1 // SPDX-License-Identifier: GPL-2.0-only
4 * Library for filesystems writers.
7 #include <linux/blkdev.h>
8 #include <linux/export.h>
9 #include <linux/pagemap.h>
10 #include <linux/slab.h>
11 #include <linux/cred.h>
12 #include <linux/mount.h>
13 #include <linux/vfs.h>
14 #include <linux/quotaops.h>
15 #include <linux/mutex.h>
16 #include <linux/namei.h>
17 #include <linux/exportfs.h>
18 #include <linux/iversion.h>
19 #include <linux/writeback.h>
20 #include <linux/buffer_head.h> /* sync_mapping_buffers */
21 #include <linux/fs_context.h>
22 #include <linux/pseudo_fs.h>
23 #include <linux/fsnotify.h>
24 #include <linux/unicode.h>
25 #include <linux/fscrypt.h>
27 #include <linux/uaccess.h>
31 int simple_getattr(struct mnt_idmap *idmap, const struct path *path,
32 struct kstat *stat, u32 request_mask,
33 unsigned int query_flags)
35 struct inode *inode = d_inode(path->dentry);
36 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
37 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
40 EXPORT_SYMBOL(simple_getattr);
42 int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
44 buf->f_type = dentry->d_sb->s_magic;
45 buf->f_bsize = PAGE_SIZE;
46 buf->f_namelen = NAME_MAX;
49 EXPORT_SYMBOL(simple_statfs);
52 * Retaining negative dentries for an in-memory filesystem just wastes
53 * memory and lookup time: arrange for them to be deleted immediately.
55 int always_delete_dentry(const struct dentry *dentry)
59 EXPORT_SYMBOL(always_delete_dentry);
61 const struct dentry_operations simple_dentry_operations = {
62 .d_delete = always_delete_dentry,
64 EXPORT_SYMBOL(simple_dentry_operations);
67 * Lookup the data. This is trivial - if the dentry didn't already
68 * exist, we know it is negative. Set d_op to delete negative dentries.
70 struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
72 if (dentry->d_name.len > NAME_MAX)
73 return ERR_PTR(-ENAMETOOLONG);
74 if (!dentry->d_sb->s_d_op)
75 d_set_d_op(dentry, &simple_dentry_operations);
79 EXPORT_SYMBOL(simple_lookup);
81 int dcache_dir_open(struct inode *inode, struct file *file)
83 file->private_data = d_alloc_cursor(file->f_path.dentry);
85 return file->private_data ? 0 : -ENOMEM;
87 EXPORT_SYMBOL(dcache_dir_open);
89 int dcache_dir_close(struct inode *inode, struct file *file)
91 dput(file->private_data);
94 EXPORT_SYMBOL(dcache_dir_close);
96 /* parent is locked at least shared */
98 * Returns an element of siblings' list.
99 * We are looking for <count>th positive after <p>; if
100 * found, dentry is grabbed and returned to caller.
101 * If no such element exists, NULL is returned.
103 static struct dentry *scan_positives(struct dentry *cursor,
108 struct dentry *dentry = cursor->d_parent, *found = NULL;
110 spin_lock(&dentry->d_lock);
111 while ((p = p->next) != &dentry->d_subdirs) {
112 struct dentry *d = list_entry(p, struct dentry, d_child);
113 // we must at least skip cursors, to avoid livelocks
114 if (d->d_flags & DCACHE_DENTRY_CURSOR)
116 if (simple_positive(d) && !--count) {
117 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
118 if (simple_positive(d))
119 found = dget_dlock(d);
120 spin_unlock(&d->d_lock);
125 if (need_resched()) {
126 list_move(&cursor->d_child, p);
127 p = &cursor->d_child;
128 spin_unlock(&dentry->d_lock);
130 spin_lock(&dentry->d_lock);
133 spin_unlock(&dentry->d_lock);
138 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
140 struct dentry *dentry = file->f_path.dentry;
143 offset += file->f_pos;
152 if (offset != file->f_pos) {
153 struct dentry *cursor = file->private_data;
154 struct dentry *to = NULL;
156 inode_lock_shared(dentry->d_inode);
159 to = scan_positives(cursor, &dentry->d_subdirs,
161 spin_lock(&dentry->d_lock);
163 list_move(&cursor->d_child, &to->d_child);
165 list_del_init(&cursor->d_child);
166 spin_unlock(&dentry->d_lock);
169 file->f_pos = offset;
171 inode_unlock_shared(dentry->d_inode);
175 EXPORT_SYMBOL(dcache_dir_lseek);
178 * Directory is locked and all positive dentries in it are safe, since
179 * for ramfs-type trees they can't go away without unlink() or rmdir(),
180 * both impossible due to the lock on directory.
183 int dcache_readdir(struct file *file, struct dir_context *ctx)
185 struct dentry *dentry = file->f_path.dentry;
186 struct dentry *cursor = file->private_data;
187 struct list_head *anchor = &dentry->d_subdirs;
188 struct dentry *next = NULL;
191 if (!dir_emit_dots(file, ctx))
196 else if (!list_empty(&cursor->d_child))
197 p = &cursor->d_child;
201 while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
202 if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
203 d_inode(next)->i_ino,
204 fs_umode_to_dtype(d_inode(next)->i_mode)))
209 spin_lock(&dentry->d_lock);
211 list_move_tail(&cursor->d_child, &next->d_child);
213 list_del_init(&cursor->d_child);
214 spin_unlock(&dentry->d_lock);
219 EXPORT_SYMBOL(dcache_readdir);
221 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
225 EXPORT_SYMBOL(generic_read_dir);
227 const 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,
235 EXPORT_SYMBOL(simple_dir_operations);
237 const struct inode_operations simple_dir_inode_operations = {
238 .lookup = simple_lookup,
240 EXPORT_SYMBOL(simple_dir_inode_operations);
242 static void offset_set(struct dentry *dentry, u32 offset)
244 dentry->d_fsdata = (void *)((uintptr_t)(offset));
247 static u32 dentry2offset(struct dentry *dentry)
249 return (u32)((uintptr_t)(dentry->d_fsdata));
252 static struct lock_class_key simple_offset_xa_lock;
255 * simple_offset_init - initialize an offset_ctx
256 * @octx: directory offset map to be initialized
259 void simple_offset_init(struct offset_ctx *octx)
261 xa_init_flags(&octx->xa, XA_FLAGS_ALLOC1);
262 lockdep_set_class(&octx->xa.xa_lock, &simple_offset_xa_lock);
264 /* 0 is '.', 1 is '..', so always start with offset 2 */
265 octx->next_offset = 2;
269 * simple_offset_add - Add an entry to a directory's offset map
270 * @octx: directory offset ctx to be updated
271 * @dentry: new dentry being added
273 * Returns zero on success. @so_ctx and the dentry offset are updated.
274 * Otherwise, a negative errno value is returned.
276 int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
278 static const struct xa_limit limit = XA_LIMIT(2, U32_MAX);
282 if (dentry2offset(dentry) != 0)
285 ret = xa_alloc_cyclic(&octx->xa, &offset, dentry, limit,
286 &octx->next_offset, GFP_KERNEL);
290 offset_set(dentry, offset);
295 * simple_offset_remove - Remove an entry to a directory's offset map
296 * @octx: directory offset ctx to be updated
297 * @dentry: dentry being removed
300 void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
304 offset = dentry2offset(dentry);
308 xa_erase(&octx->xa, offset);
309 offset_set(dentry, 0);
313 * simple_offset_rename_exchange - exchange rename with directory offsets
314 * @old_dir: parent of dentry being moved
315 * @old_dentry: dentry being moved
316 * @new_dir: destination parent
317 * @new_dentry: destination dentry
319 * Returns zero on success. Otherwise a negative errno is returned and the
320 * rename is rolled back.
322 int simple_offset_rename_exchange(struct inode *old_dir,
323 struct dentry *old_dentry,
324 struct inode *new_dir,
325 struct dentry *new_dentry)
327 struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
328 struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
329 u32 old_index = dentry2offset(old_dentry);
330 u32 new_index = dentry2offset(new_dentry);
333 simple_offset_remove(old_ctx, old_dentry);
334 simple_offset_remove(new_ctx, new_dentry);
336 ret = simple_offset_add(new_ctx, old_dentry);
340 ret = simple_offset_add(old_ctx, new_dentry);
342 simple_offset_remove(new_ctx, old_dentry);
346 ret = simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
348 simple_offset_remove(new_ctx, old_dentry);
349 simple_offset_remove(old_ctx, new_dentry);
355 offset_set(old_dentry, old_index);
356 xa_store(&old_ctx->xa, old_index, old_dentry, GFP_KERNEL);
357 offset_set(new_dentry, new_index);
358 xa_store(&new_ctx->xa, new_index, new_dentry, GFP_KERNEL);
363 * simple_offset_destroy - Release offset map
364 * @octx: directory offset ctx that is about to be destroyed
366 * During fs teardown (eg. umount), a directory's offset map might still
367 * contain entries. xa_destroy() cleans out anything that remains.
369 void simple_offset_destroy(struct offset_ctx *octx)
371 xa_destroy(&octx->xa);
375 * offset_dir_llseek - Advance the read position of a directory descriptor
376 * @file: an open directory whose position is to be updated
377 * @offset: a byte offset
378 * @whence: enumerator describing the starting position for this update
380 * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories.
382 * Returns the updated read position if successful; otherwise a
383 * negative errno is returned and the read position remains unchanged.
385 static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
389 offset += file->f_pos;
399 return vfs_setpos(file, offset, U32_MAX);
402 static struct dentry *offset_find_next(struct xa_state *xas)
404 struct dentry *child, *found = NULL;
407 child = xas_next_entry(xas, U32_MAX);
410 spin_lock(&child->d_lock);
411 if (simple_positive(child))
412 found = dget_dlock(child);
413 spin_unlock(&child->d_lock);
419 static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
421 u32 offset = dentry2offset(dentry);
422 struct inode *inode = d_inode(dentry);
424 return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset,
425 inode->i_ino, fs_umode_to_dtype(inode->i_mode));
428 static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
430 struct offset_ctx *so_ctx = inode->i_op->get_offset_ctx(inode);
431 XA_STATE(xas, &so_ctx->xa, ctx->pos);
432 struct dentry *dentry;
435 dentry = offset_find_next(&xas);
439 if (!offset_dir_emit(ctx, dentry)) {
445 ctx->pos = xas.xa_index + 1;
450 * offset_readdir - Emit entries starting at offset @ctx->pos
451 * @file: an open directory to iterate over
452 * @ctx: directory iteration context
454 * Caller must hold @file's i_rwsem to prevent insertion or removal of
455 * entries during this call.
457 * On entry, @ctx->pos contains an offset that represents the first entry
458 * to be read from the directory.
460 * The operation continues until there are no more entries to read, or
461 * until the ctx->actor indicates there is no more space in the caller's
464 * On return, @ctx->pos contains an offset that will read the next entry
465 * in this directory when offset_readdir() is called again with @ctx.
470 static int offset_readdir(struct file *file, struct dir_context *ctx)
472 struct dentry *dir = file->f_path.dentry;
474 lockdep_assert_held(&d_inode(dir)->i_rwsem);
476 if (!dir_emit_dots(file, ctx))
479 offset_iterate_dir(d_inode(dir), ctx);
483 const struct file_operations simple_offset_dir_operations = {
484 .llseek = offset_dir_llseek,
485 .iterate_shared = offset_readdir,
486 .read = generic_read_dir,
490 static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
492 struct dentry *child = NULL;
493 struct list_head *p = prev ? &prev->d_child : &parent->d_subdirs;
495 spin_lock(&parent->d_lock);
496 while ((p = p->next) != &parent->d_subdirs) {
497 struct dentry *d = container_of(p, struct dentry, d_child);
498 if (simple_positive(d)) {
499 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
500 if (simple_positive(d))
501 child = dget_dlock(d);
502 spin_unlock(&d->d_lock);
507 spin_unlock(&parent->d_lock);
512 void simple_recursive_removal(struct dentry *dentry,
513 void (*callback)(struct dentry *))
515 struct dentry *this = dget(dentry);
517 struct dentry *victim = NULL, *child;
518 struct inode *inode = this->d_inode;
522 inode->i_flags |= S_DEAD;
523 while ((child = find_next_child(this, victim)) == NULL) {
525 // update metadata while it's still locked
526 inode_set_ctime_current(inode);
530 this = this->d_parent;
531 inode = this->d_inode;
533 if (simple_positive(victim)) {
534 d_invalidate(victim); // avoid lost mounts
535 if (d_is_dir(victim))
536 fsnotify_rmdir(inode, victim);
538 fsnotify_unlink(inode, victim);
541 dput(victim); // unpin it
543 if (victim == dentry) {
544 inode_set_mtime_to_ts(inode,
545 inode_set_ctime_current(inode));
546 if (d_is_dir(dentry))
557 EXPORT_SYMBOL(simple_recursive_removal);
559 static const struct super_operations simple_super_operations = {
560 .statfs = simple_statfs,
563 static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
565 struct pseudo_fs_context *ctx = fc->fs_private;
568 s->s_maxbytes = MAX_LFS_FILESIZE;
569 s->s_blocksize = PAGE_SIZE;
570 s->s_blocksize_bits = PAGE_SHIFT;
571 s->s_magic = ctx->magic;
572 s->s_op = ctx->ops ?: &simple_super_operations;
573 s->s_xattr = ctx->xattr;
580 * since this is the first inode, make it number 1. New inodes created
581 * after this must take care not to collide with it (by passing
582 * max_reserved of 1 to iunique).
585 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
586 simple_inode_init_ts(root);
587 s->s_root = d_make_root(root);
590 s->s_d_op = ctx->dops;
594 static int pseudo_fs_get_tree(struct fs_context *fc)
596 return get_tree_nodev(fc, pseudo_fs_fill_super);
599 static void pseudo_fs_free(struct fs_context *fc)
601 kfree(fc->fs_private);
604 static const struct fs_context_operations pseudo_fs_context_ops = {
605 .free = pseudo_fs_free,
606 .get_tree = pseudo_fs_get_tree,
610 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
611 * will never be mountable)
613 struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
616 struct pseudo_fs_context *ctx;
618 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
621 fc->fs_private = ctx;
622 fc->ops = &pseudo_fs_context_ops;
623 fc->sb_flags |= SB_NOUSER;
628 EXPORT_SYMBOL(init_pseudo);
630 int simple_open(struct inode *inode, struct file *file)
632 if (inode->i_private)
633 file->private_data = inode->i_private;
636 EXPORT_SYMBOL(simple_open);
638 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
640 struct inode *inode = d_inode(old_dentry);
642 inode_set_mtime_to_ts(dir,
643 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
647 d_instantiate(dentry, inode);
650 EXPORT_SYMBOL(simple_link);
652 int simple_empty(struct dentry *dentry)
654 struct dentry *child;
657 spin_lock(&dentry->d_lock);
658 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
659 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
660 if (simple_positive(child)) {
661 spin_unlock(&child->d_lock);
664 spin_unlock(&child->d_lock);
668 spin_unlock(&dentry->d_lock);
671 EXPORT_SYMBOL(simple_empty);
673 int simple_unlink(struct inode *dir, struct dentry *dentry)
675 struct inode *inode = d_inode(dentry);
677 inode_set_mtime_to_ts(dir,
678 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
683 EXPORT_SYMBOL(simple_unlink);
685 int simple_rmdir(struct inode *dir, struct dentry *dentry)
687 if (!simple_empty(dentry))
690 drop_nlink(d_inode(dentry));
691 simple_unlink(dir, dentry);
695 EXPORT_SYMBOL(simple_rmdir);
698 * simple_rename_timestamp - update the various inode timestamps for rename
699 * @old_dir: old parent directory
700 * @old_dentry: dentry that is being renamed
701 * @new_dir: new parent directory
702 * @new_dentry: target for rename
704 * POSIX mandates that the old and new parent directories have their ctime and
705 * mtime updated, and that inodes of @old_dentry and @new_dentry (if any), have
706 * their ctime updated.
708 void simple_rename_timestamp(struct inode *old_dir, struct dentry *old_dentry,
709 struct inode *new_dir, struct dentry *new_dentry)
711 struct inode *newino = d_inode(new_dentry);
713 inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
714 if (new_dir != old_dir)
715 inode_set_mtime_to_ts(new_dir,
716 inode_set_ctime_current(new_dir));
717 inode_set_ctime_current(d_inode(old_dentry));
719 inode_set_ctime_current(newino);
721 EXPORT_SYMBOL_GPL(simple_rename_timestamp);
723 int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
724 struct inode *new_dir, struct dentry *new_dentry)
726 bool old_is_dir = d_is_dir(old_dentry);
727 bool new_is_dir = d_is_dir(new_dentry);
729 if (old_dir != new_dir && old_is_dir != new_is_dir) {
738 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
741 EXPORT_SYMBOL_GPL(simple_rename_exchange);
743 int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir,
744 struct dentry *old_dentry, struct inode *new_dir,
745 struct dentry *new_dentry, unsigned int flags)
747 int they_are_dirs = d_is_dir(old_dentry);
749 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
752 if (flags & RENAME_EXCHANGE)
753 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
755 if (!simple_empty(new_dentry))
758 if (d_really_is_positive(new_dentry)) {
759 simple_unlink(new_dir, new_dentry);
761 drop_nlink(d_inode(new_dentry));
764 } else if (they_are_dirs) {
769 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
772 EXPORT_SYMBOL(simple_rename);
775 * simple_setattr - setattr for simple filesystem
776 * @idmap: idmap of the target mount
778 * @iattr: iattr structure
780 * Returns 0 on success, -error on failure.
782 * simple_setattr is a simple ->setattr implementation without a proper
783 * implementation of size changes.
785 * It can either be used for in-memory filesystems or special files
786 * on simple regular filesystems. Anything that needs to change on-disk
787 * or wire state on size changes needs its own setattr method.
789 int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
792 struct inode *inode = d_inode(dentry);
795 error = setattr_prepare(idmap, dentry, iattr);
799 if (iattr->ia_valid & ATTR_SIZE)
800 truncate_setsize(inode, iattr->ia_size);
801 setattr_copy(idmap, inode, iattr);
802 mark_inode_dirty(inode);
805 EXPORT_SYMBOL(simple_setattr);
807 static int simple_read_folio(struct file *file, struct folio *folio)
809 folio_zero_range(folio, 0, folio_size(folio));
810 flush_dcache_folio(folio);
811 folio_mark_uptodate(folio);
816 int simple_write_begin(struct file *file, struct address_space *mapping,
817 loff_t pos, unsigned len,
818 struct page **pagep, void **fsdata)
822 folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
823 mapping_gfp_mask(mapping));
825 return PTR_ERR(folio);
827 *pagep = &folio->page;
829 if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
830 size_t from = offset_in_folio(folio, pos);
832 folio_zero_segments(folio, 0, from,
833 from + len, folio_size(folio));
837 EXPORT_SYMBOL(simple_write_begin);
840 * simple_write_end - .write_end helper for non-block-device FSes
841 * @file: See .write_end of address_space_operations
849 * simple_write_end does the minimum needed for updating a page after writing is
850 * done. It has the same API signature as the .write_end of
851 * address_space_operations vector. So it can just be set onto .write_end for
852 * FSes that don't need any other processing. i_mutex is assumed to be held.
853 * Block based filesystems should use generic_write_end().
854 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
855 * is not called, so a filesystem that actually does store data in .write_inode
856 * should extend on what's done here with a call to mark_inode_dirty() in the
857 * case that i_size has changed.
859 * Use *ONLY* with simple_read_folio()
861 static int simple_write_end(struct file *file, struct address_space *mapping,
862 loff_t pos, unsigned len, unsigned copied,
863 struct page *page, void *fsdata)
865 struct folio *folio = page_folio(page);
866 struct inode *inode = folio->mapping->host;
867 loff_t last_pos = pos + copied;
869 /* zero the stale part of the folio if we did a short copy */
870 if (!folio_test_uptodate(folio)) {
872 size_t from = offset_in_folio(folio, pos);
874 folio_zero_range(folio, from + copied, len - copied);
876 folio_mark_uptodate(folio);
879 * No need to use i_size_read() here, the i_size
880 * cannot change under us because we hold the i_mutex.
882 if (last_pos > inode->i_size)
883 i_size_write(inode, last_pos);
885 folio_mark_dirty(folio);
893 * Provides ramfs-style behavior: data in the pagecache, but no writeback.
895 const struct address_space_operations ram_aops = {
896 .read_folio = simple_read_folio,
897 .write_begin = simple_write_begin,
898 .write_end = simple_write_end,
899 .dirty_folio = noop_dirty_folio,
901 EXPORT_SYMBOL(ram_aops);
904 * the inodes created here are not hashed. If you use iunique to generate
905 * unique inode values later for this filesystem, then you must take care
906 * to pass it an appropriate max_reserved value to avoid collisions.
908 int simple_fill_super(struct super_block *s, unsigned long magic,
909 const struct tree_descr *files)
913 struct dentry *dentry;
916 s->s_blocksize = PAGE_SIZE;
917 s->s_blocksize_bits = PAGE_SHIFT;
919 s->s_op = &simple_super_operations;
922 inode = new_inode(s);
926 * because the root inode is 1, the files array must not contain an
930 inode->i_mode = S_IFDIR | 0755;
931 simple_inode_init_ts(inode);
932 inode->i_op = &simple_dir_inode_operations;
933 inode->i_fop = &simple_dir_operations;
935 root = d_make_root(inode);
938 for (i = 0; !files->name || files->name[0]; i++, files++) {
942 /* warn if it tries to conflict with the root inode */
943 if (unlikely(i == 1))
944 printk(KERN_WARNING "%s: %s passed in a files array"
945 "with an index of 1!\n", __func__,
948 dentry = d_alloc_name(root, files->name);
951 inode = new_inode(s);
956 inode->i_mode = S_IFREG | files->mode;
957 simple_inode_init_ts(inode);
958 inode->i_fop = files->ops;
960 d_add(dentry, inode);
966 shrink_dcache_parent(root);
970 EXPORT_SYMBOL(simple_fill_super);
972 static DEFINE_SPINLOCK(pin_fs_lock);
974 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
976 struct vfsmount *mnt = NULL;
977 spin_lock(&pin_fs_lock);
978 if (unlikely(!*mount)) {
979 spin_unlock(&pin_fs_lock);
980 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
983 spin_lock(&pin_fs_lock);
989 spin_unlock(&pin_fs_lock);
993 EXPORT_SYMBOL(simple_pin_fs);
995 void simple_release_fs(struct vfsmount **mount, int *count)
997 struct vfsmount *mnt;
998 spin_lock(&pin_fs_lock);
1002 spin_unlock(&pin_fs_lock);
1005 EXPORT_SYMBOL(simple_release_fs);
1008 * simple_read_from_buffer - copy data from the buffer to user space
1009 * @to: the user space buffer to read to
1010 * @count: the maximum number of bytes to read
1011 * @ppos: the current position in the buffer
1012 * @from: the buffer to read from
1013 * @available: the size of the buffer
1015 * The simple_read_from_buffer() function reads up to @count bytes from the
1016 * buffer @from at offset @ppos into the user space address starting at @to.
1018 * On success, the number of bytes read is returned and the offset @ppos is
1019 * advanced by this number, or negative value is returned on error.
1021 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
1022 const void *from, size_t available)
1029 if (pos >= available || !count)
1031 if (count > available - pos)
1032 count = available - pos;
1033 ret = copy_to_user(to, from + pos, count);
1037 *ppos = pos + count;
1040 EXPORT_SYMBOL(simple_read_from_buffer);
1043 * simple_write_to_buffer - copy data from user space to the buffer
1044 * @to: the buffer to write to
1045 * @available: the size of the buffer
1046 * @ppos: the current position in the buffer
1047 * @from: the user space buffer to read from
1048 * @count: the maximum number of bytes to read
1050 * The simple_write_to_buffer() function reads up to @count bytes from the user
1051 * space address starting at @from into the buffer @to at offset @ppos.
1053 * On success, the number of bytes written is returned and the offset @ppos is
1054 * advanced by this number, or negative value is returned on error.
1056 ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
1057 const void __user *from, size_t count)
1064 if (pos >= available || !count)
1066 if (count > available - pos)
1067 count = available - pos;
1068 res = copy_from_user(to + pos, from, count);
1072 *ppos = pos + count;
1075 EXPORT_SYMBOL(simple_write_to_buffer);
1078 * memory_read_from_buffer - copy data from the buffer
1079 * @to: the kernel space buffer to read to
1080 * @count: the maximum number of bytes to read
1081 * @ppos: the current position in the buffer
1082 * @from: the buffer to read from
1083 * @available: the size of the buffer
1085 * The memory_read_from_buffer() function reads up to @count bytes from the
1086 * buffer @from at offset @ppos into the kernel space address starting at @to.
1088 * On success, the number of bytes read is returned and the offset @ppos is
1089 * advanced by this number, or negative value is returned on error.
1091 ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
1092 const void *from, size_t available)
1098 if (pos >= available)
1100 if (count > available - pos)
1101 count = available - pos;
1102 memcpy(to, from + pos, count);
1103 *ppos = pos + count;
1107 EXPORT_SYMBOL(memory_read_from_buffer);
1110 * Transaction based IO.
1111 * The file expects a single write which triggers the transaction, and then
1112 * possibly a read which collects the result - which is stored in a
1113 * file-local buffer.
1116 void simple_transaction_set(struct file *file, size_t n)
1118 struct simple_transaction_argresp *ar = file->private_data;
1120 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1123 * The barrier ensures that ar->size will really remain zero until
1124 * ar->data is ready for reading.
1129 EXPORT_SYMBOL(simple_transaction_set);
1131 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
1133 struct simple_transaction_argresp *ar;
1134 static DEFINE_SPINLOCK(simple_transaction_lock);
1136 if (size > SIMPLE_TRANSACTION_LIMIT - 1)
1137 return ERR_PTR(-EFBIG);
1139 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
1141 return ERR_PTR(-ENOMEM);
1143 spin_lock(&simple_transaction_lock);
1145 /* only one write allowed per open */
1146 if (file->private_data) {
1147 spin_unlock(&simple_transaction_lock);
1148 free_page((unsigned long)ar);
1149 return ERR_PTR(-EBUSY);
1152 file->private_data = ar;
1154 spin_unlock(&simple_transaction_lock);
1156 if (copy_from_user(ar->data, buf, size))
1157 return ERR_PTR(-EFAULT);
1161 EXPORT_SYMBOL(simple_transaction_get);
1163 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
1165 struct simple_transaction_argresp *ar = file->private_data;
1169 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
1171 EXPORT_SYMBOL(simple_transaction_read);
1173 int simple_transaction_release(struct inode *inode, struct file *file)
1175 free_page((unsigned long)file->private_data);
1178 EXPORT_SYMBOL(simple_transaction_release);
1180 /* Simple attribute files */
1182 struct simple_attr {
1183 int (*get)(void *, u64 *);
1184 int (*set)(void *, u64);
1185 char get_buf[24]; /* enough to store a u64 and "\n\0" */
1188 const char *fmt; /* format for read operation */
1189 struct mutex mutex; /* protects access to these buffers */
1192 /* simple_attr_open is called by an actual attribute open file operation
1193 * to set the attribute specific access operations. */
1194 int simple_attr_open(struct inode *inode, struct file *file,
1195 int (*get)(void *, u64 *), int (*set)(void *, u64),
1198 struct simple_attr *attr;
1200 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1206 attr->data = inode->i_private;
1208 mutex_init(&attr->mutex);
1210 file->private_data = attr;
1212 return nonseekable_open(inode, file);
1214 EXPORT_SYMBOL_GPL(simple_attr_open);
1216 int simple_attr_release(struct inode *inode, struct file *file)
1218 kfree(file->private_data);
1221 EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */
1223 /* read from the buffer that is filled with the get function */
1224 ssize_t simple_attr_read(struct file *file, char __user *buf,
1225 size_t len, loff_t *ppos)
1227 struct simple_attr *attr;
1231 attr = file->private_data;
1236 ret = mutex_lock_interruptible(&attr->mutex);
1240 if (*ppos && attr->get_buf[0]) {
1241 /* continued read */
1242 size = strlen(attr->get_buf);
1246 ret = attr->get(attr->data, &val);
1250 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
1251 attr->fmt, (unsigned long long)val);
1254 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
1256 mutex_unlock(&attr->mutex);
1259 EXPORT_SYMBOL_GPL(simple_attr_read);
1261 /* interpret the buffer as a number to call the set function with */
1262 static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf,
1263 size_t len, loff_t *ppos, bool is_signed)
1265 struct simple_attr *attr;
1266 unsigned long long val;
1270 attr = file->private_data;
1274 ret = mutex_lock_interruptible(&attr->mutex);
1279 size = min(sizeof(attr->set_buf) - 1, len);
1280 if (copy_from_user(attr->set_buf, buf, size))
1283 attr->set_buf[size] = '\0';
1285 ret = kstrtoll(attr->set_buf, 0, &val);
1287 ret = kstrtoull(attr->set_buf, 0, &val);
1290 ret = attr->set(attr->data, val);
1292 ret = len; /* on success, claim we got the whole input */
1294 mutex_unlock(&attr->mutex);
1298 ssize_t simple_attr_write(struct file *file, const char __user *buf,
1299 size_t len, loff_t *ppos)
1301 return simple_attr_write_xsigned(file, buf, len, ppos, false);
1303 EXPORT_SYMBOL_GPL(simple_attr_write);
1305 ssize_t simple_attr_write_signed(struct file *file, const char __user *buf,
1306 size_t len, loff_t *ppos)
1308 return simple_attr_write_xsigned(file, buf, len, ppos, true);
1310 EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1313 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1314 * @sb: filesystem to do the file handle conversion on
1315 * @fid: file handle to convert
1316 * @fh_len: length of the file handle in bytes
1317 * @fh_type: type of file handle
1318 * @get_inode: filesystem callback to retrieve inode
1320 * This function decodes @fid as long as it has one of the well-known
1321 * Linux filehandle types and calls @get_inode on it to retrieve the
1322 * inode for the object specified in the file handle.
1324 struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1325 int fh_len, int fh_type, struct inode *(*get_inode)
1326 (struct super_block *sb, u64 ino, u32 gen))
1328 struct inode *inode = NULL;
1334 case FILEID_INO32_GEN:
1335 case FILEID_INO32_GEN_PARENT:
1336 inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1340 return d_obtain_alias(inode);
1342 EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1345 * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1346 * @sb: filesystem to do the file handle conversion on
1347 * @fid: file handle to convert
1348 * @fh_len: length of the file handle in bytes
1349 * @fh_type: type of file handle
1350 * @get_inode: filesystem callback to retrieve inode
1352 * This function decodes @fid as long as it has one of the well-known
1353 * Linux filehandle types and calls @get_inode on it to retrieve the
1354 * inode for the _parent_ object specified in the file handle if it
1355 * is specified in the file handle, or NULL otherwise.
1357 struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1358 int fh_len, int fh_type, struct inode *(*get_inode)
1359 (struct super_block *sb, u64 ino, u32 gen))
1361 struct inode *inode = NULL;
1367 case FILEID_INO32_GEN_PARENT:
1368 inode = get_inode(sb, fid->i32.parent_ino,
1369 (fh_len > 3 ? fid->i32.parent_gen : 0));
1373 return d_obtain_alias(inode);
1375 EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1378 * __generic_file_fsync - generic fsync implementation for simple filesystems
1380 * @file: file to synchronize
1381 * @start: start offset in bytes
1382 * @end: end offset in bytes (inclusive)
1383 * @datasync: only synchronize essential metadata if true
1385 * This is a generic implementation of the fsync method for simple
1386 * filesystems which track all non-inode metadata in the buffers list
1387 * hanging off the address_space structure.
1389 int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1392 struct inode *inode = file->f_mapping->host;
1396 err = file_write_and_wait_range(file, start, end);
1401 ret = sync_mapping_buffers(inode->i_mapping);
1402 if (!(inode->i_state & I_DIRTY_ALL))
1404 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1407 err = sync_inode_metadata(inode, 1);
1412 inode_unlock(inode);
1413 /* check and advance again to catch errors after syncing out buffers */
1414 err = file_check_and_advance_wb_err(file);
1419 EXPORT_SYMBOL(__generic_file_fsync);
1422 * generic_file_fsync - generic fsync implementation for simple filesystems
1424 * @file: file to synchronize
1425 * @start: start offset in bytes
1426 * @end: end offset in bytes (inclusive)
1427 * @datasync: only synchronize essential metadata if true
1431 int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1434 struct inode *inode = file->f_mapping->host;
1437 err = __generic_file_fsync(file, start, end, datasync);
1440 return blkdev_issue_flush(inode->i_sb->s_bdev);
1442 EXPORT_SYMBOL(generic_file_fsync);
1445 * generic_check_addressable - Check addressability of file system
1446 * @blocksize_bits: log of file system block size
1447 * @num_blocks: number of blocks in file system
1449 * Determine whether a file system with @num_blocks blocks (and a
1450 * block size of 2**@blocksize_bits) is addressable by the sector_t
1451 * and page cache of the system. Return 0 if so and -EFBIG otherwise.
1453 int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1455 u64 last_fs_block = num_blocks - 1;
1457 last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1459 if (unlikely(num_blocks == 0))
1462 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1465 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1466 (last_fs_page > (pgoff_t)(~0ULL))) {
1471 EXPORT_SYMBOL(generic_check_addressable);
1474 * No-op implementation of ->fsync for in-memory filesystems.
1476 int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1480 EXPORT_SYMBOL(noop_fsync);
1482 ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1485 * iomap based filesystems support direct I/O without need for
1486 * this callback. However, it still needs to be set in
1487 * inode->a_ops so that open/fcntl know that direct I/O is
1488 * generally supported.
1492 EXPORT_SYMBOL_GPL(noop_direct_IO);
1494 /* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1495 void kfree_link(void *p)
1499 EXPORT_SYMBOL(kfree_link);
1501 struct inode *alloc_anon_inode(struct super_block *s)
1503 static const struct address_space_operations anon_aops = {
1504 .dirty_folio = noop_dirty_folio,
1506 struct inode *inode = new_inode_pseudo(s);
1509 return ERR_PTR(-ENOMEM);
1511 inode->i_ino = get_next_ino();
1512 inode->i_mapping->a_ops = &anon_aops;
1515 * Mark the inode dirty from the very beginning,
1516 * that way it will never be moved to the dirty
1517 * list because mark_inode_dirty() will think
1518 * that it already _is_ on the dirty list.
1520 inode->i_state = I_DIRTY;
1521 inode->i_mode = S_IRUSR | S_IWUSR;
1522 inode->i_uid = current_fsuid();
1523 inode->i_gid = current_fsgid();
1524 inode->i_flags |= S_PRIVATE;
1525 simple_inode_init_ts(inode);
1528 EXPORT_SYMBOL(alloc_anon_inode);
1531 * simple_nosetlease - generic helper for prohibiting leases
1532 * @filp: file pointer
1533 * @arg: type of lease to obtain
1534 * @flp: new lease supplied for insertion
1535 * @priv: private data for lm_setup operation
1537 * Generic helper for filesystems that do not wish to allow leases to be set.
1538 * All arguments are ignored and it just returns -EINVAL.
1541 simple_nosetlease(struct file *filp, int arg, struct file_lock **flp,
1546 EXPORT_SYMBOL(simple_nosetlease);
1549 * simple_get_link - generic helper to get the target of "fast" symlinks
1550 * @dentry: not used here
1551 * @inode: the symlink inode
1552 * @done: not used here
1554 * Generic helper for filesystems to use for symlink inodes where a pointer to
1555 * the symlink target is stored in ->i_link. NOTE: this isn't normally called,
1556 * since as an optimization the path lookup code uses any non-NULL ->i_link
1557 * directly, without calling ->get_link(). But ->get_link() still must be set,
1558 * to mark the inode_operations as being for a symlink.
1560 * Return: the symlink target
1562 const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1563 struct delayed_call *done)
1565 return inode->i_link;
1567 EXPORT_SYMBOL(simple_get_link);
1569 const struct inode_operations simple_symlink_inode_operations = {
1570 .get_link = simple_get_link,
1572 EXPORT_SYMBOL(simple_symlink_inode_operations);
1575 * Operations for a permanently empty directory.
1577 static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1579 return ERR_PTR(-ENOENT);
1582 static int empty_dir_getattr(struct mnt_idmap *idmap,
1583 const struct path *path, struct kstat *stat,
1584 u32 request_mask, unsigned int query_flags)
1586 struct inode *inode = d_inode(path->dentry);
1587 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1591 static int empty_dir_setattr(struct mnt_idmap *idmap,
1592 struct dentry *dentry, struct iattr *attr)
1597 static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1602 static const struct inode_operations empty_dir_inode_operations = {
1603 .lookup = empty_dir_lookup,
1604 .permission = generic_permission,
1605 .setattr = empty_dir_setattr,
1606 .getattr = empty_dir_getattr,
1607 .listxattr = empty_dir_listxattr,
1610 static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1612 /* An empty directory has two entries . and .. at offsets 0 and 1 */
1613 return generic_file_llseek_size(file, offset, whence, 2, 2);
1616 static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1618 dir_emit_dots(file, ctx);
1622 static const struct file_operations empty_dir_operations = {
1623 .llseek = empty_dir_llseek,
1624 .read = generic_read_dir,
1625 .iterate_shared = empty_dir_readdir,
1626 .fsync = noop_fsync,
1630 void make_empty_dir_inode(struct inode *inode)
1632 set_nlink(inode, 2);
1633 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1634 inode->i_uid = GLOBAL_ROOT_UID;
1635 inode->i_gid = GLOBAL_ROOT_GID;
1638 inode->i_blkbits = PAGE_SHIFT;
1639 inode->i_blocks = 0;
1641 inode->i_op = &empty_dir_inode_operations;
1642 inode->i_opflags &= ~IOP_XATTR;
1643 inode->i_fop = &empty_dir_operations;
1646 bool is_empty_dir_inode(struct inode *inode)
1648 return (inode->i_fop == &empty_dir_operations) &&
1649 (inode->i_op == &empty_dir_inode_operations);
1652 #if IS_ENABLED(CONFIG_UNICODE)
1654 * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1655 * @dentry: dentry whose name we are checking against
1656 * @len: len of name of dentry
1657 * @str: str pointer to name of dentry
1658 * @name: Name to compare against
1660 * Return: 0 if names match, 1 if mismatch, or -ERRNO
1662 static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1663 const char *str, const struct qstr *name)
1665 const struct dentry *parent = READ_ONCE(dentry->d_parent);
1666 const struct inode *dir = READ_ONCE(parent->d_inode);
1667 const struct super_block *sb = dentry->d_sb;
1668 const struct unicode_map *um = sb->s_encoding;
1669 struct qstr qstr = QSTR_INIT(str, len);
1670 char strbuf[DNAME_INLINE_LEN];
1673 if (!dir || !IS_CASEFOLDED(dir))
1676 * If the dentry name is stored in-line, then it may be concurrently
1677 * modified by a rename. If this happens, the VFS will eventually retry
1678 * the lookup, so it doesn't matter what ->d_compare() returns.
1679 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1680 * string. Therefore, we have to copy the name into a temporary buffer.
1682 if (len <= DNAME_INLINE_LEN - 1) {
1683 memcpy(strbuf, str, len);
1686 /* prevent compiler from optimizing out the temporary buffer */
1689 ret = utf8_strncasecmp(um, name, &qstr);
1693 if (sb_has_strict_encoding(sb))
1696 if (len != name->len)
1698 return !!memcmp(str, name->name, len);
1702 * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1703 * @dentry: dentry of the parent directory
1704 * @str: qstr of name whose hash we should fill in
1706 * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1708 static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1710 const struct inode *dir = READ_ONCE(dentry->d_inode);
1711 struct super_block *sb = dentry->d_sb;
1712 const struct unicode_map *um = sb->s_encoding;
1715 if (!dir || !IS_CASEFOLDED(dir))
1718 ret = utf8_casefold_hash(um, dentry, str);
1719 if (ret < 0 && sb_has_strict_encoding(sb))
1724 static const struct dentry_operations generic_ci_dentry_ops = {
1725 .d_hash = generic_ci_d_hash,
1726 .d_compare = generic_ci_d_compare,
1730 #ifdef CONFIG_FS_ENCRYPTION
1731 static const struct dentry_operations generic_encrypted_dentry_ops = {
1732 .d_revalidate = fscrypt_d_revalidate,
1736 #if defined(CONFIG_FS_ENCRYPTION) && IS_ENABLED(CONFIG_UNICODE)
1737 static const struct dentry_operations generic_encrypted_ci_dentry_ops = {
1738 .d_hash = generic_ci_d_hash,
1739 .d_compare = generic_ci_d_compare,
1740 .d_revalidate = fscrypt_d_revalidate,
1745 * generic_set_encrypted_ci_d_ops - helper for setting d_ops for given dentry
1746 * @dentry: dentry to set ops on
1748 * Casefolded directories need d_hash and d_compare set, so that the dentries
1749 * contained in them are handled case-insensitively. Note that these operations
1750 * are needed on the parent directory rather than on the dentries in it, and
1751 * while the casefolding flag can be toggled on and off on an empty directory,
1752 * dentry_operations can't be changed later. As a result, if the filesystem has
1753 * casefolding support enabled at all, we have to give all dentries the
1754 * casefolding operations even if their inode doesn't have the casefolding flag
1755 * currently (and thus the casefolding ops would be no-ops for now).
1757 * Encryption works differently in that the only dentry operation it needs is
1758 * d_revalidate, which it only needs on dentries that have the no-key name flag.
1759 * The no-key flag can't be set "later", so we don't have to worry about that.
1761 * Finally, to maximize compatibility with overlayfs (which isn't compatible
1762 * with certain dentry operations) and to avoid taking an unnecessary
1763 * performance hit, we use custom dentry_operations for each possible
1764 * combination rather than always installing all operations.
1766 void generic_set_encrypted_ci_d_ops(struct dentry *dentry)
1768 #ifdef CONFIG_FS_ENCRYPTION
1769 bool needs_encrypt_ops = dentry->d_flags & DCACHE_NOKEY_NAME;
1771 #if IS_ENABLED(CONFIG_UNICODE)
1772 bool needs_ci_ops = dentry->d_sb->s_encoding;
1774 #if defined(CONFIG_FS_ENCRYPTION) && IS_ENABLED(CONFIG_UNICODE)
1775 if (needs_encrypt_ops && needs_ci_ops) {
1776 d_set_d_op(dentry, &generic_encrypted_ci_dentry_ops);
1780 #ifdef CONFIG_FS_ENCRYPTION
1781 if (needs_encrypt_ops) {
1782 d_set_d_op(dentry, &generic_encrypted_dentry_ops);
1786 #if IS_ENABLED(CONFIG_UNICODE)
1788 d_set_d_op(dentry, &generic_ci_dentry_ops);
1793 EXPORT_SYMBOL(generic_set_encrypted_ci_d_ops);
1796 * inode_maybe_inc_iversion - increments i_version
1797 * @inode: inode with the i_version that should be updated
1798 * @force: increment the counter even if it's not necessary?
1800 * Every time the inode is modified, the i_version field must be seen to have
1801 * changed by any observer.
1803 * If "force" is set or the QUERIED flag is set, then ensure that we increment
1804 * the value, and clear the queried flag.
1806 * In the common case where neither is set, then we can return "false" without
1807 * updating i_version.
1809 * If this function returns false, and no other metadata has changed, then we
1810 * can avoid logging the metadata.
1812 bool inode_maybe_inc_iversion(struct inode *inode, bool force)
1817 * The i_version field is not strictly ordered with any other inode
1818 * information, but the legacy inode_inc_iversion code used a spinlock
1819 * to serialize increments.
1821 * Here, we add full memory barriers to ensure that any de-facto
1822 * ordering with other info is preserved.
1824 * This barrier pairs with the barrier in inode_query_iversion()
1827 cur = inode_peek_iversion_raw(inode);
1829 /* If flag is clear then we needn't do anything */
1830 if (!force && !(cur & I_VERSION_QUERIED))
1833 /* Since lowest bit is flag, add 2 to avoid it */
1834 new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
1835 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
1838 EXPORT_SYMBOL(inode_maybe_inc_iversion);
1841 * inode_query_iversion - read i_version for later use
1842 * @inode: inode from which i_version should be read
1844 * Read the inode i_version counter. This should be used by callers that wish
1845 * to store the returned i_version for later comparison. This will guarantee
1846 * that a later query of the i_version will result in a different value if
1847 * anything has changed.
1849 * In this implementation, we fetch the current value, set the QUERIED flag and
1850 * then try to swap it into place with a cmpxchg, if it wasn't already set. If
1851 * that fails, we try again with the newly fetched value from the cmpxchg.
1853 u64 inode_query_iversion(struct inode *inode)
1857 cur = inode_peek_iversion_raw(inode);
1859 /* If flag is already set, then no need to swap */
1860 if (cur & I_VERSION_QUERIED) {
1862 * This barrier (and the implicit barrier in the
1863 * cmpxchg below) pairs with the barrier in
1864 * inode_maybe_inc_iversion().
1870 new = cur | I_VERSION_QUERIED;
1871 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
1872 return cur >> I_VERSION_QUERIED_SHIFT;
1874 EXPORT_SYMBOL(inode_query_iversion);
1876 ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter,
1877 ssize_t direct_written, ssize_t buffered_written)
1879 struct address_space *mapping = iocb->ki_filp->f_mapping;
1880 loff_t pos = iocb->ki_pos - buffered_written;
1881 loff_t end = iocb->ki_pos - 1;
1885 * If the buffered write fallback returned an error, we want to return
1886 * the number of bytes which were written by direct I/O, or the error
1887 * code if that was zero.
1889 * Note that this differs from normal direct-io semantics, which will
1890 * return -EFOO even if some bytes were written.
1892 if (unlikely(buffered_written < 0)) {
1894 return direct_written;
1895 return buffered_written;
1899 * We need to ensure that the page cache pages are written to disk and
1900 * invalidated to preserve the expected O_DIRECT semantics.
1902 err = filemap_write_and_wait_range(mapping, pos, end);
1905 * We don't know how much we wrote, so just return the number of
1906 * bytes which were direct-written
1908 iocb->ki_pos -= buffered_written;
1910 return direct_written;
1913 invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
1914 return direct_written + buffered_written;
1916 EXPORT_SYMBOL_GPL(direct_write_fallback);
1919 * simple_inode_init_ts - initialize the timestamps for a new inode
1920 * @inode: inode to be initialized
1922 * When a new inode is created, most filesystems set the timestamps to the
1923 * current time. Add a helper to do this.
1925 struct timespec64 simple_inode_init_ts(struct inode *inode)
1927 struct timespec64 ts = inode_set_ctime_current(inode);
1929 inode_set_atime_to_ts(inode, ts);
1930 inode_set_mtime_to_ts(inode, ts);
1933 EXPORT_SYMBOL(simple_inode_init_ts);