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