYMBOL(simple_pin_fs);
1073
1074void simple_release_fs(struct vfsmount **mount, int *count)
1075{
1076 struct vfsmount *mnt;
1077 spin_lock(&pin_fs_lock);
1078 mnt = *mount;
1079 if (!--*count)
1080 *mount = NULL;
1081 spin_unlock(&pin_fs_lock);
1082 mntput(mnt);
1083}
1084EXPORT_SYMBOL(simple_release_fs);
1085
1086/**
1087 * simple_read_from_buffer - copy data from the buffer to user space
1088 * @to: the user space buffer to read to
1089 * @count: the maximum number of bytes to read
1090 * @ppos: the current position in the buffer
1091 * @from: the buffer to read from
1092 * @available: the size of the buffer
1093 *
1094 * The simple_read_from_buffer() function reads up to @count bytes from the
1095 * buffer @from at offset @ppos into the user space address starting at @to.
1096 *
1097 * On success, the number of bytes read is returned and the offset @ppos is
1098 * advanced by this number, or negative value is returned on error.
1099 **/
1100ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
1101 const void *from, size_t available)
1102{
1103 loff_t pos = *ppos;
1104 size_t ret;
1105
1106 if (pos < 0)
1107 return -EINVAL;
1108 if (pos >= available || !count)
1109 return 0;
1110 if (count > available - pos)
1111 count = available - pos;
1112 ret = copy_to_user(to, from + pos, count);
1113 if (ret == count)
1114 return -EFAULT;
1115 count -= ret;
1116 *ppos = pos + count;
1117 return count;
1118}
1119EXPORT_SYMBOL(simple_read_from_buffer);
1120
1121/**
1122 * simple_write_to_buffer - copy data from user space to the buffer
1123 * @to: the buffer to write to
1124 * @available: the size of the buffer
1125 * @ppos: the current position in the buffer
1126 * @from: the user space buffer to read from
1127 * @count: the maximum number of bytes to read
1128 *
1129 * The simple_write_to_buffer() function reads up to @count bytes from the user
1130 * space address starting at @from into the buffer @to at offset @ppos.
1131 *
1132 * On success, the number of bytes written is returned and the offset @ppos is
1133 * advanced by this number, or negative value is returned on error.
1134 **/
1135ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
1136 const void __user *from, size_t count)
1137{
1138 loff_t pos = *ppos;
1139 size_t res;
1140
1141 if (pos < 0)
1142 return -EINVAL;
1143 if (pos >= available || !count)
1144 return 0;
1145 if (count > available - pos)
1146 count = available - pos;
1147 res = copy_from_user(to + pos, from, count);
1148 if (res == count)
1149 return -EFAULT;
1150 count -= res;
1151 *ppos = pos + count;
1152 return count;
1153}
1154EXPORT_SYMBOL(simple_write_to_buffer);
1155
1156/**
1157 * memory_read_from_buffer - copy data from the buffer
1158 * @to: the kernel space buffer to read to
1159 * @count: the maximum number of bytes to read
1160 * @ppos: the current position in the buffer
1161 * @from: the buffer to read from
1162 * @available: the size of the buffer
1163 *
1164 * The memory_read_from_buffer() function reads up to @count bytes from the
1165 * buffer @from at offset @ppos into the kernel space address starting at @to.
1166 *
1167 * On success, the number of bytes read is returned and the offset @ppos is
1168 * advanced by this number, or negative value is returned on error.
1169 **/
1170ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
1171 const void *from, size_t available)
1172{
1173 loff_t pos = *ppos;
1174
1175 if (pos < 0)
1176 return -EINVAL;
1177 if (pos >= available)
1178 return 0;
1179 if (count > available - pos)
1180 count = available - pos;
1181 memcpy(to, from + pos, count);
1182 *ppos = pos + count;
1183
1184 return count;
1185}
1186EXPORT_SYMBOL(memory_read_from_buffer);
1187
1188/*
1189 * Transaction based IO.
1190 * The file expects a single write which triggers the transaction, and then
1191 * possibly a read which collects the result - which is stored in a
1192 * file-local buffer.
1193 */
1194
1195void simple_transaction_set(struct file *file, size_t n)
1196{
1197 struct simple_transaction_argresp *ar = file->private_data;
1198
1199 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1200
1201 /*
1202 * The barrier ensures that ar->size will really remain zero until
1203 * ar->data is ready for reading.
1204 */
1205 smp_mb();
1206 ar->size = n;
1207}
1208EXPORT_SYMBOL(simple_transaction_set);
1209
1210char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
1211{
1212 struct simple_transaction_argresp *ar;
1213 static DEFINE_SPINLOCK(simple_transaction_lock);
1214
1215 if (size > SIMPLE_TRANSACTION_LIMIT - 1)
1216 return ERR_PTR(-EFBIG);
1217
1218 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
1219 if (!ar)
1220 return ERR_PTR(-ENOMEM);
1221
1222 spin_lock(&simple_transaction_lock);
1223
1224 /* only one write allowed per open */
1225 if (file->private_data) {
1226 spin_unlock(&simple_transaction_lock);
1227 free_page((unsigned long)ar);
1228 return ERR_PTR(-EBUSY);
1229 }
1230
1231 file->private_data = ar;
1232
1233 spin_unlock(&simple_transaction_lock);
1234
1235 if (copy_from_user(ar->data, buf, size))
1236 return ERR_PTR(-EFAULT);
1237
1238 return ar->data;
1239}
1240EXPORT_SYMBOL(simple_transaction_get);
1241
1242ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
1243{
1244 struct simple_transaction_argresp *ar = file->private_data;
1245
1246 if (!ar)
1247 return 0;
1248 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
1249}
1250EXPORT_SYMBOL(simple_transaction_read);
1251
1252int simple_transaction_release(struct inode *inode, struct file *file)
1253{
1254 free_page((unsigned long)file->private_data);
1255 return 0;
1256}
1257EXPORT_SYMBOL(simple_transaction_release);
1258
1259/* Simple attribute files */
1260
1261struct simple_attr {
1262 int (*get)(void *, u64 *);
1263 int (*set)(void *, u64);
1264 char get_buf[24]; /* enough to store a u64 and "\n\0" */
1265 char set_buf[24];
1266 void *data;
1267 const char *fmt; /* format for read operation */
1268 struct mutex mutex; /* protects access to these buffers */
1269};
1270
1271/* simple_attr_open is called by an actual attribute open file operation
1272 * to set the attribute specific access operations. */
1273int simple_attr_open(struct inode *inode, struct file *file,
1274 int (*get)(void *, u64 *), int (*set)(void *, u64),
1275 const char *fmt)
1276{
1277 struct simple_attr *attr;
1278
1279 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1280 if (!attr)
1281 return -ENOMEM;
1282
1283 attr->get = get;
1284 attr->set = set;
1285 attr->data = inode->i_private;
1286 attr->fmt = fmt;
1287 mutex_init(&attr->mutex);
1288
1289 file->private_data = attr;
1290
1291 return nonseekable_open(inode, file);
1292}
1293EXPORT_SYMBOL_GPL(simple_attr_open);
1294
1295int simple_attr_release(struct inode *inode, struct file *file)
1296{
1297 kfree(file->private_data);
1298 return 0;
1299}
1300EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */
1301
1302/* read from the buffer that is filled with the get function */
1303ssize_t simple_attr_read(struct file *file, char __user *buf,
1304 size_t len, loff_t *ppos)
1305{
1306 struct simple_attr *attr;
1307 size_t size;
1308 ssize_t ret;
1309
1310 attr = file->private_data;
1311
1312 if (!attr->get)
1313 return -EACCES;
1314
1315 ret = mutex_lock_interruptible(&attr->mutex);
1316 if (ret)
1317 return ret;
1318
1319 if (*ppos && attr->get_buf[0]) {
1320 /* continued read */
1321 size = strlen(attr->get_buf);
1322 } else {
1323 /* first read */
1324 u64 val;
1325 ret = attr->get(attr->data, &val);
1326 if (ret)
1327 goto out;
1328
1329 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
1330 attr->fmt, (unsigned long long)val);
1331 }
1332
1333 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
1334out:
1335 mutex_unlock(&attr->mutex);
1336 return ret;
1337}
1338EXPORT_SYMBOL_GPL(simple_attr_read);
1339
1340/* interpret the buffer as a number to call the set function with */
1341static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf,
1342 size_t len, loff_t *ppos, bool is_signed)
1343{
1344 struct simple_attr *attr;
1345 unsigned long long val;
1346 size_t size;
1347 ssize_t ret;
1348
1349 attr = file->private_data;
1350 if (!attr->set)
1351 return -EACCES;
1352
1353 ret = mutex_lock_interruptible(&attr->mutex);
1354 if (ret)
1355 return ret;
1356
1357 ret = -EFAULT;
1358 size = min(sizeof(attr->set_buf) - 1, len);
1359 if (copy_from_user(attr->set_buf, buf, size))
1360 goto out;
1361
1362 attr->set_buf[size] = '\0';
1363 if (is_signed)
1364 ret = kstrtoll(attr->set_buf, 0, &val);
1365 else
1366 ret = kstrtoull(attr->set_buf, 0, &val);
1367 if (ret)
1368 goto out;
1369 ret = attr->set(attr->data, val);
1370 if (ret == 0)
1371 ret = len; /* on success, claim we got the whole input */
1372out:
1373 mutex_unlock(&attr->mutex);
1374 return ret;
1375}
1376
1377ssize_t simple_attr_write(struct file *file, const char __user *buf,
1378 size_t len, loff_t *ppos)
1379{
1380 return simple_attr_write_xsigned(file, buf, len, ppos, false);
1381}
1382EXPORT_SYMBOL_GPL(simple_attr_write);
1383
1384ssize_t simple_attr_write_signed(struct file *file, const char __user *buf,
1385 size_t len, loff_t *ppos)
1386{
1387 return simple_attr_write_xsigned(file, buf, len, ppos, true);
1388}
1389EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1390
1391/**
1392 * generic_encode_ino32_fh - generic export_operations->encode_fh function
1393 * @inode: the object to encode
1394 * @fh: where to store the file handle fragment
1395 * @max_len: maximum length to store there (in 4 byte units)
1396 * @parent: parent directory inode, if wanted
1397 *
1398 * This generic encode_fh function assumes that the 32 inode number
1399 * is suitable for locating an inode, and that the generation number
1400 * can be used to check that it is still valid. It places them in the
1401 * filehandle fragment where export_decode_fh expects to find them.
1402 */
1403int generic_encode_ino32_fh(struct inode *inode, __u32 *fh, int *max_len,
1404 struct inode *parent)
1405{
1406 struct fid *fid = (void *)fh;
1407 int len = *max_len;
1408 int type = FILEID_INO32_GEN;
1409
1410 if (parent && (len < 4)) {
1411 *max_len = 4;
1412 return FILEID_INVALID;
1413 } else if (len < 2) {
1414 *max_len = 2;
1415 return FILEID_INVALID;
1416 }
1417
1418 len = 2;
1419 fid->i32.ino = inode->i_ino;
1420 fid->i32.gen = inode->i_generation;
1421 if (parent) {
1422 fid->i32.parent_ino = parent->i_ino;
1423 fid->i32.parent_gen = parent->i_generation;
1424 len = 4;
1425 type = FILEID_INO32_GEN_PARENT;
1426 }
1427 *max_len = len;
1428 return type;
1429}
1430EXPORT_SYMBOL_GPL(generic_encode_ino32_fh);
1431
1432/**
1433 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1434 * @sb: filesystem to do the file handle conversion on
1435 * @fid: file handle to convert
1436 * @fh_len: length of the file handle in bytes
1437 * @fh_type: type of file handle
1438 * @get_inode: filesystem callback to retrieve inode
1439 *
1440 * This function decodes @fid as long as it has one of the well-known
1441 * Linux filehandle types and calls @get_inode on it to retrieve the
1442 * inode for the object specified in the file handle.
1443 */
1444struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1445 int fh_len, int fh_type, struct inode *(*get_inode)
1446 (struct super_block *sb, u64 ino, u32 gen))
1447{
1448 struct inode *inode = NULL;
1449
1450 if (fh_len < 2)
1451 return NULL;
1452
1453 switch (fh_type) {
1454 case FILEID_INO32_GEN:
1455 case FILEID_INO32_GEN_PARENT:
1456 inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1457 break;
1458 }
1459
1460 return d_obtain_alias(inode);
1461}
1462EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1463
1464/**
1465 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1466 * @sb: filesystem to do the file handle conversion on
1467 * @fid: file handle to convert
1468 * @fh_len: length of the file handle in bytes
1469 * @fh_type: type of file handle
1470 * @get_inode: filesystem callback to retrieve inode
1471 *
1472 * This function decodes @fid as long as it has one of the well-known
1473 * Linux filehandle types and calls @get_inode on it to retrieve the
1474 * inode for the _parent_ object specified in the file handle if it
1475 * is specified in the file handle, or NULL otherwise.
1476 */
1477struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1478 int fh_len, int fh_type, struct inode *(*get_inode)
1479 (struct super_block *sb, u64 ino, u32 gen))
1480{
1481 struct inode *inode = NULL;
1482
1483 if (fh_len <= 2)
1484 return NULL;
1485
1486 switch (fh_type) {
1487 case FILEID_INO32_GEN_PARENT:
1488 inode = get_inode(sb, fid->i32.parent_ino,
1489 (fh_len > 3 ? fid->i32.parent_gen : 0));
1490 break;
1491 }
1492
1493 return d_obtain_alias(inode);
1494}
1495EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1496
1497/**
1498 * __generic_file_fsync - generic fsync implementation for simple filesystems
1499 *
1500 * @file: file to synchronize
1501 * @start: start offset in bytes
1502 * @end: end offset in bytes (inclusive)
1503 * @datasync: only synchronize essential metadata if true
1504 *
1505 * This is a generic implementation of the fsync method for simple
1506 * filesystems which track all non-inode metadata in the buffers list
1507 * hanging off the address_space structure.
1508 */
1509int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1510 int datasync)
1511{
1512 struct inode *inode = file->f_mapping->host;
1513 int err;
1514 int ret;
1515
1516 err = file_write_and_wait_range(file, start, end);
1517 if (err)
1518 return err;
1519
1520 inode_lock(inode);
1521 ret = sync_mapping_buffers(inode->i_mapping);
1522 if (!(inode->i_state & I_DIRTY_ALL))
1523 goto out;
1524 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1525 goto out;
1526
1527 err = sync_inode_metadata(inode, 1);
1528 if (ret == 0)
1529 ret = err;
1530
1531out:
1532 inode_unlock(inode);
1533 /* check and advance again to catch errors after syncing out buffers */
1534 err = file_check_and_advance_wb_err(file);
1535 if (ret == 0)
1536 ret = err;
1537 return ret;
1538}
1539EXPORT_SYMBOL(__generic_file_fsync);
1540
1541/**
1542 * generic_file_fsync - generic fsync implementation for simple filesystems
1543 * with flush
1544 * @file: file to synchronize
1545 * @start: start offset in bytes
1546 * @end: end offset in bytes (inclusive)
1547 * @datasync: only synchronize essential metadata if true
1548 *
1549 */
1550
1551int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1552 int datasync)
1553{
1554 struct inode *inode = file->f_mapping->host;
1555 int err;
1556
1557 err = __generic_file_fsync(file, start, end, datasync);
1558 if (err)
1559 return err;
1560 return blkdev_issue_flush(inode->i_sb->s_bdev);
1561}
1562EXPORT_SYMBOL(generic_file_fsync);
1563
1564/**
1565 * generic_check_addressable - Check addressability of file system
1566 * @blocksize_bits: log of file system block size
1567 * @num_blocks: number of blocks in file system
1568 *
1569 * Determine whether a file system with @num_blocks blocks (and a
1570 * block size of 2**@blocksize_bits) is addressable by the sector_t
1571 * and page cache of the system. Return 0 if so and -EFBIG otherwise.
1572 */
1573int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1574{
1575 u64 last_fs_block = num_blocks - 1;
1576 u64 last_fs_page =
1577 last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1578
1579 if (unlikely(num_blocks == 0))
1580 return 0;
1581
1582 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1583 return -EINVAL;
1584
1585 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1586 (last_fs_page > (pgoff_t)(~0ULL))) {
1587 return -EFBIG;
1588 }
1589 return 0;
1590}
1591EXPORT_SYMBOL(generic_check_addressable);
1592
1593/*
1594 * No-op implementation of ->fsync for in-memory filesystems.
1595 */
1596int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1597{
1598 return 0;
1599}
1600EXPORT_SYMBOL(noop_fsync);
1601
1602ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1603{
1604 /*
1605 * iomap based filesystems support direct I/O without need for
1606 * this callback. However, it still needs to be set in
1607 * inode->a_ops so that open/fcntl know that direct I/O is
1608 * generally supported.
1609 */
1610 return -EINVAL;
1611}
1612EXPORT_SYMBOL_GPL(noop_direct_IO);
1613
1614/* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1615void kfree_link(void *p)
1616{
1617 kfree(p);
1618}
1619EXPORT_SYMBOL(kfree_link);
1620
1621struct inode *alloc_anon_inode(struct super_block *s)
1622{
1623 static const struct address_space_operations anon_aops = {
1624 .dirty_folio = noop_dirty_folio,
1625 };
1626 struct inode *inode = new_inode_pseudo(s);
1627
1628 if (!inode)
1629 return ERR_PTR(-ENOMEM);
1630
1631 inode->i_ino = get_next_ino();
1632 inode->i_mapping->a_ops = &anon_aops;
1633
1634 /*
1635 * Mark the inode dirty from the very beginning,
1636 * that way it will never be moved to the dirty
1637 * list because mark_inode_dirty() will think
1638 * that it already _is_ on the dirty list.
1639 */
1640 inode->i_state = I_DIRTY;
1641 inode->i_mode = S_IRUSR | S_IWUSR;
1642 inode->i_uid = current_fsuid();
1643 inode->i_gid = current_fsgid();
1644 inode->i_flags |= S_PRIVATE;
1645 simple_inode_init_ts(inode);
1646 return inode;
1647}
1648EXPORT_SYMBOL(alloc_anon_inode);
1649
1650/**
1651 * simple_nosetlease - generic helper for prohibiting leases
1652 * @filp: file pointer
1653 * @arg: type of lease to obtain
1654 * @flp: new lease supplied for insertion
1655 * @priv: private data for lm_setup operation
1656 *
1657 * Generic helper for filesystems that do not wish to allow leases to be set.
1658 * All arguments are ignored and it just returns -EINVAL.
1659 */
1660int
1661simple_nosetlease(struct file *filp, int arg, struct file_lease **flp,
1662 void **priv)
1663{
1664 return -EINVAL;
1665}
1666EXPORT_SYMBOL(simple_nosetlease);
1667
1668/**
1669 * simple_get_link - generic helper to get the target of "fast" symlinks
1670 * @dentry: not used here
1671 * @inode: the symlink inode
1672 * @done: not used here
1673 *
1674 * Generic helper for filesystems to use for symlink inodes where a pointer to
1675 * the symlink target is stored in ->i_link. NOTE: this isn't normally called,
1676 * since as an optimization the path lookup code uses any non-NULL ->i_link
1677 * directly, without calling ->get_link(). But ->get_link() still must be set,
1678 * to mark the inode_operations as being for a symlink.
1679 *
1680 * Return: the symlink target
1681 */
1682const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1683 struct delayed_call *done)
1684{
1685 return inode->i_link;
1686}
1687EXPORT_SYMBOL(simple_get_link);
1688
1689const struct inode_operations simple_symlink_inode_operations = {
1690 .get_link = simple_get_link,
1691};
1692EXPORT_SYMBOL(simple_symlink_inode_operations);
1693
1694/*
1695 * Operations for a permanently empty directory.
1696 */
1697static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1698{
1699 return ERR_PTR(-ENOENT);
1700}
1701
1702static int empty_dir_getattr(struct mnt_idmap *idmap,
1703 const struct path *path, struct kstat *stat,
1704 u32 request_mask, unsigned int query_flags)
1705{
1706 struct inode *inode = d_inode(path->dentry);
1707 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1708 return 0;
1709}
1710
1711static int empty_dir_setattr(struct mnt_idmap *idmap,
1712 struct dentry *dentry, struct iattr *attr)
1713{
1714 return -EPERM;
1715}
1716
1717static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1718{
1719 return -EOPNOTSUPP;
1720}
1721
1722static const struct inode_operations empty_dir_inode_operations = {
1723 .lookup = empty_dir_lookup,
1724 .permission = generic_permission,
1725 .setattr = empty_dir_setattr,
1726 .getattr = empty_dir_getattr,
1727 .listxattr = empty_dir_listxattr,
1728};
1729
1730static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1731{
1732 /* An empty directory has two entries . and .. at offsets 0 and 1 */
1733 return generic_file_llseek_size(file, offset, whence, 2, 2);
1734}
1735
1736static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1737{
1738 dir_emit_dots(file, ctx);
1739 return 0;
1740}
1741
1742static const struct file_operations empty_dir_operations = {
1743 .llseek = empty_dir_llseek,
1744 .read = generic_read_dir,
1745 .iterate_shared = empty_dir_readdir,
1746 .fsync = noop_fsync,
1747};
1748
1749
1750void make_empty_dir_inode(struct inode *inode)
1751{
1752 set_nlink(inode, 2);
1753 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1754 inode->i_uid = GLOBAL_ROOT_UID;
1755 inode->i_gid = GLOBAL_ROOT_GID;
1756 inode->i_rdev = 0;
1757 inode->i_size = 0;
1758 inode->i_blkbits = PAGE_SHIFT;
1759 inode->i_blocks = 0;
1760
1761 inode->i_op = &empty_dir_inode_operations;
1762 inode->i_opflags &= ~IOP_XATTR;
1763 inode->i_fop = &empty_dir_operations;
1764}
1765
1766bool is_empty_dir_inode(struct inode *inode)
1767{
1768 return (inode->i_fop == &empty_dir_operations) &&
1769 (inode->i_op == &empty_dir_inode_operations);
1770}
1771
1772#if IS_ENABLED(CONFIG_UNICODE)
1773/**
1774 * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1775 * @dentry: dentry whose name we are checking against
1776 * @len: len of name of dentry
1777 * @str: str pointer to name of dentry
1778 * @name: Name to compare against
1779 *
1780 * Return: 0 if names match, 1 if mismatch, or -ERRNO
1781 */
1782static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1783 const char *str, const struct qstr *name)
1784{
1785 const struct dentry *parent;
1786 const struct inode *dir;
1787 char strbuf[DNAME_INLINE_LEN];
1788 struct qstr qstr;
1789
1790 /*
1791 * Attempt a case-sensitive match first. It is cheaper and
1792 * should cover most lookups, including all the sane
1793 * applications that expect a case-sensitive filesystem.
1794 *
1795 * This comparison is safe under RCU because the caller
1796 * guarantees the consistency between str and len. See
1797 * __d_lookup_rcu_op_compare() for details.
1798 */
1799 if (len == name->len && !memcmp(str, name->name, len))
1800 return 0;
1801
1802 parent = READ_ONCE(dentry->d_parent);
1803 dir = READ_ONCE(parent->d_inode);
1804 if (!dir || !IS_CASEFOLDED(dir))
1805 return 1;
1806
1807 /*
1808 * If the dentry name is stored in-line, then it may be concurrently
1809 * modified by a rename. If this happens, the VFS will eventually retry
1810 * the lookup, so it doesn't matter what ->d_compare() returns.
1811 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1812 * string. Therefore, we have to copy the name into a temporary buffer.
1813 */
1814 if (len <= DNAME_INLINE_LEN - 1) {
1815 memcpy(strbuf, str, len);
1816 strbuf[len] = 0;
1817 str = strbuf;
1818 /* prevent compiler from optimizing out the temporary buffer */
1819 barrier();
1820 }
1821 qstr.len = len;
1822 qstr.name = str;
1823
1824 return utf8_strncasecmp(dentry->d_sb->s_encoding, name, &qstr);
1825}
1826
1827/**
1828 * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1829 * @dentry: dentry of the parent directory
1830 * @str: qstr of name whose hash we should fill in
1831 *
1832 * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1833 */
1834static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1835{
1836 const struct inode *dir = READ_ONCE(dentry->d_inode);
1837 struct super_block *sb = dentry->d_sb;
1838 const struct unicode_map *um = sb->s_encoding;
1839 int ret;
1840
1841 if (!dir || !IS_CASEFOLDED(dir))
1842 return 0;
1843
1844 ret = utf8_casefold_hash(um, dentry, str);
1845 if (ret < 0 && sb_has_strict_encoding(sb))
1846 return -EINVAL;
1847 return 0;
1848}
1849
1850static const struct dentry_operations generic_ci_dentry_ops = {
1851 .d_hash = generic_ci_d_hash,
1852 .d_compare = generic_ci_d_compare,
1853#ifdef CONFIG_FS_ENCRYPTION
1854 .d_revalidate = fscrypt_d_revalidate,
1855#endif
1856};
1857#endif
1858
1859#ifdef CONFIG_FS_ENCRYPTION
1860static const struct dentry_operations generic_encrypted_dentry_ops = {
1861 .d_revalidate = fscrypt_d_revalidate,
1862};
1863#endif
1864
1865/**
1866 * generic_set_sb_d_ops - helper for choosing the set of
1867 * filesystem-wide dentry operations for the enabled features
1868 * @sb: superblock to be configured
1869 *
1870 * Filesystems supporting casefolding and/or fscrypt can call this
1871 * helper at mount-time to configure sb->s_d_op to best set of dentry
1872 * operations required for the enabled features. The helper must be
1873 * called after these have been configured, but before the root dentry
1874 * is created.
1875 */
1876void generic_set_sb_d_ops(struct super_block *sb)
1877{
1878#if IS_ENABLED(CONFIG_UNICODE)
1879 if (sb->s_encoding) {
1880 sb->s_d_op = &generic_ci_dentry_ops;
1881 return;
1882 }
1883#endif
1884#ifdef CONFIG_FS_ENCRYPTION
1885 if (sb->s_cop) {
1886 sb->s_d_op = &generic_encrypted_dentry_ops;
1887 return;
1888 }
1889#endif
1890}
1891EXPORT_SYMBOL(generic_set_sb_d_ops);
1892
1893/**
1894 * inode_maybe_inc_iversion - increments i_version
1895 * @inode: inode with the i_version that should be updated
1896 * @force: increment the counter even if it's not necessary?
1897 *
1898 * Every time the inode is modified, the i_version field must be seen to have
1899 * changed by any observer.
1900 *
1901 * If "force" is set or the QUERIED flag is set, then ensure that we increment
1902 * the value, and clear the queried flag.
1903 *
1904 * In the common case where neither is set, then we can return "false" without
1905 * updating i_version.
1906 *
1907 * If this function returns false, and no other metadata has changed, then we
1908 * can avoid logging the metadata.
1909 */
1910bool inode_maybe_inc_iversion(struct inode *inode, bool force)
1911{
1912 u64 cur, new;
1913
1914 /*
1915 * The i_version field is not strictly ordered with any other inode
1916 * information, but the legacy inode_inc_iversion code used a spinlock
1917 * to serialize increments.
1918 *
1919 * Here, we add full memory barriers to ensure that any de-facto
1920 * ordering with other info is preserved.
1921 *
1922 * This barrier pairs with the barrier in inode_query_iversion()
1923 */
1924 smp_mb();
1925 cur = inode_peek_iversion_raw(inode);
1926 do {
1927 /* If flag is clear then we needn't do anything */
1928 if (!force && !(cur & I_VERSION_QUERIED))
1929 return false;
1930
1931 /* Since lowest bit is flag, add 2 to avoid it */
1932 new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
1933 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
1934 return true;
1935}
1936EXPORT_SYMBOL(inode_maybe_inc_iversion);
1937
1938/**
1939 * inode_query_iversion - read i_version for later use
1940 * @inode: inode from which i_version should be read
1941 *
1942 * Read the inode i_version counter. This should be used by callers that wish
1943 * to store the returned i_version for later comparison. This will guarantee
1944 * that a later query of the i_version will result in a different value if
1945 * anything has changed.
1946 *
1947 * In this implementation, we fetch the current value, set the QUERIED flag and
1948 * then try to swap it into place with a cmpxchg, if it wasn't already set. If
1949 * that fails, we try again with the newly fetched value from the cmpxchg.
1950 */
1951u64 inode_query_iversion(struct inode *inode)
1952{
1953 u64 cur, new;
1954
1955 cur = inode_peek_iversion_raw(inode);
1956 do {
1957 /* If flag is already set, then no need to swap */
1958 if (cur & I_VERSION_QUERIED) {
1959 /*
1960 * This barrier (and the implicit barrier in the
1961 * cmpxchg below) pairs with the barrier in
1962 * inode_maybe_inc_iversion().
1963 */
1964 smp_mb();
1965 break;
1966 }
1967
1968 new = cur | I_VERSION_QUERIED;
1969 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
1970 return cur >> I_VERSION_QUERIED_SHIFT;
1971}
1972EXPORT_SYMBOL(inode_query_iversion);
1973
1974ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter,
1975 ssize_t direct_written, ssize_t buffered_written)
1976{
1977 struct address_space *mapping = iocb->ki_filp->f_mapping;
1978 loff_t pos = iocb->ki_pos - buffered_written;
1979 loff_t end = iocb->ki_pos - 1;
1980 int err;
1981
1982 /*
1983 * If the buffered write fallback returned an error, we want to return
1984 * the number of bytes which were written by direct I/O, or the error
1985 * code if that was zero.
1986 *
1987 * Note that this differs from normal direct-io semantics, which will
1988 * return -EFOO even if some bytes were written.
1989 */
1990 if (unlikely(buffered_written < 0)) {
1991 if (direct_written)
1992 return direct_written;
1993 return buffered_written;
1994 }
1995
1996 /*
1997 * We need to ensure that the page cache pages are written to disk and
1998 * invalidated to preserve the expected O_DIRECT semantics.
1999 */
2000 err = filemap_write_and_wait_range(mapping, pos, end);
2001 if (err < 0) {
2002 /*
2003 * We don't know how much we wrote, so just return the number of
2004 * bytes which were direct-written
2005 */
2006 iocb->ki_pos -= buffered_written;
2007 if (direct_written)
2008 return direct_written;
2009 return err;
2010 }
2011 invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
2012 return direct_written + buffered_written;
2013}
2014EXPORT_SYMBOL_GPL(direct_write_fallback);
2015
2016/**
2017 * simple_inode_init_ts - initialize the timestamps for a new inode
2018 * @inode: inode to be initialized
2019 *
2020 * When a new inode is created, most filesystems set the timestamps to the
2021 * current time. Add a helper to do this.
2022 */
2023struct timespec64 simple_inode_init_ts(struct inode *inode)
2024{
2025 struct timespec64 ts = inode_set_ctime_current(inode);
2026
2027 inode_set_atime_to_ts(inode, ts);
2028 inode_set_mtime_to_ts(inode, ts);
2029 return ts;
2030}
2031EXPORT_SYMBOL(simple_inode_init_ts);
2032
2033static inline struct dentry *get_stashed_dentry(struct dentry *stashed)
2034{
2035 struct dentry *dentry;
2036
2037 guard(rcu)();
2038 dentry = READ_ONCE(stashed);
2039 if (!dentry)
2040 return NULL;
2041 if (!lockref_get_not_dead(&dentry->d_lockref))
2042 return NULL;
2043 return dentry;
2044}
2045
2046static struct dentry *prepare_anon_dentry(struct dentry **stashed,
2047 struct super_block *sb,
2048 void *data)
2049{
2050 struct dentry *dentry;
2051 struct inode *inode;
2052 const struct stashed_operations *sops = sb->s_fs_info;
2053 int ret;
2054
2055 inode = new_inode_pseudo(sb);
2056 if (!inode) {
2057 sops->put_data(data);
2058 return ERR_PTR(-ENOMEM);
2059 }
2060
2061 inode->i_flags |= S_IMMUTABLE;
2062 inode->i_mode = S_IFREG;
2063 simple_inode_init_ts(inode);
2064
2065 ret = sops->init_inode(inode, data);
2066 if (ret < 0) {
2067 iput(inode);
2068 return ERR_PTR(ret);
2069 }
2070
2071 /* Notice when this is changed. */
2072 WARN_ON_ONCE(!S_ISREG(inode->i_mode));
2073 WARN_ON_ONCE(!IS_IMMUTABLE(inode));
2074
2075 dentry = d_alloc_anon(sb);
2076 if (!dentry) {
2077 iput(inode);
2078 return ERR_PTR(-ENOMEM);
2079 }
2080
2081 /* Store address of location where dentry's supposed to be stashed. */
2082 dentry->d_fsdata = stashed;
2083
2084 /* @data is now owned by the fs */
2085 d_instantiate(dentry, inode);
2086 return dentry;
2087}
2088
2089static struct dentry *stash_dentry(struct dentry **stashed,
2090 struct dentry *dentry)
2091{
2092 guard(rcu)();
2093 for (;;) {
2094 struct dentry *old;
2095
2096 /* Assume any old dentry was cleared out. */
2097 old = cmpxchg(stashed, NULL, dentry);
2098 if (likely(!old))
2099 return dentry;
2100
2101 /* Check if somebody else installed a reusable dentry. */
2102 if (lockref_get_not_dead(&old->d_lockref))
2103 return old;
2104
2105 /* There's an old dead dentry there, try to take it over. */
2106 if (likely(try_cmpxchg(stashed, &old, dentry)))
2107 return dentry;
2108 }
2109}
2110
2111/**
2112 * path_from_stashed - create path from stashed or new dentry
2113 * @stashed: where to retrieve or stash dentry
2114 * @mnt: mnt of the filesystems to use
2115 * @data: data to store in inode->i_private
2116 * @path: path to create
2117 *
2118 * The function tries to retrieve a stashed dentry from @stashed. If the dentry
2119 * is still valid then it will be reused. If the dentry isn't able the function
2120 * will allocate a new dentry and inode. It will then check again whether it
2121 * can reuse an existing dentry in case one has been added in the meantime or
2122 * update @stashed with the newly added dentry.
2123 *
2124 * Special-purpose helper for nsfs and pidfs.
2125 *
2126 * Return: On success zero and on failure a negative error is returned.
2127 */
2128int path_from_stashed(struct dentry **stashed, struct vfsmount *mnt, void *data,
2129 struct path *path)
2130{
2131 struct dentry *dentry;
2132 const struct stashed_operations *sops = mnt->mnt_sb->s_fs_info;
2133
2134 /* See if dentry can be reused. */
2135 path->dentry = get_stashed_dentry(*stashed);
2136 if (path->dentry) {
2137 sops->put_data(data);
2138 goto out_path;
2139 }
2140
2141 /* Allocate a new dentry. */
2142 dentry = prepare_anon_dentry(stashed, mnt->mnt_sb, data);
2143 if (IS_ERR(dentry))
2144 return PTR_ERR(dentry);
2145
2146 /* Added a new dentry. @data is now owned by the filesystem. */
2147 path->dentry = stash_dentry(stashed, dentry);
2148 if (path->dentry != dentry)
2149 dput(dentry);
2150
2151out_path:
2152 WARN_ON_ONCE(path->dentry->d_fsdata != stashed);
2153 WARN_ON_ONCE(d_inode(path->dentry)->i_private != data);
2154 path->mnt = mntget(mnt);
2155 return 0;
2156}
2157
2158void stashed_dentry_prune(struct dentry *dentry)
2159{
2160 struct dentry **stashed = dentry->d_fsdata;
2161 struct inode *inode = d_inode(dentry);
2162
2163 if (WARN_ON_ONCE(!stashed))
2164 return;
2165
2166 if (!inode)
2167 return;
2168
2169 /*
2170 * Only replace our own @dentry as someone else might've
2171 * already cleared out @dentry and stashed their own
2172 * dentry in there.
2173 */
2174 cmpxchg(stashed, dentry, NULL);
2175}