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
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}