1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * kernfs.h - pseudo filesystem decoupled from vfs locking
6 #ifndef __LINUX_KERNFS_H
7 #define __LINUX_KERNFS_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>
27 struct vm_area_struct;
28 struct vm_operations_struct;
30 struct file_system_type;
31 struct poll_table_struct;
34 struct kernfs_fs_context;
35 struct kernfs_open_node;
39 * NR_KERNFS_LOCK_BITS determines size (NR_KERNFS_LOCKS) of hash
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.
45 * At the moment size of hash table of locks is being set based on
46 * the number of CPUs as follows:
48 * NR_CPU NR_KERNFS_LOCK_BITS NR_KERNFS_LOCKS
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.
63 #define NR_KERNFS_LOCK_BITS (2 * (ilog2(NR_CPUS < 32 ? NR_CPUS : 32)))
65 #define NR_KERNFS_LOCK_BITS 1
68 #define NR_KERNFS_LOCKS (1 << NR_KERNFS_LOCK_BITS)
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.
74 * filp->private_data points to seq_file whose ->private points to
77 * kernfs_open_files are chained at kernfs_open_node->files, which is
78 * protected by kernfs_global_locks.open_file_mutex[i].
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.
84 * Hashed mutexes are safe to use here because operations using these don't
85 * rely on global exclusion.
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.
90 struct kernfs_global_locks {
91 struct mutex open_file_mutex[NR_KERNFS_LOCKS];
94 enum kernfs_node_type {
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)
105 enum kernfs_node_flag {
106 KERNFS_ACTIVATED = 0x0010,
108 KERNFS_HAS_SEQ_SHOW = 0x0040,
109 KERNFS_HAS_MMAP = 0x0080,
110 KERNFS_LOCKDEP = 0x0100,
111 KERNFS_HIDDEN = 0x0200,
112 KERNFS_SUICIDAL = 0x0400,
113 KERNFS_SUICIDED = 0x0800,
114 KERNFS_EMPTY_DIR = 0x1000,
115 KERNFS_HAS_RELEASE = 0x2000,
116 KERNFS_REMOVING = 0x4000,
119 /* @flags for kernfs_create_root() */
120 enum kernfs_root_flag {
122 * kernfs_nodes are created in the deactivated state and invisible.
123 * They require explicit kernfs_activate() to become visible. This
124 * can be used to make related nodes become visible atomically
125 * after all nodes are created successfully.
127 KERNFS_ROOT_CREATE_DEACTIVATED = 0x0001,
130 * For regular files, if the opener has CAP_DAC_OVERRIDE, open(2)
131 * succeeds regardless of the RW permissions. sysfs had an extra
132 * layer of enforcement where open(2) fails with -EACCES regardless
133 * of CAP_DAC_OVERRIDE if the permission doesn't have the
134 * respective read or write access at all (none of S_IRUGO or
135 * S_IWUGO) or the respective operation isn't implemented. The
136 * following flag enables that behavior.
138 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK = 0x0002,
141 * The filesystem supports exportfs operation, so userspace can use
142 * fhandle to access nodes of the fs.
144 KERNFS_ROOT_SUPPORT_EXPORTOP = 0x0004,
147 * Support user xattrs to be written to nodes rooted at this root.
149 KERNFS_ROOT_SUPPORT_USER_XATTR = 0x0008,
152 /* type-specific structures for kernfs_node union members */
153 struct kernfs_elem_dir {
154 unsigned long subdirs;
155 /* children rbtree starts here and goes through kn->rb */
156 struct rb_root children;
159 * The kernfs hierarchy this directory belongs to. This fits
160 * better directly in kernfs_node but is here to save space.
162 struct kernfs_root *root;
164 * Monotonic revision counter, used to identify if a directory
165 * node has changed during negative dentry revalidation.
170 struct kernfs_elem_symlink {
171 struct kernfs_node *target_kn;
174 struct kernfs_elem_attr {
175 const struct kernfs_ops *ops;
176 struct kernfs_open_node __rcu *open;
178 struct kernfs_node *notify_next; /* for kernfs_notify() */
182 * kernfs_node - the building block of kernfs hierarchy. Each and every
183 * kernfs node is represented by single kernfs_node. Most fields are
184 * private to kernfs and shouldn't be accessed directly by kernfs users.
186 * As long as count reference is held, the kernfs_node itself is
187 * accessible. Dereferencing elem or any other outer entity requires
193 #ifdef CONFIG_DEBUG_LOCK_ALLOC
194 struct lockdep_map dep_map;
197 * Use kernfs_get_parent() and kernfs_name/path() instead of
198 * accessing the following two fields directly. If the node is
199 * never moved to a different parent, it is safe to access the
202 struct kernfs_node *parent;
207 const void *ns; /* namespace tag */
208 unsigned int hash; /* ns + name hash */
210 struct kernfs_elem_dir dir;
211 struct kernfs_elem_symlink symlink;
212 struct kernfs_elem_attr attr;
218 * 64bit unique ID. On 64bit ino setups, id is the ino. On 32bit,
219 * the low 32bits are ino and upper generation.
223 unsigned short flags;
225 struct kernfs_iattrs *iattr;
229 * kernfs_syscall_ops may be specified on kernfs_create_root() to support
230 * syscalls. These optional callbacks are invoked on the matching syscalls
231 * and can perform any kernfs operations which don't necessarily have to be
232 * the exact operation requested. An active reference is held for each
233 * kernfs_node parameter.
235 struct kernfs_syscall_ops {
236 int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
238 int (*mkdir)(struct kernfs_node *parent, const char *name,
240 int (*rmdir)(struct kernfs_node *kn);
241 int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
242 const char *new_name);
243 int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
244 struct kernfs_root *root);
247 struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root);
249 struct kernfs_open_file {
250 /* published fields */
251 struct kernfs_node *kn;
253 struct seq_file *seq_file;
256 /* private fields, do not use outside kernfs proper */
258 struct mutex prealloc_mutex;
260 struct list_head list;
263 size_t atomic_write_len;
266 const struct vm_operations_struct *vm_ops;
271 * Optional open/release methods. Both are called with
272 * @of->seq_file populated.
274 int (*open)(struct kernfs_open_file *of);
275 void (*release)(struct kernfs_open_file *of);
278 * Read is handled by either seq_file or raw_read().
280 * If seq_show() is present, seq_file path is active. Other seq
281 * operations are optional and if not implemented, the behavior is
282 * equivalent to single_open(). @sf->private points to the
283 * associated kernfs_open_file.
285 * read() is bounced through kernel buffer and a read larger than
286 * PAGE_SIZE results in partial operation of PAGE_SIZE.
288 int (*seq_show)(struct seq_file *sf, void *v);
290 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
291 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
292 void (*seq_stop)(struct seq_file *sf, void *v);
294 ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
298 * write() is bounced through kernel buffer. If atomic_write_len
299 * is not set, a write larger than PAGE_SIZE results in partial
300 * operations of PAGE_SIZE chunks. If atomic_write_len is set,
301 * writes upto the specified size are executed atomically but
302 * larger ones are rejected with -E2BIG.
304 size_t atomic_write_len;
306 * "prealloc" causes a buffer to be allocated at open for
307 * all read/write requests. As ->seq_show uses seq_read()
308 * which does its own allocation, it is incompatible with
309 * ->prealloc. Provide ->read and ->write with ->prealloc.
312 ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
315 __poll_t (*poll)(struct kernfs_open_file *of,
316 struct poll_table_struct *pt);
318 int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
322 * The kernfs superblock creation/mount parameter context.
324 struct kernfs_fs_context {
325 struct kernfs_root *root; /* Root of the hierarchy being mounted */
326 void *ns_tag; /* Namespace tag of the mount (or NULL) */
327 unsigned long magic; /* File system specific magic number */
329 /* The following are set/used by kernfs_mount() */
330 bool new_sb_created; /* Set to T if we allocated a new sb */
335 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
337 return kn->flags & KERNFS_TYPE_MASK;
340 static inline ino_t kernfs_id_ino(u64 id)
342 /* id is ino if ino_t is 64bit; otherwise, low 32bits */
343 if (sizeof(ino_t) >= sizeof(u64))
349 static inline u32 kernfs_id_gen(u64 id)
351 /* gen is fixed at 1 if ino_t is 64bit; otherwise, high 32bits */
352 if (sizeof(ino_t) >= sizeof(u64))
358 static inline ino_t kernfs_ino(struct kernfs_node *kn)
360 return kernfs_id_ino(kn->id);
363 static inline ino_t kernfs_gen(struct kernfs_node *kn)
365 return kernfs_id_gen(kn->id);
369 * kernfs_enable_ns - enable namespace under a directory
370 * @kn: directory of interest, should be empty
372 * This is to be called right after @kn is created to enable namespace
373 * under it. All children of @kn must have non-NULL namespace tags and
374 * only the ones which match the super_block's tag will be visible.
376 static inline void kernfs_enable_ns(struct kernfs_node *kn)
378 WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
379 WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
380 kn->flags |= KERNFS_NS;
384 * kernfs_ns_enabled - test whether namespace is enabled
385 * @kn: the node to test
387 * Test whether namespace filtering is enabled for the children of @ns.
389 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
391 return kn->flags & KERNFS_NS;
394 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
395 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
396 char *buf, size_t buflen);
397 void pr_cont_kernfs_name(struct kernfs_node *kn);
398 void pr_cont_kernfs_path(struct kernfs_node *kn);
399 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
400 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
401 const char *name, const void *ns);
402 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
403 const char *path, const void *ns);
404 void kernfs_get(struct kernfs_node *kn);
405 void kernfs_put(struct kernfs_node *kn);
407 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
408 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
409 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
411 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
412 struct super_block *sb);
413 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
414 unsigned int flags, void *priv);
415 void kernfs_destroy_root(struct kernfs_root *root);
417 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
418 const char *name, umode_t mode,
419 kuid_t uid, kgid_t gid,
420 void *priv, const void *ns);
421 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
423 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
424 const char *name, umode_t mode,
425 kuid_t uid, kgid_t gid,
427 const struct kernfs_ops *ops,
428 void *priv, const void *ns,
429 struct lock_class_key *key);
430 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
432 struct kernfs_node *target);
433 void kernfs_activate(struct kernfs_node *kn);
434 void kernfs_show(struct kernfs_node *kn, bool show);
435 void kernfs_remove(struct kernfs_node *kn);
436 void kernfs_break_active_protection(struct kernfs_node *kn);
437 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
438 bool kernfs_remove_self(struct kernfs_node *kn);
439 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
441 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
442 const char *new_name, const void *new_ns);
443 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
444 __poll_t kernfs_generic_poll(struct kernfs_open_file *of,
445 struct poll_table_struct *pt);
446 void kernfs_notify(struct kernfs_node *kn);
448 int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
449 void *value, size_t size);
450 int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
451 const void *value, size_t size, int flags);
453 const void *kernfs_super_ns(struct super_block *sb);
454 int kernfs_get_tree(struct fs_context *fc);
455 void kernfs_free_fs_context(struct fs_context *fc);
456 void kernfs_kill_sb(struct super_block *sb);
458 void kernfs_init(void);
460 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
462 #else /* CONFIG_KERNFS */
464 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
465 { return 0; } /* whatever */
467 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
469 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
472 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
475 static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
476 struct kernfs_node *kn,
477 char *buf, size_t buflen)
480 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
481 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
483 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
486 static inline struct kernfs_node *
487 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
490 static inline struct kernfs_node *
491 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
495 static inline void kernfs_get(struct kernfs_node *kn) { }
496 static inline void kernfs_put(struct kernfs_node *kn) { }
498 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
501 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
504 static inline struct inode *
505 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
508 static inline struct kernfs_root *
509 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
511 { return ERR_PTR(-ENOSYS); }
513 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
515 static inline struct kernfs_node *
516 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
517 umode_t mode, kuid_t uid, kgid_t gid,
518 void *priv, const void *ns)
519 { return ERR_PTR(-ENOSYS); }
521 static inline struct kernfs_node *
522 __kernfs_create_file(struct kernfs_node *parent, const char *name,
523 umode_t mode, kuid_t uid, kgid_t gid,
524 loff_t size, const struct kernfs_ops *ops,
525 void *priv, const void *ns, struct lock_class_key *key)
526 { return ERR_PTR(-ENOSYS); }
528 static inline struct kernfs_node *
529 kernfs_create_link(struct kernfs_node *parent, const char *name,
530 struct kernfs_node *target)
531 { return ERR_PTR(-ENOSYS); }
533 static inline void kernfs_activate(struct kernfs_node *kn) { }
535 static inline void kernfs_remove(struct kernfs_node *kn) { }
537 static inline bool kernfs_remove_self(struct kernfs_node *kn)
540 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
541 const char *name, const void *ns)
544 static inline int kernfs_rename_ns(struct kernfs_node *kn,
545 struct kernfs_node *new_parent,
546 const char *new_name, const void *new_ns)
549 static inline int kernfs_setattr(struct kernfs_node *kn,
550 const struct iattr *iattr)
553 static inline __poll_t kernfs_generic_poll(struct kernfs_open_file *of,
554 struct poll_table_struct *pt)
557 static inline void kernfs_notify(struct kernfs_node *kn) { }
559 static inline int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
560 void *value, size_t size)
563 static inline int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
564 const void *value, size_t size, int flags)
567 static inline const void *kernfs_super_ns(struct super_block *sb)
570 static inline int kernfs_get_tree(struct fs_context *fc)
573 static inline void kernfs_free_fs_context(struct fs_context *fc) { }
575 static inline void kernfs_kill_sb(struct super_block *sb) { }
577 static inline void kernfs_init(void) { }
579 #endif /* CONFIG_KERNFS */
582 * kernfs_path - build full path of a given node
583 * @kn: kernfs_node of interest
584 * @buf: buffer to copy @kn's name into
585 * @buflen: size of @buf
587 * If @kn is NULL result will be "(null)".
589 * Returns the length of the full path. If the full length is equal to or
590 * greater than @buflen, @buf contains the truncated path with the trailing
591 * '\0'. On error, -errno is returned.
593 static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
595 return kernfs_path_from_node(kn, NULL, buf, buflen);
598 static inline struct kernfs_node *
599 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
601 return kernfs_find_and_get_ns(kn, name, NULL);
604 static inline struct kernfs_node *
605 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
607 return kernfs_walk_and_get_ns(kn, path, NULL);
610 static inline struct kernfs_node *
611 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
614 return kernfs_create_dir_ns(parent, name, mode,
615 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
619 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
622 return kernfs_remove_by_name_ns(parent, name, NULL);
625 static inline int kernfs_rename(struct kernfs_node *kn,
626 struct kernfs_node *new_parent,
627 const char *new_name)
629 return kernfs_rename_ns(kn, new_parent, new_name, NULL);
632 #endif /* __LINUX_KERNFS_H */