2 * kernfs.h - pseudo filesystem decoupled from vfs locking
4 * This file is released under the GPLv2.
7 #ifndef __LINUX_KERNFS_H
8 #define __LINUX_KERNFS_H
10 #include <linux/kernel.h>
11 #include <linux/err.h>
12 #include <linux/list.h>
13 #include <linux/mutex.h>
14 #include <linux/idr.h>
15 #include <linux/lockdep.h>
16 #include <linux/rbtree.h>
17 #include <linux/atomic.h>
18 #include <linux/uidgid.h>
19 #include <linux/wait.h>
25 struct vm_area_struct;
27 struct file_system_type;
29 struct kernfs_open_node;
32 enum kernfs_node_type {
38 #define KERNFS_TYPE_MASK 0x000f
39 #define KERNFS_FLAG_MASK ~KERNFS_TYPE_MASK
41 enum kernfs_node_flag {
42 KERNFS_ACTIVATED = 0x0010,
44 KERNFS_HAS_SEQ_SHOW = 0x0040,
45 KERNFS_HAS_MMAP = 0x0080,
46 KERNFS_LOCKDEP = 0x0100,
47 KERNFS_SUICIDAL = 0x0400,
48 KERNFS_SUICIDED = 0x0800,
49 KERNFS_EMPTY_DIR = 0x1000,
50 KERNFS_HAS_RELEASE = 0x2000,
53 /* @flags for kernfs_create_root() */
54 enum kernfs_root_flag {
56 * kernfs_nodes are created in the deactivated state and invisible.
57 * They require explicit kernfs_activate() to become visible. This
58 * can be used to make related nodes become visible atomically
59 * after all nodes are created successfully.
61 KERNFS_ROOT_CREATE_DEACTIVATED = 0x0001,
64 * For regular flies, if the opener has CAP_DAC_OVERRIDE, open(2)
65 * succeeds regardless of the RW permissions. sysfs had an extra
66 * layer of enforcement where open(2) fails with -EACCES regardless
67 * of CAP_DAC_OVERRIDE if the permission doesn't have the
68 * respective read or write access at all (none of S_IRUGO or
69 * S_IWUGO) or the respective operation isn't implemented. The
70 * following flag enables that behavior.
72 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK = 0x0002,
75 * The filesystem supports exportfs operation, so userspace can use
76 * fhandle to access nodes of the fs.
78 KERNFS_ROOT_SUPPORT_EXPORTOP = 0x0004,
81 /* type-specific structures for kernfs_node union members */
82 struct kernfs_elem_dir {
83 unsigned long subdirs;
84 /* children rbtree starts here and goes through kn->rb */
85 struct rb_root children;
88 * The kernfs hierarchy this directory belongs to. This fits
89 * better directly in kernfs_node but is here to save space.
91 struct kernfs_root *root;
94 struct kernfs_elem_symlink {
95 struct kernfs_node *target_kn;
98 struct kernfs_elem_attr {
99 const struct kernfs_ops *ops;
100 struct kernfs_open_node *open;
102 struct kernfs_node *notify_next; /* for kernfs_notify() */
105 /* represent a kernfs node */
106 union kernfs_node_id {
109 * blktrace will export this struct as a simplified 'struct
110 * fid' (which is a big data struction), so userspace can use
111 * it to find kernfs node. The layout must match the first two
112 * fields of 'struct fid' exactly.
121 * kernfs_node - the building block of kernfs hierarchy. Each and every
122 * kernfs node is represented by single kernfs_node. Most fields are
123 * private to kernfs and shouldn't be accessed directly by kernfs users.
125 * As long as s_count reference is held, the kernfs_node itself is
126 * accessible. Dereferencing elem or any other outer entity requires
132 #ifdef CONFIG_DEBUG_LOCK_ALLOC
133 struct lockdep_map dep_map;
136 * Use kernfs_get_parent() and kernfs_name/path() instead of
137 * accessing the following two fields directly. If the node is
138 * never moved to a different parent, it is safe to access the
141 struct kernfs_node *parent;
146 const void *ns; /* namespace tag */
147 unsigned int hash; /* ns + name hash */
149 struct kernfs_elem_dir dir;
150 struct kernfs_elem_symlink symlink;
151 struct kernfs_elem_attr attr;
156 union kernfs_node_id id;
157 unsigned short flags;
159 struct kernfs_iattrs *iattr;
163 * kernfs_syscall_ops may be specified on kernfs_create_root() to support
164 * syscalls. These optional callbacks are invoked on the matching syscalls
165 * and can perform any kernfs operations which don't necessarily have to be
166 * the exact operation requested. An active reference is held for each
167 * kernfs_node parameter.
169 struct kernfs_syscall_ops {
170 int (*remount_fs)(struct kernfs_root *root, int *flags, char *data);
171 int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
173 int (*mkdir)(struct kernfs_node *parent, const char *name,
175 int (*rmdir)(struct kernfs_node *kn);
176 int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
177 const char *new_name);
178 int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
179 struct kernfs_root *root);
183 /* published fields */
184 struct kernfs_node *kn;
185 unsigned int flags; /* KERNFS_ROOT_* flags */
187 /* private fields, do not use outside kernfs proper */
190 struct kernfs_syscall_ops *syscall_ops;
192 /* list of kernfs_super_info of this root, protected by kernfs_mutex */
193 struct list_head supers;
195 wait_queue_head_t deactivate_waitq;
198 struct kernfs_open_file {
199 /* published fields */
200 struct kernfs_node *kn;
202 struct seq_file *seq_file;
205 /* private fields, do not use outside kernfs proper */
207 struct mutex prealloc_mutex;
209 struct list_head list;
212 size_t atomic_write_len;
215 const struct vm_operations_struct *vm_ops;
220 * Optional open/release methods. Both are called with
221 * @of->seq_file populated.
223 int (*open)(struct kernfs_open_file *of);
224 void (*release)(struct kernfs_open_file *of);
227 * Read is handled by either seq_file or raw_read().
229 * If seq_show() is present, seq_file path is active. Other seq
230 * operations are optional and if not implemented, the behavior is
231 * equivalent to single_open(). @sf->private points to the
232 * associated kernfs_open_file.
234 * read() is bounced through kernel buffer and a read larger than
235 * PAGE_SIZE results in partial operation of PAGE_SIZE.
237 int (*seq_show)(struct seq_file *sf, void *v);
239 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
240 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
241 void (*seq_stop)(struct seq_file *sf, void *v);
243 ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
247 * write() is bounced through kernel buffer. If atomic_write_len
248 * is not set, a write larger than PAGE_SIZE results in partial
249 * operations of PAGE_SIZE chunks. If atomic_write_len is set,
250 * writes upto the specified size are executed atomically but
251 * larger ones are rejected with -E2BIG.
253 size_t atomic_write_len;
255 * "prealloc" causes a buffer to be allocated at open for
256 * all read/write requests. As ->seq_show uses seq_read()
257 * which does its own allocation, it is incompatible with
258 * ->prealloc. Provide ->read and ->write with ->prealloc.
261 ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
264 int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
266 #ifdef CONFIG_DEBUG_LOCK_ALLOC
267 struct lock_class_key lockdep_key;
273 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
275 return kn->flags & KERNFS_TYPE_MASK;
279 * kernfs_enable_ns - enable namespace under a directory
280 * @kn: directory of interest, should be empty
282 * This is to be called right after @kn is created to enable namespace
283 * under it. All children of @kn must have non-NULL namespace tags and
284 * only the ones which match the super_block's tag will be visible.
286 static inline void kernfs_enable_ns(struct kernfs_node *kn)
288 WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
289 WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
290 kn->flags |= KERNFS_NS;
294 * kernfs_ns_enabled - test whether namespace is enabled
295 * @kn: the node to test
297 * Test whether namespace filtering is enabled for the children of @ns.
299 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
301 return kn->flags & KERNFS_NS;
304 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
305 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
306 char *buf, size_t buflen);
307 void pr_cont_kernfs_name(struct kernfs_node *kn);
308 void pr_cont_kernfs_path(struct kernfs_node *kn);
309 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
310 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
311 const char *name, const void *ns);
312 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
313 const char *path, const void *ns);
314 void kernfs_get(struct kernfs_node *kn);
315 void kernfs_put(struct kernfs_node *kn);
317 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
318 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
319 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
321 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
322 struct super_block *sb);
323 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
324 unsigned int flags, void *priv);
325 void kernfs_destroy_root(struct kernfs_root *root);
327 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
328 const char *name, umode_t mode,
329 kuid_t uid, kgid_t gid,
330 void *priv, const void *ns);
331 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
333 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
334 const char *name, umode_t mode,
335 kuid_t uid, kgid_t gid,
337 const struct kernfs_ops *ops,
338 void *priv, const void *ns,
339 struct lock_class_key *key);
340 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
342 struct kernfs_node *target);
343 void kernfs_activate(struct kernfs_node *kn);
344 void kernfs_remove(struct kernfs_node *kn);
345 void kernfs_break_active_protection(struct kernfs_node *kn);
346 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
347 bool kernfs_remove_self(struct kernfs_node *kn);
348 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
350 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
351 const char *new_name, const void *new_ns);
352 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
353 void kernfs_notify(struct kernfs_node *kn);
355 const void *kernfs_super_ns(struct super_block *sb);
356 struct dentry *kernfs_mount_ns(struct file_system_type *fs_type, int flags,
357 struct kernfs_root *root, unsigned long magic,
358 bool *new_sb_created, const void *ns);
359 void kernfs_kill_sb(struct super_block *sb);
360 struct super_block *kernfs_pin_sb(struct kernfs_root *root, const void *ns);
362 void kernfs_init(void);
364 struct kernfs_node *kernfs_get_node_by_id(struct kernfs_root *root,
365 const union kernfs_node_id *id);
366 #else /* CONFIG_KERNFS */
368 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
369 { return 0; } /* whatever */
371 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
373 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
376 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
379 static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
380 struct kernfs_node *kn,
381 char *buf, size_t buflen)
384 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
385 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
387 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
390 static inline struct kernfs_node *
391 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
394 static inline struct kernfs_node *
395 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
399 static inline void kernfs_get(struct kernfs_node *kn) { }
400 static inline void kernfs_put(struct kernfs_node *kn) { }
402 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
405 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
408 static inline struct inode *
409 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
412 static inline struct kernfs_root *
413 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
415 { return ERR_PTR(-ENOSYS); }
417 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
419 static inline struct kernfs_node *
420 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
421 umode_t mode, kuid_t uid, kgid_t gid,
422 void *priv, const void *ns)
423 { return ERR_PTR(-ENOSYS); }
425 static inline struct kernfs_node *
426 __kernfs_create_file(struct kernfs_node *parent, const char *name,
427 umode_t mode, kuid_t uid, kgid_t gid,
428 loff_t size, const struct kernfs_ops *ops,
429 void *priv, const void *ns, struct lock_class_key *key)
430 { return ERR_PTR(-ENOSYS); }
432 static inline struct kernfs_node *
433 kernfs_create_link(struct kernfs_node *parent, const char *name,
434 struct kernfs_node *target)
435 { return ERR_PTR(-ENOSYS); }
437 static inline void kernfs_activate(struct kernfs_node *kn) { }
439 static inline void kernfs_remove(struct kernfs_node *kn) { }
441 static inline bool kernfs_remove_self(struct kernfs_node *kn)
444 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
445 const char *name, const void *ns)
448 static inline int kernfs_rename_ns(struct kernfs_node *kn,
449 struct kernfs_node *new_parent,
450 const char *new_name, const void *new_ns)
453 static inline int kernfs_setattr(struct kernfs_node *kn,
454 const struct iattr *iattr)
457 static inline void kernfs_notify(struct kernfs_node *kn) { }
459 static inline const void *kernfs_super_ns(struct super_block *sb)
462 static inline struct dentry *
463 kernfs_mount_ns(struct file_system_type *fs_type, int flags,
464 struct kernfs_root *root, unsigned long magic,
465 bool *new_sb_created, const void *ns)
466 { return ERR_PTR(-ENOSYS); }
468 static inline void kernfs_kill_sb(struct super_block *sb) { }
470 static inline void kernfs_init(void) { }
472 #endif /* CONFIG_KERNFS */
475 * kernfs_path - build full path of a given node
476 * @kn: kernfs_node of interest
477 * @buf: buffer to copy @kn's name into
478 * @buflen: size of @buf
480 * Builds and returns the full path of @kn in @buf of @buflen bytes. The
481 * path is built from the end of @buf so the returned pointer usually
482 * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated
483 * and %NULL is returned.
485 static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
487 return kernfs_path_from_node(kn, NULL, buf, buflen);
490 static inline struct kernfs_node *
491 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
493 return kernfs_find_and_get_ns(kn, name, NULL);
496 static inline struct kernfs_node *
497 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
499 return kernfs_walk_and_get_ns(kn, path, NULL);
502 static inline struct kernfs_node *
503 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
506 return kernfs_create_dir_ns(parent, name, mode,
507 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
511 static inline struct kernfs_node *
512 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
513 umode_t mode, kuid_t uid, kgid_t gid,
514 loff_t size, const struct kernfs_ops *ops,
515 void *priv, const void *ns)
517 struct lock_class_key *key = NULL;
519 #ifdef CONFIG_DEBUG_LOCK_ALLOC
520 key = (struct lock_class_key *)&ops->lockdep_key;
522 return __kernfs_create_file(parent, name, mode, uid, gid,
523 size, ops, priv, ns, key);
526 static inline struct kernfs_node *
527 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
528 loff_t size, const struct kernfs_ops *ops, void *priv)
530 return kernfs_create_file_ns(parent, name, mode,
531 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
532 size, ops, priv, NULL);
535 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
538 return kernfs_remove_by_name_ns(parent, name, NULL);
541 static inline int kernfs_rename(struct kernfs_node *kn,
542 struct kernfs_node *new_parent,
543 const char *new_name)
545 return kernfs_rename_ns(kn, new_parent, new_name, NULL);
548 static inline struct dentry *
549 kernfs_mount(struct file_system_type *fs_type, int flags,
550 struct kernfs_root *root, unsigned long magic,
551 bool *new_sb_created)
553 return kernfs_mount_ns(fs_type, flags, root,
554 magic, new_sb_created, NULL);
557 #endif /* __LINUX_KERNFS_H */