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_HIDDEN = 0x0200,
112 KERNFS_SUICIDAL = 0x0400,
113 KERNFS_SUICIDED = 0x0800,
114 KERNFS_EMPTY_DIR = 0x1000,
115 KERNFS_HAS_RELEASE = 0x2000,
116 KERNFS_REMOVING = 0x4000,
117};
118
119/* @flags for kernfs_create_root() */
120enum kernfs_root_flag {
121 /*
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.
126 */
127 KERNFS_ROOT_CREATE_DEACTIVATED = 0x0001,
128
129 /*
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.
137 */
138 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK = 0x0002,
139
140 /*
141 * The filesystem supports exportfs operation, so userspace can use
142 * fhandle to access nodes of the fs.
143 */
144 KERNFS_ROOT_SUPPORT_EXPORTOP = 0x0004,
145
146 /*
147 * Support user xattrs to be written to nodes rooted at this root.
148 */
149 KERNFS_ROOT_SUPPORT_USER_XATTR = 0x0008,
150
151 /*
152 * Renames must not change the parent node.
153 */
154 KERNFS_ROOT_INVARIANT_PARENT = 0x0010,
155};
156
157/* type-specific structures for kernfs_node union members */
158struct kernfs_elem_dir {
159 unsigned long subdirs;
160 /* children rbtree starts here and goes through kn->rb */
161 struct rb_root children;
162
163 /*
164 * The kernfs hierarchy this directory belongs to. This fits
165 * better directly in kernfs_node but is here to save space.
166 */
167 struct kernfs_root *root;
168 /*
169 * Monotonic revision counter, used to identify if a directory
170 * node has changed during negative dentry revalidation.
171 */
172 unsigned long rev;
173};
174
175struct kernfs_elem_symlink {
176 struct kernfs_node *target_kn;
177};
178
179struct kernfs_elem_attr {
180 const struct kernfs_ops *ops;
181 struct kernfs_open_node __rcu *open;
182 loff_t size;
183 struct kernfs_node *notify_next; /* for kernfs_notify() */
184};
185
186/*
187 * kernfs_node - the building block of kernfs hierarchy. Each and every
188 * kernfs node is represented by single kernfs_node. Most fields are
189 * private to kernfs and shouldn't be accessed directly by kernfs users.
190 *
191 * As long as count reference is held, the kernfs_node itself is
192 * accessible. Dereferencing elem or any other outer entity requires
193 * active reference.
194 */
195struct kernfs_node {
196 atomic_t count;
197 atomic_t active;
198#ifdef CONFIG_DEBUG_LOCK_ALLOC
199 struct lockdep_map dep_map;
200#endif
201 /*
202 * Use kernfs_get_parent() and kernfs_name/path() instead of
203 * accessing the following two fields directly. If the node is
204 * never moved to a different parent, it is safe to access the
205 * parent directly.
206 */
207 struct kernfs_node __rcu *__parent;
208 const char __rcu *name;
209
210 struct rb_node rb;
211
212 const void *ns; /* namespace tag */
213 unsigned int hash; /* ns + name hash */
214 unsigned short flags;
215 umode_t mode;
216
217 union {
218 struct kernfs_elem_dir dir;
219 struct kernfs_elem_symlink symlink;
220 struct kernfs_elem_attr attr;
221 };
222
223 /*
224 * 64bit unique ID. On 64bit ino setups, id is the ino. On 32bit,
225 * the low 32bits are ino and upper generation.
226 */
227 u64 id;
228
229 void *priv;
230 struct kernfs_iattrs *iattr;
231
232 struct rcu_head rcu;
233};
234
235/*
236 * kernfs_syscall_ops may be specified on kernfs_create_root() to support
237 * syscalls. These optional callbacks are invoked on the matching syscalls
238 * and can perform any kernfs operations which don't necessarily have to be
239 * the exact operation requested. An active reference is held for each
240 * kernfs_node parameter.
241 */
242struct kernfs_syscall_ops {
243 int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
244
245 int (*mkdir)(struct kernfs_node *parent, const char *name,
246 umode_t mode);
247 int (*rmdir)(struct kernfs_node *kn);
248 int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
249 const char *new_name);
250 int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
251 struct kernfs_root *root);
252};
253
254struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root);
255
256struct kernfs_open_file {
257 /* published fields */
258 struct kernfs_node *kn;
259 struct file *file;
260 struct seq_file *seq_file;
261 void *priv;
262
263 /* private fields, do not use outside kernfs proper */
264 struct mutex mutex;
265 struct mutex prealloc_mutex;
266 int event;
267 struct list_head list;
268 char *prealloc_buf;
269
270 size_t atomic_write_len;
271 bool mmapped:1;
272 bool released:1;
273 const struct vm_operations_struct *vm_ops;
274};
275
276struct kernfs_ops {
277 /*
278 * Optional open/release methods. Both are called with
279 * @of->seq_file populated.
280 */
281 int (*open)(struct kernfs_open_file *of);
282 void (*release)(struct kernfs_open_file *of);
283
284 /*
285 * Read is handled by either seq_file or raw_read().
286 *
287 * If seq_show() is present, seq_file path is active. Other seq
288 * operations are optional and if not implemented, the behavior is
289 * equivalent to single_open(). @sf->private points to the
290 * associated kernfs_open_file.
291 *
292 * read() is bounced through kernel buffer and a read larger than
293 * PAGE_SIZE results in partial operation of PAGE_SIZE.
294 */
295 int (*seq_show)(struct seq_file *sf, void *v);
296
297 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
298 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
299 void (*seq_stop)(struct seq_file *sf, void *v);
300
301 ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
302 loff_t off);
303
304 /*
305 * write() is bounced through kernel buffer. If atomic_write_len
306 * is not set, a write larger than PAGE_SIZE results in partial
307 * operations of PAGE_SIZE chunks. If atomic_write_len is set,
308 * writes upto the specified size are executed atomically but
309 * larger ones are rejected with -E2BIG.
310 */
311 size_t atomic_write_len;
312 /*
313 * "prealloc" causes a buffer to be allocated at open for
314 * all read/write requests. As ->seq_show uses seq_read()
315 * which does its own allocation, it is incompatible with
316 * ->prealloc. Provide ->read and ->write with ->prealloc.
317 */
318 bool prealloc;
319 ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
320 loff_t off);
321
322 __poll_t (*poll)(struct kernfs_open_file *of,
323 struct poll_table_struct *pt);
324
325 int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
326 loff_t (*llseek)(struct kernfs_open_file *of, loff_t offset, int whence);
327};
328
329/*
330 * The kernfs superblock creation/mount parameter context.
331 */
332struct kernfs_fs_context {
333 struct kernfs_root *root; /* Root of the hierarchy being mounted */
334 void *ns_tag; /* Namespace tag of the mount (or NULL) */
335 unsigned long magic; /* File system specific magic number */
336
337 /* The following are set/used by kernfs_mount() */
338 bool new_sb_created; /* Set to T if we allocated a new sb */
339};
340
341#ifdef CONFIG_KERNFS
342
343static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
344{
345 return kn->flags & KERNFS_TYPE_MASK;
346}
347
348static inline ino_t kernfs_id_ino(u64 id)
349{
350 /* id is ino if ino_t is 64bit; otherwise, low 32bits */
351 if (sizeof(ino_t) >= sizeof(u64))
352 return id;
353 else
354 return (u32)id;
355}
356
357static inline u32 kernfs_id_gen(u64 id)
358{
359 /* gen is fixed at 1 if ino_t is 64bit; otherwise, high 32bits */
360 if (sizeof(ino_t) >= sizeof(u64))
361 return 1;
362 else
363 return id >> 32;
364}
365
366static inline ino_t kernfs_ino(struct kernfs_node *kn)
367{
368 return kernfs_id_ino(id: kn->id);
369}
370
371static inline ino_t kernfs_gen(struct kernfs_node *kn)
372{
373 return kernfs_id_gen(id: kn->id);
374}
375
376/**
377 * kernfs_enable_ns - enable namespace under a directory
378 * @kn: directory of interest, should be empty
379 *
380 * This is to be called right after @kn is created to enable namespace
381 * under it. All children of @kn must have non-NULL namespace tags and
382 * only the ones which match the super_block's tag will be visible.
383 */
384static inline void kernfs_enable_ns(struct kernfs_node *kn)
385{
386 WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
387 WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
388 kn->flags |= KERNFS_NS;
389}
390
391/**
392 * kernfs_ns_enabled - test whether namespace is enabled
393 * @kn: the node to test
394 *
395 * Test whether namespace filtering is enabled for the children of @ns.
396 */
397static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
398{
399 return kn->flags & KERNFS_NS;
400}
401
402int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
403int kernfs_path_from_node(struct kernfs_node *kn_to, struct kernfs_node *kn_from,
404 char *buf, size_t buflen);
405void pr_cont_kernfs_name(struct kernfs_node *kn);
406void pr_cont_kernfs_path(struct kernfs_node *kn);
407struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
408struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
409 const char *name, const void *ns);
410struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
411 const char *path, const void *ns);
412void kernfs_get(struct kernfs_node *kn);
413void kernfs_put(struct kernfs_node *kn);
414
415struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
416struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
417struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
418
419struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
420 struct super_block *sb);
421struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
422 unsigned int flags, void *priv);
423void kernfs_destroy_root(struct kernfs_root *root);
424unsigned int kernfs_root_flags(struct kernfs_node *kn);
425
426struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
427 const char *name, umode_t mode,
428 kuid_t uid, kgid_t gid,
429 void *priv, const void *ns);
430struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
431 const char *name);
432struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
433 const char *name, umode_t mode,
434 kuid_t uid, kgid_t gid,
435 loff_t size,
436 const struct kernfs_ops *ops,
437 void *priv, const void *ns,
438 struct lock_class_key *key);
439struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
440 const char *name,
441 struct kernfs_node *target);
442void kernfs_activate(struct kernfs_node *kn);
443void kernfs_show(struct kernfs_node *kn, bool show);
444void kernfs_remove(struct kernfs_node *kn);
445void kernfs_break_active_protection(struct kernfs_node *kn);
446void kernfs_unbreak_active_protection(struct kernfs_node *kn);
447bool kernfs_remove_self(struct kernfs_node *kn);
448int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
449 const void *ns);
450int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
451 const char *new_name, const void *new_ns);
452int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
453__poll_t kernfs_generic_poll(struct kernfs_open_file *of,
454 struct poll_table_struct *pt);
455void kernfs_notify(struct kernfs_node *kn);
456
457int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
458 void *value, size_t size);
459int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
460 const void *value, size_t size, int flags);
461
462const void *kernfs_super_ns(struct super_block *sb);
463int kernfs_get_tree(struct fs_context *fc);
464void kernfs_free_fs_context(struct fs_context *fc);
465void kernfs_kill_sb(struct super_block *sb);
466
467void kernfs_init(void);
468
469struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
470 u64 id);
471#else /* CONFIG_KERNFS */
472
473static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
474{ return 0; } /* whatever */
475
476static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
477
478static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
479{ return false; }
480
481static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
482{ return -ENOSYS; }
483
484static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
485 struct kernfs_node *kn,
486 char *buf, size_t buflen)
487{ return -ENOSYS; }
488
489static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
490static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
491
492static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
493{ return NULL; }
494
495static inline struct kernfs_node *
496kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
497 const void *ns)
498{ return NULL; }
499static inline struct kernfs_node *
500kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
501 const void *ns)
502{ return NULL; }
503
504static inline void kernfs_get(struct kernfs_node *kn) { }
505static inline void kernfs_put(struct kernfs_node *kn) { }
506
507static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
508{ return NULL; }
509
510static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
511{ return NULL; }
512
513static inline struct inode *
514kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
515{ return NULL; }
516
517static inline struct kernfs_root *
518kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
519 void *priv)
520{ return ERR_PTR(-ENOSYS); }
521
522static inline void kernfs_destroy_root(struct kernfs_root *root) { }
523static inline unsigned int kernfs_root_flags(struct kernfs_node *kn)
524{ return 0; }
525
526static inline struct kernfs_node *
527kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
528 umode_t mode, kuid_t uid, kgid_t gid,
529 void *priv, const void *ns)
530{ return ERR_PTR(-ENOSYS); }
531
532static inline struct kernfs_node *
533__kernfs_create_file(struct kernfs_node *parent, const char *name,
534 umode_t mode, kuid_t uid, kgid_t gid,
535 loff_t size, const struct kernfs_ops *ops,
536 void *priv, const void *ns, struct lock_class_key *key)
537{ return ERR_PTR(-ENOSYS); }
538
539static inline struct kernfs_node *
540kernfs_create_link(struct kernfs_node *parent, const char *name,
541 struct kernfs_node *target)
542{ return ERR_PTR(-ENOSYS); }
543
544static inline void kernfs_activate(struct kernfs_node *kn) { }
545
546static inline void kernfs_remove(struct kernfs_node *kn) { }
547
548static inline bool kernfs_remove_self(struct kernfs_node *kn)
549{ return false; }
550
551static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
552 const char *name, const void *ns)
553{ return -ENOSYS; }
554
555static inline int kernfs_rename_ns(struct kernfs_node *kn,
556 struct kernfs_node *new_parent,
557 const char *new_name, const void *new_ns)
558{ return -ENOSYS; }
559
560static inline int kernfs_setattr(struct kernfs_node *kn,
561 const struct iattr *iattr)
562{ return -ENOSYS; }
563
564static inline __poll_t kernfs_generic_poll(struct kernfs_open_file *of,
565 struct poll_table_struct *pt)
566{ return -ENOSYS; }
567
568static inline void kernfs_notify(struct kernfs_node *kn) { }
569
570static inline int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
571 void *value, size_t size)
572{ return -ENOSYS; }
573
574static inline int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
575 const void *value, size_t size, int flags)
576{ return -ENOSYS; }
577
578static inline const void *kernfs_super_ns(struct super_block *sb)
579{ return NULL; }
580
581static inline int kernfs_get_tree(struct fs_context *fc)
582{ return -ENOSYS; }
583
584static inline void kernfs_free_fs_context(struct fs_context *fc) { }
585
586static inline void kernfs_kill_sb(struct super_block *sb) { }
587
588static inline void kernfs_init(void) { }
589
590#endif /* CONFIG_KERNFS */
591
592/**
593 * kernfs_path - build full path of a given node
594 * @kn: kernfs_node of interest
595 * @buf: buffer to copy @kn's name into
596 * @buflen: size of @buf
597 *
598 * If @kn is NULL result will be "(null)".
599 *
600 * Returns the length of the full path. If the full length is equal to or
601 * greater than @buflen, @buf contains the truncated path with the trailing
602 * '\0'. On error, -errno is returned.
603 */
604static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
605{
606 return kernfs_path_from_node(kn_to: kn, NULL, buf, buflen);
607}
608
609static inline struct kernfs_node *
610kernfs_find_and_get(struct kernfs_node *kn, const char *name)
611{
612 return kernfs_find_and_get_ns(parent: kn, name, NULL);
613}
614
615static inline struct kernfs_node *
616kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
617{
618 return kernfs_walk_and_get_ns(parent: kn, path, NULL);
619}
620
621static inline struct kernfs_node *
622kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
623 void *priv)
624{
625 return kernfs_create_dir_ns(parent, name, mode,
626 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
627 priv, NULL);
628}
629
630static inline int kernfs_remove_by_name(struct kernfs_node *parent,
631 const char *name)
632{
633 return kernfs_remove_by_name_ns(parent, name, NULL);
634}
635
636static inline int kernfs_rename(struct kernfs_node *kn,
637 struct kernfs_node *new_parent,
638 const char *new_name)
639{
640 return kernfs_rename_ns(kn, new_parent, new_name, NULL);
641}
642
643#endif /* __LINUX_KERNFS_H */
644