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_RWLOCK(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 19616 : return atomic_read(&kn->active) >= 0;
37 : }
38 :
39 : static bool kernfs_active(struct kernfs_node *kn)
40 : {
41 9808 : lockdep_assert_held(&kernfs_root(kn)->kernfs_rwsem);
42 9808 : 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 : read_lock_irqsave(&kernfs_rename_lock, flags);
200 0 : ret = kernfs_name_locked(kn, buf, buflen);
201 0 : read_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 : read_lock_irqsave(&kernfs_rename_lock, flags);
228 0 : ret = kernfs_path_from_node_locked(to, from, buf, buflen);
229 0 : read_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 : read_lock_irqsave(&kernfs_rename_lock, flags);
298 0 : parent = kn->parent;
299 0 : kernfs_get(parent);
300 0 : read_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 10409 : static unsigned int kernfs_name_hash(const char *name, const void *ns)
313 : {
314 10409 : unsigned long hash = init_name_hash(ns);
315 10409 : unsigned int len = strlen(name);
316 125596 : while (len--)
317 209556 : hash = partial_name_hash(*name++, hash);
318 10409 : hash = end_name_hash(hash);
319 10409 : hash &= 0x7fffffffU;
320 : /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
321 10409 : if (hash < 2)
322 0 : hash += 2;
323 10409 : if (hash >= INT_MAX)
324 0 : hash = INT_MAX - 1;
325 10409 : return hash;
326 : }
327 :
328 28164 : static int kernfs_name_compare(unsigned int hash, const char *name,
329 : const void *ns, const struct kernfs_node *kn)
330 : {
331 28164 : if (hash < kn->hash)
332 : return -1;
333 14015 : if (hash > kn->hash)
334 : return 1;
335 1154 : if (ns < kn->ns)
336 : return -1;
337 1154 : if (ns > kn->ns)
338 : return 1;
339 1154 : 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 25525 : 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 8759 : static int kernfs_link_sibling(struct kernfs_node *kn)
362 : {
363 8759 : struct rb_node **node = &kn->parent->dir.children.rb_node;
364 8759 : struct rb_node *parent = NULL;
365 :
366 43043 : while (*node) {
367 : struct kernfs_node *pos;
368 : int result;
369 :
370 25525 : pos = rb_to_kn(*node);
371 25525 : parent = *node;
372 25525 : result = kernfs_sd_compare(kn, pos);
373 25525 : if (result < 0)
374 13288 : node = &pos->rb.rb_left;
375 12237 : else if (result > 0)
376 12237 : node = &pos->rb.rb_right;
377 : else
378 : return -EEXIST;
379 : }
380 :
381 : /* add new node and rebalance the tree */
382 17518 : rb_link_node(&kn->rb, parent, node);
383 8759 : rb_insert_color(&kn->rb, &kn->parent->dir.children);
384 :
385 : /* successfully added, account subdir number */
386 17518 : if (kernfs_type(kn) == KERNFS_DIR)
387 1348 : kn->parent->dir.subdirs++;
388 8759 : kernfs_inc_rev(kn->parent);
389 :
390 8759 : 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 : */
406 524 : static bool kernfs_unlink_sibling(struct kernfs_node *kn)
407 : {
408 524 : if (RB_EMPTY_NODE(&kn->rb))
409 : return false;
410 :
411 1048 : if (kernfs_type(kn) == KERNFS_DIR)
412 98 : kn->parent->dir.subdirs--;
413 524 : kernfs_inc_rev(kn->parent);
414 :
415 524 : rb_erase(&kn->rb, &kn->parent->dir.children);
416 524 : RB_CLEAR_NODE(&kn->rb);
417 524 : 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 : */
474 524 : static void kernfs_drain(struct kernfs_node *kn)
475 : __releases(&kernfs_root(kn)->kernfs_rwsem)
476 : __acquires(&kernfs_root(kn)->kernfs_rwsem)
477 : {
478 524 : struct kernfs_root *root = kernfs_root(kn);
479 :
480 : lockdep_assert_held_write(&root->kernfs_rwsem);
481 524 : 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 1572 : if (atomic_read(&kn->active) == KN_DEACTIVATED_BIAS &&
490 524 : !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 15414 : void kernfs_get(struct kernfs_node *kn)
520 : {
521 15414 : if (kn) {
522 30828 : WARN_ON(!atomic_read(&kn->count));
523 15414 : atomic_inc(&kn->count);
524 : }
525 15414 : }
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 4828 : void kernfs_put(struct kernfs_node *kn)
535 : {
536 : struct kernfs_node *parent;
537 : struct kernfs_root *root;
538 :
539 9656 : if (!kn || !atomic_dec_and_test(&kn->count))
540 : return;
541 524 : 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 : */
547 524 : parent = kn->parent;
548 :
549 1048 : 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 1048 : if (kernfs_type(kn) == KERNFS_LINK)
554 93 : kernfs_put(kn->symlink.target_kn);
555 :
556 524 : kfree_const(kn->name);
557 :
558 524 : if (kn->iattr) {
559 0 : simple_xattrs_free(&kn->iattr->xattrs);
560 0 : kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
561 : }
562 524 : spin_lock(&kernfs_idr_lock);
563 1048 : idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
564 524 : spin_unlock(&kernfs_idr_lock);
565 524 : kmem_cache_free(kernfs_node_cache, kn);
566 :
567 524 : kn = parent;
568 524 : if (kn) {
569 1048 : 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 8760 : 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 8760 : name = kstrdup_const(name, GFP_KERNEL);
608 8760 : if (!name)
609 : return NULL;
610 :
611 17520 : kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
612 8760 : if (!kn)
613 : goto err_out1;
614 :
615 8760 : idr_preload(GFP_KERNEL);
616 8760 : spin_lock(&kernfs_idr_lock);
617 8760 : ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
618 8760 : if (ret >= 0 && ret < root->last_id_lowbits)
619 0 : root->id_highbits++;
620 8760 : id_highbits = root->id_highbits;
621 8760 : root->last_id_lowbits = ret;
622 8760 : spin_unlock(&kernfs_idr_lock);
623 : idr_preload_end();
624 8760 : if (ret < 0)
625 : goto err_out2;
626 :
627 8760 : kn->id = (u64)id_highbits << 32 | ret;
628 :
629 17520 : atomic_set(&kn->count, 1);
630 17520 : atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
631 8760 : RB_CLEAR_NODE(&kn->rb);
632 :
633 8760 : kn->name = name;
634 8760 : kn->mode = mode;
635 8760 : kn->flags = flags;
636 :
637 8760 : 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 8759 : 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 17518 : kn = __kernfs_new_node(kernfs_root(parent), parent,
674 : name, mode, uid, gid, flags);
675 8759 : if (kn) {
676 8759 : kernfs_get(parent);
677 8759 : kn->parent = parent;
678 : }
679 8759 : 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 8759 : int kernfs_add_one(struct kernfs_node *kn)
744 : {
745 8759 : struct kernfs_node *parent = kn->parent;
746 8759 : struct kernfs_root *root = kernfs_root(parent);
747 : struct kernfs_iattrs *ps_iattr;
748 : bool has_ns;
749 : int ret;
750 :
751 8759 : down_write(&root->kernfs_rwsem);
752 :
753 8759 : ret = -EINVAL;
754 17518 : has_ns = kernfs_ns_enabled(parent);
755 8759 : 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 17518 : if (kernfs_type(parent) != KERNFS_DIR)
760 : goto out_unlock;
761 :
762 8759 : ret = -ENOENT;
763 8759 : if (parent->flags & (KERNFS_REMOVING | KERNFS_EMPTY_DIR))
764 : goto out_unlock;
765 :
766 8759 : kn->hash = kernfs_name_hash(kn->name, kn->ns);
767 :
768 8759 : ret = kernfs_link_sibling(kn);
769 8759 : if (ret)
770 : goto out_unlock;
771 :
772 : /* Update timestamps on the parent */
773 8759 : down_write(&root->kernfs_iattr_rwsem);
774 :
775 8759 : ps_iattr = parent->iattr;
776 8759 : if (ps_iattr) {
777 0 : ktime_get_real_ts64(&ps_iattr->ia_ctime);
778 0 : ps_iattr->ia_mtime = ps_iattr->ia_ctime;
779 : }
780 :
781 8759 : up_write(&root->kernfs_iattr_rwsem);
782 8759 : up_write(&root->kernfs_rwsem);
783 :
784 : /*
785 : * Activate the new node unless CREATE_DEACTIVATED is requested.
786 : * If not activated here, the kernfs user is responsible for
787 : * activating the node with kernfs_activate(). A node which hasn't
788 : * been activated is not visible to userland and its removal won't
789 : * trigger deactivation.
790 : */
791 8759 : if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
792 8759 : kernfs_activate(kn);
793 : return 0;
794 :
795 : out_unlock:
796 0 : up_write(&root->kernfs_rwsem);
797 0 : return ret;
798 : }
799 :
800 : /**
801 : * kernfs_find_ns - find kernfs_node with the given name
802 : * @parent: kernfs_node to search under
803 : * @name: name to look for
804 : * @ns: the namespace tag to use
805 : *
806 : * Look for kernfs_node with name @name under @parent.
807 : *
808 : * Return: pointer to the found kernfs_node on success, %NULL on failure.
809 : */
810 1650 : static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
811 : const unsigned char *name,
812 : const void *ns)
813 : {
814 1650 : struct rb_node *node = parent->dir.children.rb_node;
815 3300 : bool has_ns = kernfs_ns_enabled(parent);
816 : unsigned int hash;
817 :
818 1650 : lockdep_assert_held(&kernfs_root(parent)->kernfs_rwsem);
819 :
820 1650 : if (has_ns != (bool)ns) {
821 0 : WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
822 : has_ns ? "required" : "invalid", parent->name, name);
823 0 : return NULL;
824 : }
825 :
826 1650 : hash = kernfs_name_hash(name, ns);
827 4785 : while (node) {
828 : struct kernfs_node *kn;
829 : int result;
830 :
831 2639 : kn = rb_to_kn(node);
832 2639 : result = kernfs_name_compare(hash, name, ns, kn);
833 2639 : if (result < 0)
834 861 : node = node->rb_left;
835 1778 : else if (result > 0)
836 624 : node = node->rb_right;
837 : else
838 : return kn;
839 : }
840 : return NULL;
841 : }
842 :
843 0 : static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
844 : const unsigned char *path,
845 : const void *ns)
846 : {
847 : size_t len;
848 : char *p, *name;
849 :
850 0 : lockdep_assert_held_read(&kernfs_root(parent)->kernfs_rwsem);
851 :
852 0 : spin_lock_irq(&kernfs_pr_cont_lock);
853 :
854 0 : len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
855 :
856 0 : if (len >= sizeof(kernfs_pr_cont_buf)) {
857 0 : spin_unlock_irq(&kernfs_pr_cont_lock);
858 0 : return NULL;
859 : }
860 :
861 0 : p = kernfs_pr_cont_buf;
862 :
863 0 : while ((name = strsep(&p, "/")) && parent) {
864 0 : if (*name == '\0')
865 0 : continue;
866 0 : parent = kernfs_find_ns(parent, name, ns);
867 : }
868 :
869 0 : spin_unlock_irq(&kernfs_pr_cont_lock);
870 :
871 0 : return parent;
872 : }
873 :
874 : /**
875 : * kernfs_find_and_get_ns - find and get kernfs_node with the given name
876 : * @parent: kernfs_node to search under
877 : * @name: name to look for
878 : * @ns: the namespace tag to use
879 : *
880 : * Look for kernfs_node with name @name under @parent and get a reference
881 : * if found. This function may sleep.
882 : *
883 : * Return: pointer to the found kernfs_node on success, %NULL on failure.
884 : */
885 728 : struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
886 : const char *name, const void *ns)
887 : {
888 : struct kernfs_node *kn;
889 728 : struct kernfs_root *root = kernfs_root(parent);
890 :
891 728 : down_read(&root->kernfs_rwsem);
892 728 : kn = kernfs_find_ns(parent, name, ns);
893 728 : kernfs_get(kn);
894 728 : up_read(&root->kernfs_rwsem);
895 :
896 728 : return kn;
897 : }
898 : EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
899 :
900 : /**
901 : * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
902 : * @parent: kernfs_node to search under
903 : * @path: path to look for
904 : * @ns: the namespace tag to use
905 : *
906 : * Look for kernfs_node with path @path under @parent and get a reference
907 : * if found. This function may sleep.
908 : *
909 : * Return: pointer to the found kernfs_node on success, %NULL on failure.
910 : */
911 0 : struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
912 : const char *path, const void *ns)
913 : {
914 : struct kernfs_node *kn;
915 0 : struct kernfs_root *root = kernfs_root(parent);
916 :
917 0 : down_read(&root->kernfs_rwsem);
918 0 : kn = kernfs_walk_ns(parent, path, ns);
919 0 : kernfs_get(kn);
920 0 : up_read(&root->kernfs_rwsem);
921 :
922 0 : return kn;
923 : }
924 :
925 : /**
926 : * kernfs_create_root - create a new kernfs hierarchy
927 : * @scops: optional syscall operations for the hierarchy
928 : * @flags: KERNFS_ROOT_* flags
929 : * @priv: opaque data associated with the new directory
930 : *
931 : * Return: the root of the new hierarchy on success, ERR_PTR() value on
932 : * failure.
933 : */
934 1 : struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
935 : unsigned int flags, void *priv)
936 : {
937 : struct kernfs_root *root;
938 : struct kernfs_node *kn;
939 :
940 1 : root = kzalloc(sizeof(*root), GFP_KERNEL);
941 1 : if (!root)
942 : return ERR_PTR(-ENOMEM);
943 :
944 2 : idr_init(&root->ino_idr);
945 1 : init_rwsem(&root->kernfs_rwsem);
946 1 : init_rwsem(&root->kernfs_iattr_rwsem);
947 1 : init_rwsem(&root->kernfs_supers_rwsem);
948 2 : INIT_LIST_HEAD(&root->supers);
949 :
950 : /*
951 : * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
952 : * High bits generation. The starting value for both ino and
953 : * genenration is 1. Initialize upper 32bit allocation
954 : * accordingly.
955 : */
956 : if (sizeof(ino_t) >= sizeof(u64))
957 1 : root->id_highbits = 0;
958 : else
959 : root->id_highbits = 1;
960 :
961 1 : kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
962 1 : GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
963 : KERNFS_DIR);
964 1 : if (!kn) {
965 0 : idr_destroy(&root->ino_idr);
966 0 : kfree(root);
967 0 : return ERR_PTR(-ENOMEM);
968 : }
969 :
970 1 : kn->priv = priv;
971 1 : kn->dir.root = root;
972 :
973 1 : root->syscall_ops = scops;
974 1 : root->flags = flags;
975 1 : root->kn = kn;
976 1 : init_waitqueue_head(&root->deactivate_waitq);
977 :
978 1 : if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
979 1 : kernfs_activate(kn);
980 :
981 : return root;
982 : }
983 :
984 : /**
985 : * kernfs_destroy_root - destroy a kernfs hierarchy
986 : * @root: root of the hierarchy to destroy
987 : *
988 : * Destroy the hierarchy anchored at @root by removing all existing
989 : * directories and destroying @root.
990 : */
991 0 : void kernfs_destroy_root(struct kernfs_root *root)
992 : {
993 : /*
994 : * kernfs_remove holds kernfs_rwsem from the root so the root
995 : * shouldn't be freed during the operation.
996 : */
997 0 : kernfs_get(root->kn);
998 0 : kernfs_remove(root->kn);
999 0 : kernfs_put(root->kn); /* will also free @root */
1000 0 : }
1001 :
1002 : /**
1003 : * kernfs_root_to_node - return the kernfs_node associated with a kernfs_root
1004 : * @root: root to use to lookup
1005 : *
1006 : * Return: @root's kernfs_node
1007 : */
1008 1 : struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root)
1009 : {
1010 1 : return root->kn;
1011 : }
1012 :
1013 : /**
1014 : * kernfs_create_dir_ns - create a directory
1015 : * @parent: parent in which to create a new directory
1016 : * @name: name of the new directory
1017 : * @mode: mode of the new directory
1018 : * @uid: uid of the new directory
1019 : * @gid: gid of the new directory
1020 : * @priv: opaque data associated with the new directory
1021 : * @ns: optional namespace tag of the directory
1022 : *
1023 : * Return: the created node on success, ERR_PTR() value on failure.
1024 : */
1025 1348 : struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1026 : const char *name, umode_t mode,
1027 : kuid_t uid, kgid_t gid,
1028 : void *priv, const void *ns)
1029 : {
1030 : struct kernfs_node *kn;
1031 : int rc;
1032 :
1033 : /* allocate */
1034 1348 : kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1035 : uid, gid, KERNFS_DIR);
1036 1348 : if (!kn)
1037 : return ERR_PTR(-ENOMEM);
1038 :
1039 1348 : kn->dir.root = parent->dir.root;
1040 1348 : kn->ns = ns;
1041 1348 : kn->priv = priv;
1042 :
1043 : /* link in */
1044 1348 : rc = kernfs_add_one(kn);
1045 1348 : if (!rc)
1046 : return kn;
1047 :
1048 0 : kernfs_put(kn);
1049 0 : return ERR_PTR(rc);
1050 : }
1051 :
1052 : /**
1053 : * kernfs_create_empty_dir - create an always empty directory
1054 : * @parent: parent in which to create a new directory
1055 : * @name: name of the new directory
1056 : *
1057 : * Return: the created node on success, ERR_PTR() value on failure.
1058 : */
1059 0 : struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1060 : const char *name)
1061 : {
1062 : struct kernfs_node *kn;
1063 : int rc;
1064 :
1065 : /* allocate */
1066 0 : kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1067 0 : GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1068 0 : if (!kn)
1069 : return ERR_PTR(-ENOMEM);
1070 :
1071 0 : kn->flags |= KERNFS_EMPTY_DIR;
1072 0 : kn->dir.root = parent->dir.root;
1073 0 : kn->ns = NULL;
1074 0 : kn->priv = NULL;
1075 :
1076 : /* link in */
1077 0 : rc = kernfs_add_one(kn);
1078 0 : if (!rc)
1079 : return kn;
1080 :
1081 0 : kernfs_put(kn);
1082 0 : return ERR_PTR(rc);
1083 : }
1084 :
1085 0 : static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1086 : {
1087 : struct kernfs_node *kn;
1088 : struct kernfs_root *root;
1089 :
1090 0 : if (flags & LOOKUP_RCU)
1091 : return -ECHILD;
1092 :
1093 : /* Negative hashed dentry? */
1094 0 : if (d_really_is_negative(dentry)) {
1095 : struct kernfs_node *parent;
1096 :
1097 : /* If the kernfs parent node has changed discard and
1098 : * proceed to ->lookup.
1099 : *
1100 : * There's nothing special needed here when getting the
1101 : * dentry parent, even if a concurrent rename is in
1102 : * progress. That's because the dentry is negative so
1103 : * it can only be the target of the rename and it will
1104 : * be doing a d_move() not a replace. Consequently the
1105 : * dentry d_parent won't change over the d_move().
1106 : *
1107 : * Also kernfs negative dentries transitioning from
1108 : * negative to positive during revalidate won't happen
1109 : * because they are invalidated on containing directory
1110 : * changes and the lookup re-done so that a new positive
1111 : * dentry can be properly created.
1112 : */
1113 0 : root = kernfs_root_from_sb(dentry->d_sb);
1114 0 : down_read(&root->kernfs_rwsem);
1115 0 : parent = kernfs_dentry_node(dentry->d_parent);
1116 0 : if (parent) {
1117 0 : if (kernfs_dir_changed(parent, dentry)) {
1118 0 : up_read(&root->kernfs_rwsem);
1119 0 : return 0;
1120 : }
1121 : }
1122 0 : up_read(&root->kernfs_rwsem);
1123 :
1124 : /* The kernfs parent node hasn't changed, leave the
1125 : * dentry negative and return success.
1126 : */
1127 0 : return 1;
1128 : }
1129 :
1130 0 : kn = kernfs_dentry_node(dentry);
1131 0 : root = kernfs_root(kn);
1132 0 : down_read(&root->kernfs_rwsem);
1133 :
1134 : /* The kernfs node has been deactivated */
1135 0 : if (!kernfs_active(kn))
1136 : goto out_bad;
1137 :
1138 : /* The kernfs node has been moved? */
1139 0 : if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1140 : goto out_bad;
1141 :
1142 : /* The kernfs node has been renamed */
1143 0 : if (strcmp(dentry->d_name.name, kn->name) != 0)
1144 : goto out_bad;
1145 :
1146 : /* The kernfs node has been moved to a different namespace */
1147 0 : if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1148 0 : kernfs_info(dentry->d_sb)->ns != kn->ns)
1149 : goto out_bad;
1150 :
1151 0 : up_read(&root->kernfs_rwsem);
1152 0 : return 1;
1153 : out_bad:
1154 0 : up_read(&root->kernfs_rwsem);
1155 0 : return 0;
1156 : }
1157 :
1158 : const struct dentry_operations kernfs_dops = {
1159 : .d_revalidate = kernfs_dop_revalidate,
1160 : };
1161 :
1162 0 : static struct dentry *kernfs_iop_lookup(struct inode *dir,
1163 : struct dentry *dentry,
1164 : unsigned int flags)
1165 : {
1166 0 : struct kernfs_node *parent = dir->i_private;
1167 : struct kernfs_node *kn;
1168 : struct kernfs_root *root;
1169 0 : struct inode *inode = NULL;
1170 0 : const void *ns = NULL;
1171 :
1172 0 : root = kernfs_root(parent);
1173 0 : down_read(&root->kernfs_rwsem);
1174 0 : if (kernfs_ns_enabled(parent))
1175 0 : ns = kernfs_info(dir->i_sb)->ns;
1176 :
1177 0 : kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1178 : /* attach dentry and inode */
1179 0 : if (kn) {
1180 : /* Inactive nodes are invisible to the VFS so don't
1181 : * create a negative.
1182 : */
1183 0 : if (!kernfs_active(kn)) {
1184 0 : up_read(&root->kernfs_rwsem);
1185 0 : return NULL;
1186 : }
1187 0 : inode = kernfs_get_inode(dir->i_sb, kn);
1188 0 : if (!inode)
1189 0 : inode = ERR_PTR(-ENOMEM);
1190 : }
1191 : /*
1192 : * Needed for negative dentry validation.
1193 : * The negative dentry can be created in kernfs_iop_lookup()
1194 : * or transforms from positive dentry in dentry_unlink_inode()
1195 : * called from vfs_rmdir().
1196 : */
1197 0 : if (!IS_ERR(inode))
1198 0 : kernfs_set_rev(parent, dentry);
1199 0 : up_read(&root->kernfs_rwsem);
1200 :
1201 : /* instantiate and hash (possibly negative) dentry */
1202 0 : return d_splice_alias(inode, dentry);
1203 : }
1204 :
1205 0 : static int kernfs_iop_mkdir(struct mnt_idmap *idmap,
1206 : struct inode *dir, struct dentry *dentry,
1207 : umode_t mode)
1208 : {
1209 0 : struct kernfs_node *parent = dir->i_private;
1210 0 : struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1211 : int ret;
1212 :
1213 0 : if (!scops || !scops->mkdir)
1214 : return -EPERM;
1215 :
1216 0 : if (!kernfs_get_active(parent))
1217 : return -ENODEV;
1218 :
1219 0 : ret = scops->mkdir(parent, dentry->d_name.name, mode);
1220 :
1221 0 : kernfs_put_active(parent);
1222 0 : return ret;
1223 : }
1224 :
1225 0 : static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1226 : {
1227 0 : struct kernfs_node *kn = kernfs_dentry_node(dentry);
1228 0 : struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1229 : int ret;
1230 :
1231 0 : if (!scops || !scops->rmdir)
1232 : return -EPERM;
1233 :
1234 0 : if (!kernfs_get_active(kn))
1235 : return -ENODEV;
1236 :
1237 0 : ret = scops->rmdir(kn);
1238 :
1239 0 : kernfs_put_active(kn);
1240 0 : return ret;
1241 : }
1242 :
1243 0 : static int kernfs_iop_rename(struct mnt_idmap *idmap,
1244 : struct inode *old_dir, struct dentry *old_dentry,
1245 : struct inode *new_dir, struct dentry *new_dentry,
1246 : unsigned int flags)
1247 : {
1248 0 : struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1249 0 : struct kernfs_node *new_parent = new_dir->i_private;
1250 0 : struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1251 : int ret;
1252 :
1253 0 : if (flags)
1254 : return -EINVAL;
1255 :
1256 0 : if (!scops || !scops->rename)
1257 : return -EPERM;
1258 :
1259 0 : if (!kernfs_get_active(kn))
1260 : return -ENODEV;
1261 :
1262 0 : if (!kernfs_get_active(new_parent)) {
1263 0 : kernfs_put_active(kn);
1264 0 : return -ENODEV;
1265 : }
1266 :
1267 0 : ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1268 :
1269 0 : kernfs_put_active(new_parent);
1270 0 : kernfs_put_active(kn);
1271 0 : return ret;
1272 : }
1273 :
1274 : const struct inode_operations kernfs_dir_iops = {
1275 : .lookup = kernfs_iop_lookup,
1276 : .permission = kernfs_iop_permission,
1277 : .setattr = kernfs_iop_setattr,
1278 : .getattr = kernfs_iop_getattr,
1279 : .listxattr = kernfs_iop_listxattr,
1280 :
1281 : .mkdir = kernfs_iop_mkdir,
1282 : .rmdir = kernfs_iop_rmdir,
1283 : .rename = kernfs_iop_rename,
1284 : };
1285 :
1286 : static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1287 : {
1288 : struct kernfs_node *last;
1289 :
1290 : while (true) {
1291 : struct rb_node *rbn;
1292 :
1293 9808 : last = pos;
1294 :
1295 9808 : if (kernfs_type(pos) != KERNFS_DIR)
1296 : break;
1297 :
1298 1545 : rbn = rb_first(&pos->dir.children);
1299 1545 : if (!rbn)
1300 : break;
1301 :
1302 0 : pos = rb_to_kn(rbn);
1303 : }
1304 :
1305 : return last;
1306 : }
1307 :
1308 : /**
1309 : * kernfs_next_descendant_post - find the next descendant for post-order walk
1310 : * @pos: the current position (%NULL to initiate traversal)
1311 : * @root: kernfs_node whose descendants to walk
1312 : *
1313 : * Find the next descendant to visit for post-order traversal of @root's
1314 : * descendants. @root is included in the iteration and the last node to be
1315 : * visited.
1316 : *
1317 : * Return: the next descendant to visit or %NULL when done.
1318 : */
1319 18568 : static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1320 : struct kernfs_node *root)
1321 : {
1322 : struct rb_node *rbn;
1323 :
1324 18568 : lockdep_assert_held_write(&kernfs_root(root)->kernfs_rwsem);
1325 :
1326 : /* if first iteration, visit leftmost descendant which may be root */
1327 18568 : if (!pos)
1328 : return kernfs_leftmost_descendant(root);
1329 :
1330 : /* if we visited @root, we're done */
1331 9284 : if (pos == root)
1332 : return NULL;
1333 :
1334 : /* if there's an unvisited sibling, visit its leftmost descendant */
1335 0 : rbn = rb_next(&pos->rb);
1336 0 : if (rbn)
1337 0 : return kernfs_leftmost_descendant(rb_to_kn(rbn));
1338 :
1339 : /* no sibling left, visit parent */
1340 0 : return pos->parent;
1341 : }
1342 :
1343 8760 : static void kernfs_activate_one(struct kernfs_node *kn)
1344 : {
1345 8760 : lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1346 :
1347 8760 : kn->flags |= KERNFS_ACTIVATED;
1348 :
1349 8760 : if (kernfs_active(kn) || (kn->flags & (KERNFS_HIDDEN | KERNFS_REMOVING)))
1350 : return;
1351 :
1352 8760 : WARN_ON_ONCE(kn->parent && RB_EMPTY_NODE(&kn->rb));
1353 17520 : WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1354 :
1355 8760 : atomic_sub(KN_DEACTIVATED_BIAS, &kn->active);
1356 : }
1357 :
1358 : /**
1359 : * kernfs_activate - activate a node which started deactivated
1360 : * @kn: kernfs_node whose subtree is to be activated
1361 : *
1362 : * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1363 : * needs to be explicitly activated. A node which hasn't been activated
1364 : * isn't visible to userland and deactivation is skipped during its
1365 : * removal. This is useful to construct atomic init sequences where
1366 : * creation of multiple nodes should either succeed or fail atomically.
1367 : *
1368 : * The caller is responsible for ensuring that this function is not called
1369 : * after kernfs_remove*() is invoked on @kn.
1370 : */
1371 8760 : void kernfs_activate(struct kernfs_node *kn)
1372 : {
1373 : struct kernfs_node *pos;
1374 8760 : struct kernfs_root *root = kernfs_root(kn);
1375 :
1376 8760 : down_write(&root->kernfs_rwsem);
1377 :
1378 8760 : pos = NULL;
1379 26280 : while ((pos = kernfs_next_descendant_post(pos, kn)))
1380 8760 : kernfs_activate_one(pos);
1381 :
1382 8760 : up_write(&root->kernfs_rwsem);
1383 8760 : }
1384 :
1385 : /**
1386 : * kernfs_show - show or hide a node
1387 : * @kn: kernfs_node to show or hide
1388 : * @show: whether to show or hide
1389 : *
1390 : * If @show is %false, @kn is marked hidden and deactivated. A hidden node is
1391 : * ignored in future activaitons. If %true, the mark is removed and activation
1392 : * state is restored. This function won't implicitly activate a new node in a
1393 : * %KERNFS_ROOT_CREATE_DEACTIVATED root which hasn't been activated yet.
1394 : *
1395 : * To avoid recursion complexities, directories aren't supported for now.
1396 : */
1397 0 : void kernfs_show(struct kernfs_node *kn, bool show)
1398 : {
1399 0 : struct kernfs_root *root = kernfs_root(kn);
1400 :
1401 0 : if (WARN_ON_ONCE(kernfs_type(kn) == KERNFS_DIR))
1402 : return;
1403 :
1404 0 : down_write(&root->kernfs_rwsem);
1405 :
1406 0 : if (show) {
1407 0 : kn->flags &= ~KERNFS_HIDDEN;
1408 0 : if (kn->flags & KERNFS_ACTIVATED)
1409 0 : kernfs_activate_one(kn);
1410 : } else {
1411 0 : kn->flags |= KERNFS_HIDDEN;
1412 0 : if (kernfs_active(kn))
1413 0 : atomic_add(KN_DEACTIVATED_BIAS, &kn->active);
1414 0 : kernfs_drain(kn);
1415 : }
1416 :
1417 0 : up_write(&root->kernfs_rwsem);
1418 : }
1419 :
1420 524 : static void __kernfs_remove(struct kernfs_node *kn)
1421 : {
1422 : struct kernfs_node *pos;
1423 :
1424 : /* Short-circuit if non-root @kn has already finished removal. */
1425 524 : if (!kn)
1426 : return;
1427 :
1428 524 : lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1429 :
1430 : /*
1431 : * This is for kernfs_remove_self() which plays with active ref
1432 : * after removal.
1433 : */
1434 524 : if (kn->parent && RB_EMPTY_NODE(&kn->rb))
1435 : return;
1436 :
1437 : pr_debug("kernfs %s: removing\n", kn->name);
1438 :
1439 : /* prevent new usage by marking all nodes removing and deactivating */
1440 : pos = NULL;
1441 1048 : while ((pos = kernfs_next_descendant_post(pos, kn))) {
1442 524 : pos->flags |= KERNFS_REMOVING;
1443 524 : if (kernfs_active(pos))
1444 524 : atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1445 : }
1446 :
1447 : /* deactivate and unlink the subtree node-by-node */
1448 : do {
1449 524 : pos = kernfs_leftmost_descendant(kn);
1450 :
1451 : /*
1452 : * kernfs_drain() may drop kernfs_rwsem temporarily and @pos's
1453 : * base ref could have been put by someone else by the time
1454 : * the function returns. Make sure it doesn't go away
1455 : * underneath us.
1456 : */
1457 524 : kernfs_get(pos);
1458 :
1459 524 : kernfs_drain(pos);
1460 :
1461 : /*
1462 : * kernfs_unlink_sibling() succeeds once per node. Use it
1463 : * to decide who's responsible for cleanups.
1464 : */
1465 524 : if (!pos->parent || kernfs_unlink_sibling(pos)) {
1466 524 : struct kernfs_iattrs *ps_iattr =
1467 524 : pos->parent ? pos->parent->iattr : NULL;
1468 :
1469 : /* update timestamps on the parent */
1470 524 : down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
1471 :
1472 524 : if (ps_iattr) {
1473 0 : ktime_get_real_ts64(&ps_iattr->ia_ctime);
1474 0 : ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1475 : }
1476 :
1477 524 : up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
1478 524 : kernfs_put(pos);
1479 : }
1480 :
1481 524 : kernfs_put(pos);
1482 524 : } while (pos != kn);
1483 : }
1484 :
1485 : /**
1486 : * kernfs_remove - remove a kernfs_node recursively
1487 : * @kn: the kernfs_node to remove
1488 : *
1489 : * Remove @kn along with all its subdirectories and files.
1490 : */
1491 98 : void kernfs_remove(struct kernfs_node *kn)
1492 : {
1493 : struct kernfs_root *root;
1494 :
1495 98 : if (!kn)
1496 : return;
1497 :
1498 98 : root = kernfs_root(kn);
1499 :
1500 98 : down_write(&root->kernfs_rwsem);
1501 98 : __kernfs_remove(kn);
1502 98 : up_write(&root->kernfs_rwsem);
1503 : }
1504 :
1505 : /**
1506 : * kernfs_break_active_protection - break out of active protection
1507 : * @kn: the self kernfs_node
1508 : *
1509 : * The caller must be running off of a kernfs operation which is invoked
1510 : * with an active reference - e.g. one of kernfs_ops. Each invocation of
1511 : * this function must also be matched with an invocation of
1512 : * kernfs_unbreak_active_protection().
1513 : *
1514 : * This function releases the active reference of @kn the caller is
1515 : * holding. Once this function is called, @kn may be removed at any point
1516 : * and the caller is solely responsible for ensuring that the objects it
1517 : * dereferences are accessible.
1518 : */
1519 0 : void kernfs_break_active_protection(struct kernfs_node *kn)
1520 : {
1521 : /*
1522 : * Take out ourself out of the active ref dependency chain. If
1523 : * we're called without an active ref, lockdep will complain.
1524 : */
1525 0 : kernfs_put_active(kn);
1526 0 : }
1527 :
1528 : /**
1529 : * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1530 : * @kn: the self kernfs_node
1531 : *
1532 : * If kernfs_break_active_protection() was called, this function must be
1533 : * invoked before finishing the kernfs operation. Note that while this
1534 : * function restores the active reference, it doesn't and can't actually
1535 : * restore the active protection - @kn may already or be in the process of
1536 : * being removed. Once kernfs_break_active_protection() is invoked, that
1537 : * protection is irreversibly gone for the kernfs operation instance.
1538 : *
1539 : * While this function may be called at any point after
1540 : * kernfs_break_active_protection() is invoked, its most useful location
1541 : * would be right before the enclosing kernfs operation returns.
1542 : */
1543 0 : void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1544 : {
1545 : /*
1546 : * @kn->active could be in any state; however, the increment we do
1547 : * here will be undone as soon as the enclosing kernfs operation
1548 : * finishes and this temporary bump can't break anything. If @kn
1549 : * is alive, nothing changes. If @kn is being deactivated, the
1550 : * soon-to-follow put will either finish deactivation or restore
1551 : * deactivated state. If @kn is already removed, the temporary
1552 : * bump is guaranteed to be gone before @kn is released.
1553 : */
1554 0 : atomic_inc(&kn->active);
1555 0 : if (kernfs_lockdep(kn))
1556 : rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1557 0 : }
1558 :
1559 : /**
1560 : * kernfs_remove_self - remove a kernfs_node from its own method
1561 : * @kn: the self kernfs_node to remove
1562 : *
1563 : * The caller must be running off of a kernfs operation which is invoked
1564 : * with an active reference - e.g. one of kernfs_ops. This can be used to
1565 : * implement a file operation which deletes itself.
1566 : *
1567 : * For example, the "delete" file for a sysfs device directory can be
1568 : * implemented by invoking kernfs_remove_self() on the "delete" file
1569 : * itself. This function breaks the circular dependency of trying to
1570 : * deactivate self while holding an active ref itself. It isn't necessary
1571 : * to modify the usual removal path to use kernfs_remove_self(). The
1572 : * "delete" implementation can simply invoke kernfs_remove_self() on self
1573 : * before proceeding with the usual removal path. kernfs will ignore later
1574 : * kernfs_remove() on self.
1575 : *
1576 : * kernfs_remove_self() can be called multiple times concurrently on the
1577 : * same kernfs_node. Only the first one actually performs removal and
1578 : * returns %true. All others will wait until the kernfs operation which
1579 : * won self-removal finishes and return %false. Note that the losers wait
1580 : * for the completion of not only the winning kernfs_remove_self() but also
1581 : * the whole kernfs_ops which won the arbitration. This can be used to
1582 : * guarantee, for example, all concurrent writes to a "delete" file to
1583 : * finish only after the whole operation is complete.
1584 : *
1585 : * Return: %true if @kn is removed by this call, otherwise %false.
1586 : */
1587 0 : bool kernfs_remove_self(struct kernfs_node *kn)
1588 : {
1589 : bool ret;
1590 0 : struct kernfs_root *root = kernfs_root(kn);
1591 :
1592 0 : down_write(&root->kernfs_rwsem);
1593 0 : kernfs_break_active_protection(kn);
1594 :
1595 : /*
1596 : * SUICIDAL is used to arbitrate among competing invocations. Only
1597 : * the first one will actually perform removal. When the removal
1598 : * is complete, SUICIDED is set and the active ref is restored
1599 : * while kernfs_rwsem for held exclusive. The ones which lost
1600 : * arbitration waits for SUICIDED && drained which can happen only
1601 : * after the enclosing kernfs operation which executed the winning
1602 : * instance of kernfs_remove_self() finished.
1603 : */
1604 0 : if (!(kn->flags & KERNFS_SUICIDAL)) {
1605 0 : kn->flags |= KERNFS_SUICIDAL;
1606 0 : __kernfs_remove(kn);
1607 0 : kn->flags |= KERNFS_SUICIDED;
1608 0 : ret = true;
1609 : } else {
1610 0 : wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1611 0 : DEFINE_WAIT(wait);
1612 :
1613 : while (true) {
1614 0 : prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1615 :
1616 0 : if ((kn->flags & KERNFS_SUICIDED) &&
1617 0 : atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1618 : break;
1619 :
1620 0 : up_write(&root->kernfs_rwsem);
1621 0 : schedule();
1622 0 : down_write(&root->kernfs_rwsem);
1623 : }
1624 0 : finish_wait(waitq, &wait);
1625 0 : WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1626 0 : ret = false;
1627 : }
1628 :
1629 : /*
1630 : * This must be done while kernfs_rwsem held exclusive; otherwise,
1631 : * waiting for SUICIDED && deactivated could finish prematurely.
1632 : */
1633 0 : kernfs_unbreak_active_protection(kn);
1634 :
1635 0 : up_write(&root->kernfs_rwsem);
1636 0 : return ret;
1637 : }
1638 :
1639 : /**
1640 : * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1641 : * @parent: parent of the target
1642 : * @name: name of the kernfs_node to remove
1643 : * @ns: namespace tag of the kernfs_node to remove
1644 : *
1645 : * Look for the kernfs_node with @name and @ns under @parent and remove it.
1646 : *
1647 : * Return: %0 on success, -ENOENT if such entry doesn't exist.
1648 : */
1649 922 : int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1650 : const void *ns)
1651 : {
1652 : struct kernfs_node *kn;
1653 : struct kernfs_root *root;
1654 :
1655 922 : if (!parent) {
1656 0 : WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1657 : name);
1658 0 : return -ENOENT;
1659 : }
1660 :
1661 922 : root = kernfs_root(parent);
1662 922 : down_write(&root->kernfs_rwsem);
1663 :
1664 922 : kn = kernfs_find_ns(parent, name, ns);
1665 922 : if (kn) {
1666 426 : kernfs_get(kn);
1667 426 : __kernfs_remove(kn);
1668 426 : kernfs_put(kn);
1669 : }
1670 :
1671 922 : up_write(&root->kernfs_rwsem);
1672 :
1673 922 : if (kn)
1674 : return 0;
1675 : else
1676 496 : return -ENOENT;
1677 : }
1678 :
1679 : /**
1680 : * kernfs_rename_ns - move and rename a kernfs_node
1681 : * @kn: target node
1682 : * @new_parent: new parent to put @sd under
1683 : * @new_name: new name
1684 : * @new_ns: new namespace tag
1685 : *
1686 : * Return: %0 on success, -errno on failure.
1687 : */
1688 0 : int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1689 : const char *new_name, const void *new_ns)
1690 : {
1691 : struct kernfs_node *old_parent;
1692 : struct kernfs_root *root;
1693 0 : const char *old_name = NULL;
1694 : int error;
1695 :
1696 : /* can't move or rename root */
1697 0 : if (!kn->parent)
1698 : return -EINVAL;
1699 :
1700 0 : root = kernfs_root(kn);
1701 0 : down_write(&root->kernfs_rwsem);
1702 :
1703 0 : error = -ENOENT;
1704 0 : if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1705 0 : (new_parent->flags & KERNFS_EMPTY_DIR))
1706 : goto out;
1707 :
1708 0 : error = 0;
1709 0 : if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1710 0 : (strcmp(kn->name, new_name) == 0))
1711 : goto out; /* nothing to rename */
1712 :
1713 0 : error = -EEXIST;
1714 0 : if (kernfs_find_ns(new_parent, new_name, new_ns))
1715 : goto out;
1716 :
1717 : /* rename kernfs_node */
1718 0 : if (strcmp(kn->name, new_name) != 0) {
1719 0 : error = -ENOMEM;
1720 0 : new_name = kstrdup_const(new_name, GFP_KERNEL);
1721 0 : if (!new_name)
1722 : goto out;
1723 : } else {
1724 : new_name = NULL;
1725 : }
1726 :
1727 : /*
1728 : * Move to the appropriate place in the appropriate directories rbtree.
1729 : */
1730 0 : kernfs_unlink_sibling(kn);
1731 0 : kernfs_get(new_parent);
1732 :
1733 : /* rename_lock protects ->parent and ->name accessors */
1734 0 : write_lock_irq(&kernfs_rename_lock);
1735 :
1736 0 : old_parent = kn->parent;
1737 0 : kn->parent = new_parent;
1738 :
1739 0 : kn->ns = new_ns;
1740 0 : if (new_name) {
1741 0 : old_name = kn->name;
1742 0 : kn->name = new_name;
1743 : }
1744 :
1745 0 : write_unlock_irq(&kernfs_rename_lock);
1746 :
1747 0 : kn->hash = kernfs_name_hash(kn->name, kn->ns);
1748 0 : kernfs_link_sibling(kn);
1749 :
1750 0 : kernfs_put(old_parent);
1751 0 : kfree_const(old_name);
1752 :
1753 0 : error = 0;
1754 : out:
1755 0 : up_write(&root->kernfs_rwsem);
1756 0 : return error;
1757 : }
1758 :
1759 0 : static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1760 : {
1761 0 : kernfs_put(filp->private_data);
1762 0 : return 0;
1763 : }
1764 :
1765 0 : static struct kernfs_node *kernfs_dir_pos(const void *ns,
1766 : struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1767 : {
1768 0 : if (pos) {
1769 0 : int valid = kernfs_active(pos) &&
1770 0 : pos->parent == parent && hash == pos->hash;
1771 0 : kernfs_put(pos);
1772 0 : if (!valid)
1773 0 : pos = NULL;
1774 : }
1775 0 : if (!pos && (hash > 1) && (hash < INT_MAX)) {
1776 0 : struct rb_node *node = parent->dir.children.rb_node;
1777 0 : while (node) {
1778 0 : pos = rb_to_kn(node);
1779 :
1780 0 : if (hash < pos->hash)
1781 0 : node = node->rb_left;
1782 0 : else if (hash > pos->hash)
1783 0 : node = node->rb_right;
1784 : else
1785 : break;
1786 : }
1787 : }
1788 : /* Skip over entries which are dying/dead or in the wrong namespace */
1789 0 : while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1790 0 : struct rb_node *node = rb_next(&pos->rb);
1791 0 : if (!node)
1792 : pos = NULL;
1793 : else
1794 0 : pos = rb_to_kn(node);
1795 : }
1796 0 : return pos;
1797 : }
1798 :
1799 0 : static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1800 : struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1801 : {
1802 0 : pos = kernfs_dir_pos(ns, parent, ino, pos);
1803 0 : if (pos) {
1804 : do {
1805 0 : struct rb_node *node = rb_next(&pos->rb);
1806 0 : if (!node)
1807 : pos = NULL;
1808 : else
1809 0 : pos = rb_to_kn(node);
1810 0 : } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1811 : }
1812 0 : return pos;
1813 : }
1814 :
1815 0 : static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1816 : {
1817 0 : struct dentry *dentry = file->f_path.dentry;
1818 0 : struct kernfs_node *parent = kernfs_dentry_node(dentry);
1819 0 : struct kernfs_node *pos = file->private_data;
1820 : struct kernfs_root *root;
1821 0 : const void *ns = NULL;
1822 :
1823 0 : if (!dir_emit_dots(file, ctx))
1824 : return 0;
1825 :
1826 0 : root = kernfs_root(parent);
1827 0 : down_read(&root->kernfs_rwsem);
1828 :
1829 0 : if (kernfs_ns_enabled(parent))
1830 0 : ns = kernfs_info(dentry->d_sb)->ns;
1831 :
1832 0 : for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1833 : pos;
1834 0 : pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1835 0 : const char *name = pos->name;
1836 0 : unsigned int type = fs_umode_to_dtype(pos->mode);
1837 0 : int len = strlen(name);
1838 0 : ino_t ino = kernfs_ino(pos);
1839 :
1840 0 : ctx->pos = pos->hash;
1841 0 : file->private_data = pos;
1842 0 : kernfs_get(pos);
1843 :
1844 0 : up_read(&root->kernfs_rwsem);
1845 0 : if (!dir_emit(ctx, name, len, ino, type))
1846 : return 0;
1847 0 : down_read(&root->kernfs_rwsem);
1848 : }
1849 0 : up_read(&root->kernfs_rwsem);
1850 0 : file->private_data = NULL;
1851 0 : ctx->pos = INT_MAX;
1852 0 : return 0;
1853 : }
1854 :
1855 : const struct file_operations kernfs_dir_fops = {
1856 : .read = generic_read_dir,
1857 : .iterate_shared = kernfs_fop_readdir,
1858 : .release = kernfs_dir_fop_release,
1859 : .llseek = generic_file_llseek,
1860 : };
|
