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#include <linux/cache.h>
22
23struct file;
24struct dentry;
25struct iattr;
26struct seq_file;
27struct vm_area_struct;
28struct vm_operations_struct;
29struct super_block;
30struct file_system_type;
31struct poll_table_struct;
32struct fs_context;
33
34struct kernfs_fs_context;
35struct kernfs_open_node;
36struct kernfs_iattrs;
37
38/*
39 * NR_KERNFS_LOCK_BITS determines size (NR_KERNFS_LOCKS) of hash
40 * table of locks.
41 * Having a small hash table would impact scalability, since
42 * more and more kernfs_node objects will end up using same lock
43 * and having a very large hash table would waste memory.
44 *
45 * At the moment size of hash table of locks is being set based on
46 * the number of CPUs as follows:
47 *
48 * NR_CPU      NR_KERNFS_LOCK_BITS      NR_KERNFS_LOCKS
49 *   1                  1                       2
50 *  2-3                 2                       4
51 *  4-7                 4                       16
52 *  8-15                6                       64
53 *  16-31               8                       256
54 *  32 and more         10                      1024
55 *
56 * The above relation between NR_CPU and number of locks is based
57 * on some internal experimentation which involved booting qemu
58 * with different values of smp, performing some sysfs operations
59 * on all CPUs and observing how increase in number of locks impacts
60 * completion time of these sysfs operations on each CPU.
61 */
62#ifdef CONFIG_SMP
63#define NR_KERNFS_LOCK_BITS (2 * (ilog2(NR_CPUS < 32 ? NR_CPUS : 32)))
64#else
65#define NR_KERNFS_LOCK_BITS     1
66#endif
67
68#define NR_KERNFS_LOCKS     (1 << NR_KERNFS_LOCK_BITS)
69
70/*
71 * There's one kernfs_open_file for each open file and one kernfs_open_node
72 * for each kernfs_node with one or more open files.
73 *
74 * filp->private_data points to seq_file whose ->private points to
75 * kernfs_open_file.
76 *
77 * kernfs_open_files are chained at kernfs_open_node->files, which is
78 * protected by kernfs_global_locks.open_file_mutex[i].
79 *
80 * To reduce possible contention in sysfs access, arising due to single
81 * locks, use an array of locks (e.g. open_file_mutex) and use kernfs_node
82 * object address as hash keys to get the index of these locks.
83 *
84 * Hashed mutexes are safe to use here because operations using these don't
85 * rely on global exclusion.
86 *
87 * In future we intend to replace other global locks with hashed ones as well.
88 * kernfs_global_locks acts as a holder for all such hash tables.
89 */
90struct kernfs_global_locks {
91	struct mutex open_file_mutex[NR_KERNFS_LOCKS];
92};
93
94enum kernfs_node_type {
95	KERNFS_DIR		= 0x0001,
96	KERNFS_FILE		= 0x0002,
97	KERNFS_LINK		= 0x0004,
98};
99
100#define KERNFS_TYPE_MASK		0x000f
101#define KERNFS_FLAG_MASK		~KERNFS_TYPE_MASK
102#define KERNFS_MAX_USER_XATTRS		128
103#define KERNFS_USER_XATTR_SIZE_LIMIT	(128 << 10)
104
105enum kernfs_node_flag {
106	KERNFS_ACTIVATED	= 0x0010,
107	KERNFS_NS		= 0x0020,
108	KERNFS_HAS_SEQ_SHOW	= 0x0040,
109	KERNFS_HAS_MMAP		= 0x0080,
110	KERNFS_LOCKDEP		= 0x0100,
111	KERNFS_SUICIDAL		= 0x0400,
112	KERNFS_SUICIDED		= 0x0800,
113	KERNFS_EMPTY_DIR	= 0x1000,
114	KERNFS_HAS_RELEASE	= 0x2000,
115};
116
117/* @flags for kernfs_create_root() */
118enum kernfs_root_flag {
119	/*
120	 * kernfs_nodes are created in the deactivated state and invisible.
121	 * They require explicit kernfs_activate() to become visible.  This
122	 * can be used to make related nodes become visible atomically
123	 * after all nodes are created successfully.
124	 */
125	KERNFS_ROOT_CREATE_DEACTIVATED		= 0x0001,
126
127	/*
128	 * For regular files, if the opener has CAP_DAC_OVERRIDE, open(2)
129	 * succeeds regardless of the RW permissions.  sysfs had an extra
130	 * layer of enforcement where open(2) fails with -EACCES regardless
131	 * of CAP_DAC_OVERRIDE if the permission doesn't have the
132	 * respective read or write access at all (none of S_IRUGO or
133	 * S_IWUGO) or the respective operation isn't implemented.  The
134	 * following flag enables that behavior.
135	 */
136	KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK	= 0x0002,
137
138	/*
139	 * The filesystem supports exportfs operation, so userspace can use
140	 * fhandle to access nodes of the fs.
141	 */
142	KERNFS_ROOT_SUPPORT_EXPORTOP		= 0x0004,
143
144	/*
145	 * Support user xattrs to be written to nodes rooted at this root.
146	 */
147	KERNFS_ROOT_SUPPORT_USER_XATTR		= 0x0008,
148};
149
150/* type-specific structures for kernfs_node union members */
151struct kernfs_elem_dir {
152	unsigned long		subdirs;
153	/* children rbtree starts here and goes through kn->rb */
154	struct rb_root		children;
155
156	/*
157	 * The kernfs hierarchy this directory belongs to.  This fits
158	 * better directly in kernfs_node but is here to save space.
159	 */
160	struct kernfs_root	*root;
161	/*
162	 * Monotonic revision counter, used to identify if a directory
163	 * node has changed during negative dentry revalidation.
164	 */
165	unsigned long		rev;
166};
167
168struct kernfs_elem_symlink {
169	struct kernfs_node	*target_kn;
170};
171
172struct kernfs_elem_attr {
173	const struct kernfs_ops	*ops;
174	struct kernfs_open_node __rcu	*open;
175	loff_t			size;
176	struct kernfs_node	*notify_next;	/* for kernfs_notify() */
177};
178
179/*
180 * kernfs_node - the building block of kernfs hierarchy.  Each and every
181 * kernfs node is represented by single kernfs_node.  Most fields are
182 * private to kernfs and shouldn't be accessed directly by kernfs users.
183 *
184 * As long as count reference is held, the kernfs_node itself is
185 * accessible.  Dereferencing elem or any other outer entity requires
186 * active reference.
187 */
188struct kernfs_node {
189	atomic_t		count;
190	atomic_t		active;
191#ifdef CONFIG_DEBUG_LOCK_ALLOC
192	struct lockdep_map	dep_map;
193#endif
194	/*
195	 * Use kernfs_get_parent() and kernfs_name/path() instead of
196	 * accessing the following two fields directly.  If the node is
197	 * never moved to a different parent, it is safe to access the
198	 * parent directly.
199	 */
200	struct kernfs_node	*parent;
201	const char		*name;
202
203	struct rb_node		rb;
204
205	const void		*ns;	/* namespace tag */
206	unsigned int		hash;	/* ns + name hash */
207	union {
208		struct kernfs_elem_dir		dir;
209		struct kernfs_elem_symlink	symlink;
210		struct kernfs_elem_attr		attr;
211	};
212
213	void			*priv;
214
215	/*
216	 * 64bit unique ID.  On 64bit ino setups, id is the ino.  On 32bit,
217	 * the low 32bits are ino and upper generation.
218	 */
219	u64			id;
220
221	unsigned short		flags;
222	umode_t			mode;
223	struct kernfs_iattrs	*iattr;
224};
225
226/*
227 * kernfs_syscall_ops may be specified on kernfs_create_root() to support
228 * syscalls.  These optional callbacks are invoked on the matching syscalls
229 * and can perform any kernfs operations which don't necessarily have to be
230 * the exact operation requested.  An active reference is held for each
231 * kernfs_node parameter.
232 */
233struct kernfs_syscall_ops {
234	int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
235
236	int (*mkdir)(struct kernfs_node *parent, const char *name,
237		     umode_t mode);
238	int (*rmdir)(struct kernfs_node *kn);
239	int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
240		      const char *new_name);
241	int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
242			 struct kernfs_root *root);
243};
244
245struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root);
246
247struct kernfs_open_file {
248	/* published fields */
249	struct kernfs_node	*kn;
250	struct file		*file;
251	struct seq_file		*seq_file;
252	void			*priv;
253
254	/* private fields, do not use outside kernfs proper */
255	struct mutex		mutex;
256	struct mutex		prealloc_mutex;
257	int			event;
258	struct list_head	list;
259	char			*prealloc_buf;
260
261	size_t			atomic_write_len;
262	bool			mmapped:1;
263	bool			released:1;
264	const struct vm_operations_struct *vm_ops;
265};
266
267struct kernfs_ops {
268	/*
269	 * Optional open/release methods.  Both are called with
270	 * @of->seq_file populated.
271	 */
272	int (*open)(struct kernfs_open_file *of);
273	void (*release)(struct kernfs_open_file *of);
274
275	/*
276	 * Read is handled by either seq_file or raw_read().
277	 *
278	 * If seq_show() is present, seq_file path is active.  Other seq
279	 * operations are optional and if not implemented, the behavior is
280	 * equivalent to single_open().  @sf->private points to the
281	 * associated kernfs_open_file.
282	 *
283	 * read() is bounced through kernel buffer and a read larger than
284	 * PAGE_SIZE results in partial operation of PAGE_SIZE.
285	 */
286	int (*seq_show)(struct seq_file *sf, void *v);
287
288	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
289	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
290	void (*seq_stop)(struct seq_file *sf, void *v);
291
292	ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
293			loff_t off);
294
295	/*
296	 * write() is bounced through kernel buffer.  If atomic_write_len
297	 * is not set, a write larger than PAGE_SIZE results in partial
298	 * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
299	 * writes upto the specified size are executed atomically but
300	 * larger ones are rejected with -E2BIG.
301	 */
302	size_t atomic_write_len;
303	/*
304	 * "prealloc" causes a buffer to be allocated at open for
305	 * all read/write requests.  As ->seq_show uses seq_read()
306	 * which does its own allocation, it is incompatible with
307	 * ->prealloc.  Provide ->read and ->write with ->prealloc.
308	 */
309	bool prealloc;
310	ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
311			 loff_t off);
312
313	__poll_t (*poll)(struct kernfs_open_file *of,
314			 struct poll_table_struct *pt);
315
316	int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
317};
318
319/*
320 * The kernfs superblock creation/mount parameter context.
321 */
322struct kernfs_fs_context {
323	struct kernfs_root	*root;		/* Root of the hierarchy being mounted */
324	void			*ns_tag;	/* Namespace tag of the mount (or NULL) */
325	unsigned long		magic;		/* File system specific magic number */
326
327	/* The following are set/used by kernfs_mount() */
328	bool			new_sb_created;	/* Set to T if we allocated a new sb */
329};
330
331#ifdef CONFIG_KERNFS
332
333static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
334{
335	return kn->flags & KERNFS_TYPE_MASK;
336}
337
338static inline ino_t kernfs_id_ino(u64 id)
339{
340	/* id is ino if ino_t is 64bit; otherwise, low 32bits */
341	if (sizeof(ino_t) >= sizeof(u64))
342		return id;
343	else
344		return (u32)id;
345}
346
347static inline u32 kernfs_id_gen(u64 id)
348{
349	/* gen is fixed at 1 if ino_t is 64bit; otherwise, high 32bits */
350	if (sizeof(ino_t) >= sizeof(u64))
351		return 1;
352	else
353		return id >> 32;
354}
355
356static inline ino_t kernfs_ino(struct kernfs_node *kn)
357{
358	return kernfs_id_ino(kn->id);
359}
360
361static inline ino_t kernfs_gen(struct kernfs_node *kn)
362{
363	return kernfs_id_gen(kn->id);
364}
365
366/**
367 * kernfs_enable_ns - enable namespace under a directory
368 * @kn: directory of interest, should be empty
369 *
370 * This is to be called right after @kn is created to enable namespace
371 * under it.  All children of @kn must have non-NULL namespace tags and
372 * only the ones which match the super_block's tag will be visible.
373 */
374static inline void kernfs_enable_ns(struct kernfs_node *kn)
375{
376	WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
377	WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
378	kn->flags |= KERNFS_NS;
379}
380
381/**
382 * kernfs_ns_enabled - test whether namespace is enabled
383 * @kn: the node to test
384 *
385 * Test whether namespace filtering is enabled for the children of @ns.
386 */
387static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
388{
389	return kn->flags & KERNFS_NS;
390}
391
392int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
393int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
394			  char *buf, size_t buflen);
395void pr_cont_kernfs_name(struct kernfs_node *kn);
396void pr_cont_kernfs_path(struct kernfs_node *kn);
397struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
398struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
399					   const char *name, const void *ns);
400struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
401					   const char *path, const void *ns);
402void kernfs_get(struct kernfs_node *kn);
403void kernfs_put(struct kernfs_node *kn);
404
405struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
406struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
407struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
408
409struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
410				  struct super_block *sb);
411struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
412				       unsigned int flags, void *priv);
413void kernfs_destroy_root(struct kernfs_root *root);
414
415struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
416					 const char *name, umode_t mode,
417					 kuid_t uid, kgid_t gid,
418					 void *priv, const void *ns);
419struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
420					    const char *name);
421struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
422					 const char *name, umode_t mode,
423					 kuid_t uid, kgid_t gid,
424					 loff_t size,
425					 const struct kernfs_ops *ops,
426					 void *priv, const void *ns,
427					 struct lock_class_key *key);
428struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
429				       const char *name,
430				       struct kernfs_node *target);
431void kernfs_activate(struct kernfs_node *kn);
432void kernfs_remove(struct kernfs_node *kn);
433void kernfs_break_active_protection(struct kernfs_node *kn);
434void kernfs_unbreak_active_protection(struct kernfs_node *kn);
435bool kernfs_remove_self(struct kernfs_node *kn);
436int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
437			     const void *ns);
438int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
439		     const char *new_name, const void *new_ns);
440int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
441__poll_t kernfs_generic_poll(struct kernfs_open_file *of,
442			     struct poll_table_struct *pt);
443void kernfs_notify(struct kernfs_node *kn);
444
445int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
446		     void *value, size_t size);
447int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
448		     const void *value, size_t size, int flags);
449
450const void *kernfs_super_ns(struct super_block *sb);
451int kernfs_get_tree(struct fs_context *fc);
452void kernfs_free_fs_context(struct fs_context *fc);
453void kernfs_kill_sb(struct super_block *sb);
454
455void kernfs_init(void);
456
457struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
458						   u64 id);
459#else	/* CONFIG_KERNFS */
460
461static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
462{ return 0; }	/* whatever */
463
464static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
465
466static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
467{ return false; }
468
469static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
470{ return -ENOSYS; }
471
472static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
473					struct kernfs_node *kn,
474					char *buf, size_t buflen)
475{ return -ENOSYS; }
476
477static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
478static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
479
480static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
481{ return NULL; }
482
483static inline struct kernfs_node *
484kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
485		       const void *ns)
486{ return NULL; }
487static inline struct kernfs_node *
488kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
489		       const void *ns)
490{ return NULL; }
491
492static inline void kernfs_get(struct kernfs_node *kn) { }
493static inline void kernfs_put(struct kernfs_node *kn) { }
494
495static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
496{ return NULL; }
497
498static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
499{ return NULL; }
500
501static inline struct inode *
502kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
503{ return NULL; }
504
505static inline struct kernfs_root *
506kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
507		   void *priv)
508{ return ERR_PTR(-ENOSYS); }
509
510static inline void kernfs_destroy_root(struct kernfs_root *root) { }
511
512static inline struct kernfs_node *
513kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
514		     umode_t mode, kuid_t uid, kgid_t gid,
515		     void *priv, const void *ns)
516{ return ERR_PTR(-ENOSYS); }
517
518static inline struct kernfs_node *
519__kernfs_create_file(struct kernfs_node *parent, const char *name,
520		     umode_t mode, kuid_t uid, kgid_t gid,
521		     loff_t size, const struct kernfs_ops *ops,
522		     void *priv, const void *ns, struct lock_class_key *key)
523{ return ERR_PTR(-ENOSYS); }
524
525static inline struct kernfs_node *
526kernfs_create_link(struct kernfs_node *parent, const char *name,
527		   struct kernfs_node *target)
528{ return ERR_PTR(-ENOSYS); }
529
530static inline void kernfs_activate(struct kernfs_node *kn) { }
531
532static inline void kernfs_remove(struct kernfs_node *kn) { }
533
534static inline bool kernfs_remove_self(struct kernfs_node *kn)
535{ return false; }
536
537static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
538					   const char *name, const void *ns)
539{ return -ENOSYS; }
540
541static inline int kernfs_rename_ns(struct kernfs_node *kn,
542				   struct kernfs_node *new_parent,
543				   const char *new_name, const void *new_ns)
544{ return -ENOSYS; }
545
546static inline int kernfs_setattr(struct kernfs_node *kn,
547				 const struct iattr *iattr)
548{ return -ENOSYS; }
549
550static inline void kernfs_notify(struct kernfs_node *kn) { }
551
552static inline int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
553				   void *value, size_t size)
554{ return -ENOSYS; }
555
556static inline int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
557				   const void *value, size_t size, int flags)
558{ return -ENOSYS; }
559
560static inline const void *kernfs_super_ns(struct super_block *sb)
561{ return NULL; }
562
563static inline int kernfs_get_tree(struct fs_context *fc)
564{ return -ENOSYS; }
565
566static inline void kernfs_free_fs_context(struct fs_context *fc) { }
567
568static inline void kernfs_kill_sb(struct super_block *sb) { }
569
570static inline void kernfs_init(void) { }
571
572#endif	/* CONFIG_KERNFS */
573
574/**
575 * kernfs_path - build full path of a given node
576 * @kn: kernfs_node of interest
577 * @buf: buffer to copy @kn's name into
578 * @buflen: size of @buf
579 *
580 * If @kn is NULL result will be "(null)".
581 *
582 * Returns the length of the full path.  If the full length is equal to or
583 * greater than @buflen, @buf contains the truncated path with the trailing
584 * '\0'.  On error, -errno is returned.
585 */
586static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
587{
588	return kernfs_path_from_node(kn, NULL, buf, buflen);
589}
590
591static inline struct kernfs_node *
592kernfs_find_and_get(struct kernfs_node *kn, const char *name)
593{
594	return kernfs_find_and_get_ns(kn, name, NULL);
595}
596
597static inline struct kernfs_node *
598kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
599{
600	return kernfs_walk_and_get_ns(kn, path, NULL);
601}
602
603static inline struct kernfs_node *
604kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
605		  void *priv)
606{
607	return kernfs_create_dir_ns(parent, name, mode,
608				    GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
609				    priv, NULL);
610}
611
612static inline int kernfs_remove_by_name(struct kernfs_node *parent,
613					const char *name)
614{
615	return kernfs_remove_by_name_ns(parent, name, NULL);
616}
617
618static inline int kernfs_rename(struct kernfs_node *kn,
619				struct kernfs_node *new_parent,
620				const char *new_name)
621{
622	return kernfs_rename_ns(kn, new_parent, new_name, NULL);
623}
624
625#endif	/* __LINUX_KERNFS_H */
626