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, _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, _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 if (!kn || !atomic_dec_and_test(&kn->count))
512 return;
513 root = kernfs_root(kn);
514 repeat:
515 /*
516 * Moving/renaming is always done while holding reference.
517 * kn->parent won't change beneath us.
518 */
519 parent = kn->parent;
520
521 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
522 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
523 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
524
525 if (kernfs_type(kn) == KERNFS_LINK)
526 kernfs_put(kn->symlink.target_kn);
527
528 kfree_const(kn->name);
529
530 if (kn->iattr) {
531 simple_xattrs_free(&kn->iattr->xattrs);
532 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
533 }
534 spin_lock(&kernfs_idr_lock);
535 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
536 spin_unlock(&kernfs_idr_lock);
537 kmem_cache_free(kernfs_node_cache, kn);
538
539 kn = parent;
540 if (kn) {
541 if (atomic_dec_and_test(&kn->count))
542 goto repeat;
543 } else {
544 /* just released the root kn, free @root too */
545 idr_destroy(&root->ino_idr);
546 kfree(root);
547 }
548}
549EXPORT_SYMBOL_GPL(kernfs_put);
550
551static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
552{
553 struct kernfs_node *kn;
554
555 if (flags & LOOKUP_RCU)
556 return -ECHILD;
557
558 /* Always perform fresh lookup for negatives */
559 if (d_really_is_negative(dentry))
560 goto out_bad_unlocked;
561
562 kn = kernfs_dentry_node(dentry);
563 mutex_lock(&kernfs_mutex);
564
565 /* The kernfs node has been deactivated */
566 if (!kernfs_active(kn))
567 goto out_bad;
568
569 /* The kernfs node has been moved? */
570 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
571 goto out_bad;
572
573 /* The kernfs node has been renamed */
574 if (strcmp(dentry->d_name.name, kn->name) != 0)
575 goto out_bad;
576
577 /* The kernfs node has been moved to a different namespace */
578 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
579 kernfs_info(dentry->d_sb)->ns != kn->ns)
580 goto out_bad;
581
582 mutex_unlock(&kernfs_mutex);
583 return 1;
584out_bad:
585 mutex_unlock(&kernfs_mutex);
586out_bad_unlocked:
587 return 0;
588}
589
590const struct dentry_operations kernfs_dops = {
591 .d_revalidate = kernfs_dop_revalidate,
592};
593
594/**
595 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
596 * @dentry: the dentry in question
597 *
598 * Return the kernfs_node associated with @dentry. If @dentry is not a
599 * kernfs one, %NULL is returned.
600 *
601 * While the returned kernfs_node will stay accessible as long as @dentry
602 * is accessible, the returned node can be in any state and the caller is
603 * fully responsible for determining what's accessible.
604 */
605struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
606{
607 if (dentry->d_sb->s_op == &kernfs_sops &&
608 !d_really_is_negative(dentry))
609 return kernfs_dentry_node(dentry);
610 return NULL;
611}
612
613static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
614 struct kernfs_node *parent,
615 const char *name, umode_t mode,
616 kuid_t uid, kgid_t gid,
617 unsigned flags)
618{
619 struct kernfs_node *kn;
620 u32 id_highbits;
621 int ret;
622
623 name = kstrdup_const(name, GFP_KERNEL);
624 if (!name)
625 return NULL;
626
627 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
628 if (!kn)
629 goto err_out1;
630
631 idr_preload(GFP_KERNEL);
632 spin_lock(&kernfs_idr_lock);
633 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
634 if (ret >= 0 && ret < root->last_id_lowbits)
635 root->id_highbits++;
636 id_highbits = root->id_highbits;
637 root->last_id_lowbits = ret;
638 spin_unlock(&kernfs_idr_lock);
639 idr_preload_end();
640 if (ret < 0)
641 goto err_out2;
642
643 kn->id = (u64)id_highbits << 32 | ret;
644
645 atomic_set(&kn->count, 1);
646 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
647 RB_CLEAR_NODE(&kn->rb);
648
649 kn->name = name;
650 kn->mode = mode;
651 kn->flags = flags;
652
653 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
654 struct iattr iattr = {
655 .ia_valid = ATTR_UID | ATTR_GID,
656 .ia_uid = uid,
657 .ia_gid = gid,
658 };
659
660 ret = __kernfs_setattr(kn, &iattr);
661 if (ret < 0)
662 goto err_out3;
663 }
664
665 if (parent) {
666 ret = security_kernfs_init_security(parent, kn);
667 if (ret)
668 goto err_out3;
669 }
670
671 return kn;
672
673 err_out3:
674 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
675 err_out2:
676 kmem_cache_free(kernfs_node_cache, kn);
677 err_out1:
678 kfree_const(name);
679 return NULL;
680}
681
682struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
683 const char *name, umode_t mode,
684 kuid_t uid, kgid_t gid,
685 unsigned flags)
686{
687 struct kernfs_node *kn;
688
689 kn = __kernfs_new_node(kernfs_root(parent), parent,
690 name, mode, uid, gid, flags);
691 if (kn) {
692 kernfs_get(parent);
693 kn->parent = parent;
694 }
695 return kn;
696}
697
698/*
699 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
700 * @root: the kernfs root
701 * @id: the target node id
702 *
703 * @id's lower 32bits encode ino and upper gen. If the gen portion is
704 * zero, all generations are matched.
705 *
706 * RETURNS:
707 * NULL on failure. Return a kernfs node with reference counter incremented
708 */
709struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
710 u64 id)
711{
712 struct kernfs_node *kn;
713 ino_t ino = kernfs_id_ino(id);
714 u32 gen = kernfs_id_gen(id);
715
716 spin_lock(&kernfs_idr_lock);
717
718 kn = idr_find(&root->ino_idr, (u32)ino);
719 if (!kn)
720 goto err_unlock;
721
722 if (sizeof(ino_t) >= sizeof(u64)) {
723 /* we looked up with the low 32bits, compare the whole */
724 if (kernfs_ino(kn) != ino)
725 goto err_unlock;
726 } else {
727 /* 0 matches all generations */
728 if (unlikely(gen && kernfs_gen(kn) != gen))
729 goto err_unlock;
730 }
731
732 /*
733 * ACTIVATED is protected with kernfs_mutex but it was clear when
734 * @kn was added to idr and we just wanna see it set. No need to
735 * grab kernfs_mutex.
736 */
737 if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
738 !atomic_inc_not_zero(&kn->count)))
739 goto err_unlock;
740
741 spin_unlock(&kernfs_idr_lock);
742 return kn;
743err_unlock:
744 spin_unlock(&kernfs_idr_lock);
745 return NULL;
746}
747
748/**
749 * kernfs_add_one - add kernfs_node to parent without warning
750 * @kn: kernfs_node to be added
751 *
752 * The caller must already have initialized @kn->parent. This
753 * function increments nlink of the parent's inode if @kn is a
754 * directory and link into the children list of the parent.
755 *
756 * RETURNS:
757 * 0 on success, -EEXIST if entry with the given name already
758 * exists.
759 */
760int kernfs_add_one(struct kernfs_node *kn)
761{
762 struct kernfs_node *parent = kn->parent;
763 struct kernfs_iattrs *ps_iattr;
764 bool has_ns;
765 int ret;
766
767 mutex_lock(&kernfs_mutex);
768
769 ret = -EINVAL;
770 has_ns = kernfs_ns_enabled(parent);
771 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
772 has_ns ? "required" : "invalid", parent->name, kn->name))
773 goto out_unlock;
774
775 if (kernfs_type(parent) != KERNFS_DIR)
776 goto out_unlock;
777
778 ret = -ENOENT;
779 if (parent->flags & KERNFS_EMPTY_DIR)
780 goto out_unlock;
781
782 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
783 goto out_unlock;
784
785 kn->hash = kernfs_name_hash(kn->name, kn->ns);
786
787 ret = kernfs_link_sibling(kn);
788 if (ret)
789 goto out_unlock;
790
791 /* Update timestamps on the parent */
792 ps_iattr = parent->iattr;
793 if (ps_iattr) {
794 ktime_get_real_ts64(&ps_iattr->ia_ctime);
795 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
796 }
797
798 mutex_unlock(&kernfs_mutex);
799
800 /*
801 * Activate the new node unless CREATE_DEACTIVATED is requested.
802 * If not activated here, the kernfs user is responsible for
803 * activating the node with kernfs_activate(). A node which hasn't
804 * been activated is not visible to userland and its removal won't
805 * trigger deactivation.
806 */
807 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
808 kernfs_activate(kn);
809 return 0;
810
811out_unlock:
812 mutex_unlock(&kernfs_mutex);
813 return ret;
814}
815
816/**
817 * kernfs_find_ns - find kernfs_node with the given name
818 * @parent: kernfs_node to search under
819 * @name: name to look for
820 * @ns: the namespace tag to use
821 *
822 * Look for kernfs_node with name @name under @parent. Returns pointer to
823 * the found kernfs_node on success, %NULL on failure.
824 */
825static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
826 const unsigned char *name,
827 const void *ns)
828{
829 struct rb_node *node = parent->dir.children.rb_node;
830 bool has_ns = kernfs_ns_enabled(parent);
831 unsigned int hash;
832
833 lockdep_assert_held(&kernfs_mutex);
834
835 if (has_ns != (bool)ns) {
836 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
837 has_ns ? "required" : "invalid", parent->name, name);
838 return NULL;
839 }
840
841 hash = kernfs_name_hash(name, ns);
842 while (node) {
843 struct kernfs_node *kn;
844 int result;
845
846 kn = rb_to_kn(node);
847 result = kernfs_name_compare(hash, name, ns, kn);
848 if (result < 0)
849 node = node->rb_left;
850 else if (result > 0)
851 node = node->rb_right;
852 else
853 return kn;
854 }
855 return NULL;
856}
857
858static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
859 const unsigned char *path,
860 const void *ns)
861{
862 size_t len;
863 char *p, *name;
864
865 lockdep_assert_held(&kernfs_mutex);
866
867 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
868 spin_lock_irq(&kernfs_rename_lock);
869
870 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
871
872 if (len >= sizeof(kernfs_pr_cont_buf)) {
873 spin_unlock_irq(&kernfs_rename_lock);
874 return NULL;
875 }
876
877 p = kernfs_pr_cont_buf;
878
879 while ((name = strsep(&p, "/")) && parent) {
880 if (*name == '\0')
881 continue;
882 parent = kernfs_find_ns(parent, name, ns);
883 }
884
885 spin_unlock_irq(&kernfs_rename_lock);
886
887 return parent;
888}
889
890/**
891 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
892 * @parent: kernfs_node to search under
893 * @name: name to look for
894 * @ns: the namespace tag to use
895 *
896 * Look for kernfs_node with name @name under @parent and get a reference
897 * if found. This function may sleep and returns pointer to the found
898 * kernfs_node on success, %NULL on failure.
899 */
900struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
901 const char *name, const void *ns)
902{
903 struct kernfs_node *kn;
904
905 mutex_lock(&kernfs_mutex);
906 kn = kernfs_find_ns(parent, name, ns);
907 kernfs_get(kn);
908 mutex_unlock(&kernfs_mutex);
909
910 return kn;
911}
912EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
913
914/**
915 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
916 * @parent: kernfs_node to search under
917 * @path: path to look for
918 * @ns: the namespace tag to use
919 *
920 * Look for kernfs_node with path @path under @parent and get a reference
921 * if found. This function may sleep and returns pointer to the found
922 * kernfs_node on success, %NULL on failure.
923 */
924struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
925 const char *path, const void *ns)
926{
927 struct kernfs_node *kn;
928
929 mutex_lock(&kernfs_mutex);
930 kn = kernfs_walk_ns(parent, path, ns);
931 kernfs_get(kn);
932 mutex_unlock(&kernfs_mutex);
933
934 return kn;
935}
936
937/**
938 * kernfs_create_root - create a new kernfs hierarchy
939 * @scops: optional syscall operations for the hierarchy
940 * @flags: KERNFS_ROOT_* flags
941 * @priv: opaque data associated with the new directory
942 *
943 * Returns the root of the new hierarchy on success, ERR_PTR() value on
944 * failure.
945 */
946struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
947 unsigned int flags, void *priv)
948{
949 struct kernfs_root *root;
950 struct kernfs_node *kn;
951
952 root = kzalloc(sizeof(*root), GFP_KERNEL);
953 if (!root)
954 return ERR_PTR(-ENOMEM);
955
956 idr_init(&root->ino_idr);
957 INIT_LIST_HEAD(&root->supers);
958
959 /*
960 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
961 * High bits generation. The starting value for both ino and
962 * genenration is 1. Initialize upper 32bit allocation
963 * accordingly.
964 */
965 if (sizeof(ino_t) >= sizeof(u64))
966 root->id_highbits = 0;
967 else
968 root->id_highbits = 1;
969
970 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
971 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
972 KERNFS_DIR);
973 if (!kn) {
974 idr_destroy(&root->ino_idr);
975 kfree(root);
976 return ERR_PTR(-ENOMEM);
977 }
978
979 kn->priv = priv;
980 kn->dir.root = root;
981
982 root->syscall_ops = scops;
983 root->flags = flags;
984 root->kn = kn;
985 init_waitqueue_head(&root->deactivate_waitq);
986
987 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
988 kernfs_activate(kn);
989
990 return root;
991}
992
993/**
994 * kernfs_destroy_root - destroy a kernfs hierarchy
995 * @root: root of the hierarchy to destroy
996 *
997 * Destroy the hierarchy anchored at @root by removing all existing
998 * directories and destroying @root.
999 */
1000void kernfs_destroy_root(struct kernfs_root *root)
1001{
1002 kernfs_remove(root->kn); /* will also free @root */
1003}
1004
1005/**
1006 * kernfs_create_dir_ns - create a directory
1007 * @parent: parent in which to create a new directory
1008 * @name: name of the new directory
1009 * @mode: mode of the new directory
1010 * @uid: uid of the new directory
1011 * @gid: gid of the new directory
1012 * @priv: opaque data associated with the new directory
1013 * @ns: optional namespace tag of the directory
1014 *
1015 * Returns the created node on success, ERR_PTR() value on failure.
1016 */
1017struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1018 const char *name, umode_t mode,
1019 kuid_t uid, kgid_t gid,
1020 void *priv, const void *ns)
1021{
1022 struct kernfs_node *kn;
1023 int rc;
1024
1025 /* allocate */
1026 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1027 uid, gid, KERNFS_DIR);
1028 if (!kn)
1029 return ERR_PTR(-ENOMEM);
1030
1031 kn->dir.root = parent->dir.root;
1032 kn->ns = ns;
1033 kn->priv = priv;
1034
1035 /* link in */
1036 rc = kernfs_add_one(kn);
1037 if (!rc)
1038 return kn;
1039
1040 kernfs_put(kn);
1041 return ERR_PTR(rc);
1042}
1043
1044/**
1045 * kernfs_create_empty_dir - create an always empty directory
1046 * @parent: parent in which to create a new directory
1047 * @name: name of the new directory
1048 *
1049 * Returns the created node on success, ERR_PTR() value on failure.
1050 */
1051struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1052 const char *name)
1053{
1054 struct kernfs_node *kn;
1055 int rc;
1056
1057 /* allocate */
1058 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1059 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1060 if (!kn)
1061 return ERR_PTR(-ENOMEM);
1062
1063 kn->flags |= KERNFS_EMPTY_DIR;
1064 kn->dir.root = parent->dir.root;
1065 kn->ns = NULL;
1066 kn->priv = NULL;
1067
1068 /* link in */
1069 rc = kernfs_add_one(kn);
1070 if (!rc)
1071 return kn;
1072
1073 kernfs_put(kn);
1074 return ERR_PTR(rc);
1075}
1076
1077static struct dentry *kernfs_iop_lookup(struct inode *dir,
1078 struct dentry *dentry,
1079 unsigned int flags)
1080{
1081 struct dentry *ret;
1082 struct kernfs_node *parent = dir->i_private;
1083 struct kernfs_node *kn;
1084 struct inode *inode;
1085 const void *ns = NULL;
1086
1087 mutex_lock(&kernfs_mutex);
1088
1089 if (kernfs_ns_enabled(parent))
1090 ns = kernfs_info(dir->i_sb)->ns;
1091
1092 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1093
1094 /* no such entry */
1095 if (!kn || !kernfs_active(kn)) {
1096 ret = NULL;
1097 goto out_unlock;
1098 }
1099
1100 /* attach dentry and inode */
1101 inode = kernfs_get_inode(dir->i_sb, kn);
1102 if (!inode) {
1103 ret = ERR_PTR(-ENOMEM);
1104 goto out_unlock;
1105 }
1106
1107 /* instantiate and hash dentry */
1108 ret = d_splice_alias(inode, dentry);
1109 out_unlock:
1110 mutex_unlock(&kernfs_mutex);
1111 return ret;
1112}
1113
1114static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1115 umode_t mode)
1116{
1117 struct kernfs_node *parent = dir->i_private;
1118 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1119 int ret;
1120
1121 if (!scops || !scops->mkdir)
1122 return -EPERM;
1123
1124 if (!kernfs_get_active(parent))
1125 return -ENODEV;
1126
1127 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1128
1129 kernfs_put_active(parent);
1130 return ret;
1131}
1132
1133static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1134{
1135 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1136 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1137 int ret;
1138
1139 if (!scops || !scops->rmdir)
1140 return -EPERM;
1141
1142 if (!kernfs_get_active(kn))
1143 return -ENODEV;
1144
1145 ret = scops->rmdir(kn);
1146
1147 kernfs_put_active(kn);
1148 return ret;
1149}
1150
1151static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1152 struct inode *new_dir, struct dentry *new_dentry,
1153 unsigned int flags)
1154{
1155 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1156 struct kernfs_node *new_parent = new_dir->i_private;
1157 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1158 int ret;
1159
1160 if (flags)
1161 return -EINVAL;
1162
1163 if (!scops || !scops->rename)
1164 return -EPERM;
1165
1166 if (!kernfs_get_active(kn))
1167 return -ENODEV;
1168
1169 if (!kernfs_get_active(new_parent)) {
1170 kernfs_put_active(kn);
1171 return -ENODEV;
1172 }
1173
1174 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1175
1176 kernfs_put_active(new_parent);
1177 kernfs_put_active(kn);
1178 return ret;
1179}
1180
1181const struct inode_operations kernfs_dir_iops = {
1182 .lookup = kernfs_iop_lookup,
1183 .permission = kernfs_iop_permission,
1184 .setattr = kernfs_iop_setattr,
1185 .getattr = kernfs_iop_getattr,
1186 .listxattr = kernfs_iop_listxattr,
1187
1188 .mkdir = kernfs_iop_mkdir,
1189 .rmdir = kernfs_iop_rmdir,
1190 .rename = kernfs_iop_rename,
1191};
1192
1193static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1194{
1195 struct kernfs_node *last;
1196
1197 while (true) {
1198 struct rb_node *rbn;
1199
1200 last = pos;
1201
1202 if (kernfs_type(pos) != KERNFS_DIR)
1203 break;
1204
1205 rbn = rb_first(&pos->dir.children);
1206 if (!rbn)
1207 break;
1208
1209 pos = rb_to_kn(rbn);
1210 }
1211
1212 return last;
1213}
1214
1215/**
1216 * kernfs_next_descendant_post - find the next descendant for post-order walk
1217 * @pos: the current position (%NULL to initiate traversal)
1218 * @root: kernfs_node whose descendants to walk
1219 *
1220 * Find the next descendant to visit for post-order traversal of @root's
1221 * descendants. @root is included in the iteration and the last node to be
1222 * visited.
1223 */
1224static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1225 struct kernfs_node *root)
1226{
1227 struct rb_node *rbn;
1228
1229 lockdep_assert_held(&kernfs_mutex);
1230
1231 /* if first iteration, visit leftmost descendant which may be root */
1232 if (!pos)
1233 return kernfs_leftmost_descendant(root);
1234
1235 /* if we visited @root, we're done */
1236 if (pos == root)
1237 return NULL;
1238
1239 /* if there's an unvisited sibling, visit its leftmost descendant */
1240 rbn = rb_next(&pos->rb);
1241 if (rbn)
1242 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1243
1244 /* no sibling left, visit parent */
1245 return pos->parent;
1246}
1247
1248/**
1249 * kernfs_activate - activate a node which started deactivated
1250 * @kn: kernfs_node whose subtree is to be activated
1251 *
1252 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1253 * needs to be explicitly activated. A node which hasn't been activated
1254 * isn't visible to userland and deactivation is skipped during its
1255 * removal. This is useful to construct atomic init sequences where
1256 * creation of multiple nodes should either succeed or fail atomically.
1257 *
1258 * The caller is responsible for ensuring that this function is not called
1259 * after kernfs_remove*() is invoked on @kn.
1260 */
1261void kernfs_activate(struct kernfs_node *kn)
1262{
1263 struct kernfs_node *pos;
1264
1265 mutex_lock(&kernfs_mutex);
1266
1267 pos = NULL;
1268 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1269 if (pos->flags & KERNFS_ACTIVATED)
1270 continue;
1271
1272 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1273 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1274
1275 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1276 pos->flags |= KERNFS_ACTIVATED;
1277 }
1278
1279 mutex_unlock(&kernfs_mutex);
1280}
1281
1282static void __kernfs_remove(struct kernfs_node *kn)
1283{
1284 struct kernfs_node *pos;
1285
1286 lockdep_assert_held(&kernfs_mutex);
1287
1288 /*
1289 * Short-circuit if non-root @kn has already finished removal.
1290 * This is for kernfs_remove_self() which plays with active ref
1291 * after removal.
1292 */
1293 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1294 return;
1295
1296 pr_debug("kernfs %s: removing\n", kn->name);
1297
1298 /* prevent any new usage under @kn by deactivating all nodes */
1299 pos = NULL;
1300 while ((pos = kernfs_next_descendant_post(pos, kn)))
1301 if (kernfs_active(pos))
1302 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1303
1304 /* deactivate and unlink the subtree node-by-node */
1305 do {
1306 pos = kernfs_leftmost_descendant(kn);
1307
1308 /*
1309 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1310 * base ref could have been put by someone else by the time
1311 * the function returns. Make sure it doesn't go away
1312 * underneath us.
1313 */
1314 kernfs_get(pos);
1315
1316 /*
1317 * Drain iff @kn was activated. This avoids draining and
1318 * its lockdep annotations for nodes which have never been
1319 * activated and allows embedding kernfs_remove() in create
1320 * error paths without worrying about draining.
1321 */
1322 if (kn->flags & KERNFS_ACTIVATED)
1323 kernfs_drain(pos);
1324 else
1325 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1326
1327 /*
1328 * kernfs_unlink_sibling() succeeds once per node. Use it
1329 * to decide who's responsible for cleanups.
1330 */
1331 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1332 struct kernfs_iattrs *ps_iattr =
1333 pos->parent ? pos->parent->iattr : NULL;
1334
1335 /* update timestamps on the parent */
1336 if (ps_iattr) {
1337 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1338 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1339 }
1340
1341 kernfs_put(pos);
1342 }
1343
1344 kernfs_put(pos);
1345 } while (pos != kn);
1346}
1347
1348/**
1349 * kernfs_remove - remove a kernfs_node recursively
1350 * @kn: the kernfs_node to remove
1351 *
1352 * Remove @kn along with all its subdirectories and files.
1353 */
1354void kernfs_remove(struct kernfs_node *kn)
1355{
1356 mutex_lock(&kernfs_mutex);
1357 __kernfs_remove(kn);
1358 mutex_unlock(&kernfs_mutex);
1359}
1360
1361/**
1362 * kernfs_break_active_protection - break out of active protection
1363 * @kn: the self kernfs_node
1364 *
1365 * The caller must be running off of a kernfs operation which is invoked
1366 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1367 * this function must also be matched with an invocation of
1368 * kernfs_unbreak_active_protection().
1369 *
1370 * This function releases the active reference of @kn the caller is
1371 * holding. Once this function is called, @kn may be removed at any point
1372 * and the caller is solely responsible for ensuring that the objects it
1373 * dereferences are accessible.
1374 */
1375void kernfs_break_active_protection(struct kernfs_node *kn)
1376{
1377 /*
1378 * Take out ourself out of the active ref dependency chain. If
1379 * we're called without an active ref, lockdep will complain.
1380 */
1381 kernfs_put_active(kn);
1382}
1383
1384/**
1385 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1386 * @kn: the self kernfs_node
1387 *
1388 * If kernfs_break_active_protection() was called, this function must be
1389 * invoked before finishing the kernfs operation. Note that while this
1390 * function restores the active reference, it doesn't and can't actually
1391 * restore the active protection - @kn may already or be in the process of
1392 * being removed. Once kernfs_break_active_protection() is invoked, that
1393 * protection is irreversibly gone for the kernfs operation instance.
1394 *
1395 * While this function may be called at any point after
1396 * kernfs_break_active_protection() is invoked, its most useful location
1397 * would be right before the enclosing kernfs operation returns.
1398 */
1399void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1400{
1401 /*
1402 * @kn->active could be in any state; however, the increment we do
1403 * here will be undone as soon as the enclosing kernfs operation
1404 * finishes and this temporary bump can't break anything. If @kn
1405 * is alive, nothing changes. If @kn is being deactivated, the
1406 * soon-to-follow put will either finish deactivation or restore
1407 * deactivated state. If @kn is already removed, the temporary
1408 * bump is guaranteed to be gone before @kn is released.
1409 */
1410 atomic_inc(&kn->active);
1411 if (kernfs_lockdep(kn))
1412 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1413}
1414
1415/**
1416 * kernfs_remove_self - remove a kernfs_node from its own method
1417 * @kn: the self kernfs_node to remove
1418 *
1419 * The caller must be running off of a kernfs operation which is invoked
1420 * with an active reference - e.g. one of kernfs_ops. This can be used to
1421 * implement a file operation which deletes itself.
1422 *
1423 * For example, the "delete" file for a sysfs device directory can be
1424 * implemented by invoking kernfs_remove_self() on the "delete" file
1425 * itself. This function breaks the circular dependency of trying to
1426 * deactivate self while holding an active ref itself. It isn't necessary
1427 * to modify the usual removal path to use kernfs_remove_self(). The
1428 * "delete" implementation can simply invoke kernfs_remove_self() on self
1429 * before proceeding with the usual removal path. kernfs will ignore later
1430 * kernfs_remove() on self.
1431 *
1432 * kernfs_remove_self() can be called multiple times concurrently on the
1433 * same kernfs_node. Only the first one actually performs removal and
1434 * returns %true. All others will wait until the kernfs operation which
1435 * won self-removal finishes and return %false. Note that the losers wait
1436 * for the completion of not only the winning kernfs_remove_self() but also
1437 * the whole kernfs_ops which won the arbitration. This can be used to
1438 * guarantee, for example, all concurrent writes to a "delete" file to
1439 * finish only after the whole operation is complete.
1440 */
1441bool kernfs_remove_self(struct kernfs_node *kn)
1442{
1443 bool ret;
1444
1445 mutex_lock(&kernfs_mutex);
1446 kernfs_break_active_protection(kn);
1447
1448 /*
1449 * SUICIDAL is used to arbitrate among competing invocations. Only
1450 * the first one will actually perform removal. When the removal
1451 * is complete, SUICIDED is set and the active ref is restored
1452 * while holding kernfs_mutex. The ones which lost arbitration
1453 * waits for SUICDED && drained which can happen only after the
1454 * enclosing kernfs operation which executed the winning instance
1455 * of kernfs_remove_self() finished.
1456 */
1457 if (!(kn->flags & KERNFS_SUICIDAL)) {
1458 kn->flags |= KERNFS_SUICIDAL;
1459 __kernfs_remove(kn);
1460 kn->flags |= KERNFS_SUICIDED;
1461 ret = true;
1462 } else {
1463 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1464 DEFINE_WAIT(wait);
1465
1466 while (true) {
1467 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1468
1469 if ((kn->flags & KERNFS_SUICIDED) &&
1470 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1471 break;
1472
1473 mutex_unlock(&kernfs_mutex);
1474 schedule();
1475 mutex_lock(&kernfs_mutex);
1476 }
1477 finish_wait(waitq, &wait);
1478 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1479 ret = false;
1480 }
1481
1482 /*
1483 * This must be done while holding kernfs_mutex; otherwise, waiting
1484 * for SUICIDED && deactivated could finish prematurely.
1485 */
1486 kernfs_unbreak_active_protection(kn);
1487
1488 mutex_unlock(&kernfs_mutex);
1489 return ret;
1490}
1491
1492/**
1493 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1494 * @parent: parent of the target
1495 * @name: name of the kernfs_node to remove
1496 * @ns: namespace tag of the kernfs_node to remove
1497 *
1498 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1499 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1500 */
1501int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1502 const void *ns)
1503{
1504 struct kernfs_node *kn;
1505
1506 if (!parent) {
1507 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1508 name);
1509 return -ENOENT;
1510 }
1511
1512 mutex_lock(&kernfs_mutex);
1513
1514 kn = kernfs_find_ns(parent, name, ns);
1515 if (kn)
1516 __kernfs_remove(kn);
1517
1518 mutex_unlock(&kernfs_mutex);
1519
1520 if (kn)
1521 return 0;
1522 else
1523 return -ENOENT;
1524}
1525
1526/**
1527 * kernfs_rename_ns - move and rename a kernfs_node
1528 * @kn: target node
1529 * @new_parent: new parent to put @sd under
1530 * @new_name: new name
1531 * @new_ns: new namespace tag
1532 */
1533int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1534 const char *new_name, const void *new_ns)
1535{
1536 struct kernfs_node *old_parent;
1537 const char *old_name = NULL;
1538 int error;
1539
1540 /* can't move or rename root */
1541 if (!kn->parent)
1542 return -EINVAL;
1543
1544 mutex_lock(&kernfs_mutex);
1545
1546 error = -ENOENT;
1547 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1548 (new_parent->flags & KERNFS_EMPTY_DIR))
1549 goto out;
1550
1551 error = 0;
1552 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1553 (strcmp(kn->name, new_name) == 0))
1554 goto out; /* nothing to rename */
1555
1556 error = -EEXIST;
1557 if (kernfs_find_ns(new_parent, new_name, new_ns))
1558 goto out;
1559
1560 /* rename kernfs_node */
1561 if (strcmp(kn->name, new_name) != 0) {
1562 error = -ENOMEM;
1563 new_name = kstrdup_const(new_name, GFP_KERNEL);
1564 if (!new_name)
1565 goto out;
1566 } else {
1567 new_name = NULL;
1568 }
1569
1570 /*
1571 * Move to the appropriate place in the appropriate directories rbtree.
1572 */
1573 kernfs_unlink_sibling(kn);
1574 kernfs_get(new_parent);
1575
1576 /* rename_lock protects ->parent and ->name accessors */
1577 spin_lock_irq(&kernfs_rename_lock);
1578
1579 old_parent = kn->parent;
1580 kn->parent = new_parent;
1581
1582 kn->ns = new_ns;
1583 if (new_name) {
1584 old_name = kn->name;
1585 kn->name = new_name;
1586 }
1587
1588 spin_unlock_irq(&kernfs_rename_lock);
1589
1590 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1591 kernfs_link_sibling(kn);
1592
1593 kernfs_put(old_parent);
1594 kfree_const(old_name);
1595
1596 error = 0;
1597 out:
1598 mutex_unlock(&kernfs_mutex);
1599 return error;
1600}
1601
1602/* Relationship between s_mode and the DT_xxx types */
1603static inline unsigned char dt_type(struct kernfs_node *kn)
1604{
1605 return (kn->mode >> 12) & 15;
1606}
1607
1608static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1609{
1610 kernfs_put(filp->private_data);
1611 return 0;
1612}
1613
1614static struct kernfs_node *kernfs_dir_pos(const void *ns,
1615 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1616{
1617 if (pos) {
1618 int valid = kernfs_active(pos) &&
1619 pos->parent == parent && hash == pos->hash;
1620 kernfs_put(pos);
1621 if (!valid)
1622 pos = NULL;
1623 }
1624 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1625 struct rb_node *node = parent->dir.children.rb_node;
1626 while (node) {
1627 pos = rb_to_kn(node);
1628
1629 if (hash < pos->hash)
1630 node = node->rb_left;
1631 else if (hash > pos->hash)
1632 node = node->rb_right;
1633 else
1634 break;
1635 }
1636 }
1637 /* Skip over entries which are dying/dead or in the wrong namespace */
1638 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1639 struct rb_node *node = rb_next(&pos->rb);
1640 if (!node)
1641 pos = NULL;
1642 else
1643 pos = rb_to_kn(node);
1644 }
1645 return pos;
1646}
1647
1648static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1649 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1650{
1651 pos = kernfs_dir_pos(ns, parent, ino, pos);
1652 if (pos) {
1653 do {
1654 struct rb_node *node = rb_next(&pos->rb);
1655 if (!node)
1656 pos = NULL;
1657 else
1658 pos = rb_to_kn(node);
1659 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1660 }
1661 return pos;
1662}
1663
1664static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1665{
1666 struct dentry *dentry = file->f_path.dentry;
1667 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1668 struct kernfs_node *pos = file->private_data;
1669 const void *ns = NULL;
1670
1671 if (!dir_emit_dots(file, ctx))
1672 return 0;
1673 mutex_lock(&kernfs_mutex);
1674
1675 if (kernfs_ns_enabled(parent))
1676 ns = kernfs_info(dentry->d_sb)->ns;
1677
1678 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1679 pos;
1680 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1681 const char *name = pos->name;
1682 unsigned int type = dt_type(pos);
1683 int len = strlen(name);
1684 ino_t ino = kernfs_ino(pos);
1685
1686 ctx->pos = pos->hash;
1687 file->private_data = pos;
1688 kernfs_get(pos);
1689
1690 mutex_unlock(&kernfs_mutex);
1691 if (!dir_emit(ctx, name, len, ino, type))
1692 return 0;
1693 mutex_lock(&kernfs_mutex);
1694 }
1695 mutex_unlock(&kernfs_mutex);
1696 file->private_data = NULL;
1697 ctx->pos = INT_MAX;
1698 return 0;
1699}
1700
1701const struct file_operations kernfs_dir_fops = {
1702 .read = generic_read_dir,
1703 .iterate_shared = kernfs_fop_readdir,
1704 .release = kernfs_dir_fop_release,
1705 .llseek = generic_file_llseek,
1706};
1/*
2 * fs/kernfs/dir.c - kernfs directory implementation
3 *
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
7 *
8 * This file is released under the GPLv2.
9 */
10
11#include <linux/sched.h>
12#include <linux/fs.h>
13#include <linux/namei.h>
14#include <linux/idr.h>
15#include <linux/slab.h>
16#include <linux/security.h>
17#include <linux/hash.h>
18
19#include "kernfs-internal.h"
20
21DEFINE_MUTEX(kernfs_mutex);
22static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
23static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */
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 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
45}
46
47/* kernfs_node_depth - compute depth from @from to @to */
48static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
49{
50 size_t depth = 0;
51
52 while (to->parent && to != from) {
53 depth++;
54 to = to->parent;
55 }
56 return depth;
57}
58
59static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
60 struct kernfs_node *b)
61{
62 size_t da, db;
63 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
64
65 if (ra != rb)
66 return NULL;
67
68 da = kernfs_depth(ra->kn, a);
69 db = kernfs_depth(rb->kn, b);
70
71 while (da > db) {
72 a = a->parent;
73 da--;
74 }
75 while (db > da) {
76 b = b->parent;
77 db--;
78 }
79
80 /* worst case b and a will be the same at root */
81 while (b != a) {
82 b = b->parent;
83 a = a->parent;
84 }
85
86 return a;
87}
88
89/**
90 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
91 * where kn_from is treated as root of the path.
92 * @kn_from: kernfs node which should be treated as root for the path
93 * @kn_to: kernfs node to which path is needed
94 * @buf: buffer to copy the path into
95 * @buflen: size of @buf
96 *
97 * We need to handle couple of scenarios here:
98 * [1] when @kn_from is an ancestor of @kn_to at some level
99 * kn_from: /n1/n2/n3
100 * kn_to: /n1/n2/n3/n4/n5
101 * result: /n4/n5
102 *
103 * [2] when @kn_from is on a different hierarchy and we need to find common
104 * ancestor between @kn_from and @kn_to.
105 * kn_from: /n1/n2/n3/n4
106 * kn_to: /n1/n2/n5
107 * result: /../../n5
108 * OR
109 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
110 * kn_to: /n1/n2/n3 [depth=3]
111 * result: /../..
112 *
113 * return value: length of the string. If greater than buflen,
114 * then contents of buf are undefined. On error, -1 is returned.
115 */
116static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
117 struct kernfs_node *kn_from,
118 char *buf, size_t buflen)
119{
120 struct kernfs_node *kn, *common;
121 const char parent_str[] = "/..";
122 size_t depth_from, depth_to, len = 0, nlen = 0;
123 char *p;
124 int i;
125
126 if (!kn_from)
127 kn_from = kernfs_root(kn_to)->kn;
128
129 if (kn_from == kn_to)
130 return strlcpy(buf, "/", buflen);
131
132 common = kernfs_common_ancestor(kn_from, kn_to);
133 if (WARN_ON(!common))
134 return -1;
135
136 depth_to = kernfs_depth(common, kn_to);
137 depth_from = kernfs_depth(common, kn_from);
138
139 if (buf)
140 buf[0] = '\0';
141
142 for (i = 0; i < depth_from; i++)
143 len += strlcpy(buf + len, parent_str,
144 len < buflen ? buflen - len : 0);
145
146 /* Calculate how many bytes we need for the rest */
147 for (kn = kn_to; kn != common; kn = kn->parent)
148 nlen += strlen(kn->name) + 1;
149
150 if (len + nlen >= buflen)
151 return len + nlen;
152
153 p = buf + len + nlen;
154 *p = '\0';
155 for (kn = kn_to; kn != common; kn = kn->parent) {
156 size_t tmp = strlen(kn->name);
157 p -= tmp;
158 memcpy(p, kn->name, tmp);
159 *(--p) = '/';
160 }
161
162 return len + nlen;
163}
164
165/**
166 * kernfs_name - obtain the name of a given node
167 * @kn: kernfs_node of interest
168 * @buf: buffer to copy @kn's name into
169 * @buflen: size of @buf
170 *
171 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
172 * similar to strlcpy(). It returns the length of @kn's name and if @buf
173 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
174 *
175 * This function can be called from any context.
176 */
177int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
178{
179 unsigned long flags;
180 int ret;
181
182 spin_lock_irqsave(&kernfs_rename_lock, flags);
183 ret = kernfs_name_locked(kn, buf, buflen);
184 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
185 return ret;
186}
187
188/**
189 * kernfs_path_len - determine the length of the full path of a given node
190 * @kn: kernfs_node of interest
191 *
192 * The returned length doesn't include the space for the terminating '\0'.
193 */
194size_t kernfs_path_len(struct kernfs_node *kn)
195{
196 size_t len = 0;
197 unsigned long flags;
198
199 spin_lock_irqsave(&kernfs_rename_lock, flags);
200
201 do {
202 len += strlen(kn->name) + 1;
203 kn = kn->parent;
204 } while (kn && kn->parent);
205
206 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
207
208 return len;
209}
210
211/**
212 * kernfs_path_from_node - build path of node @to relative to @from.
213 * @from: parent kernfs_node relative to which we need to build the path
214 * @to: kernfs_node of interest
215 * @buf: buffer to copy @to's path into
216 * @buflen: size of @buf
217 *
218 * Builds @to's path relative to @from in @buf. @from and @to must
219 * be on the same kernfs-root. If @from is not parent of @to, then a relative
220 * path (which includes '..'s) as needed to reach from @from to @to is
221 * returned.
222 *
223 * If @buf isn't long enough, the return value will be greater than @buflen
224 * and @buf contents are undefined.
225 */
226int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
227 char *buf, size_t buflen)
228{
229 unsigned long flags;
230 int ret;
231
232 spin_lock_irqsave(&kernfs_rename_lock, flags);
233 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
234 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
235 return ret;
236}
237EXPORT_SYMBOL_GPL(kernfs_path_from_node);
238
239/**
240 * kernfs_path - build full path of a given node
241 * @kn: kernfs_node of interest
242 * @buf: buffer to copy @kn's name into
243 * @buflen: size of @buf
244 *
245 * Builds and returns the full path of @kn in @buf of @buflen bytes. The
246 * path is built from the end of @buf so the returned pointer usually
247 * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated
248 * and %NULL is returned.
249 */
250char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
251{
252 int ret;
253
254 ret = kernfs_path_from_node(kn, NULL, buf, buflen);
255 if (ret < 0 || ret >= buflen)
256 return NULL;
257 return buf;
258}
259EXPORT_SYMBOL_GPL(kernfs_path);
260
261/**
262 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
263 * @kn: kernfs_node of interest
264 *
265 * This function can be called from any context.
266 */
267void pr_cont_kernfs_name(struct kernfs_node *kn)
268{
269 unsigned long flags;
270
271 spin_lock_irqsave(&kernfs_rename_lock, flags);
272
273 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
274 pr_cont("%s", kernfs_pr_cont_buf);
275
276 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
277}
278
279/**
280 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
281 * @kn: kernfs_node of interest
282 *
283 * This function can be called from any context.
284 */
285void pr_cont_kernfs_path(struct kernfs_node *kn)
286{
287 unsigned long flags;
288 int sz;
289
290 spin_lock_irqsave(&kernfs_rename_lock, flags);
291
292 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
293 sizeof(kernfs_pr_cont_buf));
294 if (sz < 0) {
295 pr_cont("(error)");
296 goto out;
297 }
298
299 if (sz >= sizeof(kernfs_pr_cont_buf)) {
300 pr_cont("(name too long)");
301 goto out;
302 }
303
304 pr_cont("%s", kernfs_pr_cont_buf);
305
306out:
307 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
308}
309
310/**
311 * kernfs_get_parent - determine the parent node and pin it
312 * @kn: kernfs_node of interest
313 *
314 * Determines @kn's parent, pins and returns it. This function can be
315 * called from any context.
316 */
317struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
318{
319 struct kernfs_node *parent;
320 unsigned long flags;
321
322 spin_lock_irqsave(&kernfs_rename_lock, flags);
323 parent = kn->parent;
324 kernfs_get(parent);
325 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
326
327 return parent;
328}
329
330/**
331 * kernfs_name_hash
332 * @name: Null terminated string to hash
333 * @ns: Namespace tag to hash
334 *
335 * Returns 31 bit hash of ns + name (so it fits in an off_t )
336 */
337static unsigned int kernfs_name_hash(const char *name, const void *ns)
338{
339 unsigned long hash = init_name_hash();
340 unsigned int len = strlen(name);
341 while (len--)
342 hash = partial_name_hash(*name++, hash);
343 hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31));
344 hash &= 0x7fffffffU;
345 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
346 if (hash < 2)
347 hash += 2;
348 if (hash >= INT_MAX)
349 hash = INT_MAX - 1;
350 return hash;
351}
352
353static int kernfs_name_compare(unsigned int hash, const char *name,
354 const void *ns, const struct kernfs_node *kn)
355{
356 if (hash < kn->hash)
357 return -1;
358 if (hash > kn->hash)
359 return 1;
360 if (ns < kn->ns)
361 return -1;
362 if (ns > kn->ns)
363 return 1;
364 return strcmp(name, kn->name);
365}
366
367static int kernfs_sd_compare(const struct kernfs_node *left,
368 const struct kernfs_node *right)
369{
370 return kernfs_name_compare(left->hash, left->name, left->ns, right);
371}
372
373/**
374 * kernfs_link_sibling - link kernfs_node into sibling rbtree
375 * @kn: kernfs_node of interest
376 *
377 * Link @kn into its sibling rbtree which starts from
378 * @kn->parent->dir.children.
379 *
380 * Locking:
381 * mutex_lock(kernfs_mutex)
382 *
383 * RETURNS:
384 * 0 on susccess -EEXIST on failure.
385 */
386static int kernfs_link_sibling(struct kernfs_node *kn)
387{
388 struct rb_node **node = &kn->parent->dir.children.rb_node;
389 struct rb_node *parent = NULL;
390
391 while (*node) {
392 struct kernfs_node *pos;
393 int result;
394
395 pos = rb_to_kn(*node);
396 parent = *node;
397 result = kernfs_sd_compare(kn, pos);
398 if (result < 0)
399 node = &pos->rb.rb_left;
400 else if (result > 0)
401 node = &pos->rb.rb_right;
402 else
403 return -EEXIST;
404 }
405
406 /* add new node and rebalance the tree */
407 rb_link_node(&kn->rb, parent, node);
408 rb_insert_color(&kn->rb, &kn->parent->dir.children);
409
410 /* successfully added, account subdir number */
411 if (kernfs_type(kn) == KERNFS_DIR)
412 kn->parent->dir.subdirs++;
413
414 return 0;
415}
416
417/**
418 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
419 * @kn: kernfs_node of interest
420 *
421 * Try to unlink @kn from its sibling rbtree which starts from
422 * kn->parent->dir.children. Returns %true if @kn was actually
423 * removed, %false if @kn wasn't on the rbtree.
424 *
425 * Locking:
426 * mutex_lock(kernfs_mutex)
427 */
428static bool kernfs_unlink_sibling(struct kernfs_node *kn)
429{
430 if (RB_EMPTY_NODE(&kn->rb))
431 return false;
432
433 if (kernfs_type(kn) == KERNFS_DIR)
434 kn->parent->dir.subdirs--;
435
436 rb_erase(&kn->rb, &kn->parent->dir.children);
437 RB_CLEAR_NODE(&kn->rb);
438 return true;
439}
440
441/**
442 * kernfs_get_active - get an active reference to kernfs_node
443 * @kn: kernfs_node to get an active reference to
444 *
445 * Get an active reference of @kn. This function is noop if @kn
446 * is NULL.
447 *
448 * RETURNS:
449 * Pointer to @kn on success, NULL on failure.
450 */
451struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
452{
453 if (unlikely(!kn))
454 return NULL;
455
456 if (!atomic_inc_unless_negative(&kn->active))
457 return NULL;
458
459 if (kernfs_lockdep(kn))
460 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
461 return kn;
462}
463
464/**
465 * kernfs_put_active - put an active reference to kernfs_node
466 * @kn: kernfs_node to put an active reference to
467 *
468 * Put an active reference to @kn. This function is noop if @kn
469 * is NULL.
470 */
471void kernfs_put_active(struct kernfs_node *kn)
472{
473 struct kernfs_root *root = kernfs_root(kn);
474 int v;
475
476 if (unlikely(!kn))
477 return;
478
479 if (kernfs_lockdep(kn))
480 rwsem_release(&kn->dep_map, 1, _RET_IP_);
481 v = atomic_dec_return(&kn->active);
482 if (likely(v != KN_DEACTIVATED_BIAS))
483 return;
484
485 wake_up_all(&root->deactivate_waitq);
486}
487
488/**
489 * kernfs_drain - drain kernfs_node
490 * @kn: kernfs_node to drain
491 *
492 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
493 * removers may invoke this function concurrently on @kn and all will
494 * return after draining is complete.
495 */
496static void kernfs_drain(struct kernfs_node *kn)
497 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
498{
499 struct kernfs_root *root = kernfs_root(kn);
500
501 lockdep_assert_held(&kernfs_mutex);
502 WARN_ON_ONCE(kernfs_active(kn));
503
504 mutex_unlock(&kernfs_mutex);
505
506 if (kernfs_lockdep(kn)) {
507 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
508 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
509 lock_contended(&kn->dep_map, _RET_IP_);
510 }
511
512 /* but everyone should wait for draining */
513 wait_event(root->deactivate_waitq,
514 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
515
516 if (kernfs_lockdep(kn)) {
517 lock_acquired(&kn->dep_map, _RET_IP_);
518 rwsem_release(&kn->dep_map, 1, _RET_IP_);
519 }
520
521 kernfs_unmap_bin_file(kn);
522
523 mutex_lock(&kernfs_mutex);
524}
525
526/**
527 * kernfs_get - get a reference count on a kernfs_node
528 * @kn: the target kernfs_node
529 */
530void kernfs_get(struct kernfs_node *kn)
531{
532 if (kn) {
533 WARN_ON(!atomic_read(&kn->count));
534 atomic_inc(&kn->count);
535 }
536}
537EXPORT_SYMBOL_GPL(kernfs_get);
538
539/**
540 * kernfs_put - put a reference count on a kernfs_node
541 * @kn: the target kernfs_node
542 *
543 * Put a reference count of @kn and destroy it if it reached zero.
544 */
545void kernfs_put(struct kernfs_node *kn)
546{
547 struct kernfs_node *parent;
548 struct kernfs_root *root;
549
550 if (!kn || !atomic_dec_and_test(&kn->count))
551 return;
552 root = kernfs_root(kn);
553 repeat:
554 /*
555 * Moving/renaming is always done while holding reference.
556 * kn->parent won't change beneath us.
557 */
558 parent = kn->parent;
559
560 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
561 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
562 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
563
564 if (kernfs_type(kn) == KERNFS_LINK)
565 kernfs_put(kn->symlink.target_kn);
566
567 kfree_const(kn->name);
568
569 if (kn->iattr) {
570 if (kn->iattr->ia_secdata)
571 security_release_secctx(kn->iattr->ia_secdata,
572 kn->iattr->ia_secdata_len);
573 simple_xattrs_free(&kn->iattr->xattrs);
574 }
575 kfree(kn->iattr);
576 ida_simple_remove(&root->ino_ida, kn->ino);
577 kmem_cache_free(kernfs_node_cache, kn);
578
579 kn = parent;
580 if (kn) {
581 if (atomic_dec_and_test(&kn->count))
582 goto repeat;
583 } else {
584 /* just released the root kn, free @root too */
585 ida_destroy(&root->ino_ida);
586 kfree(root);
587 }
588}
589EXPORT_SYMBOL_GPL(kernfs_put);
590
591static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
592{
593 struct kernfs_node *kn;
594
595 if (flags & LOOKUP_RCU)
596 return -ECHILD;
597
598 /* Always perform fresh lookup for negatives */
599 if (d_really_is_negative(dentry))
600 goto out_bad_unlocked;
601
602 kn = dentry->d_fsdata;
603 mutex_lock(&kernfs_mutex);
604
605 /* The kernfs node has been deactivated */
606 if (!kernfs_active(kn))
607 goto out_bad;
608
609 /* The kernfs node has been moved? */
610 if (dentry->d_parent->d_fsdata != kn->parent)
611 goto out_bad;
612
613 /* The kernfs node has been renamed */
614 if (strcmp(dentry->d_name.name, kn->name) != 0)
615 goto out_bad;
616
617 /* The kernfs node has been moved to a different namespace */
618 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
619 kernfs_info(dentry->d_sb)->ns != kn->ns)
620 goto out_bad;
621
622 mutex_unlock(&kernfs_mutex);
623 return 1;
624out_bad:
625 mutex_unlock(&kernfs_mutex);
626out_bad_unlocked:
627 return 0;
628}
629
630static void kernfs_dop_release(struct dentry *dentry)
631{
632 kernfs_put(dentry->d_fsdata);
633}
634
635const struct dentry_operations kernfs_dops = {
636 .d_revalidate = kernfs_dop_revalidate,
637 .d_release = kernfs_dop_release,
638};
639
640/**
641 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
642 * @dentry: the dentry in question
643 *
644 * Return the kernfs_node associated with @dentry. If @dentry is not a
645 * kernfs one, %NULL is returned.
646 *
647 * While the returned kernfs_node will stay accessible as long as @dentry
648 * is accessible, the returned node can be in any state and the caller is
649 * fully responsible for determining what's accessible.
650 */
651struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
652{
653 if (dentry->d_sb->s_op == &kernfs_sops)
654 return dentry->d_fsdata;
655 return NULL;
656}
657
658static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
659 const char *name, umode_t mode,
660 unsigned flags)
661{
662 struct kernfs_node *kn;
663 int ret;
664
665 name = kstrdup_const(name, GFP_KERNEL);
666 if (!name)
667 return NULL;
668
669 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
670 if (!kn)
671 goto err_out1;
672
673 ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL);
674 if (ret < 0)
675 goto err_out2;
676 kn->ino = ret;
677
678 atomic_set(&kn->count, 1);
679 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
680 RB_CLEAR_NODE(&kn->rb);
681
682 kn->name = name;
683 kn->mode = mode;
684 kn->flags = flags;
685
686 return kn;
687
688 err_out2:
689 kmem_cache_free(kernfs_node_cache, kn);
690 err_out1:
691 kfree_const(name);
692 return NULL;
693}
694
695struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
696 const char *name, umode_t mode,
697 unsigned flags)
698{
699 struct kernfs_node *kn;
700
701 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
702 if (kn) {
703 kernfs_get(parent);
704 kn->parent = parent;
705 }
706 return kn;
707}
708
709/**
710 * kernfs_add_one - add kernfs_node to parent without warning
711 * @kn: kernfs_node to be added
712 *
713 * The caller must already have initialized @kn->parent. This
714 * function increments nlink of the parent's inode if @kn is a
715 * directory and link into the children list of the parent.
716 *
717 * RETURNS:
718 * 0 on success, -EEXIST if entry with the given name already
719 * exists.
720 */
721int kernfs_add_one(struct kernfs_node *kn)
722{
723 struct kernfs_node *parent = kn->parent;
724 struct kernfs_iattrs *ps_iattr;
725 bool has_ns;
726 int ret;
727
728 mutex_lock(&kernfs_mutex);
729
730 ret = -EINVAL;
731 has_ns = kernfs_ns_enabled(parent);
732 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
733 has_ns ? "required" : "invalid", parent->name, kn->name))
734 goto out_unlock;
735
736 if (kernfs_type(parent) != KERNFS_DIR)
737 goto out_unlock;
738
739 ret = -ENOENT;
740 if (parent->flags & KERNFS_EMPTY_DIR)
741 goto out_unlock;
742
743 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
744 goto out_unlock;
745
746 kn->hash = kernfs_name_hash(kn->name, kn->ns);
747
748 ret = kernfs_link_sibling(kn);
749 if (ret)
750 goto out_unlock;
751
752 /* Update timestamps on the parent */
753 ps_iattr = parent->iattr;
754 if (ps_iattr) {
755 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
756 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
757 }
758
759 mutex_unlock(&kernfs_mutex);
760
761 /*
762 * Activate the new node unless CREATE_DEACTIVATED is requested.
763 * If not activated here, the kernfs user is responsible for
764 * activating the node with kernfs_activate(). A node which hasn't
765 * been activated is not visible to userland and its removal won't
766 * trigger deactivation.
767 */
768 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
769 kernfs_activate(kn);
770 return 0;
771
772out_unlock:
773 mutex_unlock(&kernfs_mutex);
774 return ret;
775}
776
777/**
778 * kernfs_find_ns - find kernfs_node with the given name
779 * @parent: kernfs_node to search under
780 * @name: name to look for
781 * @ns: the namespace tag to use
782 *
783 * Look for kernfs_node with name @name under @parent. Returns pointer to
784 * the found kernfs_node on success, %NULL on failure.
785 */
786static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
787 const unsigned char *name,
788 const void *ns)
789{
790 struct rb_node *node = parent->dir.children.rb_node;
791 bool has_ns = kernfs_ns_enabled(parent);
792 unsigned int hash;
793
794 lockdep_assert_held(&kernfs_mutex);
795
796 if (has_ns != (bool)ns) {
797 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
798 has_ns ? "required" : "invalid", parent->name, name);
799 return NULL;
800 }
801
802 hash = kernfs_name_hash(name, ns);
803 while (node) {
804 struct kernfs_node *kn;
805 int result;
806
807 kn = rb_to_kn(node);
808 result = kernfs_name_compare(hash, name, ns, kn);
809 if (result < 0)
810 node = node->rb_left;
811 else if (result > 0)
812 node = node->rb_right;
813 else
814 return kn;
815 }
816 return NULL;
817}
818
819static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
820 const unsigned char *path,
821 const void *ns)
822{
823 size_t len;
824 char *p, *name;
825
826 lockdep_assert_held(&kernfs_mutex);
827
828 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
829 spin_lock_irq(&kernfs_rename_lock);
830
831 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
832
833 if (len >= sizeof(kernfs_pr_cont_buf)) {
834 spin_unlock_irq(&kernfs_rename_lock);
835 return NULL;
836 }
837
838 p = kernfs_pr_cont_buf;
839
840 while ((name = strsep(&p, "/")) && parent) {
841 if (*name == '\0')
842 continue;
843 parent = kernfs_find_ns(parent, name, ns);
844 }
845
846 spin_unlock_irq(&kernfs_rename_lock);
847
848 return parent;
849}
850
851/**
852 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
853 * @parent: kernfs_node to search under
854 * @name: name to look for
855 * @ns: the namespace tag to use
856 *
857 * Look for kernfs_node with name @name under @parent and get a reference
858 * if found. This function may sleep and returns pointer to the found
859 * kernfs_node on success, %NULL on failure.
860 */
861struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
862 const char *name, const void *ns)
863{
864 struct kernfs_node *kn;
865
866 mutex_lock(&kernfs_mutex);
867 kn = kernfs_find_ns(parent, name, ns);
868 kernfs_get(kn);
869 mutex_unlock(&kernfs_mutex);
870
871 return kn;
872}
873EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
874
875/**
876 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
877 * @parent: kernfs_node to search under
878 * @path: path to look for
879 * @ns: the namespace tag to use
880 *
881 * Look for kernfs_node with path @path under @parent and get a reference
882 * if found. This function may sleep and returns pointer to the found
883 * kernfs_node on success, %NULL on failure.
884 */
885struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
886 const char *path, const void *ns)
887{
888 struct kernfs_node *kn;
889
890 mutex_lock(&kernfs_mutex);
891 kn = kernfs_walk_ns(parent, path, ns);
892 kernfs_get(kn);
893 mutex_unlock(&kernfs_mutex);
894
895 return kn;
896}
897
898/**
899 * kernfs_create_root - create a new kernfs hierarchy
900 * @scops: optional syscall operations for the hierarchy
901 * @flags: KERNFS_ROOT_* flags
902 * @priv: opaque data associated with the new directory
903 *
904 * Returns the root of the new hierarchy on success, ERR_PTR() value on
905 * failure.
906 */
907struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
908 unsigned int flags, void *priv)
909{
910 struct kernfs_root *root;
911 struct kernfs_node *kn;
912
913 root = kzalloc(sizeof(*root), GFP_KERNEL);
914 if (!root)
915 return ERR_PTR(-ENOMEM);
916
917 ida_init(&root->ino_ida);
918 INIT_LIST_HEAD(&root->supers);
919
920 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
921 KERNFS_DIR);
922 if (!kn) {
923 ida_destroy(&root->ino_ida);
924 kfree(root);
925 return ERR_PTR(-ENOMEM);
926 }
927
928 kn->priv = priv;
929 kn->dir.root = root;
930
931 root->syscall_ops = scops;
932 root->flags = flags;
933 root->kn = kn;
934 init_waitqueue_head(&root->deactivate_waitq);
935
936 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
937 kernfs_activate(kn);
938
939 return root;
940}
941
942/**
943 * kernfs_destroy_root - destroy a kernfs hierarchy
944 * @root: root of the hierarchy to destroy
945 *
946 * Destroy the hierarchy anchored at @root by removing all existing
947 * directories and destroying @root.
948 */
949void kernfs_destroy_root(struct kernfs_root *root)
950{
951 kernfs_remove(root->kn); /* will also free @root */
952}
953
954/**
955 * kernfs_create_dir_ns - create a directory
956 * @parent: parent in which to create a new directory
957 * @name: name of the new directory
958 * @mode: mode of the new directory
959 * @priv: opaque data associated with the new directory
960 * @ns: optional namespace tag of the directory
961 *
962 * Returns the created node on success, ERR_PTR() value on failure.
963 */
964struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
965 const char *name, umode_t mode,
966 void *priv, const void *ns)
967{
968 struct kernfs_node *kn;
969 int rc;
970
971 /* allocate */
972 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
973 if (!kn)
974 return ERR_PTR(-ENOMEM);
975
976 kn->dir.root = parent->dir.root;
977 kn->ns = ns;
978 kn->priv = priv;
979
980 /* link in */
981 rc = kernfs_add_one(kn);
982 if (!rc)
983 return kn;
984
985 kernfs_put(kn);
986 return ERR_PTR(rc);
987}
988
989/**
990 * kernfs_create_empty_dir - create an always empty directory
991 * @parent: parent in which to create a new directory
992 * @name: name of the new directory
993 *
994 * Returns the created node on success, ERR_PTR() value on failure.
995 */
996struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
997 const char *name)
998{
999 struct kernfs_node *kn;
1000 int rc;
1001
1002 /* allocate */
1003 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR);
1004 if (!kn)
1005 return ERR_PTR(-ENOMEM);
1006
1007 kn->flags |= KERNFS_EMPTY_DIR;
1008 kn->dir.root = parent->dir.root;
1009 kn->ns = NULL;
1010 kn->priv = NULL;
1011
1012 /* link in */
1013 rc = kernfs_add_one(kn);
1014 if (!rc)
1015 return kn;
1016
1017 kernfs_put(kn);
1018 return ERR_PTR(rc);
1019}
1020
1021static struct dentry *kernfs_iop_lookup(struct inode *dir,
1022 struct dentry *dentry,
1023 unsigned int flags)
1024{
1025 struct dentry *ret;
1026 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
1027 struct kernfs_node *kn;
1028 struct inode *inode;
1029 const void *ns = NULL;
1030
1031 mutex_lock(&kernfs_mutex);
1032
1033 if (kernfs_ns_enabled(parent))
1034 ns = kernfs_info(dir->i_sb)->ns;
1035
1036 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1037
1038 /* no such entry */
1039 if (!kn || !kernfs_active(kn)) {
1040 ret = NULL;
1041 goto out_unlock;
1042 }
1043 kernfs_get(kn);
1044 dentry->d_fsdata = kn;
1045
1046 /* attach dentry and inode */
1047 inode = kernfs_get_inode(dir->i_sb, kn);
1048 if (!inode) {
1049 ret = ERR_PTR(-ENOMEM);
1050 goto out_unlock;
1051 }
1052
1053 /* instantiate and hash dentry */
1054 ret = d_splice_alias(inode, dentry);
1055 out_unlock:
1056 mutex_unlock(&kernfs_mutex);
1057 return ret;
1058}
1059
1060static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1061 umode_t mode)
1062{
1063 struct kernfs_node *parent = dir->i_private;
1064 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1065 int ret;
1066
1067 if (!scops || !scops->mkdir)
1068 return -EPERM;
1069
1070 if (!kernfs_get_active(parent))
1071 return -ENODEV;
1072
1073 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1074
1075 kernfs_put_active(parent);
1076 return ret;
1077}
1078
1079static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1080{
1081 struct kernfs_node *kn = dentry->d_fsdata;
1082 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1083 int ret;
1084
1085 if (!scops || !scops->rmdir)
1086 return -EPERM;
1087
1088 if (!kernfs_get_active(kn))
1089 return -ENODEV;
1090
1091 ret = scops->rmdir(kn);
1092
1093 kernfs_put_active(kn);
1094 return ret;
1095}
1096
1097static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1098 struct inode *new_dir, struct dentry *new_dentry)
1099{
1100 struct kernfs_node *kn = old_dentry->d_fsdata;
1101 struct kernfs_node *new_parent = new_dir->i_private;
1102 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1103 int ret;
1104
1105 if (!scops || !scops->rename)
1106 return -EPERM;
1107
1108 if (!kernfs_get_active(kn))
1109 return -ENODEV;
1110
1111 if (!kernfs_get_active(new_parent)) {
1112 kernfs_put_active(kn);
1113 return -ENODEV;
1114 }
1115
1116 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1117
1118 kernfs_put_active(new_parent);
1119 kernfs_put_active(kn);
1120 return ret;
1121}
1122
1123const struct inode_operations kernfs_dir_iops = {
1124 .lookup = kernfs_iop_lookup,
1125 .permission = kernfs_iop_permission,
1126 .setattr = kernfs_iop_setattr,
1127 .getattr = kernfs_iop_getattr,
1128 .setxattr = kernfs_iop_setxattr,
1129 .removexattr = kernfs_iop_removexattr,
1130 .getxattr = kernfs_iop_getxattr,
1131 .listxattr = kernfs_iop_listxattr,
1132
1133 .mkdir = kernfs_iop_mkdir,
1134 .rmdir = kernfs_iop_rmdir,
1135 .rename = kernfs_iop_rename,
1136};
1137
1138static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1139{
1140 struct kernfs_node *last;
1141
1142 while (true) {
1143 struct rb_node *rbn;
1144
1145 last = pos;
1146
1147 if (kernfs_type(pos) != KERNFS_DIR)
1148 break;
1149
1150 rbn = rb_first(&pos->dir.children);
1151 if (!rbn)
1152 break;
1153
1154 pos = rb_to_kn(rbn);
1155 }
1156
1157 return last;
1158}
1159
1160/**
1161 * kernfs_next_descendant_post - find the next descendant for post-order walk
1162 * @pos: the current position (%NULL to initiate traversal)
1163 * @root: kernfs_node whose descendants to walk
1164 *
1165 * Find the next descendant to visit for post-order traversal of @root's
1166 * descendants. @root is included in the iteration and the last node to be
1167 * visited.
1168 */
1169static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1170 struct kernfs_node *root)
1171{
1172 struct rb_node *rbn;
1173
1174 lockdep_assert_held(&kernfs_mutex);
1175
1176 /* if first iteration, visit leftmost descendant which may be root */
1177 if (!pos)
1178 return kernfs_leftmost_descendant(root);
1179
1180 /* if we visited @root, we're done */
1181 if (pos == root)
1182 return NULL;
1183
1184 /* if there's an unvisited sibling, visit its leftmost descendant */
1185 rbn = rb_next(&pos->rb);
1186 if (rbn)
1187 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1188
1189 /* no sibling left, visit parent */
1190 return pos->parent;
1191}
1192
1193/**
1194 * kernfs_activate - activate a node which started deactivated
1195 * @kn: kernfs_node whose subtree is to be activated
1196 *
1197 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1198 * needs to be explicitly activated. A node which hasn't been activated
1199 * isn't visible to userland and deactivation is skipped during its
1200 * removal. This is useful to construct atomic init sequences where
1201 * creation of multiple nodes should either succeed or fail atomically.
1202 *
1203 * The caller is responsible for ensuring that this function is not called
1204 * after kernfs_remove*() is invoked on @kn.
1205 */
1206void kernfs_activate(struct kernfs_node *kn)
1207{
1208 struct kernfs_node *pos;
1209
1210 mutex_lock(&kernfs_mutex);
1211
1212 pos = NULL;
1213 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1214 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1215 continue;
1216
1217 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1218 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1219
1220 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1221 pos->flags |= KERNFS_ACTIVATED;
1222 }
1223
1224 mutex_unlock(&kernfs_mutex);
1225}
1226
1227static void __kernfs_remove(struct kernfs_node *kn)
1228{
1229 struct kernfs_node *pos;
1230
1231 lockdep_assert_held(&kernfs_mutex);
1232
1233 /*
1234 * Short-circuit if non-root @kn has already finished removal.
1235 * This is for kernfs_remove_self() which plays with active ref
1236 * after removal.
1237 */
1238 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1239 return;
1240
1241 pr_debug("kernfs %s: removing\n", kn->name);
1242
1243 /* prevent any new usage under @kn by deactivating all nodes */
1244 pos = NULL;
1245 while ((pos = kernfs_next_descendant_post(pos, kn)))
1246 if (kernfs_active(pos))
1247 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1248
1249 /* deactivate and unlink the subtree node-by-node */
1250 do {
1251 pos = kernfs_leftmost_descendant(kn);
1252
1253 /*
1254 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1255 * base ref could have been put by someone else by the time
1256 * the function returns. Make sure it doesn't go away
1257 * underneath us.
1258 */
1259 kernfs_get(pos);
1260
1261 /*
1262 * Drain iff @kn was activated. This avoids draining and
1263 * its lockdep annotations for nodes which have never been
1264 * activated and allows embedding kernfs_remove() in create
1265 * error paths without worrying about draining.
1266 */
1267 if (kn->flags & KERNFS_ACTIVATED)
1268 kernfs_drain(pos);
1269 else
1270 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1271
1272 /*
1273 * kernfs_unlink_sibling() succeeds once per node. Use it
1274 * to decide who's responsible for cleanups.
1275 */
1276 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1277 struct kernfs_iattrs *ps_iattr =
1278 pos->parent ? pos->parent->iattr : NULL;
1279
1280 /* update timestamps on the parent */
1281 if (ps_iattr) {
1282 ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME;
1283 ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME;
1284 }
1285
1286 kernfs_put(pos);
1287 }
1288
1289 kernfs_put(pos);
1290 } while (pos != kn);
1291}
1292
1293/**
1294 * kernfs_remove - remove a kernfs_node recursively
1295 * @kn: the kernfs_node to remove
1296 *
1297 * Remove @kn along with all its subdirectories and files.
1298 */
1299void kernfs_remove(struct kernfs_node *kn)
1300{
1301 mutex_lock(&kernfs_mutex);
1302 __kernfs_remove(kn);
1303 mutex_unlock(&kernfs_mutex);
1304}
1305
1306/**
1307 * kernfs_break_active_protection - break out of active protection
1308 * @kn: the self kernfs_node
1309 *
1310 * The caller must be running off of a kernfs operation which is invoked
1311 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1312 * this function must also be matched with an invocation of
1313 * kernfs_unbreak_active_protection().
1314 *
1315 * This function releases the active reference of @kn the caller is
1316 * holding. Once this function is called, @kn may be removed at any point
1317 * and the caller is solely responsible for ensuring that the objects it
1318 * dereferences are accessible.
1319 */
1320void kernfs_break_active_protection(struct kernfs_node *kn)
1321{
1322 /*
1323 * Take out ourself out of the active ref dependency chain. If
1324 * we're called without an active ref, lockdep will complain.
1325 */
1326 kernfs_put_active(kn);
1327}
1328
1329/**
1330 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1331 * @kn: the self kernfs_node
1332 *
1333 * If kernfs_break_active_protection() was called, this function must be
1334 * invoked before finishing the kernfs operation. Note that while this
1335 * function restores the active reference, it doesn't and can't actually
1336 * restore the active protection - @kn may already or be in the process of
1337 * being removed. Once kernfs_break_active_protection() is invoked, that
1338 * protection is irreversibly gone for the kernfs operation instance.
1339 *
1340 * While this function may be called at any point after
1341 * kernfs_break_active_protection() is invoked, its most useful location
1342 * would be right before the enclosing kernfs operation returns.
1343 */
1344void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1345{
1346 /*
1347 * @kn->active could be in any state; however, the increment we do
1348 * here will be undone as soon as the enclosing kernfs operation
1349 * finishes and this temporary bump can't break anything. If @kn
1350 * is alive, nothing changes. If @kn is being deactivated, the
1351 * soon-to-follow put will either finish deactivation or restore
1352 * deactivated state. If @kn is already removed, the temporary
1353 * bump is guaranteed to be gone before @kn is released.
1354 */
1355 atomic_inc(&kn->active);
1356 if (kernfs_lockdep(kn))
1357 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1358}
1359
1360/**
1361 * kernfs_remove_self - remove a kernfs_node from its own method
1362 * @kn: the self kernfs_node to remove
1363 *
1364 * The caller must be running off of a kernfs operation which is invoked
1365 * with an active reference - e.g. one of kernfs_ops. This can be used to
1366 * implement a file operation which deletes itself.
1367 *
1368 * For example, the "delete" file for a sysfs device directory can be
1369 * implemented by invoking kernfs_remove_self() on the "delete" file
1370 * itself. This function breaks the circular dependency of trying to
1371 * deactivate self while holding an active ref itself. It isn't necessary
1372 * to modify the usual removal path to use kernfs_remove_self(). The
1373 * "delete" implementation can simply invoke kernfs_remove_self() on self
1374 * before proceeding with the usual removal path. kernfs will ignore later
1375 * kernfs_remove() on self.
1376 *
1377 * kernfs_remove_self() can be called multiple times concurrently on the
1378 * same kernfs_node. Only the first one actually performs removal and
1379 * returns %true. All others will wait until the kernfs operation which
1380 * won self-removal finishes and return %false. Note that the losers wait
1381 * for the completion of not only the winning kernfs_remove_self() but also
1382 * the whole kernfs_ops which won the arbitration. This can be used to
1383 * guarantee, for example, all concurrent writes to a "delete" file to
1384 * finish only after the whole operation is complete.
1385 */
1386bool kernfs_remove_self(struct kernfs_node *kn)
1387{
1388 bool ret;
1389
1390 mutex_lock(&kernfs_mutex);
1391 kernfs_break_active_protection(kn);
1392
1393 /*
1394 * SUICIDAL is used to arbitrate among competing invocations. Only
1395 * the first one will actually perform removal. When the removal
1396 * is complete, SUICIDED is set and the active ref is restored
1397 * while holding kernfs_mutex. The ones which lost arbitration
1398 * waits for SUICDED && drained which can happen only after the
1399 * enclosing kernfs operation which executed the winning instance
1400 * of kernfs_remove_self() finished.
1401 */
1402 if (!(kn->flags & KERNFS_SUICIDAL)) {
1403 kn->flags |= KERNFS_SUICIDAL;
1404 __kernfs_remove(kn);
1405 kn->flags |= KERNFS_SUICIDED;
1406 ret = true;
1407 } else {
1408 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1409 DEFINE_WAIT(wait);
1410
1411 while (true) {
1412 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1413
1414 if ((kn->flags & KERNFS_SUICIDED) &&
1415 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1416 break;
1417
1418 mutex_unlock(&kernfs_mutex);
1419 schedule();
1420 mutex_lock(&kernfs_mutex);
1421 }
1422 finish_wait(waitq, &wait);
1423 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1424 ret = false;
1425 }
1426
1427 /*
1428 * This must be done while holding kernfs_mutex; otherwise, waiting
1429 * for SUICIDED && deactivated could finish prematurely.
1430 */
1431 kernfs_unbreak_active_protection(kn);
1432
1433 mutex_unlock(&kernfs_mutex);
1434 return ret;
1435}
1436
1437/**
1438 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1439 * @parent: parent of the target
1440 * @name: name of the kernfs_node to remove
1441 * @ns: namespace tag of the kernfs_node to remove
1442 *
1443 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1444 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1445 */
1446int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1447 const void *ns)
1448{
1449 struct kernfs_node *kn;
1450
1451 if (!parent) {
1452 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1453 name);
1454 return -ENOENT;
1455 }
1456
1457 mutex_lock(&kernfs_mutex);
1458
1459 kn = kernfs_find_ns(parent, name, ns);
1460 if (kn)
1461 __kernfs_remove(kn);
1462
1463 mutex_unlock(&kernfs_mutex);
1464
1465 if (kn)
1466 return 0;
1467 else
1468 return -ENOENT;
1469}
1470
1471/**
1472 * kernfs_rename_ns - move and rename a kernfs_node
1473 * @kn: target node
1474 * @new_parent: new parent to put @sd under
1475 * @new_name: new name
1476 * @new_ns: new namespace tag
1477 */
1478int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1479 const char *new_name, const void *new_ns)
1480{
1481 struct kernfs_node *old_parent;
1482 const char *old_name = NULL;
1483 int error;
1484
1485 /* can't move or rename root */
1486 if (!kn->parent)
1487 return -EINVAL;
1488
1489 mutex_lock(&kernfs_mutex);
1490
1491 error = -ENOENT;
1492 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1493 (new_parent->flags & KERNFS_EMPTY_DIR))
1494 goto out;
1495
1496 error = 0;
1497 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1498 (strcmp(kn->name, new_name) == 0))
1499 goto out; /* nothing to rename */
1500
1501 error = -EEXIST;
1502 if (kernfs_find_ns(new_parent, new_name, new_ns))
1503 goto out;
1504
1505 /* rename kernfs_node */
1506 if (strcmp(kn->name, new_name) != 0) {
1507 error = -ENOMEM;
1508 new_name = kstrdup_const(new_name, GFP_KERNEL);
1509 if (!new_name)
1510 goto out;
1511 } else {
1512 new_name = NULL;
1513 }
1514
1515 /*
1516 * Move to the appropriate place in the appropriate directories rbtree.
1517 */
1518 kernfs_unlink_sibling(kn);
1519 kernfs_get(new_parent);
1520
1521 /* rename_lock protects ->parent and ->name accessors */
1522 spin_lock_irq(&kernfs_rename_lock);
1523
1524 old_parent = kn->parent;
1525 kn->parent = new_parent;
1526
1527 kn->ns = new_ns;
1528 if (new_name) {
1529 old_name = kn->name;
1530 kn->name = new_name;
1531 }
1532
1533 spin_unlock_irq(&kernfs_rename_lock);
1534
1535 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1536 kernfs_link_sibling(kn);
1537
1538 kernfs_put(old_parent);
1539 kfree_const(old_name);
1540
1541 error = 0;
1542 out:
1543 mutex_unlock(&kernfs_mutex);
1544 return error;
1545}
1546
1547/* Relationship between s_mode and the DT_xxx types */
1548static inline unsigned char dt_type(struct kernfs_node *kn)
1549{
1550 return (kn->mode >> 12) & 15;
1551}
1552
1553static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1554{
1555 kernfs_put(filp->private_data);
1556 return 0;
1557}
1558
1559static struct kernfs_node *kernfs_dir_pos(const void *ns,
1560 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1561{
1562 if (pos) {
1563 int valid = kernfs_active(pos) &&
1564 pos->parent == parent && hash == pos->hash;
1565 kernfs_put(pos);
1566 if (!valid)
1567 pos = NULL;
1568 }
1569 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1570 struct rb_node *node = parent->dir.children.rb_node;
1571 while (node) {
1572 pos = rb_to_kn(node);
1573
1574 if (hash < pos->hash)
1575 node = node->rb_left;
1576 else if (hash > pos->hash)
1577 node = node->rb_right;
1578 else
1579 break;
1580 }
1581 }
1582 /* Skip over entries which are dying/dead or in the wrong namespace */
1583 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1584 struct rb_node *node = rb_next(&pos->rb);
1585 if (!node)
1586 pos = NULL;
1587 else
1588 pos = rb_to_kn(node);
1589 }
1590 return pos;
1591}
1592
1593static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1594 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1595{
1596 pos = kernfs_dir_pos(ns, parent, ino, pos);
1597 if (pos) {
1598 do {
1599 struct rb_node *node = rb_next(&pos->rb);
1600 if (!node)
1601 pos = NULL;
1602 else
1603 pos = rb_to_kn(node);
1604 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1605 }
1606 return pos;
1607}
1608
1609static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1610{
1611 struct dentry *dentry = file->f_path.dentry;
1612 struct kernfs_node *parent = dentry->d_fsdata;
1613 struct kernfs_node *pos = file->private_data;
1614 const void *ns = NULL;
1615
1616 if (!dir_emit_dots(file, ctx))
1617 return 0;
1618 mutex_lock(&kernfs_mutex);
1619
1620 if (kernfs_ns_enabled(parent))
1621 ns = kernfs_info(dentry->d_sb)->ns;
1622
1623 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1624 pos;
1625 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1626 const char *name = pos->name;
1627 unsigned int type = dt_type(pos);
1628 int len = strlen(name);
1629 ino_t ino = pos->ino;
1630
1631 ctx->pos = pos->hash;
1632 file->private_data = pos;
1633 kernfs_get(pos);
1634
1635 mutex_unlock(&kernfs_mutex);
1636 if (!dir_emit(ctx, name, len, ino, type))
1637 return 0;
1638 mutex_lock(&kernfs_mutex);
1639 }
1640 mutex_unlock(&kernfs_mutex);
1641 file->private_data = NULL;
1642 ctx->pos = INT_MAX;
1643 return 0;
1644}
1645
1646static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset,
1647 int whence)
1648{
1649 struct inode *inode = file_inode(file);
1650 loff_t ret;
1651
1652 inode_lock(inode);
1653 ret = generic_file_llseek(file, offset, whence);
1654 inode_unlock(inode);
1655
1656 return ret;
1657}
1658
1659const struct file_operations kernfs_dir_fops = {
1660 .read = generic_read_dir,
1661 .iterate = kernfs_fop_readdir,
1662 .release = kernfs_dir_fop_release,
1663 .llseek = kernfs_dir_fop_llseek,
1664};