xref: /openbmc/linux/include/linux/kernfs.h (revision 37744fee)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * kernfs.h - pseudo filesystem decoupled from vfs locking
4  */
5 
6 #ifndef __LINUX_KERNFS_H
7 #define __LINUX_KERNFS_H
8 
9 #include <linux/kernel.h>
10 #include <linux/err.h>
11 #include <linux/list.h>
12 #include <linux/mutex.h>
13 #include <linux/idr.h>
14 #include <linux/lockdep.h>
15 #include <linux/rbtree.h>
16 #include <linux/atomic.h>
17 #include <linux/uidgid.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 struct poll_table_struct;
28 struct fs_context;
29 
30 struct kernfs_fs_context;
31 struct kernfs_open_node;
32 struct kernfs_iattrs;
33 
34 enum kernfs_node_type {
35 	KERNFS_DIR		= 0x0001,
36 	KERNFS_FILE		= 0x0002,
37 	KERNFS_LINK		= 0x0004,
38 };
39 
40 #define KERNFS_TYPE_MASK		0x000f
41 #define KERNFS_FLAG_MASK		~KERNFS_TYPE_MASK
42 #define KERNFS_MAX_USER_XATTRS		128
43 #define KERNFS_USER_XATTR_SIZE_LIMIT	(128 << 10)
44 
45 enum kernfs_node_flag {
46 	KERNFS_ACTIVATED	= 0x0010,
47 	KERNFS_NS		= 0x0020,
48 	KERNFS_HAS_SEQ_SHOW	= 0x0040,
49 	KERNFS_HAS_MMAP		= 0x0080,
50 	KERNFS_LOCKDEP		= 0x0100,
51 	KERNFS_SUICIDAL		= 0x0400,
52 	KERNFS_SUICIDED		= 0x0800,
53 	KERNFS_EMPTY_DIR	= 0x1000,
54 	KERNFS_HAS_RELEASE	= 0x2000,
55 };
56 
57 /* @flags for kernfs_create_root() */
58 enum kernfs_root_flag {
59 	/*
60 	 * kernfs_nodes are created in the deactivated state and invisible.
61 	 * They require explicit kernfs_activate() to become visible.  This
62 	 * can be used to make related nodes become visible atomically
63 	 * after all nodes are created successfully.
64 	 */
65 	KERNFS_ROOT_CREATE_DEACTIVATED		= 0x0001,
66 
67 	/*
68 	 * For regular files, if the opener has CAP_DAC_OVERRIDE, open(2)
69 	 * succeeds regardless of the RW permissions.  sysfs had an extra
70 	 * layer of enforcement where open(2) fails with -EACCES regardless
71 	 * of CAP_DAC_OVERRIDE if the permission doesn't have the
72 	 * respective read or write access at all (none of S_IRUGO or
73 	 * S_IWUGO) or the respective operation isn't implemented.  The
74 	 * following flag enables that behavior.
75 	 */
76 	KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK	= 0x0002,
77 
78 	/*
79 	 * The filesystem supports exportfs operation, so userspace can use
80 	 * fhandle to access nodes of the fs.
81 	 */
82 	KERNFS_ROOT_SUPPORT_EXPORTOP		= 0x0004,
83 
84 	/*
85 	 * Support user xattrs to be written to nodes rooted at this root.
86 	 */
87 	KERNFS_ROOT_SUPPORT_USER_XATTR		= 0x0008,
88 };
89 
90 /* type-specific structures for kernfs_node union members */
91 struct kernfs_elem_dir {
92 	unsigned long		subdirs;
93 	/* children rbtree starts here and goes through kn->rb */
94 	struct rb_root		children;
95 
96 	/*
97 	 * The kernfs hierarchy this directory belongs to.  This fits
98 	 * better directly in kernfs_node but is here to save space.
99 	 */
100 	struct kernfs_root	*root;
101 };
102 
103 struct kernfs_elem_symlink {
104 	struct kernfs_node	*target_kn;
105 };
106 
107 struct kernfs_elem_attr {
108 	const struct kernfs_ops	*ops;
109 	struct kernfs_open_node	*open;
110 	loff_t			size;
111 	struct kernfs_node	*notify_next;	/* for kernfs_notify() */
112 };
113 
114 /*
115  * kernfs_node - the building block of kernfs hierarchy.  Each and every
116  * kernfs node is represented by single kernfs_node.  Most fields are
117  * private to kernfs and shouldn't be accessed directly by kernfs users.
118  *
119  * As long as s_count reference is held, the kernfs_node itself is
120  * accessible.  Dereferencing elem or any other outer entity requires
121  * active reference.
122  */
123 struct kernfs_node {
124 	atomic_t		count;
125 	atomic_t		active;
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 	struct lockdep_map	dep_map;
128 #endif
129 	/*
130 	 * Use kernfs_get_parent() and kernfs_name/path() instead of
131 	 * accessing the following two fields directly.  If the node is
132 	 * never moved to a different parent, it is safe to access the
133 	 * parent directly.
134 	 */
135 	struct kernfs_node	*parent;
136 	const char		*name;
137 
138 	struct rb_node		rb;
139 
140 	const void		*ns;	/* namespace tag */
141 	unsigned int		hash;	/* ns + name hash */
142 	union {
143 		struct kernfs_elem_dir		dir;
144 		struct kernfs_elem_symlink	symlink;
145 		struct kernfs_elem_attr		attr;
146 	};
147 
148 	void			*priv;
149 
150 	/*
151 	 * 64bit unique ID.  On 64bit ino setups, id is the ino.  On 32bit,
152 	 * the low 32bits are ino and upper generation.
153 	 */
154 	u64			id;
155 
156 	unsigned short		flags;
157 	umode_t			mode;
158 	struct kernfs_iattrs	*iattr;
159 };
160 
161 /*
162  * kernfs_syscall_ops may be specified on kernfs_create_root() to support
163  * syscalls.  These optional callbacks are invoked on the matching syscalls
164  * and can perform any kernfs operations which don't necessarily have to be
165  * the exact operation requested.  An active reference is held for each
166  * kernfs_node parameter.
167  */
168 struct kernfs_syscall_ops {
169 	int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
170 
171 	int (*mkdir)(struct kernfs_node *parent, const char *name,
172 		     umode_t mode);
173 	int (*rmdir)(struct kernfs_node *kn);
174 	int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
175 		      const char *new_name);
176 	int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
177 			 struct kernfs_root *root);
178 };
179 
180 struct kernfs_root {
181 	/* published fields */
182 	struct kernfs_node	*kn;
183 	unsigned int		flags;	/* KERNFS_ROOT_* flags */
184 
185 	/* private fields, do not use outside kernfs proper */
186 	struct idr		ino_idr;
187 	u32			last_id_lowbits;
188 	u32			id_highbits;
189 	struct kernfs_syscall_ops *syscall_ops;
190 
191 	/* list of kernfs_super_info of this root, protected by kernfs_mutex */
192 	struct list_head	supers;
193 
194 	wait_queue_head_t	deactivate_waitq;
195 };
196 
197 struct kernfs_open_file {
198 	/* published fields */
199 	struct kernfs_node	*kn;
200 	struct file		*file;
201 	struct seq_file		*seq_file;
202 	void			*priv;
203 
204 	/* private fields, do not use outside kernfs proper */
205 	struct mutex		mutex;
206 	struct mutex		prealloc_mutex;
207 	int			event;
208 	struct list_head	list;
209 	char			*prealloc_buf;
210 
211 	size_t			atomic_write_len;
212 	bool			mmapped:1;
213 	bool			released:1;
214 	const struct vm_operations_struct *vm_ops;
215 };
216 
217 struct kernfs_ops {
218 	/*
219 	 * Optional open/release methods.  Both are called with
220 	 * @of->seq_file populated.
221 	 */
222 	int (*open)(struct kernfs_open_file *of);
223 	void (*release)(struct kernfs_open_file *of);
224 
225 	/*
226 	 * Read is handled by either seq_file or raw_read().
227 	 *
228 	 * If seq_show() is present, seq_file path is active.  Other seq
229 	 * operations are optional and if not implemented, the behavior is
230 	 * equivalent to single_open().  @sf->private points to the
231 	 * associated kernfs_open_file.
232 	 *
233 	 * read() is bounced through kernel buffer and a read larger than
234 	 * PAGE_SIZE results in partial operation of PAGE_SIZE.
235 	 */
236 	int (*seq_show)(struct seq_file *sf, void *v);
237 
238 	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
239 	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
240 	void (*seq_stop)(struct seq_file *sf, void *v);
241 
242 	ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
243 			loff_t off);
244 
245 	/*
246 	 * write() is bounced through kernel buffer.  If atomic_write_len
247 	 * is not set, a write larger than PAGE_SIZE results in partial
248 	 * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
249 	 * writes upto the specified size are executed atomically but
250 	 * larger ones are rejected with -E2BIG.
251 	 */
252 	size_t atomic_write_len;
253 	/*
254 	 * "prealloc" causes a buffer to be allocated at open for
255 	 * all read/write requests.  As ->seq_show uses seq_read()
256 	 * which does its own allocation, it is incompatible with
257 	 * ->prealloc.  Provide ->read and ->write with ->prealloc.
258 	 */
259 	bool prealloc;
260 	ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
261 			 loff_t off);
262 
263 	__poll_t (*poll)(struct kernfs_open_file *of,
264 			 struct poll_table_struct *pt);
265 
266 	int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
267 
268 #ifdef CONFIG_DEBUG_LOCK_ALLOC
269 	struct lock_class_key	lockdep_key;
270 #endif
271 };
272 
273 /*
274  * The kernfs superblock creation/mount parameter context.
275  */
276 struct kernfs_fs_context {
277 	struct kernfs_root	*root;		/* Root of the hierarchy being mounted */
278 	void			*ns_tag;	/* Namespace tag of the mount (or NULL) */
279 	unsigned long		magic;		/* File system specific magic number */
280 
281 	/* The following are set/used by kernfs_mount() */
282 	bool			new_sb_created;	/* Set to T if we allocated a new sb */
283 };
284 
285 #ifdef CONFIG_KERNFS
286 
287 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
288 {
289 	return kn->flags & KERNFS_TYPE_MASK;
290 }
291 
292 static inline ino_t kernfs_id_ino(u64 id)
293 {
294 	/* id is ino if ino_t is 64bit; otherwise, low 32bits */
295 	if (sizeof(ino_t) >= sizeof(u64))
296 		return id;
297 	else
298 		return (u32)id;
299 }
300 
301 static inline u32 kernfs_id_gen(u64 id)
302 {
303 	/* gen is fixed at 1 if ino_t is 64bit; otherwise, high 32bits */
304 	if (sizeof(ino_t) >= sizeof(u64))
305 		return 1;
306 	else
307 		return id >> 32;
308 }
309 
310 static inline ino_t kernfs_ino(struct kernfs_node *kn)
311 {
312 	return kernfs_id_ino(kn->id);
313 }
314 
315 static inline ino_t kernfs_gen(struct kernfs_node *kn)
316 {
317 	return kernfs_id_gen(kn->id);
318 }
319 
320 /**
321  * kernfs_enable_ns - enable namespace under a directory
322  * @kn: directory of interest, should be empty
323  *
324  * This is to be called right after @kn is created to enable namespace
325  * under it.  All children of @kn must have non-NULL namespace tags and
326  * only the ones which match the super_block's tag will be visible.
327  */
328 static inline void kernfs_enable_ns(struct kernfs_node *kn)
329 {
330 	WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
331 	WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
332 	kn->flags |= KERNFS_NS;
333 }
334 
335 /**
336  * kernfs_ns_enabled - test whether namespace is enabled
337  * @kn: the node to test
338  *
339  * Test whether namespace filtering is enabled for the children of @ns.
340  */
341 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
342 {
343 	return kn->flags & KERNFS_NS;
344 }
345 
346 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
347 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
348 			  char *buf, size_t buflen);
349 void pr_cont_kernfs_name(struct kernfs_node *kn);
350 void pr_cont_kernfs_path(struct kernfs_node *kn);
351 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
352 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
353 					   const char *name, const void *ns);
354 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
355 					   const char *path, const void *ns);
356 void kernfs_get(struct kernfs_node *kn);
357 void kernfs_put(struct kernfs_node *kn);
358 
359 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
360 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
361 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
362 
363 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
364 				  struct super_block *sb);
365 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
366 				       unsigned int flags, void *priv);
367 void kernfs_destroy_root(struct kernfs_root *root);
368 
369 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
370 					 const char *name, umode_t mode,
371 					 kuid_t uid, kgid_t gid,
372 					 void *priv, const void *ns);
373 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
374 					    const char *name);
375 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
376 					 const char *name, umode_t mode,
377 					 kuid_t uid, kgid_t gid,
378 					 loff_t size,
379 					 const struct kernfs_ops *ops,
380 					 void *priv, const void *ns,
381 					 struct lock_class_key *key);
382 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
383 				       const char *name,
384 				       struct kernfs_node *target);
385 void kernfs_activate(struct kernfs_node *kn);
386 void kernfs_remove(struct kernfs_node *kn);
387 void kernfs_break_active_protection(struct kernfs_node *kn);
388 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
389 bool kernfs_remove_self(struct kernfs_node *kn);
390 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
391 			     const void *ns);
392 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
393 		     const char *new_name, const void *new_ns);
394 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
395 __poll_t kernfs_generic_poll(struct kernfs_open_file *of,
396 			     struct poll_table_struct *pt);
397 void kernfs_notify(struct kernfs_node *kn);
398 
399 int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
400 		     void *value, size_t size);
401 int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
402 		     const void *value, size_t size, int flags);
403 
404 const void *kernfs_super_ns(struct super_block *sb);
405 int kernfs_get_tree(struct fs_context *fc);
406 void kernfs_free_fs_context(struct fs_context *fc);
407 void kernfs_kill_sb(struct super_block *sb);
408 
409 void kernfs_init(void);
410 
411 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
412 						   u64 id);
413 #else	/* CONFIG_KERNFS */
414 
415 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
416 { return 0; }	/* whatever */
417 
418 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
419 
420 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
421 { return false; }
422 
423 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
424 { return -ENOSYS; }
425 
426 static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
427 					struct kernfs_node *kn,
428 					char *buf, size_t buflen)
429 { return -ENOSYS; }
430 
431 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
432 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
433 
434 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
435 { return NULL; }
436 
437 static inline struct kernfs_node *
438 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
439 		       const void *ns)
440 { return NULL; }
441 static inline struct kernfs_node *
442 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
443 		       const void *ns)
444 { return NULL; }
445 
446 static inline void kernfs_get(struct kernfs_node *kn) { }
447 static inline void kernfs_put(struct kernfs_node *kn) { }
448 
449 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
450 { return NULL; }
451 
452 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
453 { return NULL; }
454 
455 static inline struct inode *
456 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
457 { return NULL; }
458 
459 static inline struct kernfs_root *
460 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
461 		   void *priv)
462 { return ERR_PTR(-ENOSYS); }
463 
464 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
465 
466 static inline struct kernfs_node *
467 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
468 		     umode_t mode, kuid_t uid, kgid_t gid,
469 		     void *priv, const void *ns)
470 { return ERR_PTR(-ENOSYS); }
471 
472 static inline struct kernfs_node *
473 __kernfs_create_file(struct kernfs_node *parent, const char *name,
474 		     umode_t mode, kuid_t uid, kgid_t gid,
475 		     loff_t size, const struct kernfs_ops *ops,
476 		     void *priv, const void *ns, struct lock_class_key *key)
477 { return ERR_PTR(-ENOSYS); }
478 
479 static inline struct kernfs_node *
480 kernfs_create_link(struct kernfs_node *parent, const char *name,
481 		   struct kernfs_node *target)
482 { return ERR_PTR(-ENOSYS); }
483 
484 static inline void kernfs_activate(struct kernfs_node *kn) { }
485 
486 static inline void kernfs_remove(struct kernfs_node *kn) { }
487 
488 static inline bool kernfs_remove_self(struct kernfs_node *kn)
489 { return false; }
490 
491 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
492 					   const char *name, const void *ns)
493 { return -ENOSYS; }
494 
495 static inline int kernfs_rename_ns(struct kernfs_node *kn,
496 				   struct kernfs_node *new_parent,
497 				   const char *new_name, const void *new_ns)
498 { return -ENOSYS; }
499 
500 static inline int kernfs_setattr(struct kernfs_node *kn,
501 				 const struct iattr *iattr)
502 { return -ENOSYS; }
503 
504 static inline void kernfs_notify(struct kernfs_node *kn) { }
505 
506 static inline int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
507 				   void *value, size_t size)
508 { return -ENOSYS; }
509 
510 static inline int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
511 				   const void *value, size_t size, int flags)
512 { return -ENOSYS; }
513 
514 static inline const void *kernfs_super_ns(struct super_block *sb)
515 { return NULL; }
516 
517 static inline int kernfs_get_tree(struct fs_context *fc)
518 { return -ENOSYS; }
519 
520 static inline void kernfs_free_fs_context(struct fs_context *fc) { }
521 
522 static inline void kernfs_kill_sb(struct super_block *sb) { }
523 
524 static inline void kernfs_init(void) { }
525 
526 #endif	/* CONFIG_KERNFS */
527 
528 /**
529  * kernfs_path - build full path of a given node
530  * @kn: kernfs_node of interest
531  * @buf: buffer to copy @kn's name into
532  * @buflen: size of @buf
533  *
534  * If @kn is NULL result will be "(null)".
535  *
536  * Returns the length of the full path.  If the full length is equal to or
537  * greater than @buflen, @buf contains the truncated path with the trailing
538  * '\0'.  On error, -errno is returned.
539  */
540 static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
541 {
542 	return kernfs_path_from_node(kn, NULL, buf, buflen);
543 }
544 
545 static inline struct kernfs_node *
546 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
547 {
548 	return kernfs_find_and_get_ns(kn, name, NULL);
549 }
550 
551 static inline struct kernfs_node *
552 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
553 {
554 	return kernfs_walk_and_get_ns(kn, path, NULL);
555 }
556 
557 static inline struct kernfs_node *
558 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
559 		  void *priv)
560 {
561 	return kernfs_create_dir_ns(parent, name, mode,
562 				    GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
563 				    priv, NULL);
564 }
565 
566 static inline struct kernfs_node *
567 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
568 		      umode_t mode, kuid_t uid, kgid_t gid,
569 		      loff_t size, const struct kernfs_ops *ops,
570 		      void *priv, const void *ns)
571 {
572 	struct lock_class_key *key = NULL;
573 
574 #ifdef CONFIG_DEBUG_LOCK_ALLOC
575 	key = (struct lock_class_key *)&ops->lockdep_key;
576 #endif
577 	return __kernfs_create_file(parent, name, mode, uid, gid,
578 				    size, ops, priv, ns, key);
579 }
580 
581 static inline struct kernfs_node *
582 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
583 		   loff_t size, const struct kernfs_ops *ops, void *priv)
584 {
585 	return kernfs_create_file_ns(parent, name, mode,
586 				     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
587 				     size, ops, priv, NULL);
588 }
589 
590 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
591 					const char *name)
592 {
593 	return kernfs_remove_by_name_ns(parent, name, NULL);
594 }
595 
596 static inline int kernfs_rename(struct kernfs_node *kn,
597 				struct kernfs_node *new_parent,
598 				const char *new_name)
599 {
600 	return kernfs_rename_ns(kn, new_parent, new_name, NULL);
601 }
602 
603 #endif	/* __LINUX_KERNFS_H */
604