Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * fs/kernfs/dir.c - kernfs directory implementation
4 : *
5 : * Copyright (c) 2001-3 Patrick Mochel
6 : * Copyright (c) 2007 SUSE Linux Products GmbH
7 : * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
8 : */
9 :
10 : #include <linux/sched.h>
11 : #include <linux/fs.h>
12 : #include <linux/namei.h>
13 : #include <linux/idr.h>
14 : #include <linux/slab.h>
15 : #include <linux/security.h>
16 : #include <linux/hash.h>
17 :
18 : #include "kernfs-internal.h"
19 :
20 : static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
21 : /*
22 : * Don't use rename_lock to piggy back on pr_cont_buf. We don't want to
23 : * call pr_cont() while holding rename_lock. Because sometimes pr_cont()
24 : * will perform wakeups when releasing console_sem. Holding rename_lock
25 : * will introduce deadlock if the scheduler reads the kernfs_name in the
26 : * wakeup path.
27 : */
28 : static DEFINE_SPINLOCK(kernfs_pr_cont_lock);
29 : static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by pr_cont_lock */
30 : static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
31 :
32 : #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
33 :
34 : static bool __kernfs_active(struct kernfs_node *kn)
35 : {
36 19024 : return atomic_read(&kn->active) >= 0;
37 : }
38 :
39 : static bool kernfs_active(struct kernfs_node *kn)
40 : {
41 9512 : lockdep_assert_held(&kernfs_root(kn)->kernfs_rwsem);
42 9512 : return __kernfs_active(kn);
43 : }
44 :
45 : static bool kernfs_lockdep(struct kernfs_node *kn)
46 : {
47 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
48 : return kn->flags & KERNFS_LOCKDEP;
49 : #else
50 : return false;
51 : #endif
52 : }
53 :
54 0 : static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
55 : {
56 0 : if (!kn)
57 0 : return strlcpy(buf, "(null)", buflen);
58 :
59 0 : return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
60 : }
61 :
62 : /* kernfs_node_depth - compute depth from @from to @to */
63 : static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
64 : {
65 0 : size_t depth = 0;
66 :
67 0 : while (to->parent && to != from) {
68 0 : depth++;
69 0 : to = to->parent;
70 : }
71 : return depth;
72 : }
73 :
74 0 : static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
75 : struct kernfs_node *b)
76 : {
77 : size_t da, db;
78 0 : struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
79 :
80 0 : if (ra != rb)
81 : return NULL;
82 :
83 0 : da = kernfs_depth(ra->kn, a);
84 0 : db = kernfs_depth(rb->kn, b);
85 :
86 0 : while (da > db) {
87 0 : a = a->parent;
88 0 : da--;
89 : }
90 0 : while (db > da) {
91 0 : b = b->parent;
92 0 : db--;
93 : }
94 :
95 : /* worst case b and a will be the same at root */
96 0 : while (b != a) {
97 0 : b = b->parent;
98 0 : a = a->parent;
99 : }
100 :
101 : return a;
102 : }
103 :
104 : /**
105 : * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
106 : * where kn_from is treated as root of the path.
107 : * @kn_from: kernfs node which should be treated as root for the path
108 : * @kn_to: kernfs node to which path is needed
109 : * @buf: buffer to copy the path into
110 : * @buflen: size of @buf
111 : *
112 : * We need to handle couple of scenarios here:
113 : * [1] when @kn_from is an ancestor of @kn_to at some level
114 : * kn_from: /n1/n2/n3
115 : * kn_to: /n1/n2/n3/n4/n5
116 : * result: /n4/n5
117 : *
118 : * [2] when @kn_from is on a different hierarchy and we need to find common
119 : * ancestor between @kn_from and @kn_to.
120 : * kn_from: /n1/n2/n3/n4
121 : * kn_to: /n1/n2/n5
122 : * result: /../../n5
123 : * OR
124 : * kn_from: /n1/n2/n3/n4/n5 [depth=5]
125 : * kn_to: /n1/n2/n3 [depth=3]
126 : * result: /../..
127 : *
128 : * [3] when @kn_to is %NULL result will be "(null)"
129 : *
130 : * Return: the length of the full path. If the full length is equal to or
131 : * greater than @buflen, @buf contains the truncated path with the trailing
132 : * '\0'. On error, -errno is returned.
133 : */
134 0 : static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
135 : struct kernfs_node *kn_from,
136 : char *buf, size_t buflen)
137 : {
138 : struct kernfs_node *kn, *common;
139 0 : const char parent_str[] = "/..";
140 0 : size_t depth_from, depth_to, len = 0;
141 : int i, j;
142 :
143 0 : if (!kn_to)
144 0 : return strlcpy(buf, "(null)", buflen);
145 :
146 0 : if (!kn_from)
147 0 : kn_from = kernfs_root(kn_to)->kn;
148 :
149 0 : if (kn_from == kn_to)
150 0 : return strlcpy(buf, "/", buflen);
151 :
152 0 : common = kernfs_common_ancestor(kn_from, kn_to);
153 0 : if (WARN_ON(!common))
154 : return -EINVAL;
155 :
156 : depth_to = kernfs_depth(common, kn_to);
157 0 : depth_from = kernfs_depth(common, kn_from);
158 :
159 0 : buf[0] = '\0';
160 :
161 0 : for (i = 0; i < depth_from; i++)
162 0 : len += strlcpy(buf + len, parent_str,
163 : len < buflen ? buflen - len : 0);
164 :
165 : /* Calculate how many bytes we need for the rest */
166 0 : for (i = depth_to - 1; i >= 0; i--) {
167 0 : for (kn = kn_to, j = 0; j < i; j++)
168 0 : kn = kn->parent;
169 0 : len += strlcpy(buf + len, "/",
170 : len < buflen ? buflen - len : 0);
171 0 : len += strlcpy(buf + len, kn->name,
172 : len < buflen ? buflen - len : 0);
173 : }
174 :
175 0 : return len;
176 : }
177 :
178 : /**
179 : * kernfs_name - obtain the name of a given node
180 : * @kn: kernfs_node of interest
181 : * @buf: buffer to copy @kn's name into
182 : * @buflen: size of @buf
183 : *
184 : * Copies the name of @kn into @buf of @buflen bytes. The behavior is
185 : * similar to strlcpy().
186 : *
187 : * Fills buffer with "(null)" if @kn is %NULL.
188 : *
189 : * Return: the length of @kn's name and if @buf isn't long enough,
190 : * it's filled up to @buflen-1 and nul terminated.
191 : *
192 : * This function can be called from any context.
193 : */
194 0 : int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
195 : {
196 : unsigned long flags;
197 : int ret;
198 :
199 0 : spin_lock_irqsave(&kernfs_rename_lock, flags);
200 0 : ret = kernfs_name_locked(kn, buf, buflen);
201 0 : spin_unlock_irqrestore(&kernfs_rename_lock, flags);
202 0 : return ret;
203 : }
204 :
205 : /**
206 : * kernfs_path_from_node - build path of node @to relative to @from.
207 : * @from: parent kernfs_node relative to which we need to build the path
208 : * @to: kernfs_node of interest
209 : * @buf: buffer to copy @to's path into
210 : * @buflen: size of @buf
211 : *
212 : * Builds @to's path relative to @from in @buf. @from and @to must
213 : * be on the same kernfs-root. If @from is not parent of @to, then a relative
214 : * path (which includes '..'s) as needed to reach from @from to @to is
215 : * returned.
216 : *
217 : * Return: the length of the full path. If the full length is equal to or
218 : * greater than @buflen, @buf contains the truncated path with the trailing
219 : * '\0'. On error, -errno is returned.
220 : */
221 0 : int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
222 : char *buf, size_t buflen)
223 : {
224 : unsigned long flags;
225 : int ret;
226 :
227 0 : spin_lock_irqsave(&kernfs_rename_lock, flags);
228 0 : ret = kernfs_path_from_node_locked(to, from, buf, buflen);
229 0 : spin_unlock_irqrestore(&kernfs_rename_lock, flags);
230 0 : return ret;
231 : }
232 : EXPORT_SYMBOL_GPL(kernfs_path_from_node);
233 :
234 : /**
235 : * pr_cont_kernfs_name - pr_cont name of a kernfs_node
236 : * @kn: kernfs_node of interest
237 : *
238 : * This function can be called from any context.
239 : */
240 0 : void pr_cont_kernfs_name(struct kernfs_node *kn)
241 : {
242 : unsigned long flags;
243 :
244 0 : spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
245 :
246 0 : kernfs_name(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
247 0 : pr_cont("%s", kernfs_pr_cont_buf);
248 :
249 0 : spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
250 0 : }
251 :
252 : /**
253 : * pr_cont_kernfs_path - pr_cont path of a kernfs_node
254 : * @kn: kernfs_node of interest
255 : *
256 : * This function can be called from any context.
257 : */
258 0 : void pr_cont_kernfs_path(struct kernfs_node *kn)
259 : {
260 : unsigned long flags;
261 : int sz;
262 :
263 0 : spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
264 :
265 0 : sz = kernfs_path_from_node(kn, NULL, kernfs_pr_cont_buf,
266 : sizeof(kernfs_pr_cont_buf));
267 0 : if (sz < 0) {
268 0 : pr_cont("(error)");
269 0 : goto out;
270 : }
271 :
272 0 : if (sz >= sizeof(kernfs_pr_cont_buf)) {
273 0 : pr_cont("(name too long)");
274 0 : goto out;
275 : }
276 :
277 0 : pr_cont("%s", kernfs_pr_cont_buf);
278 :
279 : out:
280 0 : spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
281 0 : }
282 :
283 : /**
284 : * kernfs_get_parent - determine the parent node and pin it
285 : * @kn: kernfs_node of interest
286 : *
287 : * Determines @kn's parent, pins and returns it. This function can be
288 : * called from any context.
289 : *
290 : * Return: parent node of @kn
291 : */
292 0 : struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
293 : {
294 : struct kernfs_node *parent;
295 : unsigned long flags;
296 :
297 0 : spin_lock_irqsave(&kernfs_rename_lock, flags);
298 0 : parent = kn->parent;
299 0 : kernfs_get(parent);
300 0 : spin_unlock_irqrestore(&kernfs_rename_lock, flags);
301 :
302 0 : return parent;
303 : }
304 :
305 : /**
306 : * kernfs_name_hash - calculate hash of @ns + @name
307 : * @name: Null terminated string to hash
308 : * @ns: Namespace tag to hash
309 : *
310 : * Return: 31-bit hash of ns + name (so it fits in an off_t)
311 : */
312 10093 : static unsigned int kernfs_name_hash(const char *name, const void *ns)
313 : {
314 10093 : unsigned long hash = init_name_hash(ns);
315 10093 : unsigned int len = strlen(name);
316 121017 : while (len--)
317 201662 : hash = partial_name_hash(*name++, hash);
318 10093 : hash = end_name_hash(hash);
319 10093 : hash &= 0x7fffffffU;
320 : /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
321 10093 : if (hash < 2)
322 0 : hash += 2;
323 10093 : if (hash >= INT_MAX)
324 0 : hash = INT_MAX - 1;
325 10093 : return hash;
326 : }
327 :
328 27624 : static int kernfs_name_compare(unsigned int hash, const char *name,
329 : const void *ns, const struct kernfs_node *kn)
330 : {
331 27624 : if (hash < kn->hash)
332 : return -1;
333 13702 : if (hash > kn->hash)
334 : return 1;
335 1044 : if (ns < kn->ns)
336 : return -1;
337 1044 : if (ns > kn->ns)
338 : return 1;
339 1044 : return strcmp(name, kn->name);
340 : }
341 :
342 : static int kernfs_sd_compare(const struct kernfs_node *left,
343 : const struct kernfs_node *right)
344 : {
345 25301 : return kernfs_name_compare(left->hash, left->name, left->ns, right);
346 : }
347 :
348 : /**
349 : * kernfs_link_sibling - link kernfs_node into sibling rbtree
350 : * @kn: kernfs_node of interest
351 : *
352 : * Link @kn into its sibling rbtree which starts from
353 : * @kn->parent->dir.children.
354 : *
355 : * Locking:
356 : * kernfs_rwsem held exclusive
357 : *
358 : * Return:
359 : * %0 on success, -EEXIST on failure.
360 : */
361 8643 : static int kernfs_link_sibling(struct kernfs_node *kn)
362 : {
363 8643 : struct rb_node **node = &kn->parent->dir.children.rb_node;
364 8643 : struct rb_node *parent = NULL;
365 :
366 42587 : while (*node) {
367 : struct kernfs_node *pos;
368 : int result;
369 :
370 25301 : pos = rb_to_kn(*node);
371 25301 : parent = *node;
372 25301 : result = kernfs_sd_compare(kn, pos);
373 25301 : if (result < 0)
374 13183 : node = &pos->rb.rb_left;
375 12118 : else if (result > 0)
376 12118 : node = &pos->rb.rb_right;
377 : else
378 : return -EEXIST;
379 : }
380 :
381 : /* add new node and rebalance the tree */
382 17286 : rb_link_node(&kn->rb, parent, node);
383 8643 : rb_insert_color(&kn->rb, &kn->parent->dir.children);
384 :
385 : /* successfully added, account subdir number */
386 17286 : if (kernfs_type(kn) == KERNFS_DIR)
387 1332 : kn->parent->dir.subdirs++;
388 8643 : kernfs_inc_rev(kn->parent);
389 :
390 8643 : return 0;
391 : }
392 :
393 : /**
394 : * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
395 : * @kn: kernfs_node of interest
396 : *
397 : * Try to unlink @kn from its sibling rbtree which starts from
398 : * kn->parent->dir.children.
399 : *
400 : * Return: %true if @kn was actually removed,
401 : * %false if @kn wasn't on the rbtree.
402 : *
403 : * Locking:
404 : * kernfs_rwsem held exclusive
405 : */
406 434 : static bool kernfs_unlink_sibling(struct kernfs_node *kn)
407 : {
408 434 : if (RB_EMPTY_NODE(&kn->rb))
409 : return false;
410 :
411 868 : if (kernfs_type(kn) == KERNFS_DIR)
412 83 : kn->parent->dir.subdirs--;
413 434 : kernfs_inc_rev(kn->parent);
414 :
415 434 : rb_erase(&kn->rb, &kn->parent->dir.children);
416 434 : RB_CLEAR_NODE(&kn->rb);
417 434 : return true;
418 : }
419 :
420 : /**
421 : * kernfs_get_active - get an active reference to kernfs_node
422 : * @kn: kernfs_node to get an active reference to
423 : *
424 : * Get an active reference of @kn. This function is noop if @kn
425 : * is %NULL.
426 : *
427 : * Return:
428 : * Pointer to @kn on success, %NULL on failure.
429 : */
430 0 : struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
431 : {
432 0 : if (unlikely(!kn))
433 : return NULL;
434 :
435 0 : if (!atomic_inc_unless_negative(&kn->active))
436 : return NULL;
437 :
438 0 : if (kernfs_lockdep(kn))
439 : rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
440 0 : return kn;
441 : }
442 :
443 : /**
444 : * kernfs_put_active - put an active reference to kernfs_node
445 : * @kn: kernfs_node to put an active reference to
446 : *
447 : * Put an active reference to @kn. This function is noop if @kn
448 : * is %NULL.
449 : */
450 0 : void kernfs_put_active(struct kernfs_node *kn)
451 : {
452 : int v;
453 :
454 0 : if (unlikely(!kn))
455 : return;
456 :
457 0 : if (kernfs_lockdep(kn))
458 : rwsem_release(&kn->dep_map, _RET_IP_);
459 0 : v = atomic_dec_return(&kn->active);
460 0 : if (likely(v != KN_DEACTIVATED_BIAS))
461 : return;
462 :
463 0 : wake_up_all(&kernfs_root(kn)->deactivate_waitq);
464 : }
465 :
466 : /**
467 : * kernfs_drain - drain kernfs_node
468 : * @kn: kernfs_node to drain
469 : *
470 : * Drain existing usages and nuke all existing mmaps of @kn. Multiple
471 : * removers may invoke this function concurrently on @kn and all will
472 : * return after draining is complete.
473 : */
474 434 : static void kernfs_drain(struct kernfs_node *kn)
475 : __releases(&kernfs_root(kn)->kernfs_rwsem)
476 : __acquires(&kernfs_root(kn)->kernfs_rwsem)
477 : {
478 434 : struct kernfs_root *root = kernfs_root(kn);
479 :
480 : lockdep_assert_held_write(&root->kernfs_rwsem);
481 434 : WARN_ON_ONCE(kernfs_active(kn));
482 :
483 : /*
484 : * Skip draining if already fully drained. This avoids draining and its
485 : * lockdep annotations for nodes which have never been activated
486 : * allowing embedding kernfs_remove() in create error paths without
487 : * worrying about draining.
488 : */
489 1302 : if (atomic_read(&kn->active) == KN_DEACTIVATED_BIAS &&
490 434 : !kernfs_should_drain_open_files(kn))
491 : return;
492 :
493 0 : up_write(&root->kernfs_rwsem);
494 :
495 0 : if (kernfs_lockdep(kn)) {
496 : rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
497 : if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
498 : lock_contended(&kn->dep_map, _RET_IP_);
499 : }
500 :
501 0 : wait_event(root->deactivate_waitq,
502 : atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
503 :
504 0 : if (kernfs_lockdep(kn)) {
505 : lock_acquired(&kn->dep_map, _RET_IP_);
506 : rwsem_release(&kn->dep_map, _RET_IP_);
507 : }
508 :
509 0 : if (kernfs_should_drain_open_files(kn))
510 0 : kernfs_drain_open_files(kn);
511 :
512 0 : down_write(&root->kernfs_rwsem);
513 : }
514 :
515 : /**
516 : * kernfs_get - get a reference count on a kernfs_node
517 : * @kn: the target kernfs_node
518 : */
519 15033 : void kernfs_get(struct kernfs_node *kn)
520 : {
521 15033 : if (kn) {
522 30066 : WARN_ON(!atomic_read(&kn->count));
523 15033 : atomic_inc(&kn->count);
524 : }
525 15033 : }
526 : EXPORT_SYMBOL_GPL(kernfs_get);
527 :
528 : /**
529 : * kernfs_put - put a reference count on a kernfs_node
530 : * @kn: the target kernfs_node
531 : *
532 : * Put a reference count of @kn and destroy it if it reached zero.
533 : */
534 4473 : void kernfs_put(struct kernfs_node *kn)
535 : {
536 : struct kernfs_node *parent;
537 : struct kernfs_root *root;
538 :
539 8946 : if (!kn || !atomic_dec_and_test(&kn->count))
540 : return;
541 434 : root = kernfs_root(kn);
542 : repeat:
543 : /*
544 : * Moving/renaming is always done while holding reference.
545 : * kn->parent won't change beneath us.
546 : */
547 434 : parent = kn->parent;
548 :
549 868 : WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
550 : "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
551 : parent ? parent->name : "", kn->name, atomic_read(&kn->active));
552 :
553 868 : if (kernfs_type(kn) == KERNFS_LINK)
554 73 : kernfs_put(kn->symlink.target_kn);
555 :
556 434 : kfree_const(kn->name);
557 :
558 434 : if (kn->iattr) {
559 0 : simple_xattrs_free(&kn->iattr->xattrs);
560 0 : kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
561 : }
562 434 : spin_lock(&kernfs_idr_lock);
563 868 : idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
564 434 : spin_unlock(&kernfs_idr_lock);
565 434 : kmem_cache_free(kernfs_node_cache, kn);
566 :
567 434 : kn = parent;
568 434 : if (kn) {
569 868 : if (atomic_dec_and_test(&kn->count))
570 : goto repeat;
571 : } else {
572 : /* just released the root kn, free @root too */
573 0 : idr_destroy(&root->ino_idr);
574 0 : kfree(root);
575 : }
576 : }
577 : EXPORT_SYMBOL_GPL(kernfs_put);
578 :
579 : /**
580 : * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
581 : * @dentry: the dentry in question
582 : *
583 : * Return: the kernfs_node associated with @dentry. If @dentry is not a
584 : * kernfs one, %NULL is returned.
585 : *
586 : * While the returned kernfs_node will stay accessible as long as @dentry
587 : * is accessible, the returned node can be in any state and the caller is
588 : * fully responsible for determining what's accessible.
589 : */
590 0 : struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
591 : {
592 0 : if (dentry->d_sb->s_op == &kernfs_sops)
593 : return kernfs_dentry_node(dentry);
594 : return NULL;
595 : }
596 :
597 8644 : static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
598 : struct kernfs_node *parent,
599 : const char *name, umode_t mode,
600 : kuid_t uid, kgid_t gid,
601 : unsigned flags)
602 : {
603 : struct kernfs_node *kn;
604 : u32 id_highbits;
605 : int ret;
606 :
607 8644 : name = kstrdup_const(name, GFP_KERNEL);
608 8644 : if (!name)
609 : return NULL;
610 :
611 17288 : kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
612 8644 : if (!kn)
613 : goto err_out1;
614 :
615 8644 : idr_preload(GFP_KERNEL);
616 8644 : spin_lock(&kernfs_idr_lock);
617 8644 : ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
618 8644 : if (ret >= 0 && ret < root->last_id_lowbits)
619 0 : root->id_highbits++;
620 8644 : id_highbits = root->id_highbits;
621 8644 : root->last_id_lowbits = ret;
622 8644 : spin_unlock(&kernfs_idr_lock);
623 : idr_preload_end();
624 8644 : if (ret < 0)
625 : goto err_out2;
626 :
627 8644 : kn->id = (u64)id_highbits << 32 | ret;
628 :
629 17288 : atomic_set(&kn->count, 1);
630 17288 : atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
631 8644 : RB_CLEAR_NODE(&kn->rb);
632 :
633 8644 : kn->name = name;
634 8644 : kn->mode = mode;
635 8644 : kn->flags = flags;
636 :
637 8644 : if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
638 0 : struct iattr iattr = {
639 : .ia_valid = ATTR_UID | ATTR_GID,
640 : .ia_uid = uid,
641 : .ia_gid = gid,
642 : };
643 :
644 0 : ret = __kernfs_setattr(kn, &iattr);
645 0 : if (ret < 0)
646 : goto err_out3;
647 : }
648 :
649 : if (parent) {
650 : ret = security_kernfs_init_security(parent, kn);
651 : if (ret)
652 : goto err_out3;
653 : }
654 :
655 : return kn;
656 :
657 : err_out3:
658 0 : idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
659 : err_out2:
660 0 : kmem_cache_free(kernfs_node_cache, kn);
661 : err_out1:
662 0 : kfree_const(name);
663 0 : return NULL;
664 : }
665 :
666 8643 : struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
667 : const char *name, umode_t mode,
668 : kuid_t uid, kgid_t gid,
669 : unsigned flags)
670 : {
671 : struct kernfs_node *kn;
672 :
673 17286 : kn = __kernfs_new_node(kernfs_root(parent), parent,
674 : name, mode, uid, gid, flags);
675 8643 : if (kn) {
676 8643 : kernfs_get(parent);
677 8643 : kn->parent = parent;
678 : }
679 8643 : return kn;
680 : }
681 :
682 : /*
683 : * kernfs_find_and_get_node_by_id - get kernfs_node from node id
684 : * @root: the kernfs root
685 : * @id: the target node id
686 : *
687 : * @id's lower 32bits encode ino and upper gen. If the gen portion is
688 : * zero, all generations are matched.
689 : *
690 : * Return: %NULL on failure,
691 : * otherwise a kernfs node with reference counter incremented.
692 : */
693 0 : struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
694 : u64 id)
695 : {
696 : struct kernfs_node *kn;
697 0 : ino_t ino = kernfs_id_ino(id);
698 0 : u32 gen = kernfs_id_gen(id);
699 :
700 0 : spin_lock(&kernfs_idr_lock);
701 :
702 0 : kn = idr_find(&root->ino_idr, (u32)ino);
703 0 : if (!kn)
704 : goto err_unlock;
705 :
706 : if (sizeof(ino_t) >= sizeof(u64)) {
707 : /* we looked up with the low 32bits, compare the whole */
708 0 : if (kernfs_ino(kn) != ino)
709 : goto err_unlock;
710 : } else {
711 : /* 0 matches all generations */
712 : if (unlikely(gen && kernfs_gen(kn) != gen))
713 : goto err_unlock;
714 : }
715 :
716 : /*
717 : * We should fail if @kn has never been activated and guarantee success
718 : * if the caller knows that @kn is active. Both can be achieved by
719 : * __kernfs_active() which tests @kn->active without kernfs_rwsem.
720 : */
721 0 : if (unlikely(!__kernfs_active(kn) || !atomic_inc_not_zero(&kn->count)))
722 : goto err_unlock;
723 :
724 0 : spin_unlock(&kernfs_idr_lock);
725 0 : return kn;
726 : err_unlock:
727 0 : spin_unlock(&kernfs_idr_lock);
728 0 : return NULL;
729 : }
730 :
731 : /**
732 : * kernfs_add_one - add kernfs_node to parent without warning
733 : * @kn: kernfs_node to be added
734 : *
735 : * The caller must already have initialized @kn->parent. This
736 : * function increments nlink of the parent's inode if @kn is a
737 : * directory and link into the children list of the parent.
738 : *
739 : * Return:
740 : * %0 on success, -EEXIST if entry with the given name already
741 : * exists.
742 : */
743 8643 : int kernfs_add_one(struct kernfs_node *kn)
744 : {
745 8643 : struct kernfs_node *parent = kn->parent;
746 8643 : struct kernfs_root *root = kernfs_root(parent);
747 : struct kernfs_iattrs *ps_iattr;
748 : bool has_ns;
749 : int ret;
750 :
751 8643 : down_write(&root->kernfs_rwsem);
752 :
753 8643 : ret = -EINVAL;
754 17286 : has_ns = kernfs_ns_enabled(parent);
755 8643 : if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
756 : has_ns ? "required" : "invalid", parent->name, kn->name))
757 : goto out_unlock;
758 :
759 17286 : if (kernfs_type(parent) != KERNFS_DIR)
760 : goto out_unlock;
761 :
762 8643 : ret = -ENOENT;
763 8643 : if (parent->flags & (KERNFS_REMOVING | KERNFS_EMPTY_DIR))
764 : goto out_unlock;
765 :
766 8643 : kn->hash = kernfs_name_hash(kn->name, kn->ns);
767 :
768 8643 : ret = kernfs_link_sibling(kn);
769 8643 : if (ret)
770 : goto out_unlock;
771 :
772 : /* Update timestamps on the parent */
773 8643 : ps_iattr = parent->iattr;
774 8643 : if (ps_iattr) {
775 0 : ktime_get_real_ts64(&ps_iattr->ia_ctime);
776 0 : ps_iattr->ia_mtime = ps_iattr->ia_ctime;
777 : }
778 :
779 8643 : up_write(&root->kernfs_rwsem);
780 :
781 : /*
782 : * Activate the new node unless CREATE_DEACTIVATED is requested.
783 : * If not activated here, the kernfs user is responsible for
784 : * activating the node with kernfs_activate(). A node which hasn't
785 : * been activated is not visible to userland and its removal won't
786 : * trigger deactivation.
787 : */
788 8643 : if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
789 8643 : kernfs_activate(kn);
790 : return 0;
791 :
792 : out_unlock:
793 0 : up_write(&root->kernfs_rwsem);
794 0 : return ret;
795 : }
796 :
797 : /**
798 : * kernfs_find_ns - find kernfs_node with the given name
799 : * @parent: kernfs_node to search under
800 : * @name: name to look for
801 : * @ns: the namespace tag to use
802 : *
803 : * Look for kernfs_node with name @name under @parent.
804 : *
805 : * Return: pointer to the found kernfs_node on success, %NULL on failure.
806 : */
807 1450 : static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
808 : const unsigned char *name,
809 : const void *ns)
810 : {
811 1450 : struct rb_node *node = parent->dir.children.rb_node;
812 2900 : bool has_ns = kernfs_ns_enabled(parent);
813 : unsigned int hash;
814 :
815 1450 : lockdep_assert_held(&kernfs_root(parent)->kernfs_rwsem);
816 :
817 1450 : if (has_ns != (bool)ns) {
818 0 : WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
819 : has_ns ? "required" : "invalid", parent->name, name);
820 0 : return NULL;
821 : }
822 :
823 1450 : hash = kernfs_name_hash(name, ns);
824 4179 : while (node) {
825 : struct kernfs_node *kn;
826 : int result;
827 :
828 2323 : kn = rb_to_kn(node);
829 2323 : result = kernfs_name_compare(hash, name, ns, kn);
830 2323 : if (result < 0)
831 739 : node = node->rb_left;
832 1584 : else if (result > 0)
833 540 : node = node->rb_right;
834 : else
835 : return kn;
836 : }
837 : return NULL;
838 : }
839 :
840 0 : static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
841 : const unsigned char *path,
842 : const void *ns)
843 : {
844 : size_t len;
845 : char *p, *name;
846 :
847 0 : lockdep_assert_held_read(&kernfs_root(parent)->kernfs_rwsem);
848 :
849 0 : spin_lock_irq(&kernfs_pr_cont_lock);
850 :
851 0 : len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
852 :
853 0 : if (len >= sizeof(kernfs_pr_cont_buf)) {
854 0 : spin_unlock_irq(&kernfs_pr_cont_lock);
855 0 : return NULL;
856 : }
857 :
858 0 : p = kernfs_pr_cont_buf;
859 :
860 0 : while ((name = strsep(&p, "/")) && parent) {
861 0 : if (*name == '\0')
862 0 : continue;
863 0 : parent = kernfs_find_ns(parent, name, ns);
864 : }
865 :
866 0 : spin_unlock_irq(&kernfs_pr_cont_lock);
867 :
868 0 : return parent;
869 : }
870 :
871 : /**
872 : * kernfs_find_and_get_ns - find and get kernfs_node with the given name
873 : * @parent: kernfs_node to search under
874 : * @name: name to look for
875 : * @ns: the namespace tag to use
876 : *
877 : * Look for kernfs_node with name @name under @parent and get a reference
878 : * if found. This function may sleep.
879 : *
880 : * Return: pointer to the found kernfs_node on success, %NULL on failure.
881 : */
882 693 : struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
883 : const char *name, const void *ns)
884 : {
885 : struct kernfs_node *kn;
886 693 : struct kernfs_root *root = kernfs_root(parent);
887 :
888 693 : down_read(&root->kernfs_rwsem);
889 693 : kn = kernfs_find_ns(parent, name, ns);
890 693 : kernfs_get(kn);
891 693 : up_read(&root->kernfs_rwsem);
892 :
893 693 : return kn;
894 : }
895 : EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
896 :
897 : /**
898 : * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
899 : * @parent: kernfs_node to search under
900 : * @path: path to look for
901 : * @ns: the namespace tag to use
902 : *
903 : * Look for kernfs_node with path @path under @parent and get a reference
904 : * if found. This function may sleep.
905 : *
906 : * Return: pointer to the found kernfs_node on success, %NULL on failure.
907 : */
908 0 : struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
909 : const char *path, const void *ns)
910 : {
911 : struct kernfs_node *kn;
912 0 : struct kernfs_root *root = kernfs_root(parent);
913 :
914 0 : down_read(&root->kernfs_rwsem);
915 0 : kn = kernfs_walk_ns(parent, path, ns);
916 0 : kernfs_get(kn);
917 0 : up_read(&root->kernfs_rwsem);
918 :
919 0 : return kn;
920 : }
921 :
922 : /**
923 : * kernfs_create_root - create a new kernfs hierarchy
924 : * @scops: optional syscall operations for the hierarchy
925 : * @flags: KERNFS_ROOT_* flags
926 : * @priv: opaque data associated with the new directory
927 : *
928 : * Return: the root of the new hierarchy on success, ERR_PTR() value on
929 : * failure.
930 : */
931 1 : struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
932 : unsigned int flags, void *priv)
933 : {
934 : struct kernfs_root *root;
935 : struct kernfs_node *kn;
936 :
937 1 : root = kzalloc(sizeof(*root), GFP_KERNEL);
938 1 : if (!root)
939 : return ERR_PTR(-ENOMEM);
940 :
941 2 : idr_init(&root->ino_idr);
942 1 : init_rwsem(&root->kernfs_rwsem);
943 2 : INIT_LIST_HEAD(&root->supers);
944 :
945 : /*
946 : * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
947 : * High bits generation. The starting value for both ino and
948 : * genenration is 1. Initialize upper 32bit allocation
949 : * accordingly.
950 : */
951 : if (sizeof(ino_t) >= sizeof(u64))
952 1 : root->id_highbits = 0;
953 : else
954 : root->id_highbits = 1;
955 :
956 1 : kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
957 1 : GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
958 : KERNFS_DIR);
959 1 : if (!kn) {
960 0 : idr_destroy(&root->ino_idr);
961 0 : kfree(root);
962 0 : return ERR_PTR(-ENOMEM);
963 : }
964 :
965 1 : kn->priv = priv;
966 1 : kn->dir.root = root;
967 :
968 1 : root->syscall_ops = scops;
969 1 : root->flags = flags;
970 1 : root->kn = kn;
971 1 : init_waitqueue_head(&root->deactivate_waitq);
972 :
973 1 : if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
974 1 : kernfs_activate(kn);
975 :
976 : return root;
977 : }
978 :
979 : /**
980 : * kernfs_destroy_root - destroy a kernfs hierarchy
981 : * @root: root of the hierarchy to destroy
982 : *
983 : * Destroy the hierarchy anchored at @root by removing all existing
984 : * directories and destroying @root.
985 : */
986 0 : void kernfs_destroy_root(struct kernfs_root *root)
987 : {
988 : /*
989 : * kernfs_remove holds kernfs_rwsem from the root so the root
990 : * shouldn't be freed during the operation.
991 : */
992 0 : kernfs_get(root->kn);
993 0 : kernfs_remove(root->kn);
994 0 : kernfs_put(root->kn); /* will also free @root */
995 0 : }
996 :
997 : /**
998 : * kernfs_root_to_node - return the kernfs_node associated with a kernfs_root
999 : * @root: root to use to lookup
1000 : *
1001 : * Return: @root's kernfs_node
1002 : */
1003 1 : struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root)
1004 : {
1005 1 : return root->kn;
1006 : }
1007 :
1008 : /**
1009 : * kernfs_create_dir_ns - create a directory
1010 : * @parent: parent in which to create a new directory
1011 : * @name: name of the new directory
1012 : * @mode: mode of the new directory
1013 : * @uid: uid of the new directory
1014 : * @gid: gid of the new directory
1015 : * @priv: opaque data associated with the new directory
1016 : * @ns: optional namespace tag of the directory
1017 : *
1018 : * Return: the created node on success, ERR_PTR() value on failure.
1019 : */
1020 1332 : struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1021 : const char *name, umode_t mode,
1022 : kuid_t uid, kgid_t gid,
1023 : void *priv, const void *ns)
1024 : {
1025 : struct kernfs_node *kn;
1026 : int rc;
1027 :
1028 : /* allocate */
1029 1332 : kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1030 : uid, gid, KERNFS_DIR);
1031 1332 : if (!kn)
1032 : return ERR_PTR(-ENOMEM);
1033 :
1034 1332 : kn->dir.root = parent->dir.root;
1035 1332 : kn->ns = ns;
1036 1332 : kn->priv = priv;
1037 :
1038 : /* link in */
1039 1332 : rc = kernfs_add_one(kn);
1040 1332 : if (!rc)
1041 : return kn;
1042 :
1043 0 : kernfs_put(kn);
1044 0 : return ERR_PTR(rc);
1045 : }
1046 :
1047 : /**
1048 : * kernfs_create_empty_dir - create an always empty directory
1049 : * @parent: parent in which to create a new directory
1050 : * @name: name of the new directory
1051 : *
1052 : * Return: the created node on success, ERR_PTR() value on failure.
1053 : */
1054 0 : struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1055 : const char *name)
1056 : {
1057 : struct kernfs_node *kn;
1058 : int rc;
1059 :
1060 : /* allocate */
1061 0 : kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1062 0 : GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1063 0 : if (!kn)
1064 : return ERR_PTR(-ENOMEM);
1065 :
1066 0 : kn->flags |= KERNFS_EMPTY_DIR;
1067 0 : kn->dir.root = parent->dir.root;
1068 0 : kn->ns = NULL;
1069 0 : kn->priv = NULL;
1070 :
1071 : /* link in */
1072 0 : rc = kernfs_add_one(kn);
1073 0 : if (!rc)
1074 : return kn;
1075 :
1076 0 : kernfs_put(kn);
1077 0 : return ERR_PTR(rc);
1078 : }
1079 :
1080 0 : static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1081 : {
1082 : struct kernfs_node *kn;
1083 : struct kernfs_root *root;
1084 :
1085 0 : if (flags & LOOKUP_RCU)
1086 : return -ECHILD;
1087 :
1088 : /* Negative hashed dentry? */
1089 0 : if (d_really_is_negative(dentry)) {
1090 : struct kernfs_node *parent;
1091 :
1092 : /* If the kernfs parent node has changed discard and
1093 : * proceed to ->lookup.
1094 : *
1095 : * There's nothing special needed here when getting the
1096 : * dentry parent, even if a concurrent rename is in
1097 : * progress. That's because the dentry is negative so
1098 : * it can only be the target of the rename and it will
1099 : * be doing a d_move() not a replace. Consequently the
1100 : * dentry d_parent won't change over the d_move().
1101 : *
1102 : * Also kernfs negative dentries transitioning from
1103 : * negative to positive during revalidate won't happen
1104 : * because they are invalidated on containing directory
1105 : * changes and the lookup re-done so that a new positive
1106 : * dentry can be properly created.
1107 : */
1108 0 : root = kernfs_root_from_sb(dentry->d_sb);
1109 0 : down_read(&root->kernfs_rwsem);
1110 0 : parent = kernfs_dentry_node(dentry->d_parent);
1111 0 : if (parent) {
1112 0 : if (kernfs_dir_changed(parent, dentry)) {
1113 0 : up_read(&root->kernfs_rwsem);
1114 0 : return 0;
1115 : }
1116 : }
1117 0 : up_read(&root->kernfs_rwsem);
1118 :
1119 : /* The kernfs parent node hasn't changed, leave the
1120 : * dentry negative and return success.
1121 : */
1122 0 : return 1;
1123 : }
1124 :
1125 0 : kn = kernfs_dentry_node(dentry);
1126 0 : root = kernfs_root(kn);
1127 0 : down_read(&root->kernfs_rwsem);
1128 :
1129 : /* The kernfs node has been deactivated */
1130 0 : if (!kernfs_active(kn))
1131 : goto out_bad;
1132 :
1133 : /* The kernfs node has been moved? */
1134 0 : if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1135 : goto out_bad;
1136 :
1137 : /* The kernfs node has been renamed */
1138 0 : if (strcmp(dentry->d_name.name, kn->name) != 0)
1139 : goto out_bad;
1140 :
1141 : /* The kernfs node has been moved to a different namespace */
1142 0 : if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1143 0 : kernfs_info(dentry->d_sb)->ns != kn->ns)
1144 : goto out_bad;
1145 :
1146 0 : up_read(&root->kernfs_rwsem);
1147 0 : return 1;
1148 : out_bad:
1149 0 : up_read(&root->kernfs_rwsem);
1150 0 : return 0;
1151 : }
1152 :
1153 : const struct dentry_operations kernfs_dops = {
1154 : .d_revalidate = kernfs_dop_revalidate,
1155 : };
1156 :
1157 0 : static struct dentry *kernfs_iop_lookup(struct inode *dir,
1158 : struct dentry *dentry,
1159 : unsigned int flags)
1160 : {
1161 0 : struct kernfs_node *parent = dir->i_private;
1162 : struct kernfs_node *kn;
1163 : struct kernfs_root *root;
1164 0 : struct inode *inode = NULL;
1165 0 : const void *ns = NULL;
1166 :
1167 0 : root = kernfs_root(parent);
1168 0 : down_read(&root->kernfs_rwsem);
1169 0 : if (kernfs_ns_enabled(parent))
1170 0 : ns = kernfs_info(dir->i_sb)->ns;
1171 :
1172 0 : kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1173 : /* attach dentry and inode */
1174 0 : if (kn) {
1175 : /* Inactive nodes are invisible to the VFS so don't
1176 : * create a negative.
1177 : */
1178 0 : if (!kernfs_active(kn)) {
1179 0 : up_read(&root->kernfs_rwsem);
1180 0 : return NULL;
1181 : }
1182 0 : inode = kernfs_get_inode(dir->i_sb, kn);
1183 0 : if (!inode)
1184 0 : inode = ERR_PTR(-ENOMEM);
1185 : }
1186 : /*
1187 : * Needed for negative dentry validation.
1188 : * The negative dentry can be created in kernfs_iop_lookup()
1189 : * or transforms from positive dentry in dentry_unlink_inode()
1190 : * called from vfs_rmdir().
1191 : */
1192 0 : if (!IS_ERR(inode))
1193 0 : kernfs_set_rev(parent, dentry);
1194 0 : up_read(&root->kernfs_rwsem);
1195 :
1196 : /* instantiate and hash (possibly negative) dentry */
1197 0 : return d_splice_alias(inode, dentry);
1198 : }
1199 :
1200 0 : static int kernfs_iop_mkdir(struct mnt_idmap *idmap,
1201 : struct inode *dir, struct dentry *dentry,
1202 : umode_t mode)
1203 : {
1204 0 : struct kernfs_node *parent = dir->i_private;
1205 0 : struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1206 : int ret;
1207 :
1208 0 : if (!scops || !scops->mkdir)
1209 : return -EPERM;
1210 :
1211 0 : if (!kernfs_get_active(parent))
1212 : return -ENODEV;
1213 :
1214 0 : ret = scops->mkdir(parent, dentry->d_name.name, mode);
1215 :
1216 0 : kernfs_put_active(parent);
1217 0 : return ret;
1218 : }
1219 :
1220 0 : static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1221 : {
1222 0 : struct kernfs_node *kn = kernfs_dentry_node(dentry);
1223 0 : struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1224 : int ret;
1225 :
1226 0 : if (!scops || !scops->rmdir)
1227 : return -EPERM;
1228 :
1229 0 : if (!kernfs_get_active(kn))
1230 : return -ENODEV;
1231 :
1232 0 : ret = scops->rmdir(kn);
1233 :
1234 0 : kernfs_put_active(kn);
1235 0 : return ret;
1236 : }
1237 :
1238 0 : static int kernfs_iop_rename(struct mnt_idmap *idmap,
1239 : struct inode *old_dir, struct dentry *old_dentry,
1240 : struct inode *new_dir, struct dentry *new_dentry,
1241 : unsigned int flags)
1242 : {
1243 0 : struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1244 0 : struct kernfs_node *new_parent = new_dir->i_private;
1245 0 : struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1246 : int ret;
1247 :
1248 0 : if (flags)
1249 : return -EINVAL;
1250 :
1251 0 : if (!scops || !scops->rename)
1252 : return -EPERM;
1253 :
1254 0 : if (!kernfs_get_active(kn))
1255 : return -ENODEV;
1256 :
1257 0 : if (!kernfs_get_active(new_parent)) {
1258 0 : kernfs_put_active(kn);
1259 0 : return -ENODEV;
1260 : }
1261 :
1262 0 : ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1263 :
1264 0 : kernfs_put_active(new_parent);
1265 0 : kernfs_put_active(kn);
1266 0 : return ret;
1267 : }
1268 :
1269 : const struct inode_operations kernfs_dir_iops = {
1270 : .lookup = kernfs_iop_lookup,
1271 : .permission = kernfs_iop_permission,
1272 : .setattr = kernfs_iop_setattr,
1273 : .getattr = kernfs_iop_getattr,
1274 : .listxattr = kernfs_iop_listxattr,
1275 :
1276 : .mkdir = kernfs_iop_mkdir,
1277 : .rmdir = kernfs_iop_rmdir,
1278 : .rename = kernfs_iop_rename,
1279 : };
1280 :
1281 : static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1282 : {
1283 : struct kernfs_node *last;
1284 :
1285 : while (true) {
1286 : struct rb_node *rbn;
1287 :
1288 9512 : last = pos;
1289 :
1290 9512 : if (kernfs_type(pos) != KERNFS_DIR)
1291 : break;
1292 :
1293 1499 : rbn = rb_first(&pos->dir.children);
1294 1499 : if (!rbn)
1295 : break;
1296 :
1297 0 : pos = rb_to_kn(rbn);
1298 : }
1299 :
1300 : return last;
1301 : }
1302 :
1303 : /**
1304 : * kernfs_next_descendant_post - find the next descendant for post-order walk
1305 : * @pos: the current position (%NULL to initiate traversal)
1306 : * @root: kernfs_node whose descendants to walk
1307 : *
1308 : * Find the next descendant to visit for post-order traversal of @root's
1309 : * descendants. @root is included in the iteration and the last node to be
1310 : * visited.
1311 : *
1312 : * Return: the next descendant to visit or %NULL when done.
1313 : */
1314 18156 : static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1315 : struct kernfs_node *root)
1316 : {
1317 : struct rb_node *rbn;
1318 :
1319 18156 : lockdep_assert_held_write(&kernfs_root(root)->kernfs_rwsem);
1320 :
1321 : /* if first iteration, visit leftmost descendant which may be root */
1322 18156 : if (!pos)
1323 : return kernfs_leftmost_descendant(root);
1324 :
1325 : /* if we visited @root, we're done */
1326 9078 : if (pos == root)
1327 : return NULL;
1328 :
1329 : /* if there's an unvisited sibling, visit its leftmost descendant */
1330 0 : rbn = rb_next(&pos->rb);
1331 0 : if (rbn)
1332 0 : return kernfs_leftmost_descendant(rb_to_kn(rbn));
1333 :
1334 : /* no sibling left, visit parent */
1335 0 : return pos->parent;
1336 : }
1337 :
1338 8644 : static void kernfs_activate_one(struct kernfs_node *kn)
1339 : {
1340 8644 : lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1341 :
1342 8644 : kn->flags |= KERNFS_ACTIVATED;
1343 :
1344 8644 : if (kernfs_active(kn) || (kn->flags & (KERNFS_HIDDEN | KERNFS_REMOVING)))
1345 : return;
1346 :
1347 8644 : WARN_ON_ONCE(kn->parent && RB_EMPTY_NODE(&kn->rb));
1348 17288 : WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1349 :
1350 8644 : atomic_sub(KN_DEACTIVATED_BIAS, &kn->active);
1351 : }
1352 :
1353 : /**
1354 : * kernfs_activate - activate a node which started deactivated
1355 : * @kn: kernfs_node whose subtree is to be activated
1356 : *
1357 : * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1358 : * needs to be explicitly activated. A node which hasn't been activated
1359 : * isn't visible to userland and deactivation is skipped during its
1360 : * removal. This is useful to construct atomic init sequences where
1361 : * creation of multiple nodes should either succeed or fail atomically.
1362 : *
1363 : * The caller is responsible for ensuring that this function is not called
1364 : * after kernfs_remove*() is invoked on @kn.
1365 : */
1366 8644 : void kernfs_activate(struct kernfs_node *kn)
1367 : {
1368 : struct kernfs_node *pos;
1369 8644 : struct kernfs_root *root = kernfs_root(kn);
1370 :
1371 8644 : down_write(&root->kernfs_rwsem);
1372 :
1373 8644 : pos = NULL;
1374 25932 : while ((pos = kernfs_next_descendant_post(pos, kn)))
1375 8644 : kernfs_activate_one(pos);
1376 :
1377 8644 : up_write(&root->kernfs_rwsem);
1378 8644 : }
1379 :
1380 : /**
1381 : * kernfs_show - show or hide a node
1382 : * @kn: kernfs_node to show or hide
1383 : * @show: whether to show or hide
1384 : *
1385 : * If @show is %false, @kn is marked hidden and deactivated. A hidden node is
1386 : * ignored in future activaitons. If %true, the mark is removed and activation
1387 : * state is restored. This function won't implicitly activate a new node in a
1388 : * %KERNFS_ROOT_CREATE_DEACTIVATED root which hasn't been activated yet.
1389 : *
1390 : * To avoid recursion complexities, directories aren't supported for now.
1391 : */
1392 0 : void kernfs_show(struct kernfs_node *kn, bool show)
1393 : {
1394 0 : struct kernfs_root *root = kernfs_root(kn);
1395 :
1396 0 : if (WARN_ON_ONCE(kernfs_type(kn) == KERNFS_DIR))
1397 : return;
1398 :
1399 0 : down_write(&root->kernfs_rwsem);
1400 :
1401 0 : if (show) {
1402 0 : kn->flags &= ~KERNFS_HIDDEN;
1403 0 : if (kn->flags & KERNFS_ACTIVATED)
1404 0 : kernfs_activate_one(kn);
1405 : } else {
1406 0 : kn->flags |= KERNFS_HIDDEN;
1407 0 : if (kernfs_active(kn))
1408 0 : atomic_add(KN_DEACTIVATED_BIAS, &kn->active);
1409 0 : kernfs_drain(kn);
1410 : }
1411 :
1412 0 : up_write(&root->kernfs_rwsem);
1413 : }
1414 :
1415 434 : static void __kernfs_remove(struct kernfs_node *kn)
1416 : {
1417 : struct kernfs_node *pos;
1418 :
1419 : /* Short-circuit if non-root @kn has already finished removal. */
1420 434 : if (!kn)
1421 : return;
1422 :
1423 434 : lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1424 :
1425 : /*
1426 : * This is for kernfs_remove_self() which plays with active ref
1427 : * after removal.
1428 : */
1429 434 : if (kn->parent && RB_EMPTY_NODE(&kn->rb))
1430 : return;
1431 :
1432 : pr_debug("kernfs %s: removing\n", kn->name);
1433 :
1434 : /* prevent new usage by marking all nodes removing and deactivating */
1435 : pos = NULL;
1436 868 : while ((pos = kernfs_next_descendant_post(pos, kn))) {
1437 434 : pos->flags |= KERNFS_REMOVING;
1438 434 : if (kernfs_active(pos))
1439 434 : atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1440 : }
1441 :
1442 : /* deactivate and unlink the subtree node-by-node */
1443 : do {
1444 434 : pos = kernfs_leftmost_descendant(kn);
1445 :
1446 : /*
1447 : * kernfs_drain() may drop kernfs_rwsem temporarily and @pos's
1448 : * base ref could have been put by someone else by the time
1449 : * the function returns. Make sure it doesn't go away
1450 : * underneath us.
1451 : */
1452 434 : kernfs_get(pos);
1453 :
1454 434 : kernfs_drain(pos);
1455 :
1456 : /*
1457 : * kernfs_unlink_sibling() succeeds once per node. Use it
1458 : * to decide who's responsible for cleanups.
1459 : */
1460 434 : if (!pos->parent || kernfs_unlink_sibling(pos)) {
1461 434 : struct kernfs_iattrs *ps_iattr =
1462 434 : pos->parent ? pos->parent->iattr : NULL;
1463 :
1464 : /* update timestamps on the parent */
1465 434 : if (ps_iattr) {
1466 0 : ktime_get_real_ts64(&ps_iattr->ia_ctime);
1467 0 : ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1468 : }
1469 :
1470 434 : kernfs_put(pos);
1471 : }
1472 :
1473 434 : kernfs_put(pos);
1474 434 : } while (pos != kn);
1475 : }
1476 :
1477 : /**
1478 : * kernfs_remove - remove a kernfs_node recursively
1479 : * @kn: the kernfs_node to remove
1480 : *
1481 : * Remove @kn along with all its subdirectories and files.
1482 : */
1483 83 : void kernfs_remove(struct kernfs_node *kn)
1484 : {
1485 : struct kernfs_root *root;
1486 :
1487 83 : if (!kn)
1488 : return;
1489 :
1490 83 : root = kernfs_root(kn);
1491 :
1492 83 : down_write(&root->kernfs_rwsem);
1493 83 : __kernfs_remove(kn);
1494 83 : up_write(&root->kernfs_rwsem);
1495 : }
1496 :
1497 : /**
1498 : * kernfs_break_active_protection - break out of active protection
1499 : * @kn: the self kernfs_node
1500 : *
1501 : * The caller must be running off of a kernfs operation which is invoked
1502 : * with an active reference - e.g. one of kernfs_ops. Each invocation of
1503 : * this function must also be matched with an invocation of
1504 : * kernfs_unbreak_active_protection().
1505 : *
1506 : * This function releases the active reference of @kn the caller is
1507 : * holding. Once this function is called, @kn may be removed at any point
1508 : * and the caller is solely responsible for ensuring that the objects it
1509 : * dereferences are accessible.
1510 : */
1511 0 : void kernfs_break_active_protection(struct kernfs_node *kn)
1512 : {
1513 : /*
1514 : * Take out ourself out of the active ref dependency chain. If
1515 : * we're called without an active ref, lockdep will complain.
1516 : */
1517 0 : kernfs_put_active(kn);
1518 0 : }
1519 :
1520 : /**
1521 : * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1522 : * @kn: the self kernfs_node
1523 : *
1524 : * If kernfs_break_active_protection() was called, this function must be
1525 : * invoked before finishing the kernfs operation. Note that while this
1526 : * function restores the active reference, it doesn't and can't actually
1527 : * restore the active protection - @kn may already or be in the process of
1528 : * being removed. Once kernfs_break_active_protection() is invoked, that
1529 : * protection is irreversibly gone for the kernfs operation instance.
1530 : *
1531 : * While this function may be called at any point after
1532 : * kernfs_break_active_protection() is invoked, its most useful location
1533 : * would be right before the enclosing kernfs operation returns.
1534 : */
1535 0 : void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1536 : {
1537 : /*
1538 : * @kn->active could be in any state; however, the increment we do
1539 : * here will be undone as soon as the enclosing kernfs operation
1540 : * finishes and this temporary bump can't break anything. If @kn
1541 : * is alive, nothing changes. If @kn is being deactivated, the
1542 : * soon-to-follow put will either finish deactivation or restore
1543 : * deactivated state. If @kn is already removed, the temporary
1544 : * bump is guaranteed to be gone before @kn is released.
1545 : */
1546 0 : atomic_inc(&kn->active);
1547 0 : if (kernfs_lockdep(kn))
1548 : rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1549 0 : }
1550 :
1551 : /**
1552 : * kernfs_remove_self - remove a kernfs_node from its own method
1553 : * @kn: the self kernfs_node to remove
1554 : *
1555 : * The caller must be running off of a kernfs operation which is invoked
1556 : * with an active reference - e.g. one of kernfs_ops. This can be used to
1557 : * implement a file operation which deletes itself.
1558 : *
1559 : * For example, the "delete" file for a sysfs device directory can be
1560 : * implemented by invoking kernfs_remove_self() on the "delete" file
1561 : * itself. This function breaks the circular dependency of trying to
1562 : * deactivate self while holding an active ref itself. It isn't necessary
1563 : * to modify the usual removal path to use kernfs_remove_self(). The
1564 : * "delete" implementation can simply invoke kernfs_remove_self() on self
1565 : * before proceeding with the usual removal path. kernfs will ignore later
1566 : * kernfs_remove() on self.
1567 : *
1568 : * kernfs_remove_self() can be called multiple times concurrently on the
1569 : * same kernfs_node. Only the first one actually performs removal and
1570 : * returns %true. All others will wait until the kernfs operation which
1571 : * won self-removal finishes and return %false. Note that the losers wait
1572 : * for the completion of not only the winning kernfs_remove_self() but also
1573 : * the whole kernfs_ops which won the arbitration. This can be used to
1574 : * guarantee, for example, all concurrent writes to a "delete" file to
1575 : * finish only after the whole operation is complete.
1576 : *
1577 : * Return: %true if @kn is removed by this call, otherwise %false.
1578 : */
1579 0 : bool kernfs_remove_self(struct kernfs_node *kn)
1580 : {
1581 : bool ret;
1582 0 : struct kernfs_root *root = kernfs_root(kn);
1583 :
1584 0 : down_write(&root->kernfs_rwsem);
1585 0 : kernfs_break_active_protection(kn);
1586 :
1587 : /*
1588 : * SUICIDAL is used to arbitrate among competing invocations. Only
1589 : * the first one will actually perform removal. When the removal
1590 : * is complete, SUICIDED is set and the active ref is restored
1591 : * while kernfs_rwsem for held exclusive. The ones which lost
1592 : * arbitration waits for SUICIDED && drained which can happen only
1593 : * after the enclosing kernfs operation which executed the winning
1594 : * instance of kernfs_remove_self() finished.
1595 : */
1596 0 : if (!(kn->flags & KERNFS_SUICIDAL)) {
1597 0 : kn->flags |= KERNFS_SUICIDAL;
1598 0 : __kernfs_remove(kn);
1599 0 : kn->flags |= KERNFS_SUICIDED;
1600 0 : ret = true;
1601 : } else {
1602 0 : wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1603 0 : DEFINE_WAIT(wait);
1604 :
1605 : while (true) {
1606 0 : prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1607 :
1608 0 : if ((kn->flags & KERNFS_SUICIDED) &&
1609 0 : atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1610 : break;
1611 :
1612 0 : up_write(&root->kernfs_rwsem);
1613 0 : schedule();
1614 0 : down_write(&root->kernfs_rwsem);
1615 : }
1616 0 : finish_wait(waitq, &wait);
1617 0 : WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1618 0 : ret = false;
1619 : }
1620 :
1621 : /*
1622 : * This must be done while kernfs_rwsem held exclusive; otherwise,
1623 : * waiting for SUICIDED && deactivated could finish prematurely.
1624 : */
1625 0 : kernfs_unbreak_active_protection(kn);
1626 :
1627 0 : up_write(&root->kernfs_rwsem);
1628 0 : return ret;
1629 : }
1630 :
1631 : /**
1632 : * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1633 : * @parent: parent of the target
1634 : * @name: name of the kernfs_node to remove
1635 : * @ns: namespace tag of the kernfs_node to remove
1636 : *
1637 : * Look for the kernfs_node with @name and @ns under @parent and remove it.
1638 : *
1639 : * Return: %0 on success, -ENOENT if such entry doesn't exist.
1640 : */
1641 757 : int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1642 : const void *ns)
1643 : {
1644 : struct kernfs_node *kn;
1645 : struct kernfs_root *root;
1646 :
1647 757 : if (!parent) {
1648 0 : WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1649 : name);
1650 0 : return -ENOENT;
1651 : }
1652 :
1653 757 : root = kernfs_root(parent);
1654 757 : down_write(&root->kernfs_rwsem);
1655 :
1656 757 : kn = kernfs_find_ns(parent, name, ns);
1657 757 : if (kn) {
1658 351 : kernfs_get(kn);
1659 351 : __kernfs_remove(kn);
1660 351 : kernfs_put(kn);
1661 : }
1662 :
1663 757 : up_write(&root->kernfs_rwsem);
1664 :
1665 757 : if (kn)
1666 : return 0;
1667 : else
1668 406 : return -ENOENT;
1669 : }
1670 :
1671 : /**
1672 : * kernfs_rename_ns - move and rename a kernfs_node
1673 : * @kn: target node
1674 : * @new_parent: new parent to put @sd under
1675 : * @new_name: new name
1676 : * @new_ns: new namespace tag
1677 : *
1678 : * Return: %0 on success, -errno on failure.
1679 : */
1680 0 : int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1681 : const char *new_name, const void *new_ns)
1682 : {
1683 : struct kernfs_node *old_parent;
1684 : struct kernfs_root *root;
1685 0 : const char *old_name = NULL;
1686 : int error;
1687 :
1688 : /* can't move or rename root */
1689 0 : if (!kn->parent)
1690 : return -EINVAL;
1691 :
1692 0 : root = kernfs_root(kn);
1693 0 : down_write(&root->kernfs_rwsem);
1694 :
1695 0 : error = -ENOENT;
1696 0 : if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1697 0 : (new_parent->flags & KERNFS_EMPTY_DIR))
1698 : goto out;
1699 :
1700 0 : error = 0;
1701 0 : if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1702 0 : (strcmp(kn->name, new_name) == 0))
1703 : goto out; /* nothing to rename */
1704 :
1705 0 : error = -EEXIST;
1706 0 : if (kernfs_find_ns(new_parent, new_name, new_ns))
1707 : goto out;
1708 :
1709 : /* rename kernfs_node */
1710 0 : if (strcmp(kn->name, new_name) != 0) {
1711 0 : error = -ENOMEM;
1712 0 : new_name = kstrdup_const(new_name, GFP_KERNEL);
1713 0 : if (!new_name)
1714 : goto out;
1715 : } else {
1716 : new_name = NULL;
1717 : }
1718 :
1719 : /*
1720 : * Move to the appropriate place in the appropriate directories rbtree.
1721 : */
1722 0 : kernfs_unlink_sibling(kn);
1723 0 : kernfs_get(new_parent);
1724 :
1725 : /* rename_lock protects ->parent and ->name accessors */
1726 0 : spin_lock_irq(&kernfs_rename_lock);
1727 :
1728 0 : old_parent = kn->parent;
1729 0 : kn->parent = new_parent;
1730 :
1731 0 : kn->ns = new_ns;
1732 0 : if (new_name) {
1733 0 : old_name = kn->name;
1734 0 : kn->name = new_name;
1735 : }
1736 :
1737 0 : spin_unlock_irq(&kernfs_rename_lock);
1738 :
1739 0 : kn->hash = kernfs_name_hash(kn->name, kn->ns);
1740 0 : kernfs_link_sibling(kn);
1741 :
1742 0 : kernfs_put(old_parent);
1743 0 : kfree_const(old_name);
1744 :
1745 0 : error = 0;
1746 : out:
1747 0 : up_write(&root->kernfs_rwsem);
1748 0 : return error;
1749 : }
1750 :
1751 : /* Relationship between mode and the DT_xxx types */
1752 : static inline unsigned char dt_type(struct kernfs_node *kn)
1753 : {
1754 0 : return (kn->mode >> 12) & 15;
1755 : }
1756 :
1757 0 : static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1758 : {
1759 0 : kernfs_put(filp->private_data);
1760 0 : return 0;
1761 : }
1762 :
1763 0 : static struct kernfs_node *kernfs_dir_pos(const void *ns,
1764 : struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1765 : {
1766 0 : if (pos) {
1767 0 : int valid = kernfs_active(pos) &&
1768 0 : pos->parent == parent && hash == pos->hash;
1769 0 : kernfs_put(pos);
1770 0 : if (!valid)
1771 0 : pos = NULL;
1772 : }
1773 0 : if (!pos && (hash > 1) && (hash < INT_MAX)) {
1774 0 : struct rb_node *node = parent->dir.children.rb_node;
1775 0 : while (node) {
1776 0 : pos = rb_to_kn(node);
1777 :
1778 0 : if (hash < pos->hash)
1779 0 : node = node->rb_left;
1780 0 : else if (hash > pos->hash)
1781 0 : node = node->rb_right;
1782 : else
1783 : break;
1784 : }
1785 : }
1786 : /* Skip over entries which are dying/dead or in the wrong namespace */
1787 0 : while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1788 0 : struct rb_node *node = rb_next(&pos->rb);
1789 0 : if (!node)
1790 : pos = NULL;
1791 : else
1792 0 : pos = rb_to_kn(node);
1793 : }
1794 0 : return pos;
1795 : }
1796 :
1797 0 : static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1798 : struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1799 : {
1800 0 : pos = kernfs_dir_pos(ns, parent, ino, pos);
1801 0 : if (pos) {
1802 : do {
1803 0 : struct rb_node *node = rb_next(&pos->rb);
1804 0 : if (!node)
1805 : pos = NULL;
1806 : else
1807 0 : pos = rb_to_kn(node);
1808 0 : } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1809 : }
1810 0 : return pos;
1811 : }
1812 :
1813 0 : static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1814 : {
1815 0 : struct dentry *dentry = file->f_path.dentry;
1816 0 : struct kernfs_node *parent = kernfs_dentry_node(dentry);
1817 0 : struct kernfs_node *pos = file->private_data;
1818 : struct kernfs_root *root;
1819 0 : const void *ns = NULL;
1820 :
1821 0 : if (!dir_emit_dots(file, ctx))
1822 : return 0;
1823 :
1824 0 : root = kernfs_root(parent);
1825 0 : down_read(&root->kernfs_rwsem);
1826 :
1827 0 : if (kernfs_ns_enabled(parent))
1828 0 : ns = kernfs_info(dentry->d_sb)->ns;
1829 :
1830 0 : for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1831 : pos;
1832 0 : pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1833 0 : const char *name = pos->name;
1834 0 : unsigned int type = dt_type(pos);
1835 0 : int len = strlen(name);
1836 0 : ino_t ino = kernfs_ino(pos);
1837 :
1838 0 : ctx->pos = pos->hash;
1839 0 : file->private_data = pos;
1840 0 : kernfs_get(pos);
1841 :
1842 0 : up_read(&root->kernfs_rwsem);
1843 0 : if (!dir_emit(ctx, name, len, ino, type))
1844 : return 0;
1845 0 : down_read(&root->kernfs_rwsem);
1846 : }
1847 0 : up_read(&root->kernfs_rwsem);
1848 0 : file->private_data = NULL;
1849 0 : ctx->pos = INT_MAX;
1850 0 : return 0;
1851 : }
1852 :
1853 : const struct file_operations kernfs_dir_fops = {
1854 : .read = generic_read_dir,
1855 : .iterate_shared = kernfs_fop_readdir,
1856 : .release = kernfs_dir_fop_release,
1857 : .llseek = generic_file_llseek,
1858 : };
|
