Loading...
Note: File does not exist in v3.1.
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
47static char * __must_check kernfs_path_locked(struct kernfs_node *kn, char *buf,
48 size_t buflen)
49{
50 char *p = buf + buflen;
51 int len;
52
53 *--p = '\0';
54
55 do {
56 len = strlen(kn->name);
57 if (p - buf < len + 1) {
58 buf[0] = '\0';
59 p = NULL;
60 break;
61 }
62 p -= len;
63 memcpy(p, kn->name, len);
64 *--p = '/';
65 kn = kn->parent;
66 } while (kn && kn->parent);
67
68 return p;
69}
70
71/**
72 * kernfs_name - obtain the name of a given node
73 * @kn: kernfs_node of interest
74 * @buf: buffer to copy @kn's name into
75 * @buflen: size of @buf
76 *
77 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
78 * similar to strlcpy(). It returns the length of @kn's name and if @buf
79 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
80 *
81 * This function can be called from any context.
82 */
83int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
84{
85 unsigned long flags;
86 int ret;
87
88 spin_lock_irqsave(&kernfs_rename_lock, flags);
89 ret = kernfs_name_locked(kn, buf, buflen);
90 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
91 return ret;
92}
93
94/**
95 * kernfs_path - build full path of a given node
96 * @kn: kernfs_node of interest
97 * @buf: buffer to copy @kn's name into
98 * @buflen: size of @buf
99 *
100 * Builds and returns the full path of @kn in @buf of @buflen bytes. The
101 * path is built from the end of @buf so the returned pointer usually
102 * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated
103 * and %NULL is returned.
104 */
105char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
106{
107 unsigned long flags;
108 char *p;
109
110 spin_lock_irqsave(&kernfs_rename_lock, flags);
111 p = kernfs_path_locked(kn, buf, buflen);
112 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
113 return p;
114}
115EXPORT_SYMBOL_GPL(kernfs_path);
116
117/**
118 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
119 * @kn: kernfs_node of interest
120 *
121 * This function can be called from any context.
122 */
123void pr_cont_kernfs_name(struct kernfs_node *kn)
124{
125 unsigned long flags;
126
127 spin_lock_irqsave(&kernfs_rename_lock, flags);
128
129 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
130 pr_cont("%s", kernfs_pr_cont_buf);
131
132 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
133}
134
135/**
136 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
137 * @kn: kernfs_node of interest
138 *
139 * This function can be called from any context.
140 */
141void pr_cont_kernfs_path(struct kernfs_node *kn)
142{
143 unsigned long flags;
144 char *p;
145
146 spin_lock_irqsave(&kernfs_rename_lock, flags);
147
148 p = kernfs_path_locked(kn, kernfs_pr_cont_buf,
149 sizeof(kernfs_pr_cont_buf));
150 if (p)
151 pr_cont("%s", p);
152 else
153 pr_cont("<name too long>");
154
155 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
156}
157
158/**
159 * kernfs_get_parent - determine the parent node and pin it
160 * @kn: kernfs_node of interest
161 *
162 * Determines @kn's parent, pins and returns it. This function can be
163 * called from any context.
164 */
165struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
166{
167 struct kernfs_node *parent;
168 unsigned long flags;
169
170 spin_lock_irqsave(&kernfs_rename_lock, flags);
171 parent = kn->parent;
172 kernfs_get(parent);
173 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
174
175 return parent;
176}
177
178/**
179 * kernfs_name_hash
180 * @name: Null terminated string to hash
181 * @ns: Namespace tag to hash
182 *
183 * Returns 31 bit hash of ns + name (so it fits in an off_t )
184 */
185static unsigned int kernfs_name_hash(const char *name, const void *ns)
186{
187 unsigned long hash = init_name_hash();
188 unsigned int len = strlen(name);
189 while (len--)
190 hash = partial_name_hash(*name++, hash);
191 hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31));
192 hash &= 0x7fffffffU;
193 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
194 if (hash < 2)
195 hash += 2;
196 if (hash >= INT_MAX)
197 hash = INT_MAX - 1;
198 return hash;
199}
200
201static int kernfs_name_compare(unsigned int hash, const char *name,
202 const void *ns, const struct kernfs_node *kn)
203{
204 if (hash != kn->hash)
205 return hash - kn->hash;
206 if (ns != kn->ns)
207 return ns - kn->ns;
208 return strcmp(name, kn->name);
209}
210
211static int kernfs_sd_compare(const struct kernfs_node *left,
212 const struct kernfs_node *right)
213{
214 return kernfs_name_compare(left->hash, left->name, left->ns, right);
215}
216
217/**
218 * kernfs_link_sibling - link kernfs_node into sibling rbtree
219 * @kn: kernfs_node of interest
220 *
221 * Link @kn into its sibling rbtree which starts from
222 * @kn->parent->dir.children.
223 *
224 * Locking:
225 * mutex_lock(kernfs_mutex)
226 *
227 * RETURNS:
228 * 0 on susccess -EEXIST on failure.
229 */
230static int kernfs_link_sibling(struct kernfs_node *kn)
231{
232 struct rb_node **node = &kn->parent->dir.children.rb_node;
233 struct rb_node *parent = NULL;
234
235 while (*node) {
236 struct kernfs_node *pos;
237 int result;
238
239 pos = rb_to_kn(*node);
240 parent = *node;
241 result = kernfs_sd_compare(kn, pos);
242 if (result < 0)
243 node = &pos->rb.rb_left;
244 else if (result > 0)
245 node = &pos->rb.rb_right;
246 else
247 return -EEXIST;
248 }
249
250 /* add new node and rebalance the tree */
251 rb_link_node(&kn->rb, parent, node);
252 rb_insert_color(&kn->rb, &kn->parent->dir.children);
253
254 /* successfully added, account subdir number */
255 if (kernfs_type(kn) == KERNFS_DIR)
256 kn->parent->dir.subdirs++;
257
258 return 0;
259}
260
261/**
262 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
263 * @kn: kernfs_node of interest
264 *
265 * Try to unlink @kn from its sibling rbtree which starts from
266 * kn->parent->dir.children. Returns %true if @kn was actually
267 * removed, %false if @kn wasn't on the rbtree.
268 *
269 * Locking:
270 * mutex_lock(kernfs_mutex)
271 */
272static bool kernfs_unlink_sibling(struct kernfs_node *kn)
273{
274 if (RB_EMPTY_NODE(&kn->rb))
275 return false;
276
277 if (kernfs_type(kn) == KERNFS_DIR)
278 kn->parent->dir.subdirs--;
279
280 rb_erase(&kn->rb, &kn->parent->dir.children);
281 RB_CLEAR_NODE(&kn->rb);
282 return true;
283}
284
285/**
286 * kernfs_get_active - get an active reference to kernfs_node
287 * @kn: kernfs_node to get an active reference to
288 *
289 * Get an active reference of @kn. This function is noop if @kn
290 * is NULL.
291 *
292 * RETURNS:
293 * Pointer to @kn on success, NULL on failure.
294 */
295struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
296{
297 if (unlikely(!kn))
298 return NULL;
299
300 if (!atomic_inc_unless_negative(&kn->active))
301 return NULL;
302
303 if (kernfs_lockdep(kn))
304 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
305 return kn;
306}
307
308/**
309 * kernfs_put_active - put an active reference to kernfs_node
310 * @kn: kernfs_node to put an active reference to
311 *
312 * Put an active reference to @kn. This function is noop if @kn
313 * is NULL.
314 */
315void kernfs_put_active(struct kernfs_node *kn)
316{
317 struct kernfs_root *root = kernfs_root(kn);
318 int v;
319
320 if (unlikely(!kn))
321 return;
322
323 if (kernfs_lockdep(kn))
324 rwsem_release(&kn->dep_map, 1, _RET_IP_);
325 v = atomic_dec_return(&kn->active);
326 if (likely(v != KN_DEACTIVATED_BIAS))
327 return;
328
329 wake_up_all(&root->deactivate_waitq);
330}
331
332/**
333 * kernfs_drain - drain kernfs_node
334 * @kn: kernfs_node to drain
335 *
336 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
337 * removers may invoke this function concurrently on @kn and all will
338 * return after draining is complete.
339 */
340static void kernfs_drain(struct kernfs_node *kn)
341 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
342{
343 struct kernfs_root *root = kernfs_root(kn);
344
345 lockdep_assert_held(&kernfs_mutex);
346 WARN_ON_ONCE(kernfs_active(kn));
347
348 mutex_unlock(&kernfs_mutex);
349
350 if (kernfs_lockdep(kn)) {
351 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
352 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
353 lock_contended(&kn->dep_map, _RET_IP_);
354 }
355
356 /* but everyone should wait for draining */
357 wait_event(root->deactivate_waitq,
358 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
359
360 if (kernfs_lockdep(kn)) {
361 lock_acquired(&kn->dep_map, _RET_IP_);
362 rwsem_release(&kn->dep_map, 1, _RET_IP_);
363 }
364
365 kernfs_unmap_bin_file(kn);
366
367 mutex_lock(&kernfs_mutex);
368}
369
370/**
371 * kernfs_get - get a reference count on a kernfs_node
372 * @kn: the target kernfs_node
373 */
374void kernfs_get(struct kernfs_node *kn)
375{
376 if (kn) {
377 WARN_ON(!atomic_read(&kn->count));
378 atomic_inc(&kn->count);
379 }
380}
381EXPORT_SYMBOL_GPL(kernfs_get);
382
383/**
384 * kernfs_put - put a reference count on a kernfs_node
385 * @kn: the target kernfs_node
386 *
387 * Put a reference count of @kn and destroy it if it reached zero.
388 */
389void kernfs_put(struct kernfs_node *kn)
390{
391 struct kernfs_node *parent;
392 struct kernfs_root *root;
393
394 if (!kn || !atomic_dec_and_test(&kn->count))
395 return;
396 root = kernfs_root(kn);
397 repeat:
398 /*
399 * Moving/renaming is always done while holding reference.
400 * kn->parent won't change beneath us.
401 */
402 parent = kn->parent;
403
404 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
405 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
406 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
407
408 if (kernfs_type(kn) == KERNFS_LINK)
409 kernfs_put(kn->symlink.target_kn);
410 if (!(kn->flags & KERNFS_STATIC_NAME))
411 kfree(kn->name);
412 if (kn->iattr) {
413 if (kn->iattr->ia_secdata)
414 security_release_secctx(kn->iattr->ia_secdata,
415 kn->iattr->ia_secdata_len);
416 simple_xattrs_free(&kn->iattr->xattrs);
417 }
418 kfree(kn->iattr);
419 ida_simple_remove(&root->ino_ida, kn->ino);
420 kmem_cache_free(kernfs_node_cache, kn);
421
422 kn = parent;
423 if (kn) {
424 if (atomic_dec_and_test(&kn->count))
425 goto repeat;
426 } else {
427 /* just released the root kn, free @root too */
428 ida_destroy(&root->ino_ida);
429 kfree(root);
430 }
431}
432EXPORT_SYMBOL_GPL(kernfs_put);
433
434static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
435{
436 struct kernfs_node *kn;
437
438 if (flags & LOOKUP_RCU)
439 return -ECHILD;
440
441 /* Always perform fresh lookup for negatives */
442 if (!dentry->d_inode)
443 goto out_bad_unlocked;
444
445 kn = dentry->d_fsdata;
446 mutex_lock(&kernfs_mutex);
447
448 /* The kernfs node has been deactivated */
449 if (!kernfs_active(kn))
450 goto out_bad;
451
452 /* The kernfs node has been moved? */
453 if (dentry->d_parent->d_fsdata != kn->parent)
454 goto out_bad;
455
456 /* The kernfs node has been renamed */
457 if (strcmp(dentry->d_name.name, kn->name) != 0)
458 goto out_bad;
459
460 /* The kernfs node has been moved to a different namespace */
461 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
462 kernfs_info(dentry->d_sb)->ns != kn->ns)
463 goto out_bad;
464
465 mutex_unlock(&kernfs_mutex);
466out_valid:
467 return 1;
468out_bad:
469 mutex_unlock(&kernfs_mutex);
470out_bad_unlocked:
471 /*
472 * @dentry doesn't match the underlying kernfs node, drop the
473 * dentry and force lookup. If we have submounts we must allow the
474 * vfs caches to lie about the state of the filesystem to prevent
475 * leaks and other nasty things, so use check_submounts_and_drop()
476 * instead of d_drop().
477 */
478 if (check_submounts_and_drop(dentry) != 0)
479 goto out_valid;
480
481 return 0;
482}
483
484static void kernfs_dop_release(struct dentry *dentry)
485{
486 kernfs_put(dentry->d_fsdata);
487}
488
489const struct dentry_operations kernfs_dops = {
490 .d_revalidate = kernfs_dop_revalidate,
491 .d_release = kernfs_dop_release,
492};
493
494/**
495 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
496 * @dentry: the dentry in question
497 *
498 * Return the kernfs_node associated with @dentry. If @dentry is not a
499 * kernfs one, %NULL is returned.
500 *
501 * While the returned kernfs_node will stay accessible as long as @dentry
502 * is accessible, the returned node can be in any state and the caller is
503 * fully responsible for determining what's accessible.
504 */
505struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
506{
507 if (dentry->d_sb->s_op == &kernfs_sops)
508 return dentry->d_fsdata;
509 return NULL;
510}
511
512static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
513 const char *name, umode_t mode,
514 unsigned flags)
515{
516 char *dup_name = NULL;
517 struct kernfs_node *kn;
518 int ret;
519
520 if (!(flags & KERNFS_STATIC_NAME)) {
521 name = dup_name = kstrdup(name, GFP_KERNEL);
522 if (!name)
523 return NULL;
524 }
525
526 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
527 if (!kn)
528 goto err_out1;
529
530 ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL);
531 if (ret < 0)
532 goto err_out2;
533 kn->ino = ret;
534
535 atomic_set(&kn->count, 1);
536 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
537 RB_CLEAR_NODE(&kn->rb);
538
539 kn->name = name;
540 kn->mode = mode;
541 kn->flags = flags;
542
543 return kn;
544
545 err_out2:
546 kmem_cache_free(kernfs_node_cache, kn);
547 err_out1:
548 kfree(dup_name);
549 return NULL;
550}
551
552struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
553 const char *name, umode_t mode,
554 unsigned flags)
555{
556 struct kernfs_node *kn;
557
558 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
559 if (kn) {
560 kernfs_get(parent);
561 kn->parent = parent;
562 }
563 return kn;
564}
565
566/**
567 * kernfs_add_one - add kernfs_node to parent without warning
568 * @kn: kernfs_node to be added
569 *
570 * The caller must already have initialized @kn->parent. This
571 * function increments nlink of the parent's inode if @kn is a
572 * directory and link into the children list of the parent.
573 *
574 * RETURNS:
575 * 0 on success, -EEXIST if entry with the given name already
576 * exists.
577 */
578int kernfs_add_one(struct kernfs_node *kn)
579{
580 struct kernfs_node *parent = kn->parent;
581 struct kernfs_iattrs *ps_iattr;
582 bool has_ns;
583 int ret;
584
585 mutex_lock(&kernfs_mutex);
586
587 ret = -EINVAL;
588 has_ns = kernfs_ns_enabled(parent);
589 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
590 has_ns ? "required" : "invalid", parent->name, kn->name))
591 goto out_unlock;
592
593 if (kernfs_type(parent) != KERNFS_DIR)
594 goto out_unlock;
595
596 ret = -ENOENT;
597 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
598 goto out_unlock;
599
600 kn->hash = kernfs_name_hash(kn->name, kn->ns);
601
602 ret = kernfs_link_sibling(kn);
603 if (ret)
604 goto out_unlock;
605
606 /* Update timestamps on the parent */
607 ps_iattr = parent->iattr;
608 if (ps_iattr) {
609 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
610 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
611 }
612
613 mutex_unlock(&kernfs_mutex);
614
615 /*
616 * Activate the new node unless CREATE_DEACTIVATED is requested.
617 * If not activated here, the kernfs user is responsible for
618 * activating the node with kernfs_activate(). A node which hasn't
619 * been activated is not visible to userland and its removal won't
620 * trigger deactivation.
621 */
622 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
623 kernfs_activate(kn);
624 return 0;
625
626out_unlock:
627 mutex_unlock(&kernfs_mutex);
628 return ret;
629}
630
631/**
632 * kernfs_find_ns - find kernfs_node with the given name
633 * @parent: kernfs_node to search under
634 * @name: name to look for
635 * @ns: the namespace tag to use
636 *
637 * Look for kernfs_node with name @name under @parent. Returns pointer to
638 * the found kernfs_node on success, %NULL on failure.
639 */
640static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
641 const unsigned char *name,
642 const void *ns)
643{
644 struct rb_node *node = parent->dir.children.rb_node;
645 bool has_ns = kernfs_ns_enabled(parent);
646 unsigned int hash;
647
648 lockdep_assert_held(&kernfs_mutex);
649
650 if (has_ns != (bool)ns) {
651 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
652 has_ns ? "required" : "invalid", parent->name, name);
653 return NULL;
654 }
655
656 hash = kernfs_name_hash(name, ns);
657 while (node) {
658 struct kernfs_node *kn;
659 int result;
660
661 kn = rb_to_kn(node);
662 result = kernfs_name_compare(hash, name, ns, kn);
663 if (result < 0)
664 node = node->rb_left;
665 else if (result > 0)
666 node = node->rb_right;
667 else
668 return kn;
669 }
670 return NULL;
671}
672
673/**
674 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
675 * @parent: kernfs_node to search under
676 * @name: name to look for
677 * @ns: the namespace tag to use
678 *
679 * Look for kernfs_node with name @name under @parent and get a reference
680 * if found. This function may sleep and returns pointer to the found
681 * kernfs_node on success, %NULL on failure.
682 */
683struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
684 const char *name, const void *ns)
685{
686 struct kernfs_node *kn;
687
688 mutex_lock(&kernfs_mutex);
689 kn = kernfs_find_ns(parent, name, ns);
690 kernfs_get(kn);
691 mutex_unlock(&kernfs_mutex);
692
693 return kn;
694}
695EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
696
697/**
698 * kernfs_create_root - create a new kernfs hierarchy
699 * @scops: optional syscall operations for the hierarchy
700 * @flags: KERNFS_ROOT_* flags
701 * @priv: opaque data associated with the new directory
702 *
703 * Returns the root of the new hierarchy on success, ERR_PTR() value on
704 * failure.
705 */
706struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
707 unsigned int flags, void *priv)
708{
709 struct kernfs_root *root;
710 struct kernfs_node *kn;
711
712 root = kzalloc(sizeof(*root), GFP_KERNEL);
713 if (!root)
714 return ERR_PTR(-ENOMEM);
715
716 ida_init(&root->ino_ida);
717
718 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
719 KERNFS_DIR);
720 if (!kn) {
721 ida_destroy(&root->ino_ida);
722 kfree(root);
723 return ERR_PTR(-ENOMEM);
724 }
725
726 kn->priv = priv;
727 kn->dir.root = root;
728
729 root->syscall_ops = scops;
730 root->flags = flags;
731 root->kn = kn;
732 init_waitqueue_head(&root->deactivate_waitq);
733
734 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
735 kernfs_activate(kn);
736
737 return root;
738}
739
740/**
741 * kernfs_destroy_root - destroy a kernfs hierarchy
742 * @root: root of the hierarchy to destroy
743 *
744 * Destroy the hierarchy anchored at @root by removing all existing
745 * directories and destroying @root.
746 */
747void kernfs_destroy_root(struct kernfs_root *root)
748{
749 kernfs_remove(root->kn); /* will also free @root */
750}
751
752/**
753 * kernfs_create_dir_ns - create a directory
754 * @parent: parent in which to create a new directory
755 * @name: name of the new directory
756 * @mode: mode of the new directory
757 * @priv: opaque data associated with the new directory
758 * @ns: optional namespace tag of the directory
759 *
760 * Returns the created node on success, ERR_PTR() value on failure.
761 */
762struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
763 const char *name, umode_t mode,
764 void *priv, const void *ns)
765{
766 struct kernfs_node *kn;
767 int rc;
768
769 /* allocate */
770 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
771 if (!kn)
772 return ERR_PTR(-ENOMEM);
773
774 kn->dir.root = parent->dir.root;
775 kn->ns = ns;
776 kn->priv = priv;
777
778 /* link in */
779 rc = kernfs_add_one(kn);
780 if (!rc)
781 return kn;
782
783 kernfs_put(kn);
784 return ERR_PTR(rc);
785}
786
787static struct dentry *kernfs_iop_lookup(struct inode *dir,
788 struct dentry *dentry,
789 unsigned int flags)
790{
791 struct dentry *ret;
792 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
793 struct kernfs_node *kn;
794 struct inode *inode;
795 const void *ns = NULL;
796
797 mutex_lock(&kernfs_mutex);
798
799 if (kernfs_ns_enabled(parent))
800 ns = kernfs_info(dir->i_sb)->ns;
801
802 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
803
804 /* no such entry */
805 if (!kn || !kernfs_active(kn)) {
806 ret = NULL;
807 goto out_unlock;
808 }
809 kernfs_get(kn);
810 dentry->d_fsdata = kn;
811
812 /* attach dentry and inode */
813 inode = kernfs_get_inode(dir->i_sb, kn);
814 if (!inode) {
815 ret = ERR_PTR(-ENOMEM);
816 goto out_unlock;
817 }
818
819 /* instantiate and hash dentry */
820 ret = d_materialise_unique(dentry, inode);
821 out_unlock:
822 mutex_unlock(&kernfs_mutex);
823 return ret;
824}
825
826static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
827 umode_t mode)
828{
829 struct kernfs_node *parent = dir->i_private;
830 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
831 int ret;
832
833 if (!scops || !scops->mkdir)
834 return -EPERM;
835
836 if (!kernfs_get_active(parent))
837 return -ENODEV;
838
839 ret = scops->mkdir(parent, dentry->d_name.name, mode);
840
841 kernfs_put_active(parent);
842 return ret;
843}
844
845static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
846{
847 struct kernfs_node *kn = dentry->d_fsdata;
848 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
849 int ret;
850
851 if (!scops || !scops->rmdir)
852 return -EPERM;
853
854 if (!kernfs_get_active(kn))
855 return -ENODEV;
856
857 ret = scops->rmdir(kn);
858
859 kernfs_put_active(kn);
860 return ret;
861}
862
863static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
864 struct inode *new_dir, struct dentry *new_dentry)
865{
866 struct kernfs_node *kn = old_dentry->d_fsdata;
867 struct kernfs_node *new_parent = new_dir->i_private;
868 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
869 int ret;
870
871 if (!scops || !scops->rename)
872 return -EPERM;
873
874 if (!kernfs_get_active(kn))
875 return -ENODEV;
876
877 if (!kernfs_get_active(new_parent)) {
878 kernfs_put_active(kn);
879 return -ENODEV;
880 }
881
882 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
883
884 kernfs_put_active(new_parent);
885 kernfs_put_active(kn);
886 return ret;
887}
888
889const struct inode_operations kernfs_dir_iops = {
890 .lookup = kernfs_iop_lookup,
891 .permission = kernfs_iop_permission,
892 .setattr = kernfs_iop_setattr,
893 .getattr = kernfs_iop_getattr,
894 .setxattr = kernfs_iop_setxattr,
895 .removexattr = kernfs_iop_removexattr,
896 .getxattr = kernfs_iop_getxattr,
897 .listxattr = kernfs_iop_listxattr,
898
899 .mkdir = kernfs_iop_mkdir,
900 .rmdir = kernfs_iop_rmdir,
901 .rename = kernfs_iop_rename,
902};
903
904static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
905{
906 struct kernfs_node *last;
907
908 while (true) {
909 struct rb_node *rbn;
910
911 last = pos;
912
913 if (kernfs_type(pos) != KERNFS_DIR)
914 break;
915
916 rbn = rb_first(&pos->dir.children);
917 if (!rbn)
918 break;
919
920 pos = rb_to_kn(rbn);
921 }
922
923 return last;
924}
925
926/**
927 * kernfs_next_descendant_post - find the next descendant for post-order walk
928 * @pos: the current position (%NULL to initiate traversal)
929 * @root: kernfs_node whose descendants to walk
930 *
931 * Find the next descendant to visit for post-order traversal of @root's
932 * descendants. @root is included in the iteration and the last node to be
933 * visited.
934 */
935static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
936 struct kernfs_node *root)
937{
938 struct rb_node *rbn;
939
940 lockdep_assert_held(&kernfs_mutex);
941
942 /* if first iteration, visit leftmost descendant which may be root */
943 if (!pos)
944 return kernfs_leftmost_descendant(root);
945
946 /* if we visited @root, we're done */
947 if (pos == root)
948 return NULL;
949
950 /* if there's an unvisited sibling, visit its leftmost descendant */
951 rbn = rb_next(&pos->rb);
952 if (rbn)
953 return kernfs_leftmost_descendant(rb_to_kn(rbn));
954
955 /* no sibling left, visit parent */
956 return pos->parent;
957}
958
959/**
960 * kernfs_activate - activate a node which started deactivated
961 * @kn: kernfs_node whose subtree is to be activated
962 *
963 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
964 * needs to be explicitly activated. A node which hasn't been activated
965 * isn't visible to userland and deactivation is skipped during its
966 * removal. This is useful to construct atomic init sequences where
967 * creation of multiple nodes should either succeed or fail atomically.
968 *
969 * The caller is responsible for ensuring that this function is not called
970 * after kernfs_remove*() is invoked on @kn.
971 */
972void kernfs_activate(struct kernfs_node *kn)
973{
974 struct kernfs_node *pos;
975
976 mutex_lock(&kernfs_mutex);
977
978 pos = NULL;
979 while ((pos = kernfs_next_descendant_post(pos, kn))) {
980 if (!pos || (pos->flags & KERNFS_ACTIVATED))
981 continue;
982
983 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
984 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
985
986 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
987 pos->flags |= KERNFS_ACTIVATED;
988 }
989
990 mutex_unlock(&kernfs_mutex);
991}
992
993static void __kernfs_remove(struct kernfs_node *kn)
994{
995 struct kernfs_node *pos;
996
997 lockdep_assert_held(&kernfs_mutex);
998
999 /*
1000 * Short-circuit if non-root @kn has already finished removal.
1001 * This is for kernfs_remove_self() which plays with active ref
1002 * after removal.
1003 */
1004 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1005 return;
1006
1007 pr_debug("kernfs %s: removing\n", kn->name);
1008
1009 /* prevent any new usage under @kn by deactivating all nodes */
1010 pos = NULL;
1011 while ((pos = kernfs_next_descendant_post(pos, kn)))
1012 if (kernfs_active(pos))
1013 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1014
1015 /* deactivate and unlink the subtree node-by-node */
1016 do {
1017 pos = kernfs_leftmost_descendant(kn);
1018
1019 /*
1020 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1021 * base ref could have been put by someone else by the time
1022 * the function returns. Make sure it doesn't go away
1023 * underneath us.
1024 */
1025 kernfs_get(pos);
1026
1027 /*
1028 * Drain iff @kn was activated. This avoids draining and
1029 * its lockdep annotations for nodes which have never been
1030 * activated and allows embedding kernfs_remove() in create
1031 * error paths without worrying about draining.
1032 */
1033 if (kn->flags & KERNFS_ACTIVATED)
1034 kernfs_drain(pos);
1035 else
1036 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1037
1038 /*
1039 * kernfs_unlink_sibling() succeeds once per node. Use it
1040 * to decide who's responsible for cleanups.
1041 */
1042 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1043 struct kernfs_iattrs *ps_iattr =
1044 pos->parent ? pos->parent->iattr : NULL;
1045
1046 /* update timestamps on the parent */
1047 if (ps_iattr) {
1048 ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME;
1049 ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME;
1050 }
1051
1052 kernfs_put(pos);
1053 }
1054
1055 kernfs_put(pos);
1056 } while (pos != kn);
1057}
1058
1059/**
1060 * kernfs_remove - remove a kernfs_node recursively
1061 * @kn: the kernfs_node to remove
1062 *
1063 * Remove @kn along with all its subdirectories and files.
1064 */
1065void kernfs_remove(struct kernfs_node *kn)
1066{
1067 mutex_lock(&kernfs_mutex);
1068 __kernfs_remove(kn);
1069 mutex_unlock(&kernfs_mutex);
1070}
1071
1072/**
1073 * kernfs_break_active_protection - break out of active protection
1074 * @kn: the self kernfs_node
1075 *
1076 * The caller must be running off of a kernfs operation which is invoked
1077 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1078 * this function must also be matched with an invocation of
1079 * kernfs_unbreak_active_protection().
1080 *
1081 * This function releases the active reference of @kn the caller is
1082 * holding. Once this function is called, @kn may be removed at any point
1083 * and the caller is solely responsible for ensuring that the objects it
1084 * dereferences are accessible.
1085 */
1086void kernfs_break_active_protection(struct kernfs_node *kn)
1087{
1088 /*
1089 * Take out ourself out of the active ref dependency chain. If
1090 * we're called without an active ref, lockdep will complain.
1091 */
1092 kernfs_put_active(kn);
1093}
1094
1095/**
1096 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1097 * @kn: the self kernfs_node
1098 *
1099 * If kernfs_break_active_protection() was called, this function must be
1100 * invoked before finishing the kernfs operation. Note that while this
1101 * function restores the active reference, it doesn't and can't actually
1102 * restore the active protection - @kn may already or be in the process of
1103 * being removed. Once kernfs_break_active_protection() is invoked, that
1104 * protection is irreversibly gone for the kernfs operation instance.
1105 *
1106 * While this function may be called at any point after
1107 * kernfs_break_active_protection() is invoked, its most useful location
1108 * would be right before the enclosing kernfs operation returns.
1109 */
1110void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1111{
1112 /*
1113 * @kn->active could be in any state; however, the increment we do
1114 * here will be undone as soon as the enclosing kernfs operation
1115 * finishes and this temporary bump can't break anything. If @kn
1116 * is alive, nothing changes. If @kn is being deactivated, the
1117 * soon-to-follow put will either finish deactivation or restore
1118 * deactivated state. If @kn is already removed, the temporary
1119 * bump is guaranteed to be gone before @kn is released.
1120 */
1121 atomic_inc(&kn->active);
1122 if (kernfs_lockdep(kn))
1123 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1124}
1125
1126/**
1127 * kernfs_remove_self - remove a kernfs_node from its own method
1128 * @kn: the self kernfs_node to remove
1129 *
1130 * The caller must be running off of a kernfs operation which is invoked
1131 * with an active reference - e.g. one of kernfs_ops. This can be used to
1132 * implement a file operation which deletes itself.
1133 *
1134 * For example, the "delete" file for a sysfs device directory can be
1135 * implemented by invoking kernfs_remove_self() on the "delete" file
1136 * itself. This function breaks the circular dependency of trying to
1137 * deactivate self while holding an active ref itself. It isn't necessary
1138 * to modify the usual removal path to use kernfs_remove_self(). The
1139 * "delete" implementation can simply invoke kernfs_remove_self() on self
1140 * before proceeding with the usual removal path. kernfs will ignore later
1141 * kernfs_remove() on self.
1142 *
1143 * kernfs_remove_self() can be called multiple times concurrently on the
1144 * same kernfs_node. Only the first one actually performs removal and
1145 * returns %true. All others will wait until the kernfs operation which
1146 * won self-removal finishes and return %false. Note that the losers wait
1147 * for the completion of not only the winning kernfs_remove_self() but also
1148 * the whole kernfs_ops which won the arbitration. This can be used to
1149 * guarantee, for example, all concurrent writes to a "delete" file to
1150 * finish only after the whole operation is complete.
1151 */
1152bool kernfs_remove_self(struct kernfs_node *kn)
1153{
1154 bool ret;
1155
1156 mutex_lock(&kernfs_mutex);
1157 kernfs_break_active_protection(kn);
1158
1159 /*
1160 * SUICIDAL is used to arbitrate among competing invocations. Only
1161 * the first one will actually perform removal. When the removal
1162 * is complete, SUICIDED is set and the active ref is restored
1163 * while holding kernfs_mutex. The ones which lost arbitration
1164 * waits for SUICDED && drained which can happen only after the
1165 * enclosing kernfs operation which executed the winning instance
1166 * of kernfs_remove_self() finished.
1167 */
1168 if (!(kn->flags & KERNFS_SUICIDAL)) {
1169 kn->flags |= KERNFS_SUICIDAL;
1170 __kernfs_remove(kn);
1171 kn->flags |= KERNFS_SUICIDED;
1172 ret = true;
1173 } else {
1174 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1175 DEFINE_WAIT(wait);
1176
1177 while (true) {
1178 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1179
1180 if ((kn->flags & KERNFS_SUICIDED) &&
1181 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1182 break;
1183
1184 mutex_unlock(&kernfs_mutex);
1185 schedule();
1186 mutex_lock(&kernfs_mutex);
1187 }
1188 finish_wait(waitq, &wait);
1189 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1190 ret = false;
1191 }
1192
1193 /*
1194 * This must be done while holding kernfs_mutex; otherwise, waiting
1195 * for SUICIDED && deactivated could finish prematurely.
1196 */
1197 kernfs_unbreak_active_protection(kn);
1198
1199 mutex_unlock(&kernfs_mutex);
1200 return ret;
1201}
1202
1203/**
1204 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1205 * @parent: parent of the target
1206 * @name: name of the kernfs_node to remove
1207 * @ns: namespace tag of the kernfs_node to remove
1208 *
1209 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1210 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1211 */
1212int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1213 const void *ns)
1214{
1215 struct kernfs_node *kn;
1216
1217 if (!parent) {
1218 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1219 name);
1220 return -ENOENT;
1221 }
1222
1223 mutex_lock(&kernfs_mutex);
1224
1225 kn = kernfs_find_ns(parent, name, ns);
1226 if (kn)
1227 __kernfs_remove(kn);
1228
1229 mutex_unlock(&kernfs_mutex);
1230
1231 if (kn)
1232 return 0;
1233 else
1234 return -ENOENT;
1235}
1236
1237/**
1238 * kernfs_rename_ns - move and rename a kernfs_node
1239 * @kn: target node
1240 * @new_parent: new parent to put @sd under
1241 * @new_name: new name
1242 * @new_ns: new namespace tag
1243 */
1244int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1245 const char *new_name, const void *new_ns)
1246{
1247 struct kernfs_node *old_parent;
1248 const char *old_name = NULL;
1249 int error;
1250
1251 /* can't move or rename root */
1252 if (!kn->parent)
1253 return -EINVAL;
1254
1255 mutex_lock(&kernfs_mutex);
1256
1257 error = -ENOENT;
1258 if (!kernfs_active(kn) || !kernfs_active(new_parent))
1259 goto out;
1260
1261 error = 0;
1262 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1263 (strcmp(kn->name, new_name) == 0))
1264 goto out; /* nothing to rename */
1265
1266 error = -EEXIST;
1267 if (kernfs_find_ns(new_parent, new_name, new_ns))
1268 goto out;
1269
1270 /* rename kernfs_node */
1271 if (strcmp(kn->name, new_name) != 0) {
1272 error = -ENOMEM;
1273 new_name = kstrdup(new_name, GFP_KERNEL);
1274 if (!new_name)
1275 goto out;
1276 } else {
1277 new_name = NULL;
1278 }
1279
1280 /*
1281 * Move to the appropriate place in the appropriate directories rbtree.
1282 */
1283 kernfs_unlink_sibling(kn);
1284 kernfs_get(new_parent);
1285
1286 /* rename_lock protects ->parent and ->name accessors */
1287 spin_lock_irq(&kernfs_rename_lock);
1288
1289 old_parent = kn->parent;
1290 kn->parent = new_parent;
1291
1292 kn->ns = new_ns;
1293 if (new_name) {
1294 if (!(kn->flags & KERNFS_STATIC_NAME))
1295 old_name = kn->name;
1296 kn->flags &= ~KERNFS_STATIC_NAME;
1297 kn->name = new_name;
1298 }
1299
1300 spin_unlock_irq(&kernfs_rename_lock);
1301
1302 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1303 kernfs_link_sibling(kn);
1304
1305 kernfs_put(old_parent);
1306 kfree(old_name);
1307
1308 error = 0;
1309 out:
1310 mutex_unlock(&kernfs_mutex);
1311 return error;
1312}
1313
1314/* Relationship between s_mode and the DT_xxx types */
1315static inline unsigned char dt_type(struct kernfs_node *kn)
1316{
1317 return (kn->mode >> 12) & 15;
1318}
1319
1320static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1321{
1322 kernfs_put(filp->private_data);
1323 return 0;
1324}
1325
1326static struct kernfs_node *kernfs_dir_pos(const void *ns,
1327 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1328{
1329 if (pos) {
1330 int valid = kernfs_active(pos) &&
1331 pos->parent == parent && hash == pos->hash;
1332 kernfs_put(pos);
1333 if (!valid)
1334 pos = NULL;
1335 }
1336 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1337 struct rb_node *node = parent->dir.children.rb_node;
1338 while (node) {
1339 pos = rb_to_kn(node);
1340
1341 if (hash < pos->hash)
1342 node = node->rb_left;
1343 else if (hash > pos->hash)
1344 node = node->rb_right;
1345 else
1346 break;
1347 }
1348 }
1349 /* Skip over entries which are dying/dead or in the wrong namespace */
1350 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1351 struct rb_node *node = rb_next(&pos->rb);
1352 if (!node)
1353 pos = NULL;
1354 else
1355 pos = rb_to_kn(node);
1356 }
1357 return pos;
1358}
1359
1360static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1361 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1362{
1363 pos = kernfs_dir_pos(ns, parent, ino, pos);
1364 if (pos) {
1365 do {
1366 struct rb_node *node = rb_next(&pos->rb);
1367 if (!node)
1368 pos = NULL;
1369 else
1370 pos = rb_to_kn(node);
1371 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1372 }
1373 return pos;
1374}
1375
1376static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1377{
1378 struct dentry *dentry = file->f_path.dentry;
1379 struct kernfs_node *parent = dentry->d_fsdata;
1380 struct kernfs_node *pos = file->private_data;
1381 const void *ns = NULL;
1382
1383 if (!dir_emit_dots(file, ctx))
1384 return 0;
1385 mutex_lock(&kernfs_mutex);
1386
1387 if (kernfs_ns_enabled(parent))
1388 ns = kernfs_info(dentry->d_sb)->ns;
1389
1390 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1391 pos;
1392 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1393 const char *name = pos->name;
1394 unsigned int type = dt_type(pos);
1395 int len = strlen(name);
1396 ino_t ino = pos->ino;
1397
1398 ctx->pos = pos->hash;
1399 file->private_data = pos;
1400 kernfs_get(pos);
1401
1402 mutex_unlock(&kernfs_mutex);
1403 if (!dir_emit(ctx, name, len, ino, type))
1404 return 0;
1405 mutex_lock(&kernfs_mutex);
1406 }
1407 mutex_unlock(&kernfs_mutex);
1408 file->private_data = NULL;
1409 ctx->pos = INT_MAX;
1410 return 0;
1411}
1412
1413static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset,
1414 int whence)
1415{
1416 struct inode *inode = file_inode(file);
1417 loff_t ret;
1418
1419 mutex_lock(&inode->i_mutex);
1420 ret = generic_file_llseek(file, offset, whence);
1421 mutex_unlock(&inode->i_mutex);
1422
1423 return ret;
1424}
1425
1426const struct file_operations kernfs_dir_fops = {
1427 .read = generic_read_dir,
1428 .iterate = kernfs_fop_readdir,
1429 .release = kernfs_dir_fop_release,
1430 .llseek = kernfs_dir_fop_llseek,
1431};