xref: /linux-6.15/include/linux/kernfs.h (revision ef264cf0)
1 /*
2  * kernfs.h - pseudo filesystem decoupled from vfs locking
3  *
4  * This file is released under the GPLv2.
5  */
6 
7 #ifndef __LINUX_KERNFS_H
8 #define __LINUX_KERNFS_H
9 
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/wait.h>
19 
20 struct file;
21 struct dentry;
22 struct iattr;
23 struct seq_file;
24 struct vm_area_struct;
25 struct super_block;
26 struct file_system_type;
27 
28 struct kernfs_open_node;
29 struct kernfs_iattrs;
30 
31 enum kernfs_node_type {
32 	KERNFS_DIR		= 0x0001,
33 	KERNFS_FILE		= 0x0002,
34 	KERNFS_LINK		= 0x0004,
35 };
36 
37 #define KERNFS_TYPE_MASK	0x000f
38 #define KERNFS_FLAG_MASK	~KERNFS_TYPE_MASK
39 
40 enum kernfs_node_flag {
41 	KERNFS_ACTIVATED	= 0x0010,
42 	KERNFS_NS		= 0x0020,
43 	KERNFS_HAS_SEQ_SHOW	= 0x0040,
44 	KERNFS_HAS_MMAP		= 0x0080,
45 	KERNFS_LOCKDEP		= 0x0100,
46 	KERNFS_STATIC_NAME	= 0x0200,
47 	KERNFS_SUICIDAL		= 0x0400,
48 	KERNFS_SUICIDED		= 0x0800,
49 };
50 
51 /* @flags for kernfs_create_root() */
52 enum kernfs_root_flag {
53 	/*
54 	 * kernfs_nodes are created in the deactivated state and invisible.
55 	 * They require explicit kernfs_activate() to become visible.  This
56 	 * can be used to make related nodes become visible atomically
57 	 * after all nodes are created successfully.
58 	 */
59 	KERNFS_ROOT_CREATE_DEACTIVATED		= 0x0001,
60 
61 	/*
62 	 * For regular flies, if the opener has CAP_DAC_OVERRIDE, open(2)
63 	 * succeeds regardless of the RW permissions.  sysfs had an extra
64 	 * layer of enforcement where open(2) fails with -EACCES regardless
65 	 * of CAP_DAC_OVERRIDE if the permission doesn't have the
66 	 * respective read or write access at all (none of S_IRUGO or
67 	 * S_IWUGO) or the respective operation isn't implemented.  The
68 	 * following flag enables that behavior.
69 	 */
70 	KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK	= 0x0002,
71 };
72 
73 /* type-specific structures for kernfs_node union members */
74 struct kernfs_elem_dir {
75 	unsigned long		subdirs;
76 	/* children rbtree starts here and goes through kn->rb */
77 	struct rb_root		children;
78 
79 	/*
80 	 * The kernfs hierarchy this directory belongs to.  This fits
81 	 * better directly in kernfs_node but is here to save space.
82 	 */
83 	struct kernfs_root	*root;
84 };
85 
86 struct kernfs_elem_symlink {
87 	struct kernfs_node	*target_kn;
88 };
89 
90 struct kernfs_elem_attr {
91 	const struct kernfs_ops	*ops;
92 	struct kernfs_open_node	*open;
93 	loff_t			size;
94 	struct kernfs_node	*notify_next;	/* for kernfs_notify() */
95 };
96 
97 /*
98  * kernfs_node - the building block of kernfs hierarchy.  Each and every
99  * kernfs node is represented by single kernfs_node.  Most fields are
100  * private to kernfs and shouldn't be accessed directly by kernfs users.
101  *
102  * As long as s_count reference is held, the kernfs_node itself is
103  * accessible.  Dereferencing elem or any other outer entity requires
104  * active reference.
105  */
106 struct kernfs_node {
107 	atomic_t		count;
108 	atomic_t		active;
109 #ifdef CONFIG_DEBUG_LOCK_ALLOC
110 	struct lockdep_map	dep_map;
111 #endif
112 	/*
113 	 * Use kernfs_get_parent() and kernfs_name/path() instead of
114 	 * accessing the following two fields directly.  If the node is
115 	 * never moved to a different parent, it is safe to access the
116 	 * parent directly.
117 	 */
118 	struct kernfs_node	*parent;
119 	const char		*name;
120 
121 	struct rb_node		rb;
122 
123 	const void		*ns;	/* namespace tag */
124 	unsigned int		hash;	/* ns + name hash */
125 	union {
126 		struct kernfs_elem_dir		dir;
127 		struct kernfs_elem_symlink	symlink;
128 		struct kernfs_elem_attr		attr;
129 	};
130 
131 	void			*priv;
132 
133 	unsigned short		flags;
134 	umode_t			mode;
135 	unsigned int		ino;
136 	struct kernfs_iattrs	*iattr;
137 };
138 
139 /*
140  * kernfs_syscall_ops may be specified on kernfs_create_root() to support
141  * syscalls.  These optional callbacks are invoked on the matching syscalls
142  * and can perform any kernfs operations which don't necessarily have to be
143  * the exact operation requested.  An active reference is held for each
144  * kernfs_node parameter.
145  */
146 struct kernfs_syscall_ops {
147 	int (*remount_fs)(struct kernfs_root *root, int *flags, char *data);
148 	int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
149 
150 	int (*mkdir)(struct kernfs_node *parent, const char *name,
151 		     umode_t mode);
152 	int (*rmdir)(struct kernfs_node *kn);
153 	int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
154 		      const char *new_name);
155 };
156 
157 struct kernfs_root {
158 	/* published fields */
159 	struct kernfs_node	*kn;
160 	unsigned int		flags;	/* KERNFS_ROOT_* flags */
161 
162 	/* private fields, do not use outside kernfs proper */
163 	struct ida		ino_ida;
164 	struct kernfs_syscall_ops *syscall_ops;
165 
166 	/* list of kernfs_super_info of this root, protected by kernfs_mutex */
167 	struct list_head	supers;
168 
169 	wait_queue_head_t	deactivate_waitq;
170 };
171 
172 struct kernfs_open_file {
173 	/* published fields */
174 	struct kernfs_node	*kn;
175 	struct file		*file;
176 	void			*priv;
177 
178 	/* private fields, do not use outside kernfs proper */
179 	struct mutex		mutex;
180 	int			event;
181 	struct list_head	list;
182 
183 	size_t			atomic_write_len;
184 	bool			mmapped;
185 	const struct vm_operations_struct *vm_ops;
186 };
187 
188 struct kernfs_ops {
189 	/*
190 	 * Read is handled by either seq_file or raw_read().
191 	 *
192 	 * If seq_show() is present, seq_file path is active.  Other seq
193 	 * operations are optional and if not implemented, the behavior is
194 	 * equivalent to single_open().  @sf->private points to the
195 	 * associated kernfs_open_file.
196 	 *
197 	 * read() is bounced through kernel buffer and a read larger than
198 	 * PAGE_SIZE results in partial operation of PAGE_SIZE.
199 	 */
200 	int (*seq_show)(struct seq_file *sf, void *v);
201 
202 	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
203 	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
204 	void (*seq_stop)(struct seq_file *sf, void *v);
205 
206 	ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
207 			loff_t off);
208 
209 	/*
210 	 * write() is bounced through kernel buffer.  If atomic_write_len
211 	 * is not set, a write larger than PAGE_SIZE results in partial
212 	 * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
213 	 * writes upto the specified size are executed atomically but
214 	 * larger ones are rejected with -E2BIG.
215 	 */
216 	size_t atomic_write_len;
217 	ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
218 			 loff_t off);
219 
220 	int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
221 
222 #ifdef CONFIG_DEBUG_LOCK_ALLOC
223 	struct lock_class_key	lockdep_key;
224 #endif
225 };
226 
227 #ifdef CONFIG_KERNFS
228 
229 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
230 {
231 	return kn->flags & KERNFS_TYPE_MASK;
232 }
233 
234 /**
235  * kernfs_enable_ns - enable namespace under a directory
236  * @kn: directory of interest, should be empty
237  *
238  * This is to be called right after @kn is created to enable namespace
239  * under it.  All children of @kn must have non-NULL namespace tags and
240  * only the ones which match the super_block's tag will be visible.
241  */
242 static inline void kernfs_enable_ns(struct kernfs_node *kn)
243 {
244 	WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
245 	WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
246 	kn->flags |= KERNFS_NS;
247 }
248 
249 /**
250  * kernfs_ns_enabled - test whether namespace is enabled
251  * @kn: the node to test
252  *
253  * Test whether namespace filtering is enabled for the children of @ns.
254  */
255 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
256 {
257 	return kn->flags & KERNFS_NS;
258 }
259 
260 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
261 char * __must_check kernfs_path(struct kernfs_node *kn, char *buf,
262 				size_t buflen);
263 void pr_cont_kernfs_name(struct kernfs_node *kn);
264 void pr_cont_kernfs_path(struct kernfs_node *kn);
265 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
266 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
267 					   const char *name, const void *ns);
268 void kernfs_get(struct kernfs_node *kn);
269 void kernfs_put(struct kernfs_node *kn);
270 
271 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
272 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
273 
274 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
275 				       unsigned int flags, void *priv);
276 void kernfs_destroy_root(struct kernfs_root *root);
277 
278 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
279 					 const char *name, umode_t mode,
280 					 void *priv, const void *ns);
281 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
282 					 const char *name,
283 					 umode_t mode, loff_t size,
284 					 const struct kernfs_ops *ops,
285 					 void *priv, const void *ns,
286 					 bool name_is_static,
287 					 struct lock_class_key *key);
288 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
289 				       const char *name,
290 				       struct kernfs_node *target);
291 void kernfs_activate(struct kernfs_node *kn);
292 void kernfs_remove(struct kernfs_node *kn);
293 void kernfs_break_active_protection(struct kernfs_node *kn);
294 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
295 bool kernfs_remove_self(struct kernfs_node *kn);
296 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
297 			     const void *ns);
298 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
299 		     const char *new_name, const void *new_ns);
300 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
301 void kernfs_notify(struct kernfs_node *kn);
302 
303 const void *kernfs_super_ns(struct super_block *sb);
304 struct dentry *kernfs_mount_ns(struct file_system_type *fs_type, int flags,
305 			       struct kernfs_root *root, unsigned long magic,
306 			       bool *new_sb_created, const void *ns);
307 void kernfs_kill_sb(struct super_block *sb);
308 
309 void kernfs_init(void);
310 
311 #else	/* CONFIG_KERNFS */
312 
313 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
314 { return 0; }	/* whatever */
315 
316 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
317 
318 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
319 { return false; }
320 
321 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
322 { return -ENOSYS; }
323 
324 static inline char * __must_check kernfs_path(struct kernfs_node *kn, char *buf,
325 					      size_t buflen)
326 { return NULL; }
327 
328 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
329 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
330 
331 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
332 { return NULL; }
333 
334 static inline struct kernfs_node *
335 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
336 		       const void *ns)
337 { return NULL; }
338 
339 static inline void kernfs_get(struct kernfs_node *kn) { }
340 static inline void kernfs_put(struct kernfs_node *kn) { }
341 
342 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
343 { return NULL; }
344 
345 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
346 { return NULL; }
347 
348 static inline struct kernfs_root *
349 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
350 		   void *priv)
351 { return ERR_PTR(-ENOSYS); }
352 
353 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
354 
355 static inline struct kernfs_node *
356 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
357 		     umode_t mode, void *priv, const void *ns)
358 { return ERR_PTR(-ENOSYS); }
359 
360 static inline struct kernfs_node *
361 __kernfs_create_file(struct kernfs_node *parent, const char *name,
362 		     umode_t mode, loff_t size, const struct kernfs_ops *ops,
363 		     void *priv, const void *ns, bool name_is_static,
364 		     struct lock_class_key *key)
365 { return ERR_PTR(-ENOSYS); }
366 
367 static inline struct kernfs_node *
368 kernfs_create_link(struct kernfs_node *parent, const char *name,
369 		   struct kernfs_node *target)
370 { return ERR_PTR(-ENOSYS); }
371 
372 static inline void kernfs_activate(struct kernfs_node *kn) { }
373 
374 static inline void kernfs_remove(struct kernfs_node *kn) { }
375 
376 static inline bool kernfs_remove_self(struct kernfs_node *kn)
377 { return false; }
378 
379 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
380 					   const char *name, const void *ns)
381 { return -ENOSYS; }
382 
383 static inline int kernfs_rename_ns(struct kernfs_node *kn,
384 				   struct kernfs_node *new_parent,
385 				   const char *new_name, const void *new_ns)
386 { return -ENOSYS; }
387 
388 static inline int kernfs_setattr(struct kernfs_node *kn,
389 				 const struct iattr *iattr)
390 { return -ENOSYS; }
391 
392 static inline void kernfs_notify(struct kernfs_node *kn) { }
393 
394 static inline const void *kernfs_super_ns(struct super_block *sb)
395 { return NULL; }
396 
397 static inline struct dentry *
398 kernfs_mount_ns(struct file_system_type *fs_type, int flags,
399 		struct kernfs_root *root, unsigned long magic,
400 		bool *new_sb_created, const void *ns)
401 { return ERR_PTR(-ENOSYS); }
402 
403 static inline void kernfs_kill_sb(struct super_block *sb) { }
404 
405 static inline void kernfs_init(void) { }
406 
407 #endif	/* CONFIG_KERNFS */
408 
409 static inline struct kernfs_node *
410 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
411 {
412 	return kernfs_find_and_get_ns(kn, name, NULL);
413 }
414 
415 static inline struct kernfs_node *
416 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
417 		  void *priv)
418 {
419 	return kernfs_create_dir_ns(parent, name, mode, priv, NULL);
420 }
421 
422 static inline struct kernfs_node *
423 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
424 		      umode_t mode, loff_t size, const struct kernfs_ops *ops,
425 		      void *priv, const void *ns)
426 {
427 	struct lock_class_key *key = NULL;
428 
429 #ifdef CONFIG_DEBUG_LOCK_ALLOC
430 	key = (struct lock_class_key *)&ops->lockdep_key;
431 #endif
432 	return __kernfs_create_file(parent, name, mode, size, ops, priv, ns,
433 				    false, key);
434 }
435 
436 static inline struct kernfs_node *
437 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
438 		   loff_t size, const struct kernfs_ops *ops, void *priv)
439 {
440 	return kernfs_create_file_ns(parent, name, mode, size, ops, priv, NULL);
441 }
442 
443 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
444 					const char *name)
445 {
446 	return kernfs_remove_by_name_ns(parent, name, NULL);
447 }
448 
449 static inline int kernfs_rename(struct kernfs_node *kn,
450 				struct kernfs_node *new_parent,
451 				const char *new_name)
452 {
453 	return kernfs_rename_ns(kn, new_parent, new_name, NULL);
454 }
455 
456 static inline struct dentry *
457 kernfs_mount(struct file_system_type *fs_type, int flags,
458 		struct kernfs_root *root, unsigned long magic,
459 		bool *new_sb_created)
460 {
461 	return kernfs_mount_ns(fs_type, flags, root,
462 				magic, new_sb_created, NULL);
463 }
464 
465 #endif	/* __LINUX_KERNFS_H */
466