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/err.h>
10#include <linux/list.h>
11#include <linux/mutex.h>
12#include <linux/idr.h>
13#include <linux/lockdep.h>
14#include <linux/rbtree.h>
15#include <linux/atomic.h>
16#include <linux/bug.h>
17#include <linux/types.h>
18#include <linux/uidgid.h>
19#include <linux/wait.h>
20#include <linux/rwsem.h>
21
22struct file;
23struct dentry;
24struct iattr;
25struct seq_file;
26struct vm_area_struct;
27struct vm_operations_struct;
28struct super_block;
29struct file_system_type;
30struct poll_table_struct;
31struct fs_context;
32
33struct kernfs_fs_context;
34struct kernfs_open_node;
35struct kernfs_iattrs;
36
37enum kernfs_node_type {
38	KERNFS_DIR		= 0x0001,
39	KERNFS_FILE		= 0x0002,
40	KERNFS_LINK		= 0x0004,
41};
42
43#define KERNFS_TYPE_MASK		0x000f
44#define KERNFS_FLAG_MASK		~KERNFS_TYPE_MASK
45#define KERNFS_MAX_USER_XATTRS		128
46#define KERNFS_USER_XATTR_SIZE_LIMIT	(128 << 10)
47
48enum kernfs_node_flag {
49	KERNFS_ACTIVATED	= 0x0010,
50	KERNFS_NS		= 0x0020,
51	KERNFS_HAS_SEQ_SHOW	= 0x0040,
52	KERNFS_HAS_MMAP		= 0x0080,
53	KERNFS_LOCKDEP		= 0x0100,
54	KERNFS_SUICIDAL		= 0x0400,
55	KERNFS_SUICIDED		= 0x0800,
56	KERNFS_EMPTY_DIR	= 0x1000,
57	KERNFS_HAS_RELEASE	= 0x2000,
58};
59
60/* @flags for kernfs_create_root() */
61enum kernfs_root_flag {
62	/*
63	 * kernfs_nodes are created in the deactivated state and invisible.
64	 * They require explicit kernfs_activate() to become visible.  This
65	 * can be used to make related nodes become visible atomically
66	 * after all nodes are created successfully.
67	 */
68	KERNFS_ROOT_CREATE_DEACTIVATED		= 0x0001,
69
70	/*
71	 * For regular files, if the opener has CAP_DAC_OVERRIDE, open(2)
72	 * succeeds regardless of the RW permissions.  sysfs had an extra
73	 * layer of enforcement where open(2) fails with -EACCES regardless
74	 * of CAP_DAC_OVERRIDE if the permission doesn't have the
75	 * respective read or write access at all (none of S_IRUGO or
76	 * S_IWUGO) or the respective operation isn't implemented.  The
77	 * following flag enables that behavior.
78	 */
79	KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK	= 0x0002,
80
81	/*
82	 * The filesystem supports exportfs operation, so userspace can use
83	 * fhandle to access nodes of the fs.
84	 */
85	KERNFS_ROOT_SUPPORT_EXPORTOP		= 0x0004,
86
87	/*
88	 * Support user xattrs to be written to nodes rooted at this root.
89	 */
90	KERNFS_ROOT_SUPPORT_USER_XATTR		= 0x0008,
91};
92
93/* type-specific structures for kernfs_node union members */
94struct kernfs_elem_dir {
95	unsigned long		subdirs;
96	/* children rbtree starts here and goes through kn->rb */
97	struct rb_root		children;
98
99	/*
100	 * The kernfs hierarchy this directory belongs to.  This fits
101	 * better directly in kernfs_node but is here to save space.
102	 */
103	struct kernfs_root	*root;
104	/*
105	 * Monotonic revision counter, used to identify if a directory
106	 * node has changed during negative dentry revalidation.
107	 */
108	unsigned long		rev;
109};
110
111struct kernfs_elem_symlink {
112	struct kernfs_node	*target_kn;
113};
114
115struct kernfs_elem_attr {
116	const struct kernfs_ops	*ops;
117	struct kernfs_open_node	*open;
118	loff_t			size;
119	struct kernfs_node	*notify_next;	/* for kernfs_notify() */
120};
121
122/*
123 * kernfs_node - the building block of kernfs hierarchy.  Each and every
124 * kernfs node is represented by single kernfs_node.  Most fields are
125 * private to kernfs and shouldn't be accessed directly by kernfs users.
126 *
127 * As long as count reference is held, the kernfs_node itself is
128 * accessible.  Dereferencing elem or any other outer entity requires
129 * active reference.
130 */
131struct kernfs_node {
132	atomic_t		count;
133	atomic_t		active;
134#ifdef CONFIG_DEBUG_LOCK_ALLOC
135	struct lockdep_map	dep_map;
136#endif
137	/*
138	 * Use kernfs_get_parent() and kernfs_name/path() instead of
139	 * accessing the following two fields directly.  If the node is
140	 * never moved to a different parent, it is safe to access the
141	 * parent directly.
142	 */
143	struct kernfs_node	*parent;
144	const char		*name;
145
146	struct rb_node		rb;
147
148	const void		*ns;	/* namespace tag */
149	unsigned int		hash;	/* ns + name hash */
150	union {
151		struct kernfs_elem_dir		dir;
152		struct kernfs_elem_symlink	symlink;
153		struct kernfs_elem_attr		attr;
154	};
155
156	void			*priv;
157
158	/*
159	 * 64bit unique ID.  On 64bit ino setups, id is the ino.  On 32bit,
160	 * the low 32bits are ino and upper generation.
161	 */
162	u64			id;
163
164	unsigned short		flags;
165	umode_t			mode;
166	struct kernfs_iattrs	*iattr;
167};
168
169/*
170 * kernfs_syscall_ops may be specified on kernfs_create_root() to support
171 * syscalls.  These optional callbacks are invoked on the matching syscalls
172 * and can perform any kernfs operations which don't necessarily have to be
173 * the exact operation requested.  An active reference is held for each
174 * kernfs_node parameter.
175 */
176struct kernfs_syscall_ops {
177	int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
178
179	int (*mkdir)(struct kernfs_node *parent, const char *name,
180		     umode_t mode);
181	int (*rmdir)(struct kernfs_node *kn);
182	int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
183		      const char *new_name);
184	int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
185			 struct kernfs_root *root);
186};
187
188struct kernfs_root {
189	/* published fields */
190	struct kernfs_node	*kn;
191	unsigned int		flags;	/* KERNFS_ROOT_* flags */
192
193	/* private fields, do not use outside kernfs proper */
194	struct idr		ino_idr;
195	u32			last_id_lowbits;
196	u32			id_highbits;
197	struct kernfs_syscall_ops *syscall_ops;
198
199	/* list of kernfs_super_info of this root, protected by kernfs_rwsem */
200	struct list_head	supers;
201
202	wait_queue_head_t	deactivate_waitq;
203	struct rw_semaphore	kernfs_rwsem;
204};
205
206struct kernfs_open_file {
207	/* published fields */
208	struct kernfs_node	*kn;
209	struct file		*file;
210	struct seq_file		*seq_file;
211	void			*priv;
212
213	/* private fields, do not use outside kernfs proper */
214	struct mutex		mutex;
215	struct mutex		prealloc_mutex;
216	int			event;
217	struct list_head	list;
218	char			*prealloc_buf;
219
220	size_t			atomic_write_len;
221	bool			mmapped:1;
222	bool			released:1;
223	const struct vm_operations_struct *vm_ops;
224};
225
226struct kernfs_ops {
227	/*
228	 * Optional open/release methods.  Both are called with
229	 * @of->seq_file populated.
230	 */
231	int (*open)(struct kernfs_open_file *of);
232	void (*release)(struct kernfs_open_file *of);
233
234	/*
235	 * Read is handled by either seq_file or raw_read().
236	 *
237	 * If seq_show() is present, seq_file path is active.  Other seq
238	 * operations are optional and if not implemented, the behavior is
239	 * equivalent to single_open().  @sf->private points to the
240	 * associated kernfs_open_file.
241	 *
242	 * read() is bounced through kernel buffer and a read larger than
243	 * PAGE_SIZE results in partial operation of PAGE_SIZE.
244	 */
245	int (*seq_show)(struct seq_file *sf, void *v);
246
247	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
248	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
249	void (*seq_stop)(struct seq_file *sf, void *v);
250
251	ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
252			loff_t off);
253
254	/*
255	 * write() is bounced through kernel buffer.  If atomic_write_len
256	 * is not set, a write larger than PAGE_SIZE results in partial
257	 * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
258	 * writes upto the specified size are executed atomically but
259	 * larger ones are rejected with -E2BIG.
260	 */
261	size_t atomic_write_len;
262	/*
263	 * "prealloc" causes a buffer to be allocated at open for
264	 * all read/write requests.  As ->seq_show uses seq_read()
265	 * which does its own allocation, it is incompatible with
266	 * ->prealloc.  Provide ->read and ->write with ->prealloc.
267	 */
268	bool prealloc;
269	ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
270			 loff_t off);
271
272	__poll_t (*poll)(struct kernfs_open_file *of,
273			 struct poll_table_struct *pt);
274
275	int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
276};
277
278/*
279 * The kernfs superblock creation/mount parameter context.
280 */
281struct kernfs_fs_context {
282	struct kernfs_root	*root;		/* Root of the hierarchy being mounted */
283	void			*ns_tag;	/* Namespace tag of the mount (or NULL) */
284	unsigned long		magic;		/* File system specific magic number */
285
286	/* The following are set/used by kernfs_mount() */
287	bool			new_sb_created;	/* Set to T if we allocated a new sb */
288};
289
290#ifdef CONFIG_KERNFS
291
292static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
293{
294	return kn->flags & KERNFS_TYPE_MASK;
295}
296
297static inline ino_t kernfs_id_ino(u64 id)
298{
299	/* id is ino if ino_t is 64bit; otherwise, low 32bits */
300	if (sizeof(ino_t) >= sizeof(u64))
301		return id;
302	else
303		return (u32)id;
304}
305
306static inline u32 kernfs_id_gen(u64 id)
307{
308	/* gen is fixed at 1 if ino_t is 64bit; otherwise, high 32bits */
309	if (sizeof(ino_t) >= sizeof(u64))
310		return 1;
311	else
312		return id >> 32;
313}
314
315static inline ino_t kernfs_ino(struct kernfs_node *kn)
316{
317	return kernfs_id_ino(kn->id);
318}
319
320static inline ino_t kernfs_gen(struct kernfs_node *kn)
321{
322	return kernfs_id_gen(kn->id);
323}
324
325/**
326 * kernfs_enable_ns - enable namespace under a directory
327 * @kn: directory of interest, should be empty
328 *
329 * This is to be called right after @kn is created to enable namespace
330 * under it.  All children of @kn must have non-NULL namespace tags and
331 * only the ones which match the super_block's tag will be visible.
332 */
333static inline void kernfs_enable_ns(struct kernfs_node *kn)
334{
335	WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
336	WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
337	kn->flags |= KERNFS_NS;
338}
339
340/**
341 * kernfs_ns_enabled - test whether namespace is enabled
342 * @kn: the node to test
343 *
344 * Test whether namespace filtering is enabled for the children of @ns.
345 */
346static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
347{
348	return kn->flags & KERNFS_NS;
349}
350
351int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
352int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
353			  char *buf, size_t buflen);
354void pr_cont_kernfs_name(struct kernfs_node *kn);
355void pr_cont_kernfs_path(struct kernfs_node *kn);
356struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
357struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
358					   const char *name, const void *ns);
359struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
360					   const char *path, const void *ns);
361void kernfs_get(struct kernfs_node *kn);
362void kernfs_put(struct kernfs_node *kn);
363
364struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
365struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
366struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
367
368struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
369				  struct super_block *sb);
370struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
371				       unsigned int flags, void *priv);
372void kernfs_destroy_root(struct kernfs_root *root);
373
374struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
375					 const char *name, umode_t mode,
376					 kuid_t uid, kgid_t gid,
377					 void *priv, const void *ns);
378struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
379					    const char *name);
380struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
381					 const char *name, umode_t mode,
382					 kuid_t uid, kgid_t gid,
383					 loff_t size,
384					 const struct kernfs_ops *ops,
385					 void *priv, const void *ns,
386					 struct lock_class_key *key);
387struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
388				       const char *name,
389				       struct kernfs_node *target);
390void kernfs_activate(struct kernfs_node *kn);
391void kernfs_remove(struct kernfs_node *kn);
392void kernfs_break_active_protection(struct kernfs_node *kn);
393void kernfs_unbreak_active_protection(struct kernfs_node *kn);
394bool kernfs_remove_self(struct kernfs_node *kn);
395int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
396			     const void *ns);
397int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
398		     const char *new_name, const void *new_ns);
399int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
400__poll_t kernfs_generic_poll(struct kernfs_open_file *of,
401			     struct poll_table_struct *pt);
402void kernfs_notify(struct kernfs_node *kn);
403
404int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
405		     void *value, size_t size);
406int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
407		     const void *value, size_t size, int flags);
408
409const void *kernfs_super_ns(struct super_block *sb);
410int kernfs_get_tree(struct fs_context *fc);
411void kernfs_free_fs_context(struct fs_context *fc);
412void kernfs_kill_sb(struct super_block *sb);
413
414void kernfs_init(void);
415
416struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
417						   u64 id);
418#else	/* CONFIG_KERNFS */
419
420static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
421{ return 0; }	/* whatever */
422
423static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
424
425static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
426{ return false; }
427
428static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
429{ return -ENOSYS; }
430
431static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
432					struct kernfs_node *kn,
433					char *buf, size_t buflen)
434{ return -ENOSYS; }
435
436static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
437static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
438
439static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
440{ return NULL; }
441
442static inline struct kernfs_node *
443kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
444		       const void *ns)
445{ return NULL; }
446static inline struct kernfs_node *
447kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
448		       const void *ns)
449{ return NULL; }
450
451static inline void kernfs_get(struct kernfs_node *kn) { }
452static inline void kernfs_put(struct kernfs_node *kn) { }
453
454static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
455{ return NULL; }
456
457static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
458{ return NULL; }
459
460static inline struct inode *
461kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
462{ return NULL; }
463
464static inline struct kernfs_root *
465kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
466		   void *priv)
467{ return ERR_PTR(-ENOSYS); }
468
469static inline void kernfs_destroy_root(struct kernfs_root *root) { }
470
471static inline struct kernfs_node *
472kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
473		     umode_t mode, kuid_t uid, kgid_t gid,
474		     void *priv, const void *ns)
475{ return ERR_PTR(-ENOSYS); }
476
477static inline struct kernfs_node *
478__kernfs_create_file(struct kernfs_node *parent, const char *name,
479		     umode_t mode, kuid_t uid, kgid_t gid,
480		     loff_t size, const struct kernfs_ops *ops,
481		     void *priv, const void *ns, struct lock_class_key *key)
482{ return ERR_PTR(-ENOSYS); }
483
484static inline struct kernfs_node *
485kernfs_create_link(struct kernfs_node *parent, const char *name,
486		   struct kernfs_node *target)
487{ return ERR_PTR(-ENOSYS); }
488
489static inline void kernfs_activate(struct kernfs_node *kn) { }
490
491static inline void kernfs_remove(struct kernfs_node *kn) { }
492
493static inline bool kernfs_remove_self(struct kernfs_node *kn)
494{ return false; }
495
496static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
497					   const char *name, const void *ns)
498{ return -ENOSYS; }
499
500static inline int kernfs_rename_ns(struct kernfs_node *kn,
501				   struct kernfs_node *new_parent,
502				   const char *new_name, const void *new_ns)
503{ return -ENOSYS; }
504
505static inline int kernfs_setattr(struct kernfs_node *kn,
506				 const struct iattr *iattr)
507{ return -ENOSYS; }
508
509static inline void kernfs_notify(struct kernfs_node *kn) { }
510
511static inline int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
512				   void *value, size_t size)
513{ return -ENOSYS; }
514
515static inline int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
516				   const void *value, size_t size, int flags)
517{ return -ENOSYS; }
518
519static inline const void *kernfs_super_ns(struct super_block *sb)
520{ return NULL; }
521
522static inline int kernfs_get_tree(struct fs_context *fc)
523{ return -ENOSYS; }
524
525static inline void kernfs_free_fs_context(struct fs_context *fc) { }
526
527static inline void kernfs_kill_sb(struct super_block *sb) { }
528
529static inline void kernfs_init(void) { }
530
531#endif	/* CONFIG_KERNFS */
532
533/**
534 * kernfs_path - build full path of a given node
535 * @kn: kernfs_node of interest
536 * @buf: buffer to copy @kn's name into
537 * @buflen: size of @buf
538 *
539 * If @kn is NULL result will be "(null)".
540 *
541 * Returns the length of the full path.  If the full length is equal to or
542 * greater than @buflen, @buf contains the truncated path with the trailing
543 * '\0'.  On error, -errno is returned.
544 */
545static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
546{
547	return kernfs_path_from_node(kn, NULL, buf, buflen);
548}
549
550static inline struct kernfs_node *
551kernfs_find_and_get(struct kernfs_node *kn, const char *name)
552{
553	return kernfs_find_and_get_ns(kn, name, NULL);
554}
555
556static inline struct kernfs_node *
557kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
558{
559	return kernfs_walk_and_get_ns(kn, path, NULL);
560}
561
562static inline struct kernfs_node *
563kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
564		  void *priv)
565{
566	return kernfs_create_dir_ns(parent, name, mode,
567				    GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
568				    priv, NULL);
569}
570
571static inline int kernfs_remove_by_name(struct kernfs_node *parent,
572					const char *name)
573{
574	return kernfs_remove_by_name_ns(parent, name, NULL);
575}
576
577static inline int kernfs_rename(struct kernfs_node *kn,
578				struct kernfs_node *new_parent,
579				const char *new_name)
580{
581	return kernfs_rename_ns(kn, new_parent, new_name, NULL);
582}
583
584#endif	/* __LINUX_KERNFS_H */
585