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#include <linux/fsnotify.h>
23
24#include <linux/uaccess.h>
25
26#include "internal.h"
27
28int simple_getattr(const struct path *path, struct kstat *stat,
29 u32 request_mask, unsigned int query_flags)
30{
31 struct inode *inode = d_inode(path->dentry);
32 generic_fillattr(inode, stat);
33 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
34 return 0;
35}
36EXPORT_SYMBOL(simple_getattr);
37
38int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
39{
40 buf->f_type = dentry->d_sb->s_magic;
41 buf->f_bsize = PAGE_SIZE;
42 buf->f_namelen = NAME_MAX;
43 return 0;
44}
45EXPORT_SYMBOL(simple_statfs);
46
47/*
48 * Retaining negative dentries for an in-memory filesystem just wastes
49 * memory and lookup time: arrange for them to be deleted immediately.
50 */
51int always_delete_dentry(const struct dentry *dentry)
52{
53 return 1;
54}
55EXPORT_SYMBOL(always_delete_dentry);
56
57const struct dentry_operations simple_dentry_operations = {
58 .d_delete = always_delete_dentry,
59};
60EXPORT_SYMBOL(simple_dentry_operations);
61
62/*
63 * Lookup the data. This is trivial - if the dentry didn't already
64 * exist, we know it is negative. Set d_op to delete negative dentries.
65 */
66struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
67{
68 if (dentry->d_name.len > NAME_MAX)
69 return ERR_PTR(-ENAMETOOLONG);
70 if (!dentry->d_sb->s_d_op)
71 d_set_d_op(dentry, &simple_dentry_operations);
72 d_add(dentry, NULL);
73 return NULL;
74}
75EXPORT_SYMBOL(simple_lookup);
76
77int dcache_dir_open(struct inode *inode, struct file *file)
78{
79 file->private_data = d_alloc_cursor(file->f_path.dentry);
80
81 return file->private_data ? 0 : -ENOMEM;
82}
83EXPORT_SYMBOL(dcache_dir_open);
84
85int dcache_dir_close(struct inode *inode, struct file *file)
86{
87 dput(file->private_data);
88 return 0;
89}
90EXPORT_SYMBOL(dcache_dir_close);
91
92/* parent is locked at least shared */
93/*
94 * Returns an element of siblings' list.
95 * We are looking for <count>th positive after <p>; if
96 * found, dentry is grabbed and returned to caller.
97 * If no such element exists, NULL is returned.
98 */
99static struct dentry *scan_positives(struct dentry *cursor,
100 struct list_head *p,
101 loff_t count,
102 struct dentry *last)
103{
104 struct dentry *dentry = cursor->d_parent, *found = NULL;
105
106 spin_lock(&dentry->d_lock);
107 while ((p = p->next) != &dentry->d_subdirs) {
108 struct dentry *d = list_entry(p, struct dentry, d_child);
109 // we must at least skip cursors, to avoid livelocks
110 if (d->d_flags & DCACHE_DENTRY_CURSOR)
111 continue;
112 if (simple_positive(d) && !--count) {
113 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
114 if (simple_positive(d))
115 found = dget_dlock(d);
116 spin_unlock(&d->d_lock);
117 if (likely(found))
118 break;
119 count = 1;
120 }
121 if (need_resched()) {
122 list_move(&cursor->d_child, p);
123 p = &cursor->d_child;
124 spin_unlock(&dentry->d_lock);
125 cond_resched();
126 spin_lock(&dentry->d_lock);
127 }
128 }
129 spin_unlock(&dentry->d_lock);
130 dput(last);
131 return found;
132}
133
134loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
135{
136 struct dentry *dentry = file->f_path.dentry;
137 switch (whence) {
138 case 1:
139 offset += file->f_pos;
140 fallthrough;
141 case 0:
142 if (offset >= 0)
143 break;
144 fallthrough;
145 default:
146 return -EINVAL;
147 }
148 if (offset != file->f_pos) {
149 struct dentry *cursor = file->private_data;
150 struct dentry *to = NULL;
151
152 inode_lock_shared(dentry->d_inode);
153
154 if (offset > 2)
155 to = scan_positives(cursor, &dentry->d_subdirs,
156 offset - 2, NULL);
157 spin_lock(&dentry->d_lock);
158 if (to)
159 list_move(&cursor->d_child, &to->d_child);
160 else
161 list_del_init(&cursor->d_child);
162 spin_unlock(&dentry->d_lock);
163 dput(to);
164
165 file->f_pos = offset;
166
167 inode_unlock_shared(dentry->d_inode);
168 }
169 return offset;
170}
171EXPORT_SYMBOL(dcache_dir_lseek);
172
173/* Relationship between i_mode and the DT_xxx types */
174static inline unsigned char dt_type(struct inode *inode)
175{
176 return (inode->i_mode >> 12) & 15;
177}
178
179/*
180 * Directory is locked and all positive dentries in it are safe, since
181 * for ramfs-type trees they can't go away without unlink() or rmdir(),
182 * both impossible due to the lock on directory.
183 */
184
185int dcache_readdir(struct file *file, struct dir_context *ctx)
186{
187 struct dentry *dentry = file->f_path.dentry;
188 struct dentry *cursor = file->private_data;
189 struct list_head *anchor = &dentry->d_subdirs;
190 struct dentry *next = NULL;
191 struct list_head *p;
192
193 if (!dir_emit_dots(file, ctx))
194 return 0;
195
196 if (ctx->pos == 2)
197 p = anchor;
198 else if (!list_empty(&cursor->d_child))
199 p = &cursor->d_child;
200 else
201 return 0;
202
203 while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
204 if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
205 d_inode(next)->i_ino, dt_type(d_inode(next))))
206 break;
207 ctx->pos++;
208 p = &next->d_child;
209 }
210 spin_lock(&dentry->d_lock);
211 if (next)
212 list_move_tail(&cursor->d_child, &next->d_child);
213 else
214 list_del_init(&cursor->d_child);
215 spin_unlock(&dentry->d_lock);
216 dput(next);
217
218 return 0;
219}
220EXPORT_SYMBOL(dcache_readdir);
221
222ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
223{
224 return -EISDIR;
225}
226EXPORT_SYMBOL(generic_read_dir);
227
228const struct file_operations simple_dir_operations = {
229 .open = dcache_dir_open,
230 .release = dcache_dir_close,
231 .llseek = dcache_dir_lseek,
232 .read = generic_read_dir,
233 .iterate_shared = dcache_readdir,
234 .fsync = noop_fsync,
235};
236EXPORT_SYMBOL(simple_dir_operations);
237
238const struct inode_operations simple_dir_inode_operations = {
239 .lookup = simple_lookup,
240};
241EXPORT_SYMBOL(simple_dir_inode_operations);
242
243static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
244{
245 struct dentry *child = NULL;
246 struct list_head *p = prev ? &prev->d_child : &parent->d_subdirs;
247
248 spin_lock(&parent->d_lock);
249 while ((p = p->next) != &parent->d_subdirs) {
250 struct dentry *d = container_of(p, struct dentry, d_child);
251 if (simple_positive(d)) {
252 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
253 if (simple_positive(d))
254 child = dget_dlock(d);
255 spin_unlock(&d->d_lock);
256 if (likely(child))
257 break;
258 }
259 }
260 spin_unlock(&parent->d_lock);
261 dput(prev);
262 return child;
263}
264
265void simple_recursive_removal(struct dentry *dentry,
266 void (*callback)(struct dentry *))
267{
268 struct dentry *this = dget(dentry);
269 while (true) {
270 struct dentry *victim = NULL, *child;
271 struct inode *inode = this->d_inode;
272
273 inode_lock(inode);
274 if (d_is_dir(this))
275 inode->i_flags |= S_DEAD;
276 while ((child = find_next_child(this, victim)) == NULL) {
277 // kill and ascend
278 // update metadata while it's still locked
279 inode->i_ctime = current_time(inode);
280 clear_nlink(inode);
281 inode_unlock(inode);
282 victim = this;
283 this = this->d_parent;
284 inode = this->d_inode;
285 inode_lock(inode);
286 if (simple_positive(victim)) {
287 d_invalidate(victim); // avoid lost mounts
288 if (d_is_dir(victim))
289 fsnotify_rmdir(inode, victim);
290 else
291 fsnotify_unlink(inode, victim);
292 if (callback)
293 callback(victim);
294 dput(victim); // unpin it
295 }
296 if (victim == dentry) {
297 inode->i_ctime = inode->i_mtime =
298 current_time(inode);
299 if (d_is_dir(dentry))
300 drop_nlink(inode);
301 inode_unlock(inode);
302 dput(dentry);
303 return;
304 }
305 }
306 inode_unlock(inode);
307 this = child;
308 }
309}
310EXPORT_SYMBOL(simple_recursive_removal);
311
312static const struct super_operations simple_super_operations = {
313 .statfs = simple_statfs,
314};
315
316static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
317{
318 struct pseudo_fs_context *ctx = fc->fs_private;
319 struct inode *root;
320
321 s->s_maxbytes = MAX_LFS_FILESIZE;
322 s->s_blocksize = PAGE_SIZE;
323 s->s_blocksize_bits = PAGE_SHIFT;
324 s->s_magic = ctx->magic;
325 s->s_op = ctx->ops ?: &simple_super_operations;
326 s->s_xattr = ctx->xattr;
327 s->s_time_gran = 1;
328 root = new_inode(s);
329 if (!root)
330 return -ENOMEM;
331
332 /*
333 * since this is the first inode, make it number 1. New inodes created
334 * after this must take care not to collide with it (by passing
335 * max_reserved of 1 to iunique).
336 */
337 root->i_ino = 1;
338 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
339 root->i_atime = root->i_mtime = root->i_ctime = current_time(root);
340 s->s_root = d_make_root(root);
341 if (!s->s_root)
342 return -ENOMEM;
343 s->s_d_op = ctx->dops;
344 return 0;
345}
346
347static int pseudo_fs_get_tree(struct fs_context *fc)
348{
349 return get_tree_nodev(fc, pseudo_fs_fill_super);
350}
351
352static void pseudo_fs_free(struct fs_context *fc)
353{
354 kfree(fc->fs_private);
355}
356
357static const struct fs_context_operations pseudo_fs_context_ops = {
358 .free = pseudo_fs_free,
359 .get_tree = pseudo_fs_get_tree,
360};
361
362/*
363 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
364 * will never be mountable)
365 */
366struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
367 unsigned long magic)
368{
369 struct pseudo_fs_context *ctx;
370
371 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
372 if (likely(ctx)) {
373 ctx->magic = magic;
374 fc->fs_private = ctx;
375 fc->ops = &pseudo_fs_context_ops;
376 fc->sb_flags |= SB_NOUSER;
377 fc->global = true;
378 }
379 return ctx;
380}
381EXPORT_SYMBOL(init_pseudo);
382
383int simple_open(struct inode *inode, struct file *file)
384{
385 if (inode->i_private)
386 file->private_data = inode->i_private;
387 return 0;
388}
389EXPORT_SYMBOL(simple_open);
390
391int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
392{
393 struct inode *inode = d_inode(old_dentry);
394
395 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
396 inc_nlink(inode);
397 ihold(inode);
398 dget(dentry);
399 d_instantiate(dentry, inode);
400 return 0;
401}
402EXPORT_SYMBOL(simple_link);
403
404int simple_empty(struct dentry *dentry)
405{
406 struct dentry *child;
407 int ret = 0;
408
409 spin_lock(&dentry->d_lock);
410 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
411 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
412 if (simple_positive(child)) {
413 spin_unlock(&child->d_lock);
414 goto out;
415 }
416 spin_unlock(&child->d_lock);
417 }
418 ret = 1;
419out:
420 spin_unlock(&dentry->d_lock);
421 return ret;
422}
423EXPORT_SYMBOL(simple_empty);
424
425int simple_unlink(struct inode *dir, struct dentry *dentry)
426{
427 struct inode *inode = d_inode(dentry);
428
429 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
430 drop_nlink(inode);
431 dput(dentry);
432 return 0;
433}
434EXPORT_SYMBOL(simple_unlink);
435
436int simple_rmdir(struct inode *dir, struct dentry *dentry)
437{
438 if (!simple_empty(dentry))
439 return -ENOTEMPTY;
440
441 drop_nlink(d_inode(dentry));
442 simple_unlink(dir, dentry);
443 drop_nlink(dir);
444 return 0;
445}
446EXPORT_SYMBOL(simple_rmdir);
447
448int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
449 struct inode *new_dir, struct dentry *new_dentry,
450 unsigned int flags)
451{
452 struct inode *inode = d_inode(old_dentry);
453 int they_are_dirs = d_is_dir(old_dentry);
454
455 if (flags & ~RENAME_NOREPLACE)
456 return -EINVAL;
457
458 if (!simple_empty(new_dentry))
459 return -ENOTEMPTY;
460
461 if (d_really_is_positive(new_dentry)) {
462 simple_unlink(new_dir, new_dentry);
463 if (they_are_dirs) {
464 drop_nlink(d_inode(new_dentry));
465 drop_nlink(old_dir);
466 }
467 } else if (they_are_dirs) {
468 drop_nlink(old_dir);
469 inc_nlink(new_dir);
470 }
471
472 old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
473 new_dir->i_mtime = inode->i_ctime = current_time(old_dir);
474
475 return 0;
476}
477EXPORT_SYMBOL(simple_rename);
478
479/**
480 * simple_setattr - setattr for simple filesystem
481 * @dentry: dentry
482 * @iattr: iattr structure
483 *
484 * Returns 0 on success, -error on failure.
485 *
486 * simple_setattr is a simple ->setattr implementation without a proper
487 * implementation of size changes.
488 *
489 * It can either be used for in-memory filesystems or special files
490 * on simple regular filesystems. Anything that needs to change on-disk
491 * or wire state on size changes needs its own setattr method.
492 */
493int simple_setattr(struct dentry *dentry, struct iattr *iattr)
494{
495 struct inode *inode = d_inode(dentry);
496 int error;
497
498 error = setattr_prepare(dentry, iattr);
499 if (error)
500 return error;
501
502 if (iattr->ia_valid & ATTR_SIZE)
503 truncate_setsize(inode, iattr->ia_size);
504 setattr_copy(inode, iattr);
505 mark_inode_dirty(inode);
506 return 0;
507}
508EXPORT_SYMBOL(simple_setattr);
509
510int simple_readpage(struct file *file, struct page *page)
511{
512 clear_highpage(page);
513 flush_dcache_page(page);
514 SetPageUptodate(page);
515 unlock_page(page);
516 return 0;
517}
518EXPORT_SYMBOL(simple_readpage);
519
520int simple_write_begin(struct file *file, struct address_space *mapping,
521 loff_t pos, unsigned len, unsigned flags,
522 struct page **pagep, void **fsdata)
523{
524 struct page *page;
525 pgoff_t index;
526
527 index = pos >> PAGE_SHIFT;
528
529 page = grab_cache_page_write_begin(mapping, index, flags);
530 if (!page)
531 return -ENOMEM;
532
533 *pagep = page;
534
535 if (!PageUptodate(page) && (len != PAGE_SIZE)) {
536 unsigned from = pos & (PAGE_SIZE - 1);
537
538 zero_user_segments(page, 0, from, from + len, PAGE_SIZE);
539 }
540 return 0;
541}
542EXPORT_SYMBOL(simple_write_begin);
543
544/**
545 * simple_write_end - .write_end helper for non-block-device FSes
546 * @file: See .write_end of address_space_operations
547 * @mapping: "
548 * @pos: "
549 * @len: "
550 * @copied: "
551 * @page: "
552 * @fsdata: "
553 *
554 * simple_write_end does the minimum needed for updating a page after writing is
555 * done. It has the same API signature as the .write_end of
556 * address_space_operations vector. So it can just be set onto .write_end for
557 * FSes that don't need any other processing. i_mutex is assumed to be held.
558 * Block based filesystems should use generic_write_end().
559 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
560 * is not called, so a filesystem that actually does store data in .write_inode
561 * should extend on what's done here with a call to mark_inode_dirty() in the
562 * case that i_size has changed.
563 *
564 * Use *ONLY* with simple_readpage()
565 */
566int simple_write_end(struct file *file, struct address_space *mapping,
567 loff_t pos, unsigned len, unsigned copied,
568 struct page *page, void *fsdata)
569{
570 struct inode *inode = page->mapping->host;
571 loff_t last_pos = pos + copied;
572
573 /* zero the stale part of the page if we did a short copy */
574 if (!PageUptodate(page)) {
575 if (copied < len) {
576 unsigned from = pos & (PAGE_SIZE - 1);
577
578 zero_user(page, from + copied, len - copied);
579 }
580 SetPageUptodate(page);
581 }
582 /*
583 * No need to use i_size_read() here, the i_size
584 * cannot change under us because we hold the i_mutex.
585 */
586 if (last_pos > inode->i_size)
587 i_size_write(inode, last_pos);
588
589 set_page_dirty(page);
590 unlock_page(page);
591 put_page(page);
592
593 return copied;
594}
595EXPORT_SYMBOL(simple_write_end);
596
597/*
598 * the inodes created here are not hashed. If you use iunique to generate
599 * unique inode values later for this filesystem, then you must take care
600 * to pass it an appropriate max_reserved value to avoid collisions.
601 */
602int simple_fill_super(struct super_block *s, unsigned long magic,
603 const struct tree_descr *files)
604{
605 struct inode *inode;
606 struct dentry *root;
607 struct dentry *dentry;
608 int i;
609
610 s->s_blocksize = PAGE_SIZE;
611 s->s_blocksize_bits = PAGE_SHIFT;
612 s->s_magic = magic;
613 s->s_op = &simple_super_operations;
614 s->s_time_gran = 1;
615
616 inode = new_inode(s);
617 if (!inode)
618 return -ENOMEM;
619 /*
620 * because the root inode is 1, the files array must not contain an
621 * entry at index 1
622 */
623 inode->i_ino = 1;
624 inode->i_mode = S_IFDIR | 0755;
625 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
626 inode->i_op = &simple_dir_inode_operations;
627 inode->i_fop = &simple_dir_operations;
628 set_nlink(inode, 2);
629 root = d_make_root(inode);
630 if (!root)
631 return -ENOMEM;
632 for (i = 0; !files->name || files->name[0]; i++, files++) {
633 if (!files->name)
634 continue;
635
636 /* warn if it tries to conflict with the root inode */
637 if (unlikely(i == 1))
638 printk(KERN_WARNING "%s: %s passed in a files array"
639 "with an index of 1!\n", __func__,
640 s->s_type->name);
641
642 dentry = d_alloc_name(root, files->name);
643 if (!dentry)
644 goto out;
645 inode = new_inode(s);
646 if (!inode) {
647 dput(dentry);
648 goto out;
649 }
650 inode->i_mode = S_IFREG | files->mode;
651 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
652 inode->i_fop = files->ops;
653 inode->i_ino = i;
654 d_add(dentry, inode);
655 }
656 s->s_root = root;
657 return 0;
658out:
659 d_genocide(root);
660 shrink_dcache_parent(root);
661 dput(root);
662 return -ENOMEM;
663}
664EXPORT_SYMBOL(simple_fill_super);
665
666static DEFINE_SPINLOCK(pin_fs_lock);
667
668int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
669{
670 struct vfsmount *mnt = NULL;
671 spin_lock(&pin_fs_lock);
672 if (unlikely(!*mount)) {
673 spin_unlock(&pin_fs_lock);
674 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
675 if (IS_ERR(mnt))
676 return PTR_ERR(mnt);
677 spin_lock(&pin_fs_lock);
678 if (!*mount)
679 *mount = mnt;
680 }
681 mntget(*mount);
682 ++*count;
683 spin_unlock(&pin_fs_lock);
684 mntput(mnt);
685 return 0;
686}
687EXPORT_SYMBOL(simple_pin_fs);
688
689void simple_release_fs(struct vfsmount **mount, int *count)
690{
691 struct vfsmount *mnt;
692 spin_lock(&pin_fs_lock);
693 mnt = *mount;
694 if (!--*count)
695 *mount = NULL;
696 spin_unlock(&pin_fs_lock);
697 mntput(mnt);
698}
699EXPORT_SYMBOL(simple_release_fs);
700
701/**
702 * simple_read_from_buffer - copy data from the buffer to user space
703 * @to: the user 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 simple_read_from_buffer() function reads up to @count bytes from the
710 * buffer @from at offset @ppos into the user 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 simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
716 const void *from, size_t available)
717{
718 loff_t pos = *ppos;
719 size_t ret;
720
721 if (pos < 0)
722 return -EINVAL;
723 if (pos >= available || !count)
724 return 0;
725 if (count > available - pos)
726 count = available - pos;
727 ret = copy_to_user(to, from + pos, count);
728 if (ret == count)
729 return -EFAULT;
730 count -= ret;
731 *ppos = pos + count;
732 return count;
733}
734EXPORT_SYMBOL(simple_read_from_buffer);
735
736/**
737 * simple_write_to_buffer - copy data from user space to the buffer
738 * @to: the buffer to write to
739 * @available: the size of the buffer
740 * @ppos: the current position in the buffer
741 * @from: the user space buffer to read from
742 * @count: the maximum number of bytes to read
743 *
744 * The simple_write_to_buffer() function reads up to @count bytes from the user
745 * space address starting at @from into the buffer @to at offset @ppos.
746 *
747 * On success, the number of bytes written is returned and the offset @ppos is
748 * advanced by this number, or negative value is returned on error.
749 **/
750ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
751 const void __user *from, size_t count)
752{
753 loff_t pos = *ppos;
754 size_t res;
755
756 if (pos < 0)
757 return -EINVAL;
758 if (pos >= available || !count)
759 return 0;
760 if (count > available - pos)
761 count = available - pos;
762 res = copy_from_user(to + pos, from, count);
763 if (res == count)
764 return -EFAULT;
765 count -= res;
766 *ppos = pos + count;
767 return count;
768}
769EXPORT_SYMBOL(simple_write_to_buffer);
770
771/**
772 * memory_read_from_buffer - copy data from the buffer
773 * @to: the kernel space buffer to read to
774 * @count: the maximum number of bytes to read
775 * @ppos: the current position in the buffer
776 * @from: the buffer to read from
777 * @available: the size of the buffer
778 *
779 * The memory_read_from_buffer() function reads up to @count bytes from the
780 * buffer @from at offset @ppos into the kernel space address starting at @to.
781 *
782 * On success, the number of bytes read is returned and the offset @ppos is
783 * advanced by this number, or negative value is returned on error.
784 **/
785ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
786 const void *from, size_t available)
787{
788 loff_t pos = *ppos;
789
790 if (pos < 0)
791 return -EINVAL;
792 if (pos >= available)
793 return 0;
794 if (count > available - pos)
795 count = available - pos;
796 memcpy(to, from + pos, count);
797 *ppos = pos + count;
798
799 return count;
800}
801EXPORT_SYMBOL(memory_read_from_buffer);
802
803/*
804 * Transaction based IO.
805 * The file expects a single write which triggers the transaction, and then
806 * possibly a read which collects the result - which is stored in a
807 * file-local buffer.
808 */
809
810void simple_transaction_set(struct file *file, size_t n)
811{
812 struct simple_transaction_argresp *ar = file->private_data;
813
814 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
815
816 /*
817 * The barrier ensures that ar->size will really remain zero until
818 * ar->data is ready for reading.
819 */
820 smp_mb();
821 ar->size = n;
822}
823EXPORT_SYMBOL(simple_transaction_set);
824
825char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
826{
827 struct simple_transaction_argresp *ar;
828 static DEFINE_SPINLOCK(simple_transaction_lock);
829
830 if (size > SIMPLE_TRANSACTION_LIMIT - 1)
831 return ERR_PTR(-EFBIG);
832
833 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
834 if (!ar)
835 return ERR_PTR(-ENOMEM);
836
837 spin_lock(&simple_transaction_lock);
838
839 /* only one write allowed per open */
840 if (file->private_data) {
841 spin_unlock(&simple_transaction_lock);
842 free_page((unsigned long)ar);
843 return ERR_PTR(-EBUSY);
844 }
845
846 file->private_data = ar;
847
848 spin_unlock(&simple_transaction_lock);
849
850 if (copy_from_user(ar->data, buf, size))
851 return ERR_PTR(-EFAULT);
852
853 return ar->data;
854}
855EXPORT_SYMBOL(simple_transaction_get);
856
857ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
858{
859 struct simple_transaction_argresp *ar = file->private_data;
860
861 if (!ar)
862 return 0;
863 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
864}
865EXPORT_SYMBOL(simple_transaction_read);
866
867int simple_transaction_release(struct inode *inode, struct file *file)
868{
869 free_page((unsigned long)file->private_data);
870 return 0;
871}
872EXPORT_SYMBOL(simple_transaction_release);
873
874/* Simple attribute files */
875
876struct simple_attr {
877 int (*get)(void *, u64 *);
878 int (*set)(void *, u64);
879 char get_buf[24]; /* enough to store a u64 and "\n\0" */
880 char set_buf[24];
881 void *data;
882 const char *fmt; /* format for read operation */
883 struct mutex mutex; /* protects access to these buffers */
884};
885
886/* simple_attr_open is called by an actual attribute open file operation
887 * to set the attribute specific access operations. */
888int simple_attr_open(struct inode *inode, struct file *file,
889 int (*get)(void *, u64 *), int (*set)(void *, u64),
890 const char *fmt)
891{
892 struct simple_attr *attr;
893
894 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
895 if (!attr)
896 return -ENOMEM;
897
898 attr->get = get;
899 attr->set = set;
900 attr->data = inode->i_private;
901 attr->fmt = fmt;
902 mutex_init(&attr->mutex);
903
904 file->private_data = attr;
905
906 return nonseekable_open(inode, file);
907}
908EXPORT_SYMBOL_GPL(simple_attr_open);
909
910int simple_attr_release(struct inode *inode, struct file *file)
911{
912 kfree(file->private_data);
913 return 0;
914}
915EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */
916
917/* read from the buffer that is filled with the get function */
918ssize_t simple_attr_read(struct file *file, char __user *buf,
919 size_t len, loff_t *ppos)
920{
921 struct simple_attr *attr;
922 size_t size;
923 ssize_t ret;
924
925 attr = file->private_data;
926
927 if (!attr->get)
928 return -EACCES;
929
930 ret = mutex_lock_interruptible(&attr->mutex);
931 if (ret)
932 return ret;
933
934 if (*ppos && attr->get_buf[0]) {
935 /* continued read */
936 size = strlen(attr->get_buf);
937 } else {
938 /* first read */
939 u64 val;
940 ret = attr->get(attr->data, &val);
941 if (ret)
942 goto out;
943
944 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
945 attr->fmt, (unsigned long long)val);
946 }
947
948 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
949out:
950 mutex_unlock(&attr->mutex);
951 return ret;
952}
953EXPORT_SYMBOL_GPL(simple_attr_read);
954
955/* interpret the buffer as a number to call the set function with */
956ssize_t simple_attr_write(struct file *file, const char __user *buf,
957 size_t len, loff_t *ppos)
958{
959 struct simple_attr *attr;
960 u64 val;
961 size_t size;
962 ssize_t ret;
963
964 attr = file->private_data;
965 if (!attr->set)
966 return -EACCES;
967
968 ret = mutex_lock_interruptible(&attr->mutex);
969 if (ret)
970 return ret;
971
972 ret = -EFAULT;
973 size = min(sizeof(attr->set_buf) - 1, len);
974 if (copy_from_user(attr->set_buf, buf, size))
975 goto out;
976
977 attr->set_buf[size] = '\0';
978 val = simple_strtoll(attr->set_buf, NULL, 0);
979 ret = attr->set(attr->data, val);
980 if (ret == 0)
981 ret = len; /* on success, claim we got the whole input */
982out:
983 mutex_unlock(&attr->mutex);
984 return ret;
985}
986EXPORT_SYMBOL_GPL(simple_attr_write);
987
988/**
989 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
990 * @sb: filesystem to do the file handle conversion on
991 * @fid: file handle to convert
992 * @fh_len: length of the file handle in bytes
993 * @fh_type: type of file handle
994 * @get_inode: filesystem callback to retrieve inode
995 *
996 * This function decodes @fid as long as it has one of the well-known
997 * Linux filehandle types and calls @get_inode on it to retrieve the
998 * inode for the object specified in the file handle.
999 */
1000struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1001 int fh_len, int fh_type, struct inode *(*get_inode)
1002 (struct super_block *sb, u64 ino, u32 gen))
1003{
1004 struct inode *inode = NULL;
1005
1006 if (fh_len < 2)
1007 return NULL;
1008
1009 switch (fh_type) {
1010 case FILEID_INO32_GEN:
1011 case FILEID_INO32_GEN_PARENT:
1012 inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1013 break;
1014 }
1015
1016 return d_obtain_alias(inode);
1017}
1018EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1019
1020/**
1021 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1022 * @sb: filesystem to do the file handle conversion on
1023 * @fid: file handle to convert
1024 * @fh_len: length of the file handle in bytes
1025 * @fh_type: type of file handle
1026 * @get_inode: filesystem callback to retrieve inode
1027 *
1028 * This function decodes @fid as long as it has one of the well-known
1029 * Linux filehandle types and calls @get_inode on it to retrieve the
1030 * inode for the _parent_ object specified in the file handle if it
1031 * is specified in the file handle, or NULL otherwise.
1032 */
1033struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1034 int fh_len, int fh_type, struct inode *(*get_inode)
1035 (struct super_block *sb, u64 ino, u32 gen))
1036{
1037 struct inode *inode = NULL;
1038
1039 if (fh_len <= 2)
1040 return NULL;
1041
1042 switch (fh_type) {
1043 case FILEID_INO32_GEN_PARENT:
1044 inode = get_inode(sb, fid->i32.parent_ino,
1045 (fh_len > 3 ? fid->i32.parent_gen : 0));
1046 break;
1047 }
1048
1049 return d_obtain_alias(inode);
1050}
1051EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1052
1053/**
1054 * __generic_file_fsync - generic fsync implementation for simple filesystems
1055 *
1056 * @file: file to synchronize
1057 * @start: start offset in bytes
1058 * @end: end offset in bytes (inclusive)
1059 * @datasync: only synchronize essential metadata if true
1060 *
1061 * This is a generic implementation of the fsync method for simple
1062 * filesystems which track all non-inode metadata in the buffers list
1063 * hanging off the address_space structure.
1064 */
1065int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1066 int datasync)
1067{
1068 struct inode *inode = file->f_mapping->host;
1069 int err;
1070 int ret;
1071
1072 err = file_write_and_wait_range(file, start, end);
1073 if (err)
1074 return err;
1075
1076 inode_lock(inode);
1077 ret = sync_mapping_buffers(inode->i_mapping);
1078 if (!(inode->i_state & I_DIRTY_ALL))
1079 goto out;
1080 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1081 goto out;
1082
1083 err = sync_inode_metadata(inode, 1);
1084 if (ret == 0)
1085 ret = err;
1086
1087out:
1088 inode_unlock(inode);
1089 /* check and advance again to catch errors after syncing out buffers */
1090 err = file_check_and_advance_wb_err(file);
1091 if (ret == 0)
1092 ret = err;
1093 return ret;
1094}
1095EXPORT_SYMBOL(__generic_file_fsync);
1096
1097/**
1098 * generic_file_fsync - generic fsync implementation for simple filesystems
1099 * with flush
1100 * @file: file to synchronize
1101 * @start: start offset in bytes
1102 * @end: end offset in bytes (inclusive)
1103 * @datasync: only synchronize essential metadata if true
1104 *
1105 */
1106
1107int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1108 int datasync)
1109{
1110 struct inode *inode = file->f_mapping->host;
1111 int err;
1112
1113 err = __generic_file_fsync(file, start, end, datasync);
1114 if (err)
1115 return err;
1116 return blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL);
1117}
1118EXPORT_SYMBOL(generic_file_fsync);
1119
1120/**
1121 * generic_check_addressable - Check addressability of file system
1122 * @blocksize_bits: log of file system block size
1123 * @num_blocks: number of blocks in file system
1124 *
1125 * Determine whether a file system with @num_blocks blocks (and a
1126 * block size of 2**@blocksize_bits) is addressable by the sector_t
1127 * and page cache of the system. Return 0 if so and -EFBIG otherwise.
1128 */
1129int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1130{
1131 u64 last_fs_block = num_blocks - 1;
1132 u64 last_fs_page =
1133 last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1134
1135 if (unlikely(num_blocks == 0))
1136 return 0;
1137
1138 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1139 return -EINVAL;
1140
1141 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1142 (last_fs_page > (pgoff_t)(~0ULL))) {
1143 return -EFBIG;
1144 }
1145 return 0;
1146}
1147EXPORT_SYMBOL(generic_check_addressable);
1148
1149/*
1150 * No-op implementation of ->fsync for in-memory filesystems.
1151 */
1152int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1153{
1154 return 0;
1155}
1156EXPORT_SYMBOL(noop_fsync);
1157
1158int noop_set_page_dirty(struct page *page)
1159{
1160 /*
1161 * Unlike __set_page_dirty_no_writeback that handles dirty page
1162 * tracking in the page object, dax does all dirty tracking in
1163 * the inode address_space in response to mkwrite faults. In the
1164 * dax case we only need to worry about potentially dirty CPU
1165 * caches, not dirty page cache pages to write back.
1166 *
1167 * This callback is defined to prevent fallback to
1168 * __set_page_dirty_buffers() in set_page_dirty().
1169 */
1170 return 0;
1171}
1172EXPORT_SYMBOL_GPL(noop_set_page_dirty);
1173
1174void noop_invalidatepage(struct page *page, unsigned int offset,
1175 unsigned int length)
1176{
1177 /*
1178 * There is no page cache to invalidate in the dax case, however
1179 * we need this callback defined to prevent falling back to
1180 * block_invalidatepage() in do_invalidatepage().
1181 */
1182}
1183EXPORT_SYMBOL_GPL(noop_invalidatepage);
1184
1185ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1186{
1187 /*
1188 * iomap based filesystems support direct I/O without need for
1189 * this callback. However, it still needs to be set in
1190 * inode->a_ops so that open/fcntl know that direct I/O is
1191 * generally supported.
1192 */
1193 return -EINVAL;
1194}
1195EXPORT_SYMBOL_GPL(noop_direct_IO);
1196
1197/* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1198void kfree_link(void *p)
1199{
1200 kfree(p);
1201}
1202EXPORT_SYMBOL(kfree_link);
1203
1204/*
1205 * nop .set_page_dirty method so that people can use .page_mkwrite on
1206 * anon inodes.
1207 */
1208static int anon_set_page_dirty(struct page *page)
1209{
1210 return 0;
1211};
1212
1213/*
1214 * A single inode exists for all anon_inode files. Contrary to pipes,
1215 * anon_inode inodes have no associated per-instance data, so we need
1216 * only allocate one of them.
1217 */
1218struct inode *alloc_anon_inode(struct super_block *s)
1219{
1220 static const struct address_space_operations anon_aops = {
1221 .set_page_dirty = anon_set_page_dirty,
1222 };
1223 struct inode *inode = new_inode_pseudo(s);
1224
1225 if (!inode)
1226 return ERR_PTR(-ENOMEM);
1227
1228 inode->i_ino = get_next_ino();
1229 inode->i_mapping->a_ops = &anon_aops;
1230
1231 /*
1232 * Mark the inode dirty from the very beginning,
1233 * that way it will never be moved to the dirty
1234 * list because mark_inode_dirty() will think
1235 * that it already _is_ on the dirty list.
1236 */
1237 inode->i_state = I_DIRTY;
1238 inode->i_mode = S_IRUSR | S_IWUSR;
1239 inode->i_uid = current_fsuid();
1240 inode->i_gid = current_fsgid();
1241 inode->i_flags |= S_PRIVATE;
1242 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
1243 return inode;
1244}
1245EXPORT_SYMBOL(alloc_anon_inode);
1246
1247/**
1248 * simple_nosetlease - generic helper for prohibiting leases
1249 * @filp: file pointer
1250 * @arg: type of lease to obtain
1251 * @flp: new lease supplied for insertion
1252 * @priv: private data for lm_setup operation
1253 *
1254 * Generic helper for filesystems that do not wish to allow leases to be set.
1255 * All arguments are ignored and it just returns -EINVAL.
1256 */
1257int
1258simple_nosetlease(struct file *filp, long arg, struct file_lock **flp,
1259 void **priv)
1260{
1261 return -EINVAL;
1262}
1263EXPORT_SYMBOL(simple_nosetlease);
1264
1265/**
1266 * simple_get_link - generic helper to get the target of "fast" symlinks
1267 * @dentry: not used here
1268 * @inode: the symlink inode
1269 * @done: not used here
1270 *
1271 * Generic helper for filesystems to use for symlink inodes where a pointer to
1272 * the symlink target is stored in ->i_link. NOTE: this isn't normally called,
1273 * since as an optimization the path lookup code uses any non-NULL ->i_link
1274 * directly, without calling ->get_link(). But ->get_link() still must be set,
1275 * to mark the inode_operations as being for a symlink.
1276 *
1277 * Return: the symlink target
1278 */
1279const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1280 struct delayed_call *done)
1281{
1282 return inode->i_link;
1283}
1284EXPORT_SYMBOL(simple_get_link);
1285
1286const struct inode_operations simple_symlink_inode_operations = {
1287 .get_link = simple_get_link,
1288};
1289EXPORT_SYMBOL(simple_symlink_inode_operations);
1290
1291/*
1292 * Operations for a permanently empty directory.
1293 */
1294static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1295{
1296 return ERR_PTR(-ENOENT);
1297}
1298
1299static int empty_dir_getattr(const struct path *path, struct kstat *stat,
1300 u32 request_mask, unsigned int query_flags)
1301{
1302 struct inode *inode = d_inode(path->dentry);
1303 generic_fillattr(inode, stat);
1304 return 0;
1305}
1306
1307static int empty_dir_setattr(struct dentry *dentry, struct iattr *attr)
1308{
1309 return -EPERM;
1310}
1311
1312static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1313{
1314 return -EOPNOTSUPP;
1315}
1316
1317static const struct inode_operations empty_dir_inode_operations = {
1318 .lookup = empty_dir_lookup,
1319 .permission = generic_permission,
1320 .setattr = empty_dir_setattr,
1321 .getattr = empty_dir_getattr,
1322 .listxattr = empty_dir_listxattr,
1323};
1324
1325static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1326{
1327 /* An empty directory has two entries . and .. at offsets 0 and 1 */
1328 return generic_file_llseek_size(file, offset, whence, 2, 2);
1329}
1330
1331static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1332{
1333 dir_emit_dots(file, ctx);
1334 return 0;
1335}
1336
1337static const struct file_operations empty_dir_operations = {
1338 .llseek = empty_dir_llseek,
1339 .read = generic_read_dir,
1340 .iterate_shared = empty_dir_readdir,
1341 .fsync = noop_fsync,
1342};
1343
1344
1345void make_empty_dir_inode(struct inode *inode)
1346{
1347 set_nlink(inode, 2);
1348 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1349 inode->i_uid = GLOBAL_ROOT_UID;
1350 inode->i_gid = GLOBAL_ROOT_GID;
1351 inode->i_rdev = 0;
1352 inode->i_size = 0;
1353 inode->i_blkbits = PAGE_SHIFT;
1354 inode->i_blocks = 0;
1355
1356 inode->i_op = &empty_dir_inode_operations;
1357 inode->i_opflags &= ~IOP_XATTR;
1358 inode->i_fop = &empty_dir_operations;
1359}
1360
1361bool is_empty_dir_inode(struct inode *inode)
1362{
1363 return (inode->i_fop == &empty_dir_operations) &&
1364 (inode->i_op == &empty_dir_inode_operations);
1365}
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#include <linux/fsnotify.h>
23#include <linux/unicode.h>
24#include <linux/fscrypt.h>
25
26#include <linux/uaccess.h>
27
28#include "internal.h"
29
30int simple_getattr(struct user_namespace *mnt_userns, const struct path *path,
31 struct kstat *stat, u32 request_mask,
32 unsigned int query_flags)
33{
34 struct inode *inode = d_inode(path->dentry);
35 generic_fillattr(&init_user_ns, inode, stat);
36 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
37 return 0;
38}
39EXPORT_SYMBOL(simple_getattr);
40
41int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
42{
43 buf->f_type = dentry->d_sb->s_magic;
44 buf->f_bsize = PAGE_SIZE;
45 buf->f_namelen = NAME_MAX;
46 return 0;
47}
48EXPORT_SYMBOL(simple_statfs);
49
50/*
51 * Retaining negative dentries for an in-memory filesystem just wastes
52 * memory and lookup time: arrange for them to be deleted immediately.
53 */
54int always_delete_dentry(const struct dentry *dentry)
55{
56 return 1;
57}
58EXPORT_SYMBOL(always_delete_dentry);
59
60const struct dentry_operations simple_dentry_operations = {
61 .d_delete = always_delete_dentry,
62};
63EXPORT_SYMBOL(simple_dentry_operations);
64
65/*
66 * Lookup the data. This is trivial - if the dentry didn't already
67 * exist, we know it is negative. Set d_op to delete negative dentries.
68 */
69struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
70{
71 if (dentry->d_name.len > NAME_MAX)
72 return ERR_PTR(-ENAMETOOLONG);
73 if (!dentry->d_sb->s_d_op)
74 d_set_d_op(dentry, &simple_dentry_operations);
75 d_add(dentry, NULL);
76 return NULL;
77}
78EXPORT_SYMBOL(simple_lookup);
79
80int dcache_dir_open(struct inode *inode, struct file *file)
81{
82 file->private_data = d_alloc_cursor(file->f_path.dentry);
83
84 return file->private_data ? 0 : -ENOMEM;
85}
86EXPORT_SYMBOL(dcache_dir_open);
87
88int dcache_dir_close(struct inode *inode, struct file *file)
89{
90 dput(file->private_data);
91 return 0;
92}
93EXPORT_SYMBOL(dcache_dir_close);
94
95/* parent is locked at least shared */
96/*
97 * Returns an element of siblings' list.
98 * We are looking for <count>th positive after <p>; if
99 * found, dentry is grabbed and returned to caller.
100 * If no such element exists, NULL is returned.
101 */
102static struct dentry *scan_positives(struct dentry *cursor,
103 struct list_head *p,
104 loff_t count,
105 struct dentry *last)
106{
107 struct dentry *dentry = cursor->d_parent, *found = NULL;
108
109 spin_lock(&dentry->d_lock);
110 while ((p = p->next) != &dentry->d_subdirs) {
111 struct dentry *d = list_entry(p, struct dentry, d_child);
112 // we must at least skip cursors, to avoid livelocks
113 if (d->d_flags & DCACHE_DENTRY_CURSOR)
114 continue;
115 if (simple_positive(d) && !--count) {
116 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
117 if (simple_positive(d))
118 found = dget_dlock(d);
119 spin_unlock(&d->d_lock);
120 if (likely(found))
121 break;
122 count = 1;
123 }
124 if (need_resched()) {
125 list_move(&cursor->d_child, p);
126 p = &cursor->d_child;
127 spin_unlock(&dentry->d_lock);
128 cond_resched();
129 spin_lock(&dentry->d_lock);
130 }
131 }
132 spin_unlock(&dentry->d_lock);
133 dput(last);
134 return found;
135}
136
137loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
138{
139 struct dentry *dentry = file->f_path.dentry;
140 switch (whence) {
141 case 1:
142 offset += file->f_pos;
143 fallthrough;
144 case 0:
145 if (offset >= 0)
146 break;
147 fallthrough;
148 default:
149 return -EINVAL;
150 }
151 if (offset != file->f_pos) {
152 struct dentry *cursor = file->private_data;
153 struct dentry *to = NULL;
154
155 inode_lock_shared(dentry->d_inode);
156
157 if (offset > 2)
158 to = scan_positives(cursor, &dentry->d_subdirs,
159 offset - 2, NULL);
160 spin_lock(&dentry->d_lock);
161 if (to)
162 list_move(&cursor->d_child, &to->d_child);
163 else
164 list_del_init(&cursor->d_child);
165 spin_unlock(&dentry->d_lock);
166 dput(to);
167
168 file->f_pos = offset;
169
170 inode_unlock_shared(dentry->d_inode);
171 }
172 return offset;
173}
174EXPORT_SYMBOL(dcache_dir_lseek);
175
176/* Relationship between i_mode and the DT_xxx types */
177static inline unsigned char dt_type(struct inode *inode)
178{
179 return (inode->i_mode >> 12) & 15;
180}
181
182/*
183 * Directory is locked and all positive dentries in it are safe, since
184 * for ramfs-type trees they can't go away without unlink() or rmdir(),
185 * both impossible due to the lock on directory.
186 */
187
188int dcache_readdir(struct file *file, struct dir_context *ctx)
189{
190 struct dentry *dentry = file->f_path.dentry;
191 struct dentry *cursor = file->private_data;
192 struct list_head *anchor = &dentry->d_subdirs;
193 struct dentry *next = NULL;
194 struct list_head *p;
195
196 if (!dir_emit_dots(file, ctx))
197 return 0;
198
199 if (ctx->pos == 2)
200 p = anchor;
201 else if (!list_empty(&cursor->d_child))
202 p = &cursor->d_child;
203 else
204 return 0;
205
206 while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
207 if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
208 d_inode(next)->i_ino, dt_type(d_inode(next))))
209 break;
210 ctx->pos++;
211 p = &next->d_child;
212 }
213 spin_lock(&dentry->d_lock);
214 if (next)
215 list_move_tail(&cursor->d_child, &next->d_child);
216 else
217 list_del_init(&cursor->d_child);
218 spin_unlock(&dentry->d_lock);
219 dput(next);
220
221 return 0;
222}
223EXPORT_SYMBOL(dcache_readdir);
224
225ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
226{
227 return -EISDIR;
228}
229EXPORT_SYMBOL(generic_read_dir);
230
231const struct file_operations simple_dir_operations = {
232 .open = dcache_dir_open,
233 .release = dcache_dir_close,
234 .llseek = dcache_dir_lseek,
235 .read = generic_read_dir,
236 .iterate_shared = dcache_readdir,
237 .fsync = noop_fsync,
238};
239EXPORT_SYMBOL(simple_dir_operations);
240
241const struct inode_operations simple_dir_inode_operations = {
242 .lookup = simple_lookup,
243};
244EXPORT_SYMBOL(simple_dir_inode_operations);
245
246static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
247{
248 struct dentry *child = NULL;
249 struct list_head *p = prev ? &prev->d_child : &parent->d_subdirs;
250
251 spin_lock(&parent->d_lock);
252 while ((p = p->next) != &parent->d_subdirs) {
253 struct dentry *d = container_of(p, struct dentry, d_child);
254 if (simple_positive(d)) {
255 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
256 if (simple_positive(d))
257 child = dget_dlock(d);
258 spin_unlock(&d->d_lock);
259 if (likely(child))
260 break;
261 }
262 }
263 spin_unlock(&parent->d_lock);
264 dput(prev);
265 return child;
266}
267
268void simple_recursive_removal(struct dentry *dentry,
269 void (*callback)(struct dentry *))
270{
271 struct dentry *this = dget(dentry);
272 while (true) {
273 struct dentry *victim = NULL, *child;
274 struct inode *inode = this->d_inode;
275
276 inode_lock(inode);
277 if (d_is_dir(this))
278 inode->i_flags |= S_DEAD;
279 while ((child = find_next_child(this, victim)) == NULL) {
280 // kill and ascend
281 // update metadata while it's still locked
282 inode->i_ctime = current_time(inode);
283 clear_nlink(inode);
284 inode_unlock(inode);
285 victim = this;
286 this = this->d_parent;
287 inode = this->d_inode;
288 inode_lock(inode);
289 if (simple_positive(victim)) {
290 d_invalidate(victim); // avoid lost mounts
291 if (d_is_dir(victim))
292 fsnotify_rmdir(inode, victim);
293 else
294 fsnotify_unlink(inode, victim);
295 if (callback)
296 callback(victim);
297 dput(victim); // unpin it
298 }
299 if (victim == dentry) {
300 inode->i_ctime = inode->i_mtime =
301 current_time(inode);
302 if (d_is_dir(dentry))
303 drop_nlink(inode);
304 inode_unlock(inode);
305 dput(dentry);
306 return;
307 }
308 }
309 inode_unlock(inode);
310 this = child;
311 }
312}
313EXPORT_SYMBOL(simple_recursive_removal);
314
315static const struct super_operations simple_super_operations = {
316 .statfs = simple_statfs,
317};
318
319static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
320{
321 struct pseudo_fs_context *ctx = fc->fs_private;
322 struct inode *root;
323
324 s->s_maxbytes = MAX_LFS_FILESIZE;
325 s->s_blocksize = PAGE_SIZE;
326 s->s_blocksize_bits = PAGE_SHIFT;
327 s->s_magic = ctx->magic;
328 s->s_op = ctx->ops ?: &simple_super_operations;
329 s->s_xattr = ctx->xattr;
330 s->s_time_gran = 1;
331 root = new_inode(s);
332 if (!root)
333 return -ENOMEM;
334
335 /*
336 * since this is the first inode, make it number 1. New inodes created
337 * after this must take care not to collide with it (by passing
338 * max_reserved of 1 to iunique).
339 */
340 root->i_ino = 1;
341 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
342 root->i_atime = root->i_mtime = root->i_ctime = current_time(root);
343 s->s_root = d_make_root(root);
344 if (!s->s_root)
345 return -ENOMEM;
346 s->s_d_op = ctx->dops;
347 return 0;
348}
349
350static int pseudo_fs_get_tree(struct fs_context *fc)
351{
352 return get_tree_nodev(fc, pseudo_fs_fill_super);
353}
354
355static void pseudo_fs_free(struct fs_context *fc)
356{
357 kfree(fc->fs_private);
358}
359
360static const struct fs_context_operations pseudo_fs_context_ops = {
361 .free = pseudo_fs_free,
362 .get_tree = pseudo_fs_get_tree,
363};
364
365/*
366 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
367 * will never be mountable)
368 */
369struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
370 unsigned long magic)
371{
372 struct pseudo_fs_context *ctx;
373
374 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
375 if (likely(ctx)) {
376 ctx->magic = magic;
377 fc->fs_private = ctx;
378 fc->ops = &pseudo_fs_context_ops;
379 fc->sb_flags |= SB_NOUSER;
380 fc->global = true;
381 }
382 return ctx;
383}
384EXPORT_SYMBOL(init_pseudo);
385
386int simple_open(struct inode *inode, struct file *file)
387{
388 if (inode->i_private)
389 file->private_data = inode->i_private;
390 return 0;
391}
392EXPORT_SYMBOL(simple_open);
393
394int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
395{
396 struct inode *inode = d_inode(old_dentry);
397
398 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
399 inc_nlink(inode);
400 ihold(inode);
401 dget(dentry);
402 d_instantiate(dentry, inode);
403 return 0;
404}
405EXPORT_SYMBOL(simple_link);
406
407int simple_empty(struct dentry *dentry)
408{
409 struct dentry *child;
410 int ret = 0;
411
412 spin_lock(&dentry->d_lock);
413 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
414 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
415 if (simple_positive(child)) {
416 spin_unlock(&child->d_lock);
417 goto out;
418 }
419 spin_unlock(&child->d_lock);
420 }
421 ret = 1;
422out:
423 spin_unlock(&dentry->d_lock);
424 return ret;
425}
426EXPORT_SYMBOL(simple_empty);
427
428int simple_unlink(struct inode *dir, struct dentry *dentry)
429{
430 struct inode *inode = d_inode(dentry);
431
432 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
433 drop_nlink(inode);
434 dput(dentry);
435 return 0;
436}
437EXPORT_SYMBOL(simple_unlink);
438
439int simple_rmdir(struct inode *dir, struct dentry *dentry)
440{
441 if (!simple_empty(dentry))
442 return -ENOTEMPTY;
443
444 drop_nlink(d_inode(dentry));
445 simple_unlink(dir, dentry);
446 drop_nlink(dir);
447 return 0;
448}
449EXPORT_SYMBOL(simple_rmdir);
450
451int simple_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
452 struct dentry *old_dentry, struct inode *new_dir,
453 struct dentry *new_dentry, unsigned int flags)
454{
455 struct inode *inode = d_inode(old_dentry);
456 int they_are_dirs = d_is_dir(old_dentry);
457
458 if (flags & ~RENAME_NOREPLACE)
459 return -EINVAL;
460
461 if (!simple_empty(new_dentry))
462 return -ENOTEMPTY;
463
464 if (d_really_is_positive(new_dentry)) {
465 simple_unlink(new_dir, new_dentry);
466 if (they_are_dirs) {
467 drop_nlink(d_inode(new_dentry));
468 drop_nlink(old_dir);
469 }
470 } else if (they_are_dirs) {
471 drop_nlink(old_dir);
472 inc_nlink(new_dir);
473 }
474
475 old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
476 new_dir->i_mtime = inode->i_ctime = current_time(old_dir);
477
478 return 0;
479}
480EXPORT_SYMBOL(simple_rename);
481
482/**
483 * simple_setattr - setattr for simple filesystem
484 * @mnt_userns: user namespace of the target mount
485 * @dentry: dentry
486 * @iattr: iattr structure
487 *
488 * Returns 0 on success, -error on failure.
489 *
490 * simple_setattr is a simple ->setattr implementation without a proper
491 * implementation of size changes.
492 *
493 * It can either be used for in-memory filesystems or special files
494 * on simple regular filesystems. Anything that needs to change on-disk
495 * or wire state on size changes needs its own setattr method.
496 */
497int simple_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
498 struct iattr *iattr)
499{
500 struct inode *inode = d_inode(dentry);
501 int error;
502
503 error = setattr_prepare(mnt_userns, dentry, iattr);
504 if (error)
505 return error;
506
507 if (iattr->ia_valid & ATTR_SIZE)
508 truncate_setsize(inode, iattr->ia_size);
509 setattr_copy(mnt_userns, inode, iattr);
510 mark_inode_dirty(inode);
511 return 0;
512}
513EXPORT_SYMBOL(simple_setattr);
514
515static int simple_readpage(struct file *file, struct page *page)
516{
517 clear_highpage(page);
518 flush_dcache_page(page);
519 SetPageUptodate(page);
520 unlock_page(page);
521 return 0;
522}
523
524int simple_write_begin(struct file *file, struct address_space *mapping,
525 loff_t pos, unsigned len, unsigned flags,
526 struct page **pagep, void **fsdata)
527{
528 struct page *page;
529 pgoff_t index;
530
531 index = pos >> PAGE_SHIFT;
532
533 page = grab_cache_page_write_begin(mapping, index, flags);
534 if (!page)
535 return -ENOMEM;
536
537 *pagep = page;
538
539 if (!PageUptodate(page) && (len != PAGE_SIZE)) {
540 unsigned from = pos & (PAGE_SIZE - 1);
541
542 zero_user_segments(page, 0, from, from + len, PAGE_SIZE);
543 }
544 return 0;
545}
546EXPORT_SYMBOL(simple_write_begin);
547
548/**
549 * simple_write_end - .write_end helper for non-block-device FSes
550 * @file: See .write_end of address_space_operations
551 * @mapping: "
552 * @pos: "
553 * @len: "
554 * @copied: "
555 * @page: "
556 * @fsdata: "
557 *
558 * simple_write_end does the minimum needed for updating a page after writing is
559 * done. It has the same API signature as the .write_end of
560 * address_space_operations vector. So it can just be set onto .write_end for
561 * FSes that don't need any other processing. i_mutex is assumed to be held.
562 * Block based filesystems should use generic_write_end().
563 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
564 * is not called, so a filesystem that actually does store data in .write_inode
565 * should extend on what's done here with a call to mark_inode_dirty() in the
566 * case that i_size has changed.
567 *
568 * Use *ONLY* with simple_readpage()
569 */
570static int simple_write_end(struct file *file, struct address_space *mapping,
571 loff_t pos, unsigned len, unsigned copied,
572 struct page *page, void *fsdata)
573{
574 struct inode *inode = page->mapping->host;
575 loff_t last_pos = pos + copied;
576
577 /* zero the stale part of the page if we did a short copy */
578 if (!PageUptodate(page)) {
579 if (copied < len) {
580 unsigned from = pos & (PAGE_SIZE - 1);
581
582 zero_user(page, from + copied, len - copied);
583 }
584 SetPageUptodate(page);
585 }
586 /*
587 * No need to use i_size_read() here, the i_size
588 * cannot change under us because we hold the i_mutex.
589 */
590 if (last_pos > inode->i_size)
591 i_size_write(inode, last_pos);
592
593 set_page_dirty(page);
594 unlock_page(page);
595 put_page(page);
596
597 return copied;
598}
599
600/*
601 * Provides ramfs-style behavior: data in the pagecache, but no writeback.
602 */
603const struct address_space_operations ram_aops = {
604 .readpage = simple_readpage,
605 .write_begin = simple_write_begin,
606 .write_end = simple_write_end,
607 .set_page_dirty = __set_page_dirty_no_writeback,
608};
609EXPORT_SYMBOL(ram_aops);
610
611/*
612 * the inodes created here are not hashed. If you use iunique to generate
613 * unique inode values later for this filesystem, then you must take care
614 * to pass it an appropriate max_reserved value to avoid collisions.
615 */
616int simple_fill_super(struct super_block *s, unsigned long magic,
617 const struct tree_descr *files)
618{
619 struct inode *inode;
620 struct dentry *root;
621 struct dentry *dentry;
622 int i;
623
624 s->s_blocksize = PAGE_SIZE;
625 s->s_blocksize_bits = PAGE_SHIFT;
626 s->s_magic = magic;
627 s->s_op = &simple_super_operations;
628 s->s_time_gran = 1;
629
630 inode = new_inode(s);
631 if (!inode)
632 return -ENOMEM;
633 /*
634 * because the root inode is 1, the files array must not contain an
635 * entry at index 1
636 */
637 inode->i_ino = 1;
638 inode->i_mode = S_IFDIR | 0755;
639 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
640 inode->i_op = &simple_dir_inode_operations;
641 inode->i_fop = &simple_dir_operations;
642 set_nlink(inode, 2);
643 root = d_make_root(inode);
644 if (!root)
645 return -ENOMEM;
646 for (i = 0; !files->name || files->name[0]; i++, files++) {
647 if (!files->name)
648 continue;
649
650 /* warn if it tries to conflict with the root inode */
651 if (unlikely(i == 1))
652 printk(KERN_WARNING "%s: %s passed in a files array"
653 "with an index of 1!\n", __func__,
654 s->s_type->name);
655
656 dentry = d_alloc_name(root, files->name);
657 if (!dentry)
658 goto out;
659 inode = new_inode(s);
660 if (!inode) {
661 dput(dentry);
662 goto out;
663 }
664 inode->i_mode = S_IFREG | files->mode;
665 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
666 inode->i_fop = files->ops;
667 inode->i_ino = i;
668 d_add(dentry, inode);
669 }
670 s->s_root = root;
671 return 0;
672out:
673 d_genocide(root);
674 shrink_dcache_parent(root);
675 dput(root);
676 return -ENOMEM;
677}
678EXPORT_SYMBOL(simple_fill_super);
679
680static DEFINE_SPINLOCK(pin_fs_lock);
681
682int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
683{
684 struct vfsmount *mnt = NULL;
685 spin_lock(&pin_fs_lock);
686 if (unlikely(!*mount)) {
687 spin_unlock(&pin_fs_lock);
688 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
689 if (IS_ERR(mnt))
690 return PTR_ERR(mnt);
691 spin_lock(&pin_fs_lock);
692 if (!*mount)
693 *mount = mnt;
694 }
695 mntget(*mount);
696 ++*count;
697 spin_unlock(&pin_fs_lock);
698 mntput(mnt);
699 return 0;
700}
701EXPORT_SYMBOL(simple_pin_fs);
702
703void simple_release_fs(struct vfsmount **mount, int *count)
704{
705 struct vfsmount *mnt;
706 spin_lock(&pin_fs_lock);
707 mnt = *mount;
708 if (!--*count)
709 *mount = NULL;
710 spin_unlock(&pin_fs_lock);
711 mntput(mnt);
712}
713EXPORT_SYMBOL(simple_release_fs);
714
715/**
716 * simple_read_from_buffer - copy data from the buffer to user space
717 * @to: the user space buffer to read to
718 * @count: the maximum number of bytes to read
719 * @ppos: the current position in the buffer
720 * @from: the buffer to read from
721 * @available: the size of the buffer
722 *
723 * The simple_read_from_buffer() function reads up to @count bytes from the
724 * buffer @from at offset @ppos into the user space address starting at @to.
725 *
726 * On success, the number of bytes read is returned and the offset @ppos is
727 * advanced by this number, or negative value is returned on error.
728 **/
729ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
730 const void *from, size_t available)
731{
732 loff_t pos = *ppos;
733 size_t ret;
734
735 if (pos < 0)
736 return -EINVAL;
737 if (pos >= available || !count)
738 return 0;
739 if (count > available - pos)
740 count = available - pos;
741 ret = copy_to_user(to, from + pos, count);
742 if (ret == count)
743 return -EFAULT;
744 count -= ret;
745 *ppos = pos + count;
746 return count;
747}
748EXPORT_SYMBOL(simple_read_from_buffer);
749
750/**
751 * simple_write_to_buffer - copy data from user space to the buffer
752 * @to: the buffer to write to
753 * @available: the size of the buffer
754 * @ppos: the current position in the buffer
755 * @from: the user space buffer to read from
756 * @count: the maximum number of bytes to read
757 *
758 * The simple_write_to_buffer() function reads up to @count bytes from the user
759 * space address starting at @from into the buffer @to at offset @ppos.
760 *
761 * On success, the number of bytes written is returned and the offset @ppos is
762 * advanced by this number, or negative value is returned on error.
763 **/
764ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
765 const void __user *from, size_t count)
766{
767 loff_t pos = *ppos;
768 size_t res;
769
770 if (pos < 0)
771 return -EINVAL;
772 if (pos >= available || !count)
773 return 0;
774 if (count > available - pos)
775 count = available - pos;
776 res = copy_from_user(to + pos, from, count);
777 if (res == count)
778 return -EFAULT;
779 count -= res;
780 *ppos = pos + count;
781 return count;
782}
783EXPORT_SYMBOL(simple_write_to_buffer);
784
785/**
786 * memory_read_from_buffer - copy data from the buffer
787 * @to: the kernel space buffer to read to
788 * @count: the maximum number of bytes to read
789 * @ppos: the current position in the buffer
790 * @from: the buffer to read from
791 * @available: the size of the buffer
792 *
793 * The memory_read_from_buffer() function reads up to @count bytes from the
794 * buffer @from at offset @ppos into the kernel space address starting at @to.
795 *
796 * On success, the number of bytes read is returned and the offset @ppos is
797 * advanced by this number, or negative value is returned on error.
798 **/
799ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
800 const void *from, size_t available)
801{
802 loff_t pos = *ppos;
803
804 if (pos < 0)
805 return -EINVAL;
806 if (pos >= available)
807 return 0;
808 if (count > available - pos)
809 count = available - pos;
810 memcpy(to, from + pos, count);
811 *ppos = pos + count;
812
813 return count;
814}
815EXPORT_SYMBOL(memory_read_from_buffer);
816
817/*
818 * Transaction based IO.
819 * The file expects a single write which triggers the transaction, and then
820 * possibly a read which collects the result - which is stored in a
821 * file-local buffer.
822 */
823
824void simple_transaction_set(struct file *file, size_t n)
825{
826 struct simple_transaction_argresp *ar = file->private_data;
827
828 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
829
830 /*
831 * The barrier ensures that ar->size will really remain zero until
832 * ar->data is ready for reading.
833 */
834 smp_mb();
835 ar->size = n;
836}
837EXPORT_SYMBOL(simple_transaction_set);
838
839char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
840{
841 struct simple_transaction_argresp *ar;
842 static DEFINE_SPINLOCK(simple_transaction_lock);
843
844 if (size > SIMPLE_TRANSACTION_LIMIT - 1)
845 return ERR_PTR(-EFBIG);
846
847 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
848 if (!ar)
849 return ERR_PTR(-ENOMEM);
850
851 spin_lock(&simple_transaction_lock);
852
853 /* only one write allowed per open */
854 if (file->private_data) {
855 spin_unlock(&simple_transaction_lock);
856 free_page((unsigned long)ar);
857 return ERR_PTR(-EBUSY);
858 }
859
860 file->private_data = ar;
861
862 spin_unlock(&simple_transaction_lock);
863
864 if (copy_from_user(ar->data, buf, size))
865 return ERR_PTR(-EFAULT);
866
867 return ar->data;
868}
869EXPORT_SYMBOL(simple_transaction_get);
870
871ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
872{
873 struct simple_transaction_argresp *ar = file->private_data;
874
875 if (!ar)
876 return 0;
877 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
878}
879EXPORT_SYMBOL(simple_transaction_read);
880
881int simple_transaction_release(struct inode *inode, struct file *file)
882{
883 free_page((unsigned long)file->private_data);
884 return 0;
885}
886EXPORT_SYMBOL(simple_transaction_release);
887
888/* Simple attribute files */
889
890struct simple_attr {
891 int (*get)(void *, u64 *);
892 int (*set)(void *, u64);
893 char get_buf[24]; /* enough to store a u64 and "\n\0" */
894 char set_buf[24];
895 void *data;
896 const char *fmt; /* format for read operation */
897 struct mutex mutex; /* protects access to these buffers */
898};
899
900/* simple_attr_open is called by an actual attribute open file operation
901 * to set the attribute specific access operations. */
902int simple_attr_open(struct inode *inode, struct file *file,
903 int (*get)(void *, u64 *), int (*set)(void *, u64),
904 const char *fmt)
905{
906 struct simple_attr *attr;
907
908 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
909 if (!attr)
910 return -ENOMEM;
911
912 attr->get = get;
913 attr->set = set;
914 attr->data = inode->i_private;
915 attr->fmt = fmt;
916 mutex_init(&attr->mutex);
917
918 file->private_data = attr;
919
920 return nonseekable_open(inode, file);
921}
922EXPORT_SYMBOL_GPL(simple_attr_open);
923
924int simple_attr_release(struct inode *inode, struct file *file)
925{
926 kfree(file->private_data);
927 return 0;
928}
929EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */
930
931/* read from the buffer that is filled with the get function */
932ssize_t simple_attr_read(struct file *file, char __user *buf,
933 size_t len, loff_t *ppos)
934{
935 struct simple_attr *attr;
936 size_t size;
937 ssize_t ret;
938
939 attr = file->private_data;
940
941 if (!attr->get)
942 return -EACCES;
943
944 ret = mutex_lock_interruptible(&attr->mutex);
945 if (ret)
946 return ret;
947
948 if (*ppos && attr->get_buf[0]) {
949 /* continued read */
950 size = strlen(attr->get_buf);
951 } else {
952 /* first read */
953 u64 val;
954 ret = attr->get(attr->data, &val);
955 if (ret)
956 goto out;
957
958 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
959 attr->fmt, (unsigned long long)val);
960 }
961
962 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
963out:
964 mutex_unlock(&attr->mutex);
965 return ret;
966}
967EXPORT_SYMBOL_GPL(simple_attr_read);
968
969/* interpret the buffer as a number to call the set function with */
970ssize_t simple_attr_write(struct file *file, const char __user *buf,
971 size_t len, loff_t *ppos)
972{
973 struct simple_attr *attr;
974 unsigned long long val;
975 size_t size;
976 ssize_t ret;
977
978 attr = file->private_data;
979 if (!attr->set)
980 return -EACCES;
981
982 ret = mutex_lock_interruptible(&attr->mutex);
983 if (ret)
984 return ret;
985
986 ret = -EFAULT;
987 size = min(sizeof(attr->set_buf) - 1, len);
988 if (copy_from_user(attr->set_buf, buf, size))
989 goto out;
990
991 attr->set_buf[size] = '\0';
992 ret = kstrtoull(attr->set_buf, 0, &val);
993 if (ret)
994 goto out;
995 ret = attr->set(attr->data, val);
996 if (ret == 0)
997 ret = len; /* on success, claim we got the whole input */
998out:
999 mutex_unlock(&attr->mutex);
1000 return ret;
1001}
1002EXPORT_SYMBOL_GPL(simple_attr_write);
1003
1004/**
1005 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1006 * @sb: filesystem to do the file handle conversion on
1007 * @fid: file handle to convert
1008 * @fh_len: length of the file handle in bytes
1009 * @fh_type: type of file handle
1010 * @get_inode: filesystem callback to retrieve inode
1011 *
1012 * This function decodes @fid as long as it has one of the well-known
1013 * Linux filehandle types and calls @get_inode on it to retrieve the
1014 * inode for the object specified in the file handle.
1015 */
1016struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1017 int fh_len, int fh_type, struct inode *(*get_inode)
1018 (struct super_block *sb, u64 ino, u32 gen))
1019{
1020 struct inode *inode = NULL;
1021
1022 if (fh_len < 2)
1023 return NULL;
1024
1025 switch (fh_type) {
1026 case FILEID_INO32_GEN:
1027 case FILEID_INO32_GEN_PARENT:
1028 inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1029 break;
1030 }
1031
1032 return d_obtain_alias(inode);
1033}
1034EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1035
1036/**
1037 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1038 * @sb: filesystem to do the file handle conversion on
1039 * @fid: file handle to convert
1040 * @fh_len: length of the file handle in bytes
1041 * @fh_type: type of file handle
1042 * @get_inode: filesystem callback to retrieve inode
1043 *
1044 * This function decodes @fid as long as it has one of the well-known
1045 * Linux filehandle types and calls @get_inode on it to retrieve the
1046 * inode for the _parent_ object specified in the file handle if it
1047 * is specified in the file handle, or NULL otherwise.
1048 */
1049struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1050 int fh_len, int fh_type, struct inode *(*get_inode)
1051 (struct super_block *sb, u64 ino, u32 gen))
1052{
1053 struct inode *inode = NULL;
1054
1055 if (fh_len <= 2)
1056 return NULL;
1057
1058 switch (fh_type) {
1059 case FILEID_INO32_GEN_PARENT:
1060 inode = get_inode(sb, fid->i32.parent_ino,
1061 (fh_len > 3 ? fid->i32.parent_gen : 0));
1062 break;
1063 }
1064
1065 return d_obtain_alias(inode);
1066}
1067EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1068
1069/**
1070 * __generic_file_fsync - generic fsync implementation for simple filesystems
1071 *
1072 * @file: file to synchronize
1073 * @start: start offset in bytes
1074 * @end: end offset in bytes (inclusive)
1075 * @datasync: only synchronize essential metadata if true
1076 *
1077 * This is a generic implementation of the fsync method for simple
1078 * filesystems which track all non-inode metadata in the buffers list
1079 * hanging off the address_space structure.
1080 */
1081int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1082 int datasync)
1083{
1084 struct inode *inode = file->f_mapping->host;
1085 int err;
1086 int ret;
1087
1088 err = file_write_and_wait_range(file, start, end);
1089 if (err)
1090 return err;
1091
1092 inode_lock(inode);
1093 ret = sync_mapping_buffers(inode->i_mapping);
1094 if (!(inode->i_state & I_DIRTY_ALL))
1095 goto out;
1096 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1097 goto out;
1098
1099 err = sync_inode_metadata(inode, 1);
1100 if (ret == 0)
1101 ret = err;
1102
1103out:
1104 inode_unlock(inode);
1105 /* check and advance again to catch errors after syncing out buffers */
1106 err = file_check_and_advance_wb_err(file);
1107 if (ret == 0)
1108 ret = err;
1109 return ret;
1110}
1111EXPORT_SYMBOL(__generic_file_fsync);
1112
1113/**
1114 * generic_file_fsync - generic fsync implementation for simple filesystems
1115 * with flush
1116 * @file: file to synchronize
1117 * @start: start offset in bytes
1118 * @end: end offset in bytes (inclusive)
1119 * @datasync: only synchronize essential metadata if true
1120 *
1121 */
1122
1123int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1124 int datasync)
1125{
1126 struct inode *inode = file->f_mapping->host;
1127 int err;
1128
1129 err = __generic_file_fsync(file, start, end, datasync);
1130 if (err)
1131 return err;
1132 return blkdev_issue_flush(inode->i_sb->s_bdev);
1133}
1134EXPORT_SYMBOL(generic_file_fsync);
1135
1136/**
1137 * generic_check_addressable - Check addressability of file system
1138 * @blocksize_bits: log of file system block size
1139 * @num_blocks: number of blocks in file system
1140 *
1141 * Determine whether a file system with @num_blocks blocks (and a
1142 * block size of 2**@blocksize_bits) is addressable by the sector_t
1143 * and page cache of the system. Return 0 if so and -EFBIG otherwise.
1144 */
1145int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1146{
1147 u64 last_fs_block = num_blocks - 1;
1148 u64 last_fs_page =
1149 last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1150
1151 if (unlikely(num_blocks == 0))
1152 return 0;
1153
1154 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1155 return -EINVAL;
1156
1157 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1158 (last_fs_page > (pgoff_t)(~0ULL))) {
1159 return -EFBIG;
1160 }
1161 return 0;
1162}
1163EXPORT_SYMBOL(generic_check_addressable);
1164
1165/*
1166 * No-op implementation of ->fsync for in-memory filesystems.
1167 */
1168int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1169{
1170 return 0;
1171}
1172EXPORT_SYMBOL(noop_fsync);
1173
1174void noop_invalidatepage(struct page *page, unsigned int offset,
1175 unsigned int length)
1176{
1177 /*
1178 * There is no page cache to invalidate in the dax case, however
1179 * we need this callback defined to prevent falling back to
1180 * block_invalidatepage() in do_invalidatepage().
1181 */
1182}
1183EXPORT_SYMBOL_GPL(noop_invalidatepage);
1184
1185ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1186{
1187 /*
1188 * iomap based filesystems support direct I/O without need for
1189 * this callback. However, it still needs to be set in
1190 * inode->a_ops so that open/fcntl know that direct I/O is
1191 * generally supported.
1192 */
1193 return -EINVAL;
1194}
1195EXPORT_SYMBOL_GPL(noop_direct_IO);
1196
1197/* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1198void kfree_link(void *p)
1199{
1200 kfree(p);
1201}
1202EXPORT_SYMBOL(kfree_link);
1203
1204struct inode *alloc_anon_inode(struct super_block *s)
1205{
1206 static const struct address_space_operations anon_aops = {
1207 .set_page_dirty = __set_page_dirty_no_writeback,
1208 };
1209 struct inode *inode = new_inode_pseudo(s);
1210
1211 if (!inode)
1212 return ERR_PTR(-ENOMEM);
1213
1214 inode->i_ino = get_next_ino();
1215 inode->i_mapping->a_ops = &anon_aops;
1216
1217 /*
1218 * Mark the inode dirty from the very beginning,
1219 * that way it will never be moved to the dirty
1220 * list because mark_inode_dirty() will think
1221 * that it already _is_ on the dirty list.
1222 */
1223 inode->i_state = I_DIRTY;
1224 inode->i_mode = S_IRUSR | S_IWUSR;
1225 inode->i_uid = current_fsuid();
1226 inode->i_gid = current_fsgid();
1227 inode->i_flags |= S_PRIVATE;
1228 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
1229 return inode;
1230}
1231EXPORT_SYMBOL(alloc_anon_inode);
1232
1233/**
1234 * simple_nosetlease - generic helper for prohibiting leases
1235 * @filp: file pointer
1236 * @arg: type of lease to obtain
1237 * @flp: new lease supplied for insertion
1238 * @priv: private data for lm_setup operation
1239 *
1240 * Generic helper for filesystems that do not wish to allow leases to be set.
1241 * All arguments are ignored and it just returns -EINVAL.
1242 */
1243int
1244simple_nosetlease(struct file *filp, long arg, struct file_lock **flp,
1245 void **priv)
1246{
1247 return -EINVAL;
1248}
1249EXPORT_SYMBOL(simple_nosetlease);
1250
1251/**
1252 * simple_get_link - generic helper to get the target of "fast" symlinks
1253 * @dentry: not used here
1254 * @inode: the symlink inode
1255 * @done: not used here
1256 *
1257 * Generic helper for filesystems to use for symlink inodes where a pointer to
1258 * the symlink target is stored in ->i_link. NOTE: this isn't normally called,
1259 * since as an optimization the path lookup code uses any non-NULL ->i_link
1260 * directly, without calling ->get_link(). But ->get_link() still must be set,
1261 * to mark the inode_operations as being for a symlink.
1262 *
1263 * Return: the symlink target
1264 */
1265const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1266 struct delayed_call *done)
1267{
1268 return inode->i_link;
1269}
1270EXPORT_SYMBOL(simple_get_link);
1271
1272const struct inode_operations simple_symlink_inode_operations = {
1273 .get_link = simple_get_link,
1274};
1275EXPORT_SYMBOL(simple_symlink_inode_operations);
1276
1277/*
1278 * Operations for a permanently empty directory.
1279 */
1280static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1281{
1282 return ERR_PTR(-ENOENT);
1283}
1284
1285static int empty_dir_getattr(struct user_namespace *mnt_userns,
1286 const struct path *path, struct kstat *stat,
1287 u32 request_mask, unsigned int query_flags)
1288{
1289 struct inode *inode = d_inode(path->dentry);
1290 generic_fillattr(&init_user_ns, inode, stat);
1291 return 0;
1292}
1293
1294static int empty_dir_setattr(struct user_namespace *mnt_userns,
1295 struct dentry *dentry, struct iattr *attr)
1296{
1297 return -EPERM;
1298}
1299
1300static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1301{
1302 return -EOPNOTSUPP;
1303}
1304
1305static const struct inode_operations empty_dir_inode_operations = {
1306 .lookup = empty_dir_lookup,
1307 .permission = generic_permission,
1308 .setattr = empty_dir_setattr,
1309 .getattr = empty_dir_getattr,
1310 .listxattr = empty_dir_listxattr,
1311};
1312
1313static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1314{
1315 /* An empty directory has two entries . and .. at offsets 0 and 1 */
1316 return generic_file_llseek_size(file, offset, whence, 2, 2);
1317}
1318
1319static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1320{
1321 dir_emit_dots(file, ctx);
1322 return 0;
1323}
1324
1325static const struct file_operations empty_dir_operations = {
1326 .llseek = empty_dir_llseek,
1327 .read = generic_read_dir,
1328 .iterate_shared = empty_dir_readdir,
1329 .fsync = noop_fsync,
1330};
1331
1332
1333void make_empty_dir_inode(struct inode *inode)
1334{
1335 set_nlink(inode, 2);
1336 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1337 inode->i_uid = GLOBAL_ROOT_UID;
1338 inode->i_gid = GLOBAL_ROOT_GID;
1339 inode->i_rdev = 0;
1340 inode->i_size = 0;
1341 inode->i_blkbits = PAGE_SHIFT;
1342 inode->i_blocks = 0;
1343
1344 inode->i_op = &empty_dir_inode_operations;
1345 inode->i_opflags &= ~IOP_XATTR;
1346 inode->i_fop = &empty_dir_operations;
1347}
1348
1349bool is_empty_dir_inode(struct inode *inode)
1350{
1351 return (inode->i_fop == &empty_dir_operations) &&
1352 (inode->i_op == &empty_dir_inode_operations);
1353}
1354
1355#ifdef CONFIG_UNICODE
1356/*
1357 * Determine if the name of a dentry should be casefolded.
1358 *
1359 * Return: if names will need casefolding
1360 */
1361static bool needs_casefold(const struct inode *dir)
1362{
1363 return IS_CASEFOLDED(dir) && dir->i_sb->s_encoding;
1364}
1365
1366/**
1367 * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1368 * @dentry: dentry whose name we are checking against
1369 * @len: len of name of dentry
1370 * @str: str pointer to name of dentry
1371 * @name: Name to compare against
1372 *
1373 * Return: 0 if names match, 1 if mismatch, or -ERRNO
1374 */
1375static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1376 const char *str, const struct qstr *name)
1377{
1378 const struct dentry *parent = READ_ONCE(dentry->d_parent);
1379 const struct inode *dir = READ_ONCE(parent->d_inode);
1380 const struct super_block *sb = dentry->d_sb;
1381 const struct unicode_map *um = sb->s_encoding;
1382 struct qstr qstr = QSTR_INIT(str, len);
1383 char strbuf[DNAME_INLINE_LEN];
1384 int ret;
1385
1386 if (!dir || !needs_casefold(dir))
1387 goto fallback;
1388 /*
1389 * If the dentry name is stored in-line, then it may be concurrently
1390 * modified by a rename. If this happens, the VFS will eventually retry
1391 * the lookup, so it doesn't matter what ->d_compare() returns.
1392 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1393 * string. Therefore, we have to copy the name into a temporary buffer.
1394 */
1395 if (len <= DNAME_INLINE_LEN - 1) {
1396 memcpy(strbuf, str, len);
1397 strbuf[len] = 0;
1398 qstr.name = strbuf;
1399 /* prevent compiler from optimizing out the temporary buffer */
1400 barrier();
1401 }
1402 ret = utf8_strncasecmp(um, name, &qstr);
1403 if (ret >= 0)
1404 return ret;
1405
1406 if (sb_has_strict_encoding(sb))
1407 return -EINVAL;
1408fallback:
1409 if (len != name->len)
1410 return 1;
1411 return !!memcmp(str, name->name, len);
1412}
1413
1414/**
1415 * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1416 * @dentry: dentry of the parent directory
1417 * @str: qstr of name whose hash we should fill in
1418 *
1419 * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1420 */
1421static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1422{
1423 const struct inode *dir = READ_ONCE(dentry->d_inode);
1424 struct super_block *sb = dentry->d_sb;
1425 const struct unicode_map *um = sb->s_encoding;
1426 int ret = 0;
1427
1428 if (!dir || !needs_casefold(dir))
1429 return 0;
1430
1431 ret = utf8_casefold_hash(um, dentry, str);
1432 if (ret < 0 && sb_has_strict_encoding(sb))
1433 return -EINVAL;
1434 return 0;
1435}
1436
1437static const struct dentry_operations generic_ci_dentry_ops = {
1438 .d_hash = generic_ci_d_hash,
1439 .d_compare = generic_ci_d_compare,
1440};
1441#endif
1442
1443#ifdef CONFIG_FS_ENCRYPTION
1444static const struct dentry_operations generic_encrypted_dentry_ops = {
1445 .d_revalidate = fscrypt_d_revalidate,
1446};
1447#endif
1448
1449#if defined(CONFIG_FS_ENCRYPTION) && defined(CONFIG_UNICODE)
1450static const struct dentry_operations generic_encrypted_ci_dentry_ops = {
1451 .d_hash = generic_ci_d_hash,
1452 .d_compare = generic_ci_d_compare,
1453 .d_revalidate = fscrypt_d_revalidate,
1454};
1455#endif
1456
1457/**
1458 * generic_set_encrypted_ci_d_ops - helper for setting d_ops for given dentry
1459 * @dentry: dentry to set ops on
1460 *
1461 * Casefolded directories need d_hash and d_compare set, so that the dentries
1462 * contained in them are handled case-insensitively. Note that these operations
1463 * are needed on the parent directory rather than on the dentries in it, and
1464 * while the casefolding flag can be toggled on and off on an empty directory,
1465 * dentry_operations can't be changed later. As a result, if the filesystem has
1466 * casefolding support enabled at all, we have to give all dentries the
1467 * casefolding operations even if their inode doesn't have the casefolding flag
1468 * currently (and thus the casefolding ops would be no-ops for now).
1469 *
1470 * Encryption works differently in that the only dentry operation it needs is
1471 * d_revalidate, which it only needs on dentries that have the no-key name flag.
1472 * The no-key flag can't be set "later", so we don't have to worry about that.
1473 *
1474 * Finally, to maximize compatibility with overlayfs (which isn't compatible
1475 * with certain dentry operations) and to avoid taking an unnecessary
1476 * performance hit, we use custom dentry_operations for each possible
1477 * combination rather than always installing all operations.
1478 */
1479void generic_set_encrypted_ci_d_ops(struct dentry *dentry)
1480{
1481#ifdef CONFIG_FS_ENCRYPTION
1482 bool needs_encrypt_ops = dentry->d_flags & DCACHE_NOKEY_NAME;
1483#endif
1484#ifdef CONFIG_UNICODE
1485 bool needs_ci_ops = dentry->d_sb->s_encoding;
1486#endif
1487#if defined(CONFIG_FS_ENCRYPTION) && defined(CONFIG_UNICODE)
1488 if (needs_encrypt_ops && needs_ci_ops) {
1489 d_set_d_op(dentry, &generic_encrypted_ci_dentry_ops);
1490 return;
1491 }
1492#endif
1493#ifdef CONFIG_FS_ENCRYPTION
1494 if (needs_encrypt_ops) {
1495 d_set_d_op(dentry, &generic_encrypted_dentry_ops);
1496 return;
1497 }
1498#endif
1499#ifdef CONFIG_UNICODE
1500 if (needs_ci_ops) {
1501 d_set_d_op(dentry, &generic_ci_dentry_ops);
1502 return;
1503 }
1504#endif
1505}
1506EXPORT_SYMBOL(generic_set_encrypted_ci_d_ops);