Loading...
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
20DEFINE_MUTEX(kernfs_mutex);
21static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
22static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */
23static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
24
25#define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
26
27static bool kernfs_active(struct kernfs_node *kn)
28{
29 lockdep_assert_held(&kernfs_mutex);
30 return atomic_read(&kn->active) >= 0;
31}
32
33static bool kernfs_lockdep(struct kernfs_node *kn)
34{
35#ifdef CONFIG_DEBUG_LOCK_ALLOC
36 return kn->flags & KERNFS_LOCKDEP;
37#else
38 return false;
39#endif
40}
41
42static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
43{
44 if (!kn)
45 return strlcpy(buf, "(null)", buflen);
46
47 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
48}
49
50/* kernfs_node_depth - compute depth from @from to @to */
51static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
52{
53 size_t depth = 0;
54
55 while (to->parent && to != from) {
56 depth++;
57 to = to->parent;
58 }
59 return depth;
60}
61
62static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
63 struct kernfs_node *b)
64{
65 size_t da, db;
66 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
67
68 if (ra != rb)
69 return NULL;
70
71 da = kernfs_depth(ra->kn, a);
72 db = kernfs_depth(rb->kn, b);
73
74 while (da > db) {
75 a = a->parent;
76 da--;
77 }
78 while (db > da) {
79 b = b->parent;
80 db--;
81 }
82
83 /* worst case b and a will be the same at root */
84 while (b != a) {
85 b = b->parent;
86 a = a->parent;
87 }
88
89 return a;
90}
91
92/**
93 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
94 * where kn_from is treated as root of the path.
95 * @kn_from: kernfs node which should be treated as root for the path
96 * @kn_to: kernfs node to which path is needed
97 * @buf: buffer to copy the path into
98 * @buflen: size of @buf
99 *
100 * We need to handle couple of scenarios here:
101 * [1] when @kn_from is an ancestor of @kn_to at some level
102 * kn_from: /n1/n2/n3
103 * kn_to: /n1/n2/n3/n4/n5
104 * result: /n4/n5
105 *
106 * [2] when @kn_from is on a different hierarchy and we need to find common
107 * ancestor between @kn_from and @kn_to.
108 * kn_from: /n1/n2/n3/n4
109 * kn_to: /n1/n2/n5
110 * result: /../../n5
111 * OR
112 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
113 * kn_to: /n1/n2/n3 [depth=3]
114 * result: /../..
115 *
116 * [3] when @kn_to is NULL result will be "(null)"
117 *
118 * Returns the length of the full path. If the full length is equal to or
119 * greater than @buflen, @buf contains the truncated path with the trailing
120 * '\0'. On error, -errno is returned.
121 */
122static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
123 struct kernfs_node *kn_from,
124 char *buf, size_t buflen)
125{
126 struct kernfs_node *kn, *common;
127 const char parent_str[] = "/..";
128 size_t depth_from, depth_to, len = 0;
129 int i, j;
130
131 if (!kn_to)
132 return strlcpy(buf, "(null)", buflen);
133
134 if (!kn_from)
135 kn_from = kernfs_root(kn_to)->kn;
136
137 if (kn_from == kn_to)
138 return strlcpy(buf, "/", buflen);
139
140 if (!buf)
141 return -EINVAL;
142
143 common = kernfs_common_ancestor(kn_from, kn_to);
144 if (WARN_ON(!common))
145 return -EINVAL;
146
147 depth_to = kernfs_depth(common, kn_to);
148 depth_from = kernfs_depth(common, kn_from);
149
150 buf[0] = '\0';
151
152 for (i = 0; i < depth_from; i++)
153 len += strlcpy(buf + len, parent_str,
154 len < buflen ? buflen - len : 0);
155
156 /* Calculate how many bytes we need for the rest */
157 for (i = depth_to - 1; i >= 0; i--) {
158 for (kn = kn_to, j = 0; j < i; j++)
159 kn = kn->parent;
160 len += strlcpy(buf + len, "/",
161 len < buflen ? buflen - len : 0);
162 len += strlcpy(buf + len, kn->name,
163 len < buflen ? buflen - len : 0);
164 }
165
166 return len;
167}
168
169/**
170 * kernfs_name - obtain the name of a given node
171 * @kn: kernfs_node of interest
172 * @buf: buffer to copy @kn's name into
173 * @buflen: size of @buf
174 *
175 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
176 * similar to strlcpy(). It returns the length of @kn's name and if @buf
177 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
178 *
179 * Fills buffer with "(null)" if @kn is NULL.
180 *
181 * This function can be called from any context.
182 */
183int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
184{
185 unsigned long flags;
186 int ret;
187
188 spin_lock_irqsave(&kernfs_rename_lock, flags);
189 ret = kernfs_name_locked(kn, buf, buflen);
190 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
191 return ret;
192}
193
194/**
195 * kernfs_path_from_node - build path of node @to relative to @from.
196 * @from: parent kernfs_node relative to which we need to build the path
197 * @to: kernfs_node of interest
198 * @buf: buffer to copy @to's path into
199 * @buflen: size of @buf
200 *
201 * Builds @to's path relative to @from in @buf. @from and @to must
202 * be on the same kernfs-root. If @from is not parent of @to, then a relative
203 * path (which includes '..'s) as needed to reach from @from to @to is
204 * returned.
205 *
206 * Returns the length of the full path. If the full length is equal to or
207 * greater than @buflen, @buf contains the truncated path with the trailing
208 * '\0'. On error, -errno is returned.
209 */
210int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
211 char *buf, size_t buflen)
212{
213 unsigned long flags;
214 int ret;
215
216 spin_lock_irqsave(&kernfs_rename_lock, flags);
217 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
218 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
219 return ret;
220}
221EXPORT_SYMBOL_GPL(kernfs_path_from_node);
222
223/**
224 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
225 * @kn: kernfs_node of interest
226 *
227 * This function can be called from any context.
228 */
229void pr_cont_kernfs_name(struct kernfs_node *kn)
230{
231 unsigned long flags;
232
233 spin_lock_irqsave(&kernfs_rename_lock, flags);
234
235 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
236 pr_cont("%s", kernfs_pr_cont_buf);
237
238 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
239}
240
241/**
242 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
243 * @kn: kernfs_node of interest
244 *
245 * This function can be called from any context.
246 */
247void pr_cont_kernfs_path(struct kernfs_node *kn)
248{
249 unsigned long flags;
250 int sz;
251
252 spin_lock_irqsave(&kernfs_rename_lock, flags);
253
254 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
255 sizeof(kernfs_pr_cont_buf));
256 if (sz < 0) {
257 pr_cont("(error)");
258 goto out;
259 }
260
261 if (sz >= sizeof(kernfs_pr_cont_buf)) {
262 pr_cont("(name too long)");
263 goto out;
264 }
265
266 pr_cont("%s", kernfs_pr_cont_buf);
267
268out:
269 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
270}
271
272/**
273 * kernfs_get_parent - determine the parent node and pin it
274 * @kn: kernfs_node of interest
275 *
276 * Determines @kn's parent, pins and returns it. This function can be
277 * called from any context.
278 */
279struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
280{
281 struct kernfs_node *parent;
282 unsigned long flags;
283
284 spin_lock_irqsave(&kernfs_rename_lock, flags);
285 parent = kn->parent;
286 kernfs_get(parent);
287 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
288
289 return parent;
290}
291
292/**
293 * kernfs_name_hash
294 * @name: Null terminated string to hash
295 * @ns: Namespace tag to hash
296 *
297 * Returns 31 bit hash of ns + name (so it fits in an off_t )
298 */
299static unsigned int kernfs_name_hash(const char *name, const void *ns)
300{
301 unsigned long hash = init_name_hash(ns);
302 unsigned int len = strlen(name);
303 while (len--)
304 hash = partial_name_hash(*name++, hash);
305 hash = end_name_hash(hash);
306 hash &= 0x7fffffffU;
307 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
308 if (hash < 2)
309 hash += 2;
310 if (hash >= INT_MAX)
311 hash = INT_MAX - 1;
312 return hash;
313}
314
315static int kernfs_name_compare(unsigned int hash, const char *name,
316 const void *ns, const struct kernfs_node *kn)
317{
318 if (hash < kn->hash)
319 return -1;
320 if (hash > kn->hash)
321 return 1;
322 if (ns < kn->ns)
323 return -1;
324 if (ns > kn->ns)
325 return 1;
326 return strcmp(name, kn->name);
327}
328
329static int kernfs_sd_compare(const struct kernfs_node *left,
330 const struct kernfs_node *right)
331{
332 return kernfs_name_compare(left->hash, left->name, left->ns, right);
333}
334
335/**
336 * kernfs_link_sibling - link kernfs_node into sibling rbtree
337 * @kn: kernfs_node of interest
338 *
339 * Link @kn into its sibling rbtree which starts from
340 * @kn->parent->dir.children.
341 *
342 * Locking:
343 * mutex_lock(kernfs_mutex)
344 *
345 * RETURNS:
346 * 0 on susccess -EEXIST on failure.
347 */
348static int kernfs_link_sibling(struct kernfs_node *kn)
349{
350 struct rb_node **node = &kn->parent->dir.children.rb_node;
351 struct rb_node *parent = NULL;
352
353 while (*node) {
354 struct kernfs_node *pos;
355 int result;
356
357 pos = rb_to_kn(*node);
358 parent = *node;
359 result = kernfs_sd_compare(kn, pos);
360 if (result < 0)
361 node = &pos->rb.rb_left;
362 else if (result > 0)
363 node = &pos->rb.rb_right;
364 else
365 return -EEXIST;
366 }
367
368 /* add new node and rebalance the tree */
369 rb_link_node(&kn->rb, parent, node);
370 rb_insert_color(&kn->rb, &kn->parent->dir.children);
371
372 /* successfully added, account subdir number */
373 if (kernfs_type(kn) == KERNFS_DIR)
374 kn->parent->dir.subdirs++;
375
376 return 0;
377}
378
379/**
380 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
381 * @kn: kernfs_node of interest
382 *
383 * Try to unlink @kn from its sibling rbtree which starts from
384 * kn->parent->dir.children. Returns %true if @kn was actually
385 * removed, %false if @kn wasn't on the rbtree.
386 *
387 * Locking:
388 * mutex_lock(kernfs_mutex)
389 */
390static bool kernfs_unlink_sibling(struct kernfs_node *kn)
391{
392 if (RB_EMPTY_NODE(&kn->rb))
393 return false;
394
395 if (kernfs_type(kn) == KERNFS_DIR)
396 kn->parent->dir.subdirs--;
397
398 rb_erase(&kn->rb, &kn->parent->dir.children);
399 RB_CLEAR_NODE(&kn->rb);
400 return true;
401}
402
403/**
404 * kernfs_get_active - get an active reference to kernfs_node
405 * @kn: kernfs_node to get an active reference to
406 *
407 * Get an active reference of @kn. This function is noop if @kn
408 * is NULL.
409 *
410 * RETURNS:
411 * Pointer to @kn on success, NULL on failure.
412 */
413struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
414{
415 if (unlikely(!kn))
416 return NULL;
417
418 if (!atomic_inc_unless_negative(&kn->active))
419 return NULL;
420
421 if (kernfs_lockdep(kn))
422 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
423 return kn;
424}
425
426/**
427 * kernfs_put_active - put an active reference to kernfs_node
428 * @kn: kernfs_node to put an active reference to
429 *
430 * Put an active reference to @kn. This function is noop if @kn
431 * is NULL.
432 */
433void kernfs_put_active(struct kernfs_node *kn)
434{
435 int v;
436
437 if (unlikely(!kn))
438 return;
439
440 if (kernfs_lockdep(kn))
441 rwsem_release(&kn->dep_map, 1, _RET_IP_);
442 v = atomic_dec_return(&kn->active);
443 if (likely(v != KN_DEACTIVATED_BIAS))
444 return;
445
446 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
447}
448
449/**
450 * kernfs_drain - drain kernfs_node
451 * @kn: kernfs_node to drain
452 *
453 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
454 * removers may invoke this function concurrently on @kn and all will
455 * return after draining is complete.
456 */
457static void kernfs_drain(struct kernfs_node *kn)
458 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
459{
460 struct kernfs_root *root = kernfs_root(kn);
461
462 lockdep_assert_held(&kernfs_mutex);
463 WARN_ON_ONCE(kernfs_active(kn));
464
465 mutex_unlock(&kernfs_mutex);
466
467 if (kernfs_lockdep(kn)) {
468 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
469 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
470 lock_contended(&kn->dep_map, _RET_IP_);
471 }
472
473 /* but everyone should wait for draining */
474 wait_event(root->deactivate_waitq,
475 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
476
477 if (kernfs_lockdep(kn)) {
478 lock_acquired(&kn->dep_map, _RET_IP_);
479 rwsem_release(&kn->dep_map, 1, _RET_IP_);
480 }
481
482 kernfs_drain_open_files(kn);
483
484 mutex_lock(&kernfs_mutex);
485}
486
487/**
488 * kernfs_get - get a reference count on a kernfs_node
489 * @kn: the target kernfs_node
490 */
491void kernfs_get(struct kernfs_node *kn)
492{
493 if (kn) {
494 WARN_ON(!atomic_read(&kn->count));
495 atomic_inc(&kn->count);
496 }
497}
498EXPORT_SYMBOL_GPL(kernfs_get);
499
500/**
501 * kernfs_put - put a reference count on a kernfs_node
502 * @kn: the target kernfs_node
503 *
504 * Put a reference count of @kn and destroy it if it reached zero.
505 */
506void kernfs_put(struct kernfs_node *kn)
507{
508 struct kernfs_node *parent;
509 struct kernfs_root *root;
510
511 /*
512 * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino
513 * depends on this to filter reused stale node
514 */
515 if (!kn || !atomic_dec_and_test(&kn->count))
516 return;
517 root = kernfs_root(kn);
518 repeat:
519 /*
520 * Moving/renaming is always done while holding reference.
521 * kn->parent won't change beneath us.
522 */
523 parent = kn->parent;
524
525 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
526 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
527 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
528
529 if (kernfs_type(kn) == KERNFS_LINK)
530 kernfs_put(kn->symlink.target_kn);
531
532 kfree_const(kn->name);
533
534 if (kn->iattr) {
535 simple_xattrs_free(&kn->iattr->xattrs);
536 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
537 }
538 spin_lock(&kernfs_idr_lock);
539 idr_remove(&root->ino_idr, kn->id.ino);
540 spin_unlock(&kernfs_idr_lock);
541 kmem_cache_free(kernfs_node_cache, kn);
542
543 kn = parent;
544 if (kn) {
545 if (atomic_dec_and_test(&kn->count))
546 goto repeat;
547 } else {
548 /* just released the root kn, free @root too */
549 idr_destroy(&root->ino_idr);
550 kfree(root);
551 }
552}
553EXPORT_SYMBOL_GPL(kernfs_put);
554
555static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
556{
557 struct kernfs_node *kn;
558
559 if (flags & LOOKUP_RCU)
560 return -ECHILD;
561
562 /* Always perform fresh lookup for negatives */
563 if (d_really_is_negative(dentry))
564 goto out_bad_unlocked;
565
566 kn = kernfs_dentry_node(dentry);
567 mutex_lock(&kernfs_mutex);
568
569 /* The kernfs node has been deactivated */
570 if (!kernfs_active(kn))
571 goto out_bad;
572
573 /* The kernfs node has been moved? */
574 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
575 goto out_bad;
576
577 /* The kernfs node has been renamed */
578 if (strcmp(dentry->d_name.name, kn->name) != 0)
579 goto out_bad;
580
581 /* The kernfs node has been moved to a different namespace */
582 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
583 kernfs_info(dentry->d_sb)->ns != kn->ns)
584 goto out_bad;
585
586 mutex_unlock(&kernfs_mutex);
587 return 1;
588out_bad:
589 mutex_unlock(&kernfs_mutex);
590out_bad_unlocked:
591 return 0;
592}
593
594const struct dentry_operations kernfs_dops = {
595 .d_revalidate = kernfs_dop_revalidate,
596};
597
598/**
599 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
600 * @dentry: the dentry in question
601 *
602 * Return the kernfs_node associated with @dentry. If @dentry is not a
603 * kernfs one, %NULL is returned.
604 *
605 * While the returned kernfs_node will stay accessible as long as @dentry
606 * is accessible, the returned node can be in any state and the caller is
607 * fully responsible for determining what's accessible.
608 */
609struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
610{
611 if (dentry->d_sb->s_op == &kernfs_sops &&
612 !d_really_is_negative(dentry))
613 return kernfs_dentry_node(dentry);
614 return NULL;
615}
616
617static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
618 struct kernfs_node *parent,
619 const char *name, umode_t mode,
620 kuid_t uid, kgid_t gid,
621 unsigned flags)
622{
623 struct kernfs_node *kn;
624 u32 gen;
625 int cursor;
626 int ret;
627
628 name = kstrdup_const(name, GFP_KERNEL);
629 if (!name)
630 return NULL;
631
632 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
633 if (!kn)
634 goto err_out1;
635
636 idr_preload(GFP_KERNEL);
637 spin_lock(&kernfs_idr_lock);
638 cursor = idr_get_cursor(&root->ino_idr);
639 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
640 if (ret >= 0 && ret < cursor)
641 root->next_generation++;
642 gen = root->next_generation;
643 spin_unlock(&kernfs_idr_lock);
644 idr_preload_end();
645 if (ret < 0)
646 goto err_out2;
647 kn->id.ino = ret;
648 kn->id.generation = gen;
649
650 /*
651 * set ino first. This RELEASE is paired with atomic_inc_not_zero in
652 * kernfs_find_and_get_node_by_ino
653 */
654 atomic_set_release(&kn->count, 1);
655 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
656 RB_CLEAR_NODE(&kn->rb);
657
658 kn->name = name;
659 kn->mode = mode;
660 kn->flags = flags;
661
662 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
663 struct iattr iattr = {
664 .ia_valid = ATTR_UID | ATTR_GID,
665 .ia_uid = uid,
666 .ia_gid = gid,
667 };
668
669 ret = __kernfs_setattr(kn, &iattr);
670 if (ret < 0)
671 goto err_out3;
672 }
673
674 if (parent) {
675 ret = security_kernfs_init_security(parent, kn);
676 if (ret)
677 goto err_out3;
678 }
679
680 return kn;
681
682 err_out3:
683 idr_remove(&root->ino_idr, kn->id.ino);
684 err_out2:
685 kmem_cache_free(kernfs_node_cache, kn);
686 err_out1:
687 kfree_const(name);
688 return NULL;
689}
690
691struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
692 const char *name, umode_t mode,
693 kuid_t uid, kgid_t gid,
694 unsigned flags)
695{
696 struct kernfs_node *kn;
697
698 kn = __kernfs_new_node(kernfs_root(parent), parent,
699 name, mode, uid, gid, flags);
700 if (kn) {
701 kernfs_get(parent);
702 kn->parent = parent;
703 }
704 return kn;
705}
706
707/*
708 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
709 * @root: the kernfs root
710 * @ino: inode number
711 *
712 * RETURNS:
713 * NULL on failure. Return a kernfs node with reference counter incremented
714 */
715struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
716 unsigned int ino)
717{
718 struct kernfs_node *kn;
719
720 rcu_read_lock();
721 kn = idr_find(&root->ino_idr, ino);
722 if (!kn)
723 goto out;
724
725 /*
726 * Since kernfs_node is freed in RCU, it's possible an old node for ino
727 * is freed, but reused before RCU grace period. But a freed node (see
728 * kernfs_put) or an incompletedly initialized node (see
729 * __kernfs_new_node) should have 'count' 0. We can use this fact to
730 * filter out such node.
731 */
732 if (!atomic_inc_not_zero(&kn->count)) {
733 kn = NULL;
734 goto out;
735 }
736
737 /*
738 * The node could be a new node or a reused node. If it's a new node,
739 * we are ok. If it's reused because of RCU (because of
740 * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
741 * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
742 * hence we can use 'ino' to filter stale node.
743 */
744 if (kn->id.ino != ino)
745 goto out;
746 rcu_read_unlock();
747
748 return kn;
749out:
750 rcu_read_unlock();
751 kernfs_put(kn);
752 return NULL;
753}
754
755/**
756 * kernfs_add_one - add kernfs_node to parent without warning
757 * @kn: kernfs_node to be added
758 *
759 * The caller must already have initialized @kn->parent. This
760 * function increments nlink of the parent's inode if @kn is a
761 * directory and link into the children list of the parent.
762 *
763 * RETURNS:
764 * 0 on success, -EEXIST if entry with the given name already
765 * exists.
766 */
767int kernfs_add_one(struct kernfs_node *kn)
768{
769 struct kernfs_node *parent = kn->parent;
770 struct kernfs_iattrs *ps_iattr;
771 bool has_ns;
772 int ret;
773
774 mutex_lock(&kernfs_mutex);
775
776 ret = -EINVAL;
777 has_ns = kernfs_ns_enabled(parent);
778 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
779 has_ns ? "required" : "invalid", parent->name, kn->name))
780 goto out_unlock;
781
782 if (kernfs_type(parent) != KERNFS_DIR)
783 goto out_unlock;
784
785 ret = -ENOENT;
786 if (parent->flags & KERNFS_EMPTY_DIR)
787 goto out_unlock;
788
789 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
790 goto out_unlock;
791
792 kn->hash = kernfs_name_hash(kn->name, kn->ns);
793
794 ret = kernfs_link_sibling(kn);
795 if (ret)
796 goto out_unlock;
797
798 /* Update timestamps on the parent */
799 ps_iattr = parent->iattr;
800 if (ps_iattr) {
801 ktime_get_real_ts64(&ps_iattr->ia_ctime);
802 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
803 }
804
805 mutex_unlock(&kernfs_mutex);
806
807 /*
808 * Activate the new node unless CREATE_DEACTIVATED is requested.
809 * If not activated here, the kernfs user is responsible for
810 * activating the node with kernfs_activate(). A node which hasn't
811 * been activated is not visible to userland and its removal won't
812 * trigger deactivation.
813 */
814 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
815 kernfs_activate(kn);
816 return 0;
817
818out_unlock:
819 mutex_unlock(&kernfs_mutex);
820 return ret;
821}
822
823/**
824 * kernfs_find_ns - find kernfs_node with the given name
825 * @parent: kernfs_node to search under
826 * @name: name to look for
827 * @ns: the namespace tag to use
828 *
829 * Look for kernfs_node with name @name under @parent. Returns pointer to
830 * the found kernfs_node on success, %NULL on failure.
831 */
832static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
833 const unsigned char *name,
834 const void *ns)
835{
836 struct rb_node *node = parent->dir.children.rb_node;
837 bool has_ns = kernfs_ns_enabled(parent);
838 unsigned int hash;
839
840 lockdep_assert_held(&kernfs_mutex);
841
842 if (has_ns != (bool)ns) {
843 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
844 has_ns ? "required" : "invalid", parent->name, name);
845 return NULL;
846 }
847
848 hash = kernfs_name_hash(name, ns);
849 while (node) {
850 struct kernfs_node *kn;
851 int result;
852
853 kn = rb_to_kn(node);
854 result = kernfs_name_compare(hash, name, ns, kn);
855 if (result < 0)
856 node = node->rb_left;
857 else if (result > 0)
858 node = node->rb_right;
859 else
860 return kn;
861 }
862 return NULL;
863}
864
865static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
866 const unsigned char *path,
867 const void *ns)
868{
869 size_t len;
870 char *p, *name;
871
872 lockdep_assert_held(&kernfs_mutex);
873
874 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
875 spin_lock_irq(&kernfs_rename_lock);
876
877 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
878
879 if (len >= sizeof(kernfs_pr_cont_buf)) {
880 spin_unlock_irq(&kernfs_rename_lock);
881 return NULL;
882 }
883
884 p = kernfs_pr_cont_buf;
885
886 while ((name = strsep(&p, "/")) && parent) {
887 if (*name == '\0')
888 continue;
889 parent = kernfs_find_ns(parent, name, ns);
890 }
891
892 spin_unlock_irq(&kernfs_rename_lock);
893
894 return parent;
895}
896
897/**
898 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
899 * @parent: kernfs_node to search under
900 * @name: name to look for
901 * @ns: the namespace tag to use
902 *
903 * Look for kernfs_node with name @name under @parent and get a reference
904 * if found. This function may sleep and returns pointer to the found
905 * kernfs_node on success, %NULL on failure.
906 */
907struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
908 const char *name, const void *ns)
909{
910 struct kernfs_node *kn;
911
912 mutex_lock(&kernfs_mutex);
913 kn = kernfs_find_ns(parent, name, ns);
914 kernfs_get(kn);
915 mutex_unlock(&kernfs_mutex);
916
917 return kn;
918}
919EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
920
921/**
922 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
923 * @parent: kernfs_node to search under
924 * @path: path to look for
925 * @ns: the namespace tag to use
926 *
927 * Look for kernfs_node with path @path under @parent and get a reference
928 * if found. This function may sleep and returns pointer to the found
929 * kernfs_node on success, %NULL on failure.
930 */
931struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
932 const char *path, const void *ns)
933{
934 struct kernfs_node *kn;
935
936 mutex_lock(&kernfs_mutex);
937 kn = kernfs_walk_ns(parent, path, ns);
938 kernfs_get(kn);
939 mutex_unlock(&kernfs_mutex);
940
941 return kn;
942}
943
944/**
945 * kernfs_create_root - create a new kernfs hierarchy
946 * @scops: optional syscall operations for the hierarchy
947 * @flags: KERNFS_ROOT_* flags
948 * @priv: opaque data associated with the new directory
949 *
950 * Returns the root of the new hierarchy on success, ERR_PTR() value on
951 * failure.
952 */
953struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
954 unsigned int flags, void *priv)
955{
956 struct kernfs_root *root;
957 struct kernfs_node *kn;
958
959 root = kzalloc(sizeof(*root), GFP_KERNEL);
960 if (!root)
961 return ERR_PTR(-ENOMEM);
962
963 idr_init(&root->ino_idr);
964 INIT_LIST_HEAD(&root->supers);
965 root->next_generation = 1;
966
967 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
968 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
969 KERNFS_DIR);
970 if (!kn) {
971 idr_destroy(&root->ino_idr);
972 kfree(root);
973 return ERR_PTR(-ENOMEM);
974 }
975
976 kn->priv = priv;
977 kn->dir.root = root;
978
979 root->syscall_ops = scops;
980 root->flags = flags;
981 root->kn = kn;
982 init_waitqueue_head(&root->deactivate_waitq);
983
984 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
985 kernfs_activate(kn);
986
987 return root;
988}
989
990/**
991 * kernfs_destroy_root - destroy a kernfs hierarchy
992 * @root: root of the hierarchy to destroy
993 *
994 * Destroy the hierarchy anchored at @root by removing all existing
995 * directories and destroying @root.
996 */
997void kernfs_destroy_root(struct kernfs_root *root)
998{
999 kernfs_remove(root->kn); /* will also free @root */
1000}
1001
1002/**
1003 * kernfs_create_dir_ns - create a directory
1004 * @parent: parent in which to create a new directory
1005 * @name: name of the new directory
1006 * @mode: mode of the new directory
1007 * @uid: uid of the new directory
1008 * @gid: gid of the new directory
1009 * @priv: opaque data associated with the new directory
1010 * @ns: optional namespace tag of the directory
1011 *
1012 * Returns the created node on success, ERR_PTR() value on failure.
1013 */
1014struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1015 const char *name, umode_t mode,
1016 kuid_t uid, kgid_t gid,
1017 void *priv, const void *ns)
1018{
1019 struct kernfs_node *kn;
1020 int rc;
1021
1022 /* allocate */
1023 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1024 uid, gid, KERNFS_DIR);
1025 if (!kn)
1026 return ERR_PTR(-ENOMEM);
1027
1028 kn->dir.root = parent->dir.root;
1029 kn->ns = ns;
1030 kn->priv = priv;
1031
1032 /* link in */
1033 rc = kernfs_add_one(kn);
1034 if (!rc)
1035 return kn;
1036
1037 kernfs_put(kn);
1038 return ERR_PTR(rc);
1039}
1040
1041/**
1042 * kernfs_create_empty_dir - create an always empty directory
1043 * @parent: parent in which to create a new directory
1044 * @name: name of the new directory
1045 *
1046 * Returns the created node on success, ERR_PTR() value on failure.
1047 */
1048struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1049 const char *name)
1050{
1051 struct kernfs_node *kn;
1052 int rc;
1053
1054 /* allocate */
1055 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1056 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1057 if (!kn)
1058 return ERR_PTR(-ENOMEM);
1059
1060 kn->flags |= KERNFS_EMPTY_DIR;
1061 kn->dir.root = parent->dir.root;
1062 kn->ns = NULL;
1063 kn->priv = NULL;
1064
1065 /* link in */
1066 rc = kernfs_add_one(kn);
1067 if (!rc)
1068 return kn;
1069
1070 kernfs_put(kn);
1071 return ERR_PTR(rc);
1072}
1073
1074static struct dentry *kernfs_iop_lookup(struct inode *dir,
1075 struct dentry *dentry,
1076 unsigned int flags)
1077{
1078 struct dentry *ret;
1079 struct kernfs_node *parent = dir->i_private;
1080 struct kernfs_node *kn;
1081 struct inode *inode;
1082 const void *ns = NULL;
1083
1084 mutex_lock(&kernfs_mutex);
1085
1086 if (kernfs_ns_enabled(parent))
1087 ns = kernfs_info(dir->i_sb)->ns;
1088
1089 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1090
1091 /* no such entry */
1092 if (!kn || !kernfs_active(kn)) {
1093 ret = NULL;
1094 goto out_unlock;
1095 }
1096
1097 /* attach dentry and inode */
1098 inode = kernfs_get_inode(dir->i_sb, kn);
1099 if (!inode) {
1100 ret = ERR_PTR(-ENOMEM);
1101 goto out_unlock;
1102 }
1103
1104 /* instantiate and hash dentry */
1105 ret = d_splice_alias(inode, dentry);
1106 out_unlock:
1107 mutex_unlock(&kernfs_mutex);
1108 return ret;
1109}
1110
1111static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1112 umode_t mode)
1113{
1114 struct kernfs_node *parent = dir->i_private;
1115 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1116 int ret;
1117
1118 if (!scops || !scops->mkdir)
1119 return -EPERM;
1120
1121 if (!kernfs_get_active(parent))
1122 return -ENODEV;
1123
1124 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1125
1126 kernfs_put_active(parent);
1127 return ret;
1128}
1129
1130static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1131{
1132 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1133 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1134 int ret;
1135
1136 if (!scops || !scops->rmdir)
1137 return -EPERM;
1138
1139 if (!kernfs_get_active(kn))
1140 return -ENODEV;
1141
1142 ret = scops->rmdir(kn);
1143
1144 kernfs_put_active(kn);
1145 return ret;
1146}
1147
1148static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1149 struct inode *new_dir, struct dentry *new_dentry,
1150 unsigned int flags)
1151{
1152 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1153 struct kernfs_node *new_parent = new_dir->i_private;
1154 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1155 int ret;
1156
1157 if (flags)
1158 return -EINVAL;
1159
1160 if (!scops || !scops->rename)
1161 return -EPERM;
1162
1163 if (!kernfs_get_active(kn))
1164 return -ENODEV;
1165
1166 if (!kernfs_get_active(new_parent)) {
1167 kernfs_put_active(kn);
1168 return -ENODEV;
1169 }
1170
1171 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1172
1173 kernfs_put_active(new_parent);
1174 kernfs_put_active(kn);
1175 return ret;
1176}
1177
1178const struct inode_operations kernfs_dir_iops = {
1179 .lookup = kernfs_iop_lookup,
1180 .permission = kernfs_iop_permission,
1181 .setattr = kernfs_iop_setattr,
1182 .getattr = kernfs_iop_getattr,
1183 .listxattr = kernfs_iop_listxattr,
1184
1185 .mkdir = kernfs_iop_mkdir,
1186 .rmdir = kernfs_iop_rmdir,
1187 .rename = kernfs_iop_rename,
1188};
1189
1190static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1191{
1192 struct kernfs_node *last;
1193
1194 while (true) {
1195 struct rb_node *rbn;
1196
1197 last = pos;
1198
1199 if (kernfs_type(pos) != KERNFS_DIR)
1200 break;
1201
1202 rbn = rb_first(&pos->dir.children);
1203 if (!rbn)
1204 break;
1205
1206 pos = rb_to_kn(rbn);
1207 }
1208
1209 return last;
1210}
1211
1212/**
1213 * kernfs_next_descendant_post - find the next descendant for post-order walk
1214 * @pos: the current position (%NULL to initiate traversal)
1215 * @root: kernfs_node whose descendants to walk
1216 *
1217 * Find the next descendant to visit for post-order traversal of @root's
1218 * descendants. @root is included in the iteration and the last node to be
1219 * visited.
1220 */
1221static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1222 struct kernfs_node *root)
1223{
1224 struct rb_node *rbn;
1225
1226 lockdep_assert_held(&kernfs_mutex);
1227
1228 /* if first iteration, visit leftmost descendant which may be root */
1229 if (!pos)
1230 return kernfs_leftmost_descendant(root);
1231
1232 /* if we visited @root, we're done */
1233 if (pos == root)
1234 return NULL;
1235
1236 /* if there's an unvisited sibling, visit its leftmost descendant */
1237 rbn = rb_next(&pos->rb);
1238 if (rbn)
1239 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1240
1241 /* no sibling left, visit parent */
1242 return pos->parent;
1243}
1244
1245/**
1246 * kernfs_activate - activate a node which started deactivated
1247 * @kn: kernfs_node whose subtree is to be activated
1248 *
1249 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1250 * needs to be explicitly activated. A node which hasn't been activated
1251 * isn't visible to userland and deactivation is skipped during its
1252 * removal. This is useful to construct atomic init sequences where
1253 * creation of multiple nodes should either succeed or fail atomically.
1254 *
1255 * The caller is responsible for ensuring that this function is not called
1256 * after kernfs_remove*() is invoked on @kn.
1257 */
1258void kernfs_activate(struct kernfs_node *kn)
1259{
1260 struct kernfs_node *pos;
1261
1262 mutex_lock(&kernfs_mutex);
1263
1264 pos = NULL;
1265 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1266 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1267 continue;
1268
1269 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1270 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1271
1272 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1273 pos->flags |= KERNFS_ACTIVATED;
1274 }
1275
1276 mutex_unlock(&kernfs_mutex);
1277}
1278
1279static void __kernfs_remove(struct kernfs_node *kn)
1280{
1281 struct kernfs_node *pos;
1282
1283 lockdep_assert_held(&kernfs_mutex);
1284
1285 /*
1286 * Short-circuit if non-root @kn has already finished removal.
1287 * This is for kernfs_remove_self() which plays with active ref
1288 * after removal.
1289 */
1290 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1291 return;
1292
1293 pr_debug("kernfs %s: removing\n", kn->name);
1294
1295 /* prevent any new usage under @kn by deactivating all nodes */
1296 pos = NULL;
1297 while ((pos = kernfs_next_descendant_post(pos, kn)))
1298 if (kernfs_active(pos))
1299 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1300
1301 /* deactivate and unlink the subtree node-by-node */
1302 do {
1303 pos = kernfs_leftmost_descendant(kn);
1304
1305 /*
1306 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1307 * base ref could have been put by someone else by the time
1308 * the function returns. Make sure it doesn't go away
1309 * underneath us.
1310 */
1311 kernfs_get(pos);
1312
1313 /*
1314 * Drain iff @kn was activated. This avoids draining and
1315 * its lockdep annotations for nodes which have never been
1316 * activated and allows embedding kernfs_remove() in create
1317 * error paths without worrying about draining.
1318 */
1319 if (kn->flags & KERNFS_ACTIVATED)
1320 kernfs_drain(pos);
1321 else
1322 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1323
1324 /*
1325 * kernfs_unlink_sibling() succeeds once per node. Use it
1326 * to decide who's responsible for cleanups.
1327 */
1328 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1329 struct kernfs_iattrs *ps_iattr =
1330 pos->parent ? pos->parent->iattr : NULL;
1331
1332 /* update timestamps on the parent */
1333 if (ps_iattr) {
1334 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1335 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1336 }
1337
1338 kernfs_put(pos);
1339 }
1340
1341 kernfs_put(pos);
1342 } while (pos != kn);
1343}
1344
1345/**
1346 * kernfs_remove - remove a kernfs_node recursively
1347 * @kn: the kernfs_node to remove
1348 *
1349 * Remove @kn along with all its subdirectories and files.
1350 */
1351void kernfs_remove(struct kernfs_node *kn)
1352{
1353 mutex_lock(&kernfs_mutex);
1354 __kernfs_remove(kn);
1355 mutex_unlock(&kernfs_mutex);
1356}
1357
1358/**
1359 * kernfs_break_active_protection - break out of active protection
1360 * @kn: the self kernfs_node
1361 *
1362 * The caller must be running off of a kernfs operation which is invoked
1363 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1364 * this function must also be matched with an invocation of
1365 * kernfs_unbreak_active_protection().
1366 *
1367 * This function releases the active reference of @kn the caller is
1368 * holding. Once this function is called, @kn may be removed at any point
1369 * and the caller is solely responsible for ensuring that the objects it
1370 * dereferences are accessible.
1371 */
1372void kernfs_break_active_protection(struct kernfs_node *kn)
1373{
1374 /*
1375 * Take out ourself out of the active ref dependency chain. If
1376 * we're called without an active ref, lockdep will complain.
1377 */
1378 kernfs_put_active(kn);
1379}
1380
1381/**
1382 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1383 * @kn: the self kernfs_node
1384 *
1385 * If kernfs_break_active_protection() was called, this function must be
1386 * invoked before finishing the kernfs operation. Note that while this
1387 * function restores the active reference, it doesn't and can't actually
1388 * restore the active protection - @kn may already or be in the process of
1389 * being removed. Once kernfs_break_active_protection() is invoked, that
1390 * protection is irreversibly gone for the kernfs operation instance.
1391 *
1392 * While this function may be called at any point after
1393 * kernfs_break_active_protection() is invoked, its most useful location
1394 * would be right before the enclosing kernfs operation returns.
1395 */
1396void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1397{
1398 /*
1399 * @kn->active could be in any state; however, the increment we do
1400 * here will be undone as soon as the enclosing kernfs operation
1401 * finishes and this temporary bump can't break anything. If @kn
1402 * is alive, nothing changes. If @kn is being deactivated, the
1403 * soon-to-follow put will either finish deactivation or restore
1404 * deactivated state. If @kn is already removed, the temporary
1405 * bump is guaranteed to be gone before @kn is released.
1406 */
1407 atomic_inc(&kn->active);
1408 if (kernfs_lockdep(kn))
1409 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1410}
1411
1412/**
1413 * kernfs_remove_self - remove a kernfs_node from its own method
1414 * @kn: the self kernfs_node to remove
1415 *
1416 * The caller must be running off of a kernfs operation which is invoked
1417 * with an active reference - e.g. one of kernfs_ops. This can be used to
1418 * implement a file operation which deletes itself.
1419 *
1420 * For example, the "delete" file for a sysfs device directory can be
1421 * implemented by invoking kernfs_remove_self() on the "delete" file
1422 * itself. This function breaks the circular dependency of trying to
1423 * deactivate self while holding an active ref itself. It isn't necessary
1424 * to modify the usual removal path to use kernfs_remove_self(). The
1425 * "delete" implementation can simply invoke kernfs_remove_self() on self
1426 * before proceeding with the usual removal path. kernfs will ignore later
1427 * kernfs_remove() on self.
1428 *
1429 * kernfs_remove_self() can be called multiple times concurrently on the
1430 * same kernfs_node. Only the first one actually performs removal and
1431 * returns %true. All others will wait until the kernfs operation which
1432 * won self-removal finishes and return %false. Note that the losers wait
1433 * for the completion of not only the winning kernfs_remove_self() but also
1434 * the whole kernfs_ops which won the arbitration. This can be used to
1435 * guarantee, for example, all concurrent writes to a "delete" file to
1436 * finish only after the whole operation is complete.
1437 */
1438bool kernfs_remove_self(struct kernfs_node *kn)
1439{
1440 bool ret;
1441
1442 mutex_lock(&kernfs_mutex);
1443 kernfs_break_active_protection(kn);
1444
1445 /*
1446 * SUICIDAL is used to arbitrate among competing invocations. Only
1447 * the first one will actually perform removal. When the removal
1448 * is complete, SUICIDED is set and the active ref is restored
1449 * while holding kernfs_mutex. The ones which lost arbitration
1450 * waits for SUICDED && drained which can happen only after the
1451 * enclosing kernfs operation which executed the winning instance
1452 * of kernfs_remove_self() finished.
1453 */
1454 if (!(kn->flags & KERNFS_SUICIDAL)) {
1455 kn->flags |= KERNFS_SUICIDAL;
1456 __kernfs_remove(kn);
1457 kn->flags |= KERNFS_SUICIDED;
1458 ret = true;
1459 } else {
1460 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1461 DEFINE_WAIT(wait);
1462
1463 while (true) {
1464 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1465
1466 if ((kn->flags & KERNFS_SUICIDED) &&
1467 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1468 break;
1469
1470 mutex_unlock(&kernfs_mutex);
1471 schedule();
1472 mutex_lock(&kernfs_mutex);
1473 }
1474 finish_wait(waitq, &wait);
1475 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1476 ret = false;
1477 }
1478
1479 /*
1480 * This must be done while holding kernfs_mutex; otherwise, waiting
1481 * for SUICIDED && deactivated could finish prematurely.
1482 */
1483 kernfs_unbreak_active_protection(kn);
1484
1485 mutex_unlock(&kernfs_mutex);
1486 return ret;
1487}
1488
1489/**
1490 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1491 * @parent: parent of the target
1492 * @name: name of the kernfs_node to remove
1493 * @ns: namespace tag of the kernfs_node to remove
1494 *
1495 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1496 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1497 */
1498int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1499 const void *ns)
1500{
1501 struct kernfs_node *kn;
1502
1503 if (!parent) {
1504 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1505 name);
1506 return -ENOENT;
1507 }
1508
1509 mutex_lock(&kernfs_mutex);
1510
1511 kn = kernfs_find_ns(parent, name, ns);
1512 if (kn)
1513 __kernfs_remove(kn);
1514
1515 mutex_unlock(&kernfs_mutex);
1516
1517 if (kn)
1518 return 0;
1519 else
1520 return -ENOENT;
1521}
1522
1523/**
1524 * kernfs_rename_ns - move and rename a kernfs_node
1525 * @kn: target node
1526 * @new_parent: new parent to put @sd under
1527 * @new_name: new name
1528 * @new_ns: new namespace tag
1529 */
1530int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1531 const char *new_name, const void *new_ns)
1532{
1533 struct kernfs_node *old_parent;
1534 const char *old_name = NULL;
1535 int error;
1536
1537 /* can't move or rename root */
1538 if (!kn->parent)
1539 return -EINVAL;
1540
1541 mutex_lock(&kernfs_mutex);
1542
1543 error = -ENOENT;
1544 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1545 (new_parent->flags & KERNFS_EMPTY_DIR))
1546 goto out;
1547
1548 error = 0;
1549 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1550 (strcmp(kn->name, new_name) == 0))
1551 goto out; /* nothing to rename */
1552
1553 error = -EEXIST;
1554 if (kernfs_find_ns(new_parent, new_name, new_ns))
1555 goto out;
1556
1557 /* rename kernfs_node */
1558 if (strcmp(kn->name, new_name) != 0) {
1559 error = -ENOMEM;
1560 new_name = kstrdup_const(new_name, GFP_KERNEL);
1561 if (!new_name)
1562 goto out;
1563 } else {
1564 new_name = NULL;
1565 }
1566
1567 /*
1568 * Move to the appropriate place in the appropriate directories rbtree.
1569 */
1570 kernfs_unlink_sibling(kn);
1571 kernfs_get(new_parent);
1572
1573 /* rename_lock protects ->parent and ->name accessors */
1574 spin_lock_irq(&kernfs_rename_lock);
1575
1576 old_parent = kn->parent;
1577 kn->parent = new_parent;
1578
1579 kn->ns = new_ns;
1580 if (new_name) {
1581 old_name = kn->name;
1582 kn->name = new_name;
1583 }
1584
1585 spin_unlock_irq(&kernfs_rename_lock);
1586
1587 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1588 kernfs_link_sibling(kn);
1589
1590 kernfs_put(old_parent);
1591 kfree_const(old_name);
1592
1593 error = 0;
1594 out:
1595 mutex_unlock(&kernfs_mutex);
1596 return error;
1597}
1598
1599/* Relationship between s_mode and the DT_xxx types */
1600static inline unsigned char dt_type(struct kernfs_node *kn)
1601{
1602 return (kn->mode >> 12) & 15;
1603}
1604
1605static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1606{
1607 kernfs_put(filp->private_data);
1608 return 0;
1609}
1610
1611static struct kernfs_node *kernfs_dir_pos(const void *ns,
1612 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1613{
1614 if (pos) {
1615 int valid = kernfs_active(pos) &&
1616 pos->parent == parent && hash == pos->hash;
1617 kernfs_put(pos);
1618 if (!valid)
1619 pos = NULL;
1620 }
1621 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1622 struct rb_node *node = parent->dir.children.rb_node;
1623 while (node) {
1624 pos = rb_to_kn(node);
1625
1626 if (hash < pos->hash)
1627 node = node->rb_left;
1628 else if (hash > pos->hash)
1629 node = node->rb_right;
1630 else
1631 break;
1632 }
1633 }
1634 /* Skip over entries which are dying/dead or in the wrong namespace */
1635 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1636 struct rb_node *node = rb_next(&pos->rb);
1637 if (!node)
1638 pos = NULL;
1639 else
1640 pos = rb_to_kn(node);
1641 }
1642 return pos;
1643}
1644
1645static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1646 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1647{
1648 pos = kernfs_dir_pos(ns, parent, ino, pos);
1649 if (pos) {
1650 do {
1651 struct rb_node *node = rb_next(&pos->rb);
1652 if (!node)
1653 pos = NULL;
1654 else
1655 pos = rb_to_kn(node);
1656 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1657 }
1658 return pos;
1659}
1660
1661static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1662{
1663 struct dentry *dentry = file->f_path.dentry;
1664 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1665 struct kernfs_node *pos = file->private_data;
1666 const void *ns = NULL;
1667
1668 if (!dir_emit_dots(file, ctx))
1669 return 0;
1670 mutex_lock(&kernfs_mutex);
1671
1672 if (kernfs_ns_enabled(parent))
1673 ns = kernfs_info(dentry->d_sb)->ns;
1674
1675 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1676 pos;
1677 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1678 const char *name = pos->name;
1679 unsigned int type = dt_type(pos);
1680 int len = strlen(name);
1681 ino_t ino = pos->id.ino;
1682
1683 ctx->pos = pos->hash;
1684 file->private_data = pos;
1685 kernfs_get(pos);
1686
1687 mutex_unlock(&kernfs_mutex);
1688 if (!dir_emit(ctx, name, len, ino, type))
1689 return 0;
1690 mutex_lock(&kernfs_mutex);
1691 }
1692 mutex_unlock(&kernfs_mutex);
1693 file->private_data = NULL;
1694 ctx->pos = INT_MAX;
1695 return 0;
1696}
1697
1698const struct file_operations kernfs_dir_fops = {
1699 .read = generic_read_dir,
1700 .iterate_shared = kernfs_fop_readdir,
1701 .release = kernfs_dir_fop_release,
1702 .llseek = generic_file_llseek,
1703};
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
20static 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 */
28static DEFINE_SPINLOCK(kernfs_pr_cont_lock);
29static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by pr_cont_lock */
30static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
31
32#define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
33
34static bool __kernfs_active(struct kernfs_node *kn)
35{
36 return atomic_read(&kn->active) >= 0;
37}
38
39static bool kernfs_active(struct kernfs_node *kn)
40{
41 lockdep_assert_held(&kernfs_root(kn)->kernfs_rwsem);
42 return __kernfs_active(kn);
43}
44
45static 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
54static 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 */
63static 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
74static 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 * 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 */
134static 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().
189 *
190 * Fills buffer with "(null)" if @kn is %NULL.
191 *
192 * Return: the length of @kn's name and if @buf isn't long enough,
193 * it's filled up to @buflen-1 and nul terminated.
194 *
195 * This function can be called from any context.
196 */
197int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
198{
199 unsigned long flags;
200 int ret;
201
202 spin_lock_irqsave(&kernfs_rename_lock, flags);
203 ret = kernfs_name_locked(kn, buf, buflen);
204 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
205 return ret;
206}
207
208/**
209 * kernfs_path_from_node - build path of node @to relative to @from.
210 * @from: parent kernfs_node relative to which we need to build the path
211 * @to: kernfs_node of interest
212 * @buf: buffer to copy @to's path into
213 * @buflen: size of @buf
214 *
215 * Builds @to's path relative to @from in @buf. @from and @to must
216 * be on the same kernfs-root. If @from is not parent of @to, then a relative
217 * path (which includes '..'s) as needed to reach from @from to @to is
218 * returned.
219 *
220 * Return: the length of the full path. If the full length is equal to or
221 * greater than @buflen, @buf contains the truncated path with the trailing
222 * '\0'. On error, -errno is returned.
223 */
224int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
225 char *buf, size_t buflen)
226{
227 unsigned long flags;
228 int ret;
229
230 spin_lock_irqsave(&kernfs_rename_lock, flags);
231 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
232 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
233 return ret;
234}
235EXPORT_SYMBOL_GPL(kernfs_path_from_node);
236
237/**
238 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
239 * @kn: kernfs_node of interest
240 *
241 * This function can be called from any context.
242 */
243void pr_cont_kernfs_name(struct kernfs_node *kn)
244{
245 unsigned long flags;
246
247 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
248
249 kernfs_name(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
250 pr_cont("%s", kernfs_pr_cont_buf);
251
252 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
253}
254
255/**
256 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
257 * @kn: kernfs_node of interest
258 *
259 * This function can be called from any context.
260 */
261void pr_cont_kernfs_path(struct kernfs_node *kn)
262{
263 unsigned long flags;
264 int sz;
265
266 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
267
268 sz = kernfs_path_from_node(kn, NULL, kernfs_pr_cont_buf,
269 sizeof(kernfs_pr_cont_buf));
270 if (sz < 0) {
271 pr_cont("(error)");
272 goto out;
273 }
274
275 if (sz >= sizeof(kernfs_pr_cont_buf)) {
276 pr_cont("(name too long)");
277 goto out;
278 }
279
280 pr_cont("%s", kernfs_pr_cont_buf);
281
282out:
283 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
284}
285
286/**
287 * kernfs_get_parent - determine the parent node and pin it
288 * @kn: kernfs_node of interest
289 *
290 * Determines @kn's parent, pins and returns it. This function can be
291 * called from any context.
292 *
293 * Return: parent node of @kn
294 */
295struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
296{
297 struct kernfs_node *parent;
298 unsigned long flags;
299
300 spin_lock_irqsave(&kernfs_rename_lock, flags);
301 parent = kn->parent;
302 kernfs_get(parent);
303 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
304
305 return parent;
306}
307
308/**
309 * kernfs_name_hash - calculate hash of @ns + @name
310 * @name: Null terminated string to hash
311 * @ns: Namespace tag to hash
312 *
313 * Return: 31-bit hash of ns + name (so it fits in an off_t)
314 */
315static unsigned int kernfs_name_hash(const char *name, const void *ns)
316{
317 unsigned long hash = init_name_hash(ns);
318 unsigned int len = strlen(name);
319 while (len--)
320 hash = partial_name_hash(*name++, hash);
321 hash = end_name_hash(hash);
322 hash &= 0x7fffffffU;
323 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
324 if (hash < 2)
325 hash += 2;
326 if (hash >= INT_MAX)
327 hash = INT_MAX - 1;
328 return hash;
329}
330
331static int kernfs_name_compare(unsigned int hash, const char *name,
332 const void *ns, const struct kernfs_node *kn)
333{
334 if (hash < kn->hash)
335 return -1;
336 if (hash > kn->hash)
337 return 1;
338 if (ns < kn->ns)
339 return -1;
340 if (ns > kn->ns)
341 return 1;
342 return strcmp(name, kn->name);
343}
344
345static int kernfs_sd_compare(const struct kernfs_node *left,
346 const struct kernfs_node *right)
347{
348 return kernfs_name_compare(left->hash, left->name, left->ns, right);
349}
350
351/**
352 * kernfs_link_sibling - link kernfs_node into sibling rbtree
353 * @kn: kernfs_node of interest
354 *
355 * Link @kn into its sibling rbtree which starts from
356 * @kn->parent->dir.children.
357 *
358 * Locking:
359 * kernfs_rwsem held exclusive
360 *
361 * Return:
362 * %0 on success, -EEXIST on failure.
363 */
364static int kernfs_link_sibling(struct kernfs_node *kn)
365{
366 struct rb_node **node = &kn->parent->dir.children.rb_node;
367 struct rb_node *parent = NULL;
368
369 while (*node) {
370 struct kernfs_node *pos;
371 int result;
372
373 pos = rb_to_kn(*node);
374 parent = *node;
375 result = kernfs_sd_compare(kn, pos);
376 if (result < 0)
377 node = &pos->rb.rb_left;
378 else if (result > 0)
379 node = &pos->rb.rb_right;
380 else
381 return -EEXIST;
382 }
383
384 /* add new node and rebalance the tree */
385 rb_link_node(&kn->rb, parent, node);
386 rb_insert_color(&kn->rb, &kn->parent->dir.children);
387
388 /* successfully added, account subdir number */
389 if (kernfs_type(kn) == KERNFS_DIR)
390 kn->parent->dir.subdirs++;
391 kernfs_inc_rev(kn->parent);
392
393 return 0;
394}
395
396/**
397 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
398 * @kn: kernfs_node of interest
399 *
400 * Try to unlink @kn from its sibling rbtree which starts from
401 * kn->parent->dir.children.
402 *
403 * Return: %true if @kn was actually removed,
404 * %false if @kn wasn't on the rbtree.
405 *
406 * Locking:
407 * kernfs_rwsem held exclusive
408 */
409static bool kernfs_unlink_sibling(struct kernfs_node *kn)
410{
411 if (RB_EMPTY_NODE(&kn->rb))
412 return false;
413
414 if (kernfs_type(kn) == KERNFS_DIR)
415 kn->parent->dir.subdirs--;
416 kernfs_inc_rev(kn->parent);
417
418 rb_erase(&kn->rb, &kn->parent->dir.children);
419 RB_CLEAR_NODE(&kn->rb);
420 return true;
421}
422
423/**
424 * kernfs_get_active - get an active reference to kernfs_node
425 * @kn: kernfs_node to get an active reference to
426 *
427 * Get an active reference of @kn. This function is noop if @kn
428 * is %NULL.
429 *
430 * Return:
431 * Pointer to @kn on success, %NULL on failure.
432 */
433struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
434{
435 if (unlikely(!kn))
436 return NULL;
437
438 if (!atomic_inc_unless_negative(&kn->active))
439 return NULL;
440
441 if (kernfs_lockdep(kn))
442 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
443 return kn;
444}
445
446/**
447 * kernfs_put_active - put an active reference to kernfs_node
448 * @kn: kernfs_node to put an active reference to
449 *
450 * Put an active reference to @kn. This function is noop if @kn
451 * is %NULL.
452 */
453void kernfs_put_active(struct kernfs_node *kn)
454{
455 int v;
456
457 if (unlikely(!kn))
458 return;
459
460 if (kernfs_lockdep(kn))
461 rwsem_release(&kn->dep_map, _RET_IP_);
462 v = atomic_dec_return(&kn->active);
463 if (likely(v != KN_DEACTIVATED_BIAS))
464 return;
465
466 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
467}
468
469/**
470 * kernfs_drain - drain kernfs_node
471 * @kn: kernfs_node to drain
472 *
473 * Drain existing usages and nuke all existing mmaps of @kn. Multiple
474 * removers may invoke this function concurrently on @kn and all will
475 * return after draining is complete.
476 */
477static void kernfs_drain(struct kernfs_node *kn)
478 __releases(&kernfs_root(kn)->kernfs_rwsem)
479 __acquires(&kernfs_root(kn)->kernfs_rwsem)
480{
481 struct kernfs_root *root = kernfs_root(kn);
482
483 lockdep_assert_held_write(&root->kernfs_rwsem);
484 WARN_ON_ONCE(kernfs_active(kn));
485
486 /*
487 * Skip draining if already fully drained. This avoids draining and its
488 * lockdep annotations for nodes which have never been activated
489 * allowing embedding kernfs_remove() in create error paths without
490 * worrying about draining.
491 */
492 if (atomic_read(&kn->active) == KN_DEACTIVATED_BIAS &&
493 !kernfs_should_drain_open_files(kn))
494 return;
495
496 up_write(&root->kernfs_rwsem);
497
498 if (kernfs_lockdep(kn)) {
499 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
500 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
501 lock_contended(&kn->dep_map, _RET_IP_);
502 }
503
504 wait_event(root->deactivate_waitq,
505 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
506
507 if (kernfs_lockdep(kn)) {
508 lock_acquired(&kn->dep_map, _RET_IP_);
509 rwsem_release(&kn->dep_map, _RET_IP_);
510 }
511
512 if (kernfs_should_drain_open_files(kn))
513 kernfs_drain_open_files(kn);
514
515 down_write(&root->kernfs_rwsem);
516}
517
518/**
519 * kernfs_get - get a reference count on a kernfs_node
520 * @kn: the target kernfs_node
521 */
522void kernfs_get(struct kernfs_node *kn)
523{
524 if (kn) {
525 WARN_ON(!atomic_read(&kn->count));
526 atomic_inc(&kn->count);
527 }
528}
529EXPORT_SYMBOL_GPL(kernfs_get);
530
531/**
532 * kernfs_put - put a reference count on a kernfs_node
533 * @kn: the target kernfs_node
534 *
535 * Put a reference count of @kn and destroy it if it reached zero.
536 */
537void kernfs_put(struct kernfs_node *kn)
538{
539 struct kernfs_node *parent;
540 struct kernfs_root *root;
541
542 if (!kn || !atomic_dec_and_test(&kn->count))
543 return;
544 root = kernfs_root(kn);
545 repeat:
546 /*
547 * Moving/renaming is always done while holding reference.
548 * kn->parent won't change beneath us.
549 */
550 parent = kn->parent;
551
552 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
553 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
554 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
555
556 if (kernfs_type(kn) == KERNFS_LINK)
557 kernfs_put(kn->symlink.target_kn);
558
559 kfree_const(kn->name);
560
561 if (kn->iattr) {
562 simple_xattrs_free(&kn->iattr->xattrs);
563 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
564 }
565 spin_lock(&kernfs_idr_lock);
566 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
567 spin_unlock(&kernfs_idr_lock);
568 kmem_cache_free(kernfs_node_cache, kn);
569
570 kn = parent;
571 if (kn) {
572 if (atomic_dec_and_test(&kn->count))
573 goto repeat;
574 } else {
575 /* just released the root kn, free @root too */
576 idr_destroy(&root->ino_idr);
577 kfree(root);
578 }
579}
580EXPORT_SYMBOL_GPL(kernfs_put);
581
582/**
583 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
584 * @dentry: the dentry in question
585 *
586 * Return: the kernfs_node associated with @dentry. If @dentry is not a
587 * kernfs one, %NULL is returned.
588 *
589 * While the returned kernfs_node will stay accessible as long as @dentry
590 * is accessible, the returned node can be in any state and the caller is
591 * fully responsible for determining what's accessible.
592 */
593struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
594{
595 if (dentry->d_sb->s_op == &kernfs_sops)
596 return kernfs_dentry_node(dentry);
597 return NULL;
598}
599
600static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
601 struct kernfs_node *parent,
602 const char *name, umode_t mode,
603 kuid_t uid, kgid_t gid,
604 unsigned flags)
605{
606 struct kernfs_node *kn;
607 u32 id_highbits;
608 int ret;
609
610 name = kstrdup_const(name, GFP_KERNEL);
611 if (!name)
612 return NULL;
613
614 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
615 if (!kn)
616 goto err_out1;
617
618 idr_preload(GFP_KERNEL);
619 spin_lock(&kernfs_idr_lock);
620 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
621 if (ret >= 0 && ret < root->last_id_lowbits)
622 root->id_highbits++;
623 id_highbits = root->id_highbits;
624 root->last_id_lowbits = ret;
625 spin_unlock(&kernfs_idr_lock);
626 idr_preload_end();
627 if (ret < 0)
628 goto err_out2;
629
630 kn->id = (u64)id_highbits << 32 | ret;
631
632 atomic_set(&kn->count, 1);
633 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
634 RB_CLEAR_NODE(&kn->rb);
635
636 kn->name = name;
637 kn->mode = mode;
638 kn->flags = flags;
639
640 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
641 struct iattr iattr = {
642 .ia_valid = ATTR_UID | ATTR_GID,
643 .ia_uid = uid,
644 .ia_gid = gid,
645 };
646
647 ret = __kernfs_setattr(kn, &iattr);
648 if (ret < 0)
649 goto err_out3;
650 }
651
652 if (parent) {
653 ret = security_kernfs_init_security(parent, kn);
654 if (ret)
655 goto err_out3;
656 }
657
658 return kn;
659
660 err_out3:
661 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
662 err_out2:
663 kmem_cache_free(kernfs_node_cache, kn);
664 err_out1:
665 kfree_const(name);
666 return NULL;
667}
668
669struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
670 const char *name, umode_t mode,
671 kuid_t uid, kgid_t gid,
672 unsigned flags)
673{
674 struct kernfs_node *kn;
675
676 kn = __kernfs_new_node(kernfs_root(parent), parent,
677 name, mode, uid, gid, flags);
678 if (kn) {
679 kernfs_get(parent);
680 kn->parent = parent;
681 }
682 return kn;
683}
684
685/*
686 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
687 * @root: the kernfs root
688 * @id: the target node id
689 *
690 * @id's lower 32bits encode ino and upper gen. If the gen portion is
691 * zero, all generations are matched.
692 *
693 * Return: %NULL on failure,
694 * otherwise a kernfs node with reference counter incremented.
695 */
696struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
697 u64 id)
698{
699 struct kernfs_node *kn;
700 ino_t ino = kernfs_id_ino(id);
701 u32 gen = kernfs_id_gen(id);
702
703 spin_lock(&kernfs_idr_lock);
704
705 kn = idr_find(&root->ino_idr, (u32)ino);
706 if (!kn)
707 goto err_unlock;
708
709 if (sizeof(ino_t) >= sizeof(u64)) {
710 /* we looked up with the low 32bits, compare the whole */
711 if (kernfs_ino(kn) != ino)
712 goto err_unlock;
713 } else {
714 /* 0 matches all generations */
715 if (unlikely(gen && kernfs_gen(kn) != gen))
716 goto err_unlock;
717 }
718
719 /*
720 * We should fail if @kn has never been activated and guarantee success
721 * if the caller knows that @kn is active. Both can be achieved by
722 * __kernfs_active() which tests @kn->active without kernfs_rwsem.
723 */
724 if (unlikely(!__kernfs_active(kn) || !atomic_inc_not_zero(&kn->count)))
725 goto err_unlock;
726
727 spin_unlock(&kernfs_idr_lock);
728 return kn;
729err_unlock:
730 spin_unlock(&kernfs_idr_lock);
731 return NULL;
732}
733
734/**
735 * kernfs_add_one - add kernfs_node to parent without warning
736 * @kn: kernfs_node to be added
737 *
738 * The caller must already have initialized @kn->parent. This
739 * function increments nlink of the parent's inode if @kn is a
740 * directory and link into the children list of the parent.
741 *
742 * Return:
743 * %0 on success, -EEXIST if entry with the given name already
744 * exists.
745 */
746int kernfs_add_one(struct kernfs_node *kn)
747{
748 struct kernfs_node *parent = kn->parent;
749 struct kernfs_root *root = kernfs_root(parent);
750 struct kernfs_iattrs *ps_iattr;
751 bool has_ns;
752 int ret;
753
754 down_write(&root->kernfs_rwsem);
755
756 ret = -EINVAL;
757 has_ns = kernfs_ns_enabled(parent);
758 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
759 has_ns ? "required" : "invalid", parent->name, kn->name))
760 goto out_unlock;
761
762 if (kernfs_type(parent) != KERNFS_DIR)
763 goto out_unlock;
764
765 ret = -ENOENT;
766 if (parent->flags & (KERNFS_REMOVING | KERNFS_EMPTY_DIR))
767 goto out_unlock;
768
769 kn->hash = kernfs_name_hash(kn->name, kn->ns);
770
771 ret = kernfs_link_sibling(kn);
772 if (ret)
773 goto out_unlock;
774
775 /* Update timestamps on the parent */
776 ps_iattr = parent->iattr;
777 if (ps_iattr) {
778 ktime_get_real_ts64(&ps_iattr->ia_ctime);
779 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
780 }
781
782 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 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
792 kernfs_activate(kn);
793 return 0;
794
795out_unlock:
796 up_write(&root->kernfs_rwsem);
797 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 */
810static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
811 const unsigned char *name,
812 const void *ns)
813{
814 struct rb_node *node = parent->dir.children.rb_node;
815 bool has_ns = kernfs_ns_enabled(parent);
816 unsigned int hash;
817
818 lockdep_assert_held(&kernfs_root(parent)->kernfs_rwsem);
819
820 if (has_ns != (bool)ns) {
821 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
822 has_ns ? "required" : "invalid", parent->name, name);
823 return NULL;
824 }
825
826 hash = kernfs_name_hash(name, ns);
827 while (node) {
828 struct kernfs_node *kn;
829 int result;
830
831 kn = rb_to_kn(node);
832 result = kernfs_name_compare(hash, name, ns, kn);
833 if (result < 0)
834 node = node->rb_left;
835 else if (result > 0)
836 node = node->rb_right;
837 else
838 return kn;
839 }
840 return NULL;
841}
842
843static 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 lockdep_assert_held_read(&kernfs_root(parent)->kernfs_rwsem);
851
852 spin_lock_irq(&kernfs_pr_cont_lock);
853
854 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
855
856 if (len >= sizeof(kernfs_pr_cont_buf)) {
857 spin_unlock_irq(&kernfs_pr_cont_lock);
858 return NULL;
859 }
860
861 p = kernfs_pr_cont_buf;
862
863 while ((name = strsep(&p, "/")) && parent) {
864 if (*name == '\0')
865 continue;
866 parent = kernfs_find_ns(parent, name, ns);
867 }
868
869 spin_unlock_irq(&kernfs_pr_cont_lock);
870
871 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 */
885struct 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 struct kernfs_root *root = kernfs_root(parent);
890
891 down_read(&root->kernfs_rwsem);
892 kn = kernfs_find_ns(parent, name, ns);
893 kernfs_get(kn);
894 up_read(&root->kernfs_rwsem);
895
896 return kn;
897}
898EXPORT_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 */
911struct 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 struct kernfs_root *root = kernfs_root(parent);
916
917 down_read(&root->kernfs_rwsem);
918 kn = kernfs_walk_ns(parent, path, ns);
919 kernfs_get(kn);
920 up_read(&root->kernfs_rwsem);
921
922 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 */
934struct 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 root = kzalloc(sizeof(*root), GFP_KERNEL);
941 if (!root)
942 return ERR_PTR(-ENOMEM);
943
944 idr_init(&root->ino_idr);
945 init_rwsem(&root->kernfs_rwsem);
946 INIT_LIST_HEAD(&root->supers);
947
948 /*
949 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
950 * High bits generation. The starting value for both ino and
951 * genenration is 1. Initialize upper 32bit allocation
952 * accordingly.
953 */
954 if (sizeof(ino_t) >= sizeof(u64))
955 root->id_highbits = 0;
956 else
957 root->id_highbits = 1;
958
959 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
960 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
961 KERNFS_DIR);
962 if (!kn) {
963 idr_destroy(&root->ino_idr);
964 kfree(root);
965 return ERR_PTR(-ENOMEM);
966 }
967
968 kn->priv = priv;
969 kn->dir.root = root;
970
971 root->syscall_ops = scops;
972 root->flags = flags;
973 root->kn = kn;
974 init_waitqueue_head(&root->deactivate_waitq);
975
976 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
977 kernfs_activate(kn);
978
979 return root;
980}
981
982/**
983 * kernfs_destroy_root - destroy a kernfs hierarchy
984 * @root: root of the hierarchy to destroy
985 *
986 * Destroy the hierarchy anchored at @root by removing all existing
987 * directories and destroying @root.
988 */
989void kernfs_destroy_root(struct kernfs_root *root)
990{
991 /*
992 * kernfs_remove holds kernfs_rwsem from the root so the root
993 * shouldn't be freed during the operation.
994 */
995 kernfs_get(root->kn);
996 kernfs_remove(root->kn);
997 kernfs_put(root->kn); /* will also free @root */
998}
999
1000/**
1001 * kernfs_root_to_node - return the kernfs_node associated with a kernfs_root
1002 * @root: root to use to lookup
1003 *
1004 * Return: @root's kernfs_node
1005 */
1006struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root)
1007{
1008 return root->kn;
1009}
1010
1011/**
1012 * kernfs_create_dir_ns - create a directory
1013 * @parent: parent in which to create a new directory
1014 * @name: name of the new directory
1015 * @mode: mode of the new directory
1016 * @uid: uid of the new directory
1017 * @gid: gid of the new directory
1018 * @priv: opaque data associated with the new directory
1019 * @ns: optional namespace tag of the directory
1020 *
1021 * Return: the created node on success, ERR_PTR() value on failure.
1022 */
1023struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1024 const char *name, umode_t mode,
1025 kuid_t uid, kgid_t gid,
1026 void *priv, const void *ns)
1027{
1028 struct kernfs_node *kn;
1029 int rc;
1030
1031 /* allocate */
1032 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1033 uid, gid, KERNFS_DIR);
1034 if (!kn)
1035 return ERR_PTR(-ENOMEM);
1036
1037 kn->dir.root = parent->dir.root;
1038 kn->ns = ns;
1039 kn->priv = priv;
1040
1041 /* link in */
1042 rc = kernfs_add_one(kn);
1043 if (!rc)
1044 return kn;
1045
1046 kernfs_put(kn);
1047 return ERR_PTR(rc);
1048}
1049
1050/**
1051 * kernfs_create_empty_dir - create an always empty directory
1052 * @parent: parent in which to create a new directory
1053 * @name: name of the new directory
1054 *
1055 * Return: the created node on success, ERR_PTR() value on failure.
1056 */
1057struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1058 const char *name)
1059{
1060 struct kernfs_node *kn;
1061 int rc;
1062
1063 /* allocate */
1064 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1065 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1066 if (!kn)
1067 return ERR_PTR(-ENOMEM);
1068
1069 kn->flags |= KERNFS_EMPTY_DIR;
1070 kn->dir.root = parent->dir.root;
1071 kn->ns = NULL;
1072 kn->priv = NULL;
1073
1074 /* link in */
1075 rc = kernfs_add_one(kn);
1076 if (!rc)
1077 return kn;
1078
1079 kernfs_put(kn);
1080 return ERR_PTR(rc);
1081}
1082
1083static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1084{
1085 struct kernfs_node *kn;
1086 struct kernfs_root *root;
1087
1088 if (flags & LOOKUP_RCU)
1089 return -ECHILD;
1090
1091 /* Negative hashed dentry? */
1092 if (d_really_is_negative(dentry)) {
1093 struct kernfs_node *parent;
1094
1095 /* If the kernfs parent node has changed discard and
1096 * proceed to ->lookup.
1097 *
1098 * There's nothing special needed here when getting the
1099 * dentry parent, even if a concurrent rename is in
1100 * progress. That's because the dentry is negative so
1101 * it can only be the target of the rename and it will
1102 * be doing a d_move() not a replace. Consequently the
1103 * dentry d_parent won't change over the d_move().
1104 *
1105 * Also kernfs negative dentries transitioning from
1106 * negative to positive during revalidate won't happen
1107 * because they are invalidated on containing directory
1108 * changes and the lookup re-done so that a new positive
1109 * dentry can be properly created.
1110 */
1111 root = kernfs_root_from_sb(dentry->d_sb);
1112 down_read(&root->kernfs_rwsem);
1113 parent = kernfs_dentry_node(dentry->d_parent);
1114 if (parent) {
1115 if (kernfs_dir_changed(parent, dentry)) {
1116 up_read(&root->kernfs_rwsem);
1117 return 0;
1118 }
1119 }
1120 up_read(&root->kernfs_rwsem);
1121
1122 /* The kernfs parent node hasn't changed, leave the
1123 * dentry negative and return success.
1124 */
1125 return 1;
1126 }
1127
1128 kn = kernfs_dentry_node(dentry);
1129 root = kernfs_root(kn);
1130 down_read(&root->kernfs_rwsem);
1131
1132 /* The kernfs node has been deactivated */
1133 if (!kernfs_active(kn))
1134 goto out_bad;
1135
1136 /* The kernfs node has been moved? */
1137 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1138 goto out_bad;
1139
1140 /* The kernfs node has been renamed */
1141 if (strcmp(dentry->d_name.name, kn->name) != 0)
1142 goto out_bad;
1143
1144 /* The kernfs node has been moved to a different namespace */
1145 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1146 kernfs_info(dentry->d_sb)->ns != kn->ns)
1147 goto out_bad;
1148
1149 up_read(&root->kernfs_rwsem);
1150 return 1;
1151out_bad:
1152 up_read(&root->kernfs_rwsem);
1153 return 0;
1154}
1155
1156const struct dentry_operations kernfs_dops = {
1157 .d_revalidate = kernfs_dop_revalidate,
1158};
1159
1160static struct dentry *kernfs_iop_lookup(struct inode *dir,
1161 struct dentry *dentry,
1162 unsigned int flags)
1163{
1164 struct kernfs_node *parent = dir->i_private;
1165 struct kernfs_node *kn;
1166 struct kernfs_root *root;
1167 struct inode *inode = NULL;
1168 const void *ns = NULL;
1169
1170 root = kernfs_root(parent);
1171 down_read(&root->kernfs_rwsem);
1172 if (kernfs_ns_enabled(parent))
1173 ns = kernfs_info(dir->i_sb)->ns;
1174
1175 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1176 /* attach dentry and inode */
1177 if (kn) {
1178 /* Inactive nodes are invisible to the VFS so don't
1179 * create a negative.
1180 */
1181 if (!kernfs_active(kn)) {
1182 up_read(&root->kernfs_rwsem);
1183 return NULL;
1184 }
1185 inode = kernfs_get_inode(dir->i_sb, kn);
1186 if (!inode)
1187 inode = ERR_PTR(-ENOMEM);
1188 }
1189 /*
1190 * Needed for negative dentry validation.
1191 * The negative dentry can be created in kernfs_iop_lookup()
1192 * or transforms from positive dentry in dentry_unlink_inode()
1193 * called from vfs_rmdir().
1194 */
1195 if (!IS_ERR(inode))
1196 kernfs_set_rev(parent, dentry);
1197 up_read(&root->kernfs_rwsem);
1198
1199 /* instantiate and hash (possibly negative) dentry */
1200 return d_splice_alias(inode, dentry);
1201}
1202
1203static int kernfs_iop_mkdir(struct user_namespace *mnt_userns,
1204 struct inode *dir, struct dentry *dentry,
1205 umode_t mode)
1206{
1207 struct kernfs_node *parent = dir->i_private;
1208 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1209 int ret;
1210
1211 if (!scops || !scops->mkdir)
1212 return -EPERM;
1213
1214 if (!kernfs_get_active(parent))
1215 return -ENODEV;
1216
1217 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1218
1219 kernfs_put_active(parent);
1220 return ret;
1221}
1222
1223static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1224{
1225 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1226 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1227 int ret;
1228
1229 if (!scops || !scops->rmdir)
1230 return -EPERM;
1231
1232 if (!kernfs_get_active(kn))
1233 return -ENODEV;
1234
1235 ret = scops->rmdir(kn);
1236
1237 kernfs_put_active(kn);
1238 return ret;
1239}
1240
1241static int kernfs_iop_rename(struct user_namespace *mnt_userns,
1242 struct inode *old_dir, struct dentry *old_dentry,
1243 struct inode *new_dir, struct dentry *new_dentry,
1244 unsigned int flags)
1245{
1246 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1247 struct kernfs_node *new_parent = new_dir->i_private;
1248 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1249 int ret;
1250
1251 if (flags)
1252 return -EINVAL;
1253
1254 if (!scops || !scops->rename)
1255 return -EPERM;
1256
1257 if (!kernfs_get_active(kn))
1258 return -ENODEV;
1259
1260 if (!kernfs_get_active(new_parent)) {
1261 kernfs_put_active(kn);
1262 return -ENODEV;
1263 }
1264
1265 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1266
1267 kernfs_put_active(new_parent);
1268 kernfs_put_active(kn);
1269 return ret;
1270}
1271
1272const struct inode_operations kernfs_dir_iops = {
1273 .lookup = kernfs_iop_lookup,
1274 .permission = kernfs_iop_permission,
1275 .setattr = kernfs_iop_setattr,
1276 .getattr = kernfs_iop_getattr,
1277 .listxattr = kernfs_iop_listxattr,
1278
1279 .mkdir = kernfs_iop_mkdir,
1280 .rmdir = kernfs_iop_rmdir,
1281 .rename = kernfs_iop_rename,
1282};
1283
1284static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1285{
1286 struct kernfs_node *last;
1287
1288 while (true) {
1289 struct rb_node *rbn;
1290
1291 last = pos;
1292
1293 if (kernfs_type(pos) != KERNFS_DIR)
1294 break;
1295
1296 rbn = rb_first(&pos->dir.children);
1297 if (!rbn)
1298 break;
1299
1300 pos = rb_to_kn(rbn);
1301 }
1302
1303 return last;
1304}
1305
1306/**
1307 * kernfs_next_descendant_post - find the next descendant for post-order walk
1308 * @pos: the current position (%NULL to initiate traversal)
1309 * @root: kernfs_node whose descendants to walk
1310 *
1311 * Find the next descendant to visit for post-order traversal of @root's
1312 * descendants. @root is included in the iteration and the last node to be
1313 * visited.
1314 *
1315 * Return: the next descendant to visit or %NULL when done.
1316 */
1317static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1318 struct kernfs_node *root)
1319{
1320 struct rb_node *rbn;
1321
1322 lockdep_assert_held_write(&kernfs_root(root)->kernfs_rwsem);
1323
1324 /* if first iteration, visit leftmost descendant which may be root */
1325 if (!pos)
1326 return kernfs_leftmost_descendant(root);
1327
1328 /* if we visited @root, we're done */
1329 if (pos == root)
1330 return NULL;
1331
1332 /* if there's an unvisited sibling, visit its leftmost descendant */
1333 rbn = rb_next(&pos->rb);
1334 if (rbn)
1335 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1336
1337 /* no sibling left, visit parent */
1338 return pos->parent;
1339}
1340
1341static void kernfs_activate_one(struct kernfs_node *kn)
1342{
1343 lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1344
1345 kn->flags |= KERNFS_ACTIVATED;
1346
1347 if (kernfs_active(kn) || (kn->flags & (KERNFS_HIDDEN | KERNFS_REMOVING)))
1348 return;
1349
1350 WARN_ON_ONCE(kn->parent && RB_EMPTY_NODE(&kn->rb));
1351 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1352
1353 atomic_sub(KN_DEACTIVATED_BIAS, &kn->active);
1354}
1355
1356/**
1357 * kernfs_activate - activate a node which started deactivated
1358 * @kn: kernfs_node whose subtree is to be activated
1359 *
1360 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1361 * needs to be explicitly activated. A node which hasn't been activated
1362 * isn't visible to userland and deactivation is skipped during its
1363 * removal. This is useful to construct atomic init sequences where
1364 * creation of multiple nodes should either succeed or fail atomically.
1365 *
1366 * The caller is responsible for ensuring that this function is not called
1367 * after kernfs_remove*() is invoked on @kn.
1368 */
1369void kernfs_activate(struct kernfs_node *kn)
1370{
1371 struct kernfs_node *pos;
1372 struct kernfs_root *root = kernfs_root(kn);
1373
1374 down_write(&root->kernfs_rwsem);
1375
1376 pos = NULL;
1377 while ((pos = kernfs_next_descendant_post(pos, kn)))
1378 kernfs_activate_one(pos);
1379
1380 up_write(&root->kernfs_rwsem);
1381}
1382
1383/**
1384 * kernfs_show - show or hide a node
1385 * @kn: kernfs_node to show or hide
1386 * @show: whether to show or hide
1387 *
1388 * If @show is %false, @kn is marked hidden and deactivated. A hidden node is
1389 * ignored in future activaitons. If %true, the mark is removed and activation
1390 * state is restored. This function won't implicitly activate a new node in a
1391 * %KERNFS_ROOT_CREATE_DEACTIVATED root which hasn't been activated yet.
1392 *
1393 * To avoid recursion complexities, directories aren't supported for now.
1394 */
1395void kernfs_show(struct kernfs_node *kn, bool show)
1396{
1397 struct kernfs_root *root = kernfs_root(kn);
1398
1399 if (WARN_ON_ONCE(kernfs_type(kn) == KERNFS_DIR))
1400 return;
1401
1402 down_write(&root->kernfs_rwsem);
1403
1404 if (show) {
1405 kn->flags &= ~KERNFS_HIDDEN;
1406 if (kn->flags & KERNFS_ACTIVATED)
1407 kernfs_activate_one(kn);
1408 } else {
1409 kn->flags |= KERNFS_HIDDEN;
1410 if (kernfs_active(kn))
1411 atomic_add(KN_DEACTIVATED_BIAS, &kn->active);
1412 kernfs_drain(kn);
1413 }
1414
1415 up_write(&root->kernfs_rwsem);
1416}
1417
1418static void __kernfs_remove(struct kernfs_node *kn)
1419{
1420 struct kernfs_node *pos;
1421
1422 /* Short-circuit if non-root @kn has already finished removal. */
1423 if (!kn)
1424 return;
1425
1426 lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1427
1428 /*
1429 * This is for kernfs_remove_self() which plays with active ref
1430 * after removal.
1431 */
1432 if (kn->parent && RB_EMPTY_NODE(&kn->rb))
1433 return;
1434
1435 pr_debug("kernfs %s: removing\n", kn->name);
1436
1437 /* prevent new usage by marking all nodes removing and deactivating */
1438 pos = NULL;
1439 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1440 pos->flags |= KERNFS_REMOVING;
1441 if (kernfs_active(pos))
1442 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1443 }
1444
1445 /* deactivate and unlink the subtree node-by-node */
1446 do {
1447 pos = kernfs_leftmost_descendant(kn);
1448
1449 /*
1450 * kernfs_drain() may drop kernfs_rwsem temporarily and @pos's
1451 * base ref could have been put by someone else by the time
1452 * the function returns. Make sure it doesn't go away
1453 * underneath us.
1454 */
1455 kernfs_get(pos);
1456
1457 kernfs_drain(pos);
1458
1459 /*
1460 * kernfs_unlink_sibling() succeeds once per node. Use it
1461 * to decide who's responsible for cleanups.
1462 */
1463 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1464 struct kernfs_iattrs *ps_iattr =
1465 pos->parent ? pos->parent->iattr : NULL;
1466
1467 /* update timestamps on the parent */
1468 if (ps_iattr) {
1469 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1470 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1471 }
1472
1473 kernfs_put(pos);
1474 }
1475
1476 kernfs_put(pos);
1477 } while (pos != kn);
1478}
1479
1480/**
1481 * kernfs_remove - remove a kernfs_node recursively
1482 * @kn: the kernfs_node to remove
1483 *
1484 * Remove @kn along with all its subdirectories and files.
1485 */
1486void kernfs_remove(struct kernfs_node *kn)
1487{
1488 struct kernfs_root *root;
1489
1490 if (!kn)
1491 return;
1492
1493 root = kernfs_root(kn);
1494
1495 down_write(&root->kernfs_rwsem);
1496 __kernfs_remove(kn);
1497 up_write(&root->kernfs_rwsem);
1498}
1499
1500/**
1501 * kernfs_break_active_protection - break out of active protection
1502 * @kn: the self kernfs_node
1503 *
1504 * The caller must be running off of a kernfs operation which is invoked
1505 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1506 * this function must also be matched with an invocation of
1507 * kernfs_unbreak_active_protection().
1508 *
1509 * This function releases the active reference of @kn the caller is
1510 * holding. Once this function is called, @kn may be removed at any point
1511 * and the caller is solely responsible for ensuring that the objects it
1512 * dereferences are accessible.
1513 */
1514void kernfs_break_active_protection(struct kernfs_node *kn)
1515{
1516 /*
1517 * Take out ourself out of the active ref dependency chain. If
1518 * we're called without an active ref, lockdep will complain.
1519 */
1520 kernfs_put_active(kn);
1521}
1522
1523/**
1524 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1525 * @kn: the self kernfs_node
1526 *
1527 * If kernfs_break_active_protection() was called, this function must be
1528 * invoked before finishing the kernfs operation. Note that while this
1529 * function restores the active reference, it doesn't and can't actually
1530 * restore the active protection - @kn may already or be in the process of
1531 * being removed. Once kernfs_break_active_protection() is invoked, that
1532 * protection is irreversibly gone for the kernfs operation instance.
1533 *
1534 * While this function may be called at any point after
1535 * kernfs_break_active_protection() is invoked, its most useful location
1536 * would be right before the enclosing kernfs operation returns.
1537 */
1538void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1539{
1540 /*
1541 * @kn->active could be in any state; however, the increment we do
1542 * here will be undone as soon as the enclosing kernfs operation
1543 * finishes and this temporary bump can't break anything. If @kn
1544 * is alive, nothing changes. If @kn is being deactivated, the
1545 * soon-to-follow put will either finish deactivation or restore
1546 * deactivated state. If @kn is already removed, the temporary
1547 * bump is guaranteed to be gone before @kn is released.
1548 */
1549 atomic_inc(&kn->active);
1550 if (kernfs_lockdep(kn))
1551 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1552}
1553
1554/**
1555 * kernfs_remove_self - remove a kernfs_node from its own method
1556 * @kn: the self kernfs_node to remove
1557 *
1558 * The caller must be running off of a kernfs operation which is invoked
1559 * with an active reference - e.g. one of kernfs_ops. This can be used to
1560 * implement a file operation which deletes itself.
1561 *
1562 * For example, the "delete" file for a sysfs device directory can be
1563 * implemented by invoking kernfs_remove_self() on the "delete" file
1564 * itself. This function breaks the circular dependency of trying to
1565 * deactivate self while holding an active ref itself. It isn't necessary
1566 * to modify the usual removal path to use kernfs_remove_self(). The
1567 * "delete" implementation can simply invoke kernfs_remove_self() on self
1568 * before proceeding with the usual removal path. kernfs will ignore later
1569 * kernfs_remove() on self.
1570 *
1571 * kernfs_remove_self() can be called multiple times concurrently on the
1572 * same kernfs_node. Only the first one actually performs removal and
1573 * returns %true. All others will wait until the kernfs operation which
1574 * won self-removal finishes and return %false. Note that the losers wait
1575 * for the completion of not only the winning kernfs_remove_self() but also
1576 * the whole kernfs_ops which won the arbitration. This can be used to
1577 * guarantee, for example, all concurrent writes to a "delete" file to
1578 * finish only after the whole operation is complete.
1579 *
1580 * Return: %true if @kn is removed by this call, otherwise %false.
1581 */
1582bool kernfs_remove_self(struct kernfs_node *kn)
1583{
1584 bool ret;
1585 struct kernfs_root *root = kernfs_root(kn);
1586
1587 down_write(&root->kernfs_rwsem);
1588 kernfs_break_active_protection(kn);
1589
1590 /*
1591 * SUICIDAL is used to arbitrate among competing invocations. Only
1592 * the first one will actually perform removal. When the removal
1593 * is complete, SUICIDED is set and the active ref is restored
1594 * while kernfs_rwsem for held exclusive. The ones which lost
1595 * arbitration waits for SUICIDED && drained which can happen only
1596 * after the enclosing kernfs operation which executed the winning
1597 * instance of kernfs_remove_self() finished.
1598 */
1599 if (!(kn->flags & KERNFS_SUICIDAL)) {
1600 kn->flags |= KERNFS_SUICIDAL;
1601 __kernfs_remove(kn);
1602 kn->flags |= KERNFS_SUICIDED;
1603 ret = true;
1604 } else {
1605 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1606 DEFINE_WAIT(wait);
1607
1608 while (true) {
1609 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1610
1611 if ((kn->flags & KERNFS_SUICIDED) &&
1612 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1613 break;
1614
1615 up_write(&root->kernfs_rwsem);
1616 schedule();
1617 down_write(&root->kernfs_rwsem);
1618 }
1619 finish_wait(waitq, &wait);
1620 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1621 ret = false;
1622 }
1623
1624 /*
1625 * This must be done while kernfs_rwsem held exclusive; otherwise,
1626 * waiting for SUICIDED && deactivated could finish prematurely.
1627 */
1628 kernfs_unbreak_active_protection(kn);
1629
1630 up_write(&root->kernfs_rwsem);
1631 return ret;
1632}
1633
1634/**
1635 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1636 * @parent: parent of the target
1637 * @name: name of the kernfs_node to remove
1638 * @ns: namespace tag of the kernfs_node to remove
1639 *
1640 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1641 *
1642 * Return: %0 on success, -ENOENT if such entry doesn't exist.
1643 */
1644int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1645 const void *ns)
1646{
1647 struct kernfs_node *kn;
1648 struct kernfs_root *root;
1649
1650 if (!parent) {
1651 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1652 name);
1653 return -ENOENT;
1654 }
1655
1656 root = kernfs_root(parent);
1657 down_write(&root->kernfs_rwsem);
1658
1659 kn = kernfs_find_ns(parent, name, ns);
1660 if (kn) {
1661 kernfs_get(kn);
1662 __kernfs_remove(kn);
1663 kernfs_put(kn);
1664 }
1665
1666 up_write(&root->kernfs_rwsem);
1667
1668 if (kn)
1669 return 0;
1670 else
1671 return -ENOENT;
1672}
1673
1674/**
1675 * kernfs_rename_ns - move and rename a kernfs_node
1676 * @kn: target node
1677 * @new_parent: new parent to put @sd under
1678 * @new_name: new name
1679 * @new_ns: new namespace tag
1680 *
1681 * Return: %0 on success, -errno on failure.
1682 */
1683int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1684 const char *new_name, const void *new_ns)
1685{
1686 struct kernfs_node *old_parent;
1687 struct kernfs_root *root;
1688 const char *old_name = NULL;
1689 int error;
1690
1691 /* can't move or rename root */
1692 if (!kn->parent)
1693 return -EINVAL;
1694
1695 root = kernfs_root(kn);
1696 down_write(&root->kernfs_rwsem);
1697
1698 error = -ENOENT;
1699 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1700 (new_parent->flags & KERNFS_EMPTY_DIR))
1701 goto out;
1702
1703 error = 0;
1704 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1705 (strcmp(kn->name, new_name) == 0))
1706 goto out; /* nothing to rename */
1707
1708 error = -EEXIST;
1709 if (kernfs_find_ns(new_parent, new_name, new_ns))
1710 goto out;
1711
1712 /* rename kernfs_node */
1713 if (strcmp(kn->name, new_name) != 0) {
1714 error = -ENOMEM;
1715 new_name = kstrdup_const(new_name, GFP_KERNEL);
1716 if (!new_name)
1717 goto out;
1718 } else {
1719 new_name = NULL;
1720 }
1721
1722 /*
1723 * Move to the appropriate place in the appropriate directories rbtree.
1724 */
1725 kernfs_unlink_sibling(kn);
1726 kernfs_get(new_parent);
1727
1728 /* rename_lock protects ->parent and ->name accessors */
1729 spin_lock_irq(&kernfs_rename_lock);
1730
1731 old_parent = kn->parent;
1732 kn->parent = new_parent;
1733
1734 kn->ns = new_ns;
1735 if (new_name) {
1736 old_name = kn->name;
1737 kn->name = new_name;
1738 }
1739
1740 spin_unlock_irq(&kernfs_rename_lock);
1741
1742 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1743 kernfs_link_sibling(kn);
1744
1745 kernfs_put(old_parent);
1746 kfree_const(old_name);
1747
1748 error = 0;
1749 out:
1750 up_write(&root->kernfs_rwsem);
1751 return error;
1752}
1753
1754/* Relationship between mode and the DT_xxx types */
1755static inline unsigned char dt_type(struct kernfs_node *kn)
1756{
1757 return (kn->mode >> 12) & 15;
1758}
1759
1760static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1761{
1762 kernfs_put(filp->private_data);
1763 return 0;
1764}
1765
1766static struct kernfs_node *kernfs_dir_pos(const void *ns,
1767 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1768{
1769 if (pos) {
1770 int valid = kernfs_active(pos) &&
1771 pos->parent == parent && hash == pos->hash;
1772 kernfs_put(pos);
1773 if (!valid)
1774 pos = NULL;
1775 }
1776 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1777 struct rb_node *node = parent->dir.children.rb_node;
1778 while (node) {
1779 pos = rb_to_kn(node);
1780
1781 if (hash < pos->hash)
1782 node = node->rb_left;
1783 else if (hash > pos->hash)
1784 node = node->rb_right;
1785 else
1786 break;
1787 }
1788 }
1789 /* Skip over entries which are dying/dead or in the wrong namespace */
1790 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1791 struct rb_node *node = rb_next(&pos->rb);
1792 if (!node)
1793 pos = NULL;
1794 else
1795 pos = rb_to_kn(node);
1796 }
1797 return pos;
1798}
1799
1800static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1801 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1802{
1803 pos = kernfs_dir_pos(ns, parent, ino, pos);
1804 if (pos) {
1805 do {
1806 struct rb_node *node = rb_next(&pos->rb);
1807 if (!node)
1808 pos = NULL;
1809 else
1810 pos = rb_to_kn(node);
1811 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1812 }
1813 return pos;
1814}
1815
1816static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1817{
1818 struct dentry *dentry = file->f_path.dentry;
1819 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1820 struct kernfs_node *pos = file->private_data;
1821 struct kernfs_root *root;
1822 const void *ns = NULL;
1823
1824 if (!dir_emit_dots(file, ctx))
1825 return 0;
1826
1827 root = kernfs_root(parent);
1828 down_read(&root->kernfs_rwsem);
1829
1830 if (kernfs_ns_enabled(parent))
1831 ns = kernfs_info(dentry->d_sb)->ns;
1832
1833 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1834 pos;
1835 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1836 const char *name = pos->name;
1837 unsigned int type = dt_type(pos);
1838 int len = strlen(name);
1839 ino_t ino = kernfs_ino(pos);
1840
1841 ctx->pos = pos->hash;
1842 file->private_data = pos;
1843 kernfs_get(pos);
1844
1845 up_read(&root->kernfs_rwsem);
1846 if (!dir_emit(ctx, name, len, ino, type))
1847 return 0;
1848 down_read(&root->kernfs_rwsem);
1849 }
1850 up_read(&root->kernfs_rwsem);
1851 file->private_data = NULL;
1852 ctx->pos = INT_MAX;
1853 return 0;
1854}
1855
1856const struct file_operations kernfs_dir_fops = {
1857 .read = generic_read_dir,
1858 .iterate_shared = kernfs_fop_readdir,
1859 .release = kernfs_dir_fop_release,
1860 .llseek = generic_file_llseek,
1861};