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