1/*
2 * kernfs.h - pseudo filesystem decoupled from vfs locking
3 *
4 * This file is released under the GPLv2.
5 */
6
7#ifndef __LINUX_KERNFS_H
8#define __LINUX_KERNFS_H
9
10#include <linux/kernel.h>
11#include <linux/err.h>
12#include <linux/list.h>
13#include <linux/mutex.h>
14#include <linux/idr.h>
15#include <linux/lockdep.h>
16#include <linux/rbtree.h>
17#include <linux/atomic.h>
18#include <linux/uidgid.h>
19#include <linux/wait.h>
20
21struct file;
22struct dentry;
23struct iattr;
24struct seq_file;
25struct vm_area_struct;
26struct super_block;
27struct file_system_type;
28struct poll_table_struct;
29struct fs_context;
30
31struct kernfs_fs_context;
32struct kernfs_open_node;
33struct kernfs_iattrs;
34
35enum kernfs_node_type {
36 KERNFS_DIR = 0x0001,
37 KERNFS_FILE = 0x0002,
38 KERNFS_LINK = 0x0004,
39};
40
41#define KERNFS_TYPE_MASK 0x000f
42#define KERNFS_FLAG_MASK ~KERNFS_TYPE_MASK
43
44enum kernfs_node_flag {
45 KERNFS_ACTIVATED = 0x0010,
46 KERNFS_NS = 0x0020,
47 KERNFS_HAS_SEQ_SHOW = 0x0040,
48 KERNFS_HAS_MMAP = 0x0080,
49 KERNFS_LOCKDEP = 0x0100,
50 KERNFS_SUICIDAL = 0x0400,
51 KERNFS_SUICIDED = 0x0800,
52 KERNFS_EMPTY_DIR = 0x1000,
53 KERNFS_HAS_RELEASE = 0x2000,
54};
55
56/* @flags for kernfs_create_root() */
57enum kernfs_root_flag {
58 /*
59 * kernfs_nodes are created in the deactivated state and invisible.
60 * They require explicit kernfs_activate() to become visible. This
61 * can be used to make related nodes become visible atomically
62 * after all nodes are created successfully.
63 */
64 KERNFS_ROOT_CREATE_DEACTIVATED = 0x0001,
65
66 /*
67 * For regular flies, if the opener has CAP_DAC_OVERRIDE, open(2)
68 * succeeds regardless of the RW permissions. sysfs had an extra
69 * layer of enforcement where open(2) fails with -EACCES regardless
70 * of CAP_DAC_OVERRIDE if the permission doesn't have the
71 * respective read or write access at all (none of S_IRUGO or
72 * S_IWUGO) or the respective operation isn't implemented. The
73 * following flag enables that behavior.
74 */
75 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK = 0x0002,
76
77 /*
78 * The filesystem supports exportfs operation, so userspace can use
79 * fhandle to access nodes of the fs.
80 */
81 KERNFS_ROOT_SUPPORT_EXPORTOP = 0x0004,
82};
83
84/* type-specific structures for kernfs_node union members */
85struct kernfs_elem_dir {
86 unsigned long subdirs;
87 /* children rbtree starts here and goes through kn->rb */
88 struct rb_root children;
89
90 /*
91 * The kernfs hierarchy this directory belongs to. This fits
92 * better directly in kernfs_node but is here to save space.
93 */
94 struct kernfs_root *root;
95};
96
97struct kernfs_elem_symlink {
98 struct kernfs_node *target_kn;
99};
100
101struct kernfs_elem_attr {
102 const struct kernfs_ops *ops;
103 struct kernfs_open_node *open;
104 loff_t size;
105 struct kernfs_node *notify_next; /* for kernfs_notify() */
106};
107
108/* represent a kernfs node */
109union kernfs_node_id {
110 struct {
111 /*
112 * blktrace will export this struct as a simplified 'struct
113 * fid' (which is a big data struction), so userspace can use
114 * it to find kernfs node. The layout must match the first two
115 * fields of 'struct fid' exactly.
116 */
117 u32 ino;
118 u32 generation;
119 };
120 u64 id;
121};
122
123/*
124 * kernfs_node - the building block of kernfs hierarchy. Each and every
125 * kernfs node is represented by single kernfs_node. Most fields are
126 * private to kernfs and shouldn't be accessed directly by kernfs users.
127 *
128 * As long as s_count reference is held, the kernfs_node itself is
129 * accessible. Dereferencing elem or any other outer entity requires
130 * active reference.
131 */
132struct kernfs_node {
133 atomic_t count;
134 atomic_t active;
135#ifdef CONFIG_DEBUG_LOCK_ALLOC
136 struct lockdep_map dep_map;
137#endif
138 /*
139 * Use kernfs_get_parent() and kernfs_name/path() instead of
140 * accessing the following two fields directly. If the node is
141 * never moved to a different parent, it is safe to access the
142 * parent directly.
143 */
144 struct kernfs_node *parent;
145 const char *name;
146
147 struct rb_node rb;
148
149 const void *ns; /* namespace tag */
150 unsigned int hash; /* ns + name hash */
151 union {
152 struct kernfs_elem_dir dir;
153 struct kernfs_elem_symlink symlink;
154 struct kernfs_elem_attr attr;
155 };
156
157 void *priv;
158
159 union kernfs_node_id id;
160 unsigned short flags;
161 umode_t mode;
162 struct kernfs_iattrs *iattr;
163};
164
165/*
166 * kernfs_syscall_ops may be specified on kernfs_create_root() to support
167 * syscalls. These optional callbacks are invoked on the matching syscalls
168 * and can perform any kernfs operations which don't necessarily have to be
169 * the exact operation requested. An active reference is held for each
170 * kernfs_node parameter.
171 */
172struct kernfs_syscall_ops {
173 int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
174
175 int (*mkdir)(struct kernfs_node *parent, const char *name,
176 umode_t mode);
177 int (*rmdir)(struct kernfs_node *kn);
178 int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
179 const char *new_name);
180 int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
181 struct kernfs_root *root);
182};
183
184struct kernfs_root {
185 /* published fields */
186 struct kernfs_node *kn;
187 unsigned int flags; /* KERNFS_ROOT_* flags */
188
189 /* private fields, do not use outside kernfs proper */
190 struct idr ino_idr;
191 u32 next_generation;
192 struct kernfs_syscall_ops *syscall_ops;
193
194 /* list of kernfs_super_info of this root, protected by kernfs_mutex */
195 struct list_head supers;
196
197 wait_queue_head_t deactivate_waitq;
198};
199
200struct kernfs_open_file {
201 /* published fields */
202 struct kernfs_node *kn;
203 struct file *file;
204 struct seq_file *seq_file;
205 void *priv;
206
207 /* private fields, do not use outside kernfs proper */
208 struct mutex mutex;
209 struct mutex prealloc_mutex;
210 int event;
211 struct list_head list;
212 char *prealloc_buf;
213
214 size_t atomic_write_len;
215 bool mmapped:1;
216 bool released:1;
217 const struct vm_operations_struct *vm_ops;
218};
219
220struct kernfs_ops {
221 /*
222 * Optional open/release methods. Both are called with
223 * @of->seq_file populated.
224 */
225 int (*open)(struct kernfs_open_file *of);
226 void (*release)(struct kernfs_open_file *of);
227
228 /*
229 * Read is handled by either seq_file or raw_read().
230 *
231 * If seq_show() is present, seq_file path is active. Other seq
232 * operations are optional and if not implemented, the behavior is
233 * equivalent to single_open(). @sf->private points to the
234 * associated kernfs_open_file.
235 *
236 * read() is bounced through kernel buffer and a read larger than
237 * PAGE_SIZE results in partial operation of PAGE_SIZE.
238 */
239 int (*seq_show)(struct seq_file *sf, void *v);
240
241 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
242 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
243 void (*seq_stop)(struct seq_file *sf, void *v);
244
245 ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
246 loff_t off);
247
248 /*
249 * write() is bounced through kernel buffer. If atomic_write_len
250 * is not set, a write larger than PAGE_SIZE results in partial
251 * operations of PAGE_SIZE chunks. If atomic_write_len is set,
252 * writes upto the specified size are executed atomically but
253 * larger ones are rejected with -E2BIG.
254 */
255 size_t atomic_write_len;
256 /*
257 * "prealloc" causes a buffer to be allocated at open for
258 * all read/write requests. As ->seq_show uses seq_read()
259 * which does its own allocation, it is incompatible with
260 * ->prealloc. Provide ->read and ->write with ->prealloc.
261 */
262 bool prealloc;
263 ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
264 loff_t off);
265
266 __poll_t (*poll)(struct kernfs_open_file *of,
267 struct poll_table_struct *pt);
268
269 int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
270
271#ifdef CONFIG_DEBUG_LOCK_ALLOC
272 struct lock_class_key lockdep_key;
273#endif
274};
275
276/*
277 * The kernfs superblock creation/mount parameter context.
278 */
279struct kernfs_fs_context {
280 struct kernfs_root *root; /* Root of the hierarchy being mounted */
281 void *ns_tag; /* Namespace tag of the mount (or NULL) */
282 unsigned long magic; /* File system specific magic number */
283
284 /* The following are set/used by kernfs_mount() */
285 bool new_sb_created; /* Set to T if we allocated a new sb */
286};
287
288#ifdef CONFIG_KERNFS
289
290static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
291{
292 return kn->flags & KERNFS_TYPE_MASK;
293}
294
295/**
296 * kernfs_enable_ns - enable namespace under a directory
297 * @kn: directory of interest, should be empty
298 *
299 * This is to be called right after @kn is created to enable namespace
300 * under it. All children of @kn must have non-NULL namespace tags and
301 * only the ones which match the super_block's tag will be visible.
302 */
303static inline void kernfs_enable_ns(struct kernfs_node *kn)
304{
305 WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
306 WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
307 kn->flags |= KERNFS_NS;
308}
309
310/**
311 * kernfs_ns_enabled - test whether namespace is enabled
312 * @kn: the node to test
313 *
314 * Test whether namespace filtering is enabled for the children of @ns.
315 */
316static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
317{
318 return kn->flags & KERNFS_NS;
319}
320
321int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
322int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
323 char *buf, size_t buflen);
324void pr_cont_kernfs_name(struct kernfs_node *kn);
325void pr_cont_kernfs_path(struct kernfs_node *kn);
326struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
327struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
328 const char *name, const void *ns);
329struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
330 const char *path, const void *ns);
331void kernfs_get(struct kernfs_node *kn);
332void kernfs_put(struct kernfs_node *kn);
333
334struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
335struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
336struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
337
338struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
339 struct super_block *sb);
340struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
341 unsigned int flags, void *priv);
342void kernfs_destroy_root(struct kernfs_root *root);
343
344struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
345 const char *name, umode_t mode,
346 kuid_t uid, kgid_t gid,
347 void *priv, const void *ns);
348struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
349 const char *name);
350struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
351 const char *name, umode_t mode,
352 kuid_t uid, kgid_t gid,
353 loff_t size,
354 const struct kernfs_ops *ops,
355 void *priv, const void *ns,
356 struct lock_class_key *key);
357struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
358 const char *name,
359 struct kernfs_node *target);
360void kernfs_activate(struct kernfs_node *kn);
361void kernfs_remove(struct kernfs_node *kn);
362void kernfs_break_active_protection(struct kernfs_node *kn);
363void kernfs_unbreak_active_protection(struct kernfs_node *kn);
364bool kernfs_remove_self(struct kernfs_node *kn);
365int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
366 const void *ns);
367int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
368 const char *new_name, const void *new_ns);
369int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
370__poll_t kernfs_generic_poll(struct kernfs_open_file *of,
371 struct poll_table_struct *pt);
372void kernfs_notify(struct kernfs_node *kn);
373
374const void *kernfs_super_ns(struct super_block *sb);
375int kernfs_get_tree(struct fs_context *fc);
376void kernfs_free_fs_context(struct fs_context *fc);
377void kernfs_kill_sb(struct super_block *sb);
378
379void kernfs_init(void);
380
381struct kernfs_node *kernfs_get_node_by_id(struct kernfs_root *root,
382 const union kernfs_node_id *id);
383#else /* CONFIG_KERNFS */
384
385static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
386{ return 0; } /* whatever */
387
388static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
389
390static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
391{ return false; }
392
393static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
394{ return -ENOSYS; }
395
396static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
397 struct kernfs_node *kn,
398 char *buf, size_t buflen)
399{ return -ENOSYS; }
400
401static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
402static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
403
404static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
405{ return NULL; }
406
407static inline struct kernfs_node *
408kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
409 const void *ns)
410{ return NULL; }
411static inline struct kernfs_node *
412kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
413 const void *ns)
414{ return NULL; }
415
416static inline void kernfs_get(struct kernfs_node *kn) { }
417static inline void kernfs_put(struct kernfs_node *kn) { }
418
419static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
420{ return NULL; }
421
422static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
423{ return NULL; }
424
425static inline struct inode *
426kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
427{ return NULL; }
428
429static inline struct kernfs_root *
430kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
431 void *priv)
432{ return ERR_PTR(-ENOSYS); }
433
434static inline void kernfs_destroy_root(struct kernfs_root *root) { }
435
436static inline struct kernfs_node *
437kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
438 umode_t mode, kuid_t uid, kgid_t gid,
439 void *priv, const void *ns)
440{ return ERR_PTR(-ENOSYS); }
441
442static inline struct kernfs_node *
443__kernfs_create_file(struct kernfs_node *parent, const char *name,
444 umode_t mode, kuid_t uid, kgid_t gid,
445 loff_t size, const struct kernfs_ops *ops,
446 void *priv, const void *ns, struct lock_class_key *key)
447{ return ERR_PTR(-ENOSYS); }
448
449static inline struct kernfs_node *
450kernfs_create_link(struct kernfs_node *parent, const char *name,
451 struct kernfs_node *target)
452{ return ERR_PTR(-ENOSYS); }
453
454static inline void kernfs_activate(struct kernfs_node *kn) { }
455
456static inline void kernfs_remove(struct kernfs_node *kn) { }
457
458static inline bool kernfs_remove_self(struct kernfs_node *kn)
459{ return false; }
460
461static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
462 const char *name, const void *ns)
463{ return -ENOSYS; }
464
465static inline int kernfs_rename_ns(struct kernfs_node *kn,
466 struct kernfs_node *new_parent,
467 const char *new_name, const void *new_ns)
468{ return -ENOSYS; }
469
470static inline int kernfs_setattr(struct kernfs_node *kn,
471 const struct iattr *iattr)
472{ return -ENOSYS; }
473
474static inline void kernfs_notify(struct kernfs_node *kn) { }
475
476static inline const void *kernfs_super_ns(struct super_block *sb)
477{ return NULL; }
478
479static inline int kernfs_get_tree(struct fs_context *fc)
480{ return -ENOSYS; }
481
482static inline void kernfs_free_fs_context(struct fs_context *fc) { }
483
484static inline void kernfs_kill_sb(struct super_block *sb) { }
485
486static inline void kernfs_init(void) { }
487
488#endif /* CONFIG_KERNFS */
489
490/**
491 * kernfs_path - build full path of a given node
492 * @kn: kernfs_node of interest
493 * @buf: buffer to copy @kn's name into
494 * @buflen: size of @buf
495 *
496 * If @kn is NULL result will be "(null)".
497 *
498 * Returns the length of the full path. If the full length is equal to or
499 * greater than @buflen, @buf contains the truncated path with the trailing
500 * '\0'. On error, -errno is returned.
501 */
502static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
503{
504 return kernfs_path_from_node(kn, NULL, buf, buflen);
505}
506
507static inline struct kernfs_node *
508kernfs_find_and_get(struct kernfs_node *kn, const char *name)
509{
510 return kernfs_find_and_get_ns(kn, name, NULL);
511}
512
513static inline struct kernfs_node *
514kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
515{
516 return kernfs_walk_and_get_ns(kn, path, NULL);
517}
518
519static inline struct kernfs_node *
520kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
521 void *priv)
522{
523 return kernfs_create_dir_ns(parent, name, mode,
524 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
525 priv, NULL);
526}
527
528static inline struct kernfs_node *
529kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
530 umode_t mode, kuid_t uid, kgid_t gid,
531 loff_t size, const struct kernfs_ops *ops,
532 void *priv, const void *ns)
533{
534 struct lock_class_key *key = NULL;
535
536#ifdef CONFIG_DEBUG_LOCK_ALLOC
537 key = (struct lock_class_key *)&ops->lockdep_key;
538#endif
539 return __kernfs_create_file(parent, name, mode, uid, gid,
540 size, ops, priv, ns, key);
541}
542
543static inline struct kernfs_node *
544kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
545 loff_t size, const struct kernfs_ops *ops, void *priv)
546{
547 return kernfs_create_file_ns(parent, name, mode,
548 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
549 size, ops, priv, NULL);
550}
551
552static inline int kernfs_remove_by_name(struct kernfs_node *parent,
553 const char *name)
554{
555 return kernfs_remove_by_name_ns(parent, name, NULL);
556}
557
558static inline int kernfs_rename(struct kernfs_node *kn,
559 struct kernfs_node *new_parent,
560 const char *new_name)
561{
562 return kernfs_rename_ns(kn, new_parent, new_name, NULL);
563}
564
565#endif /* __LINUX_KERNFS_H */
566